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 struct btrfs_path *btrfs_alloc_path(void)
43 struct btrfs_path *path;
44 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
51 * set all locked nodes in the path to blocking locks. This should
52 * be done before scheduling
54 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
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]);
64 * reset all the locked nodes in the patch to spinning locks.
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
71 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
72 struct extent_buffer *held)
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
84 btrfs_set_lock_blocking(held);
85 btrfs_set_path_blocking(p);
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]);
93 #ifdef CONFIG_DEBUG_LOCK_ALLOC
95 btrfs_clear_lock_blocking(held);
99 /* this also releases the path */
100 void btrfs_free_path(struct btrfs_path *p)
102 btrfs_release_path(NULL, p);
103 kmem_cache_free(btrfs_path_cachep, p);
107 * path release drops references on the extent buffers in the path
108 * and it drops any locks held by this path
110 * It is safe to call this on paths that no locks or extent buffers held.
112 noinline void btrfs_release_path(struct btrfs_root *root, struct btrfs_path *p)
116 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
121 btrfs_tree_unlock(p->nodes[i]);
124 free_extent_buffer(p->nodes[i]);
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
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.
139 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
141 struct extent_buffer *eb;
142 spin_lock(&root->node_lock);
144 extent_buffer_get(eb);
145 spin_unlock(&root->node_lock);
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.
153 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
155 struct extent_buffer *eb;
158 eb = btrfs_root_node(root);
161 spin_lock(&root->node_lock);
162 if (eb == root->node) {
163 spin_unlock(&root->node_lock);
166 spin_unlock(&root->node_lock);
168 btrfs_tree_unlock(eb);
169 free_extent_buffer(eb);
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.
178 static void add_root_to_dirty_list(struct btrfs_root *root)
180 if (root->track_dirty && list_empty(&root->dirty_list)) {
181 list_add(&root->dirty_list,
182 &root->fs_info->dirty_cowonly_roots);
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.
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)
196 struct extent_buffer *cow;
200 struct btrfs_root *new_root;
202 new_root = kmalloc(sizeof(*new_root), GFP_NOFS);
206 memcpy(new_root, root, sizeof(*new_root));
207 new_root->root_key.objectid = new_root_objectid;
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);
213 level = btrfs_header_level(buf);
214 nritems = btrfs_header_nritems(buf);
216 cow = btrfs_alloc_free_block(trans, new_root, buf->len, 0,
217 new_root_objectid, trans->transid,
218 level, buf->start, 0);
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);
230 write_extent_buffer(cow, root->fs_info->fsid,
231 (unsigned long)btrfs_header_fsid(cow),
234 WARN_ON(btrfs_header_generation(buf) > trans->transid);
235 ret = btrfs_inc_ref(trans, new_root, buf, cow, NULL);
241 btrfs_mark_buffer_dirty(cow);
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
252 * search_start -- an allocation hint for the new block
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.
258 * prealloc_dest -- if you have already reserved a destination for the cow,
259 * this uses that block instead of allocating a new one.
260 * btrfs_alloc_reserved_extent is used to finish the allocation.
262 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
263 struct btrfs_root *root,
264 struct extent_buffer *buf,
265 struct extent_buffer *parent, int parent_slot,
266 struct extent_buffer **cow_ret,
267 u64 search_start, u64 empty_size,
271 struct extent_buffer *cow;
280 btrfs_assert_tree_locked(buf);
283 parent_start = parent->start;
287 WARN_ON(root->ref_cows && trans->transid !=
288 root->fs_info->running_transaction->transid);
289 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
291 level = btrfs_header_level(buf);
292 nritems = btrfs_header_nritems(buf);
295 struct btrfs_key ins;
297 ins.objectid = prealloc_dest;
298 ins.offset = buf->len;
299 ins.type = BTRFS_EXTENT_ITEM_KEY;
301 ret = btrfs_alloc_reserved_extent(trans, root, parent_start,
302 root->root_key.objectid,
303 trans->transid, level, &ins);
305 cow = btrfs_init_new_buffer(trans, root, prealloc_dest,
308 cow = btrfs_alloc_free_block(trans, root, buf->len,
310 root->root_key.objectid,
311 trans->transid, level,
312 search_start, empty_size);
317 /* cow is set to blocking by btrfs_init_new_buffer */
319 copy_extent_buffer(cow, buf, 0, 0, cow->len);
320 btrfs_set_header_bytenr(cow, cow->start);
321 btrfs_set_header_generation(cow, trans->transid);
322 btrfs_set_header_owner(cow, root->root_key.objectid);
323 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
325 write_extent_buffer(cow, root->fs_info->fsid,
326 (unsigned long)btrfs_header_fsid(cow),
329 WARN_ON(btrfs_header_generation(buf) > trans->transid);
330 if (btrfs_header_generation(buf) != trans->transid) {
332 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
336 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
338 } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
340 * There are only two places that can drop reference to
341 * tree blocks owned by living reloc trees, one is here,
342 * the other place is btrfs_drop_subtree. In both places,
343 * we check reference count while tree block is locked.
344 * Furthermore, if reference count is one, it won't get
345 * increased by someone else.
348 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
352 ret = btrfs_update_ref(trans, root, buf, cow,
354 clean_tree_block(trans, root, buf);
356 ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
360 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
363 clean_tree_block(trans, root, buf);
366 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
367 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
371 if (buf == root->node) {
372 WARN_ON(parent && parent != buf);
374 spin_lock(&root->node_lock);
376 extent_buffer_get(cow);
377 spin_unlock(&root->node_lock);
379 if (buf != root->commit_root) {
380 btrfs_free_extent(trans, root, buf->start,
381 buf->len, buf->start,
382 root->root_key.objectid,
383 btrfs_header_generation(buf),
386 free_extent_buffer(buf);
387 add_root_to_dirty_list(root);
389 btrfs_set_node_blockptr(parent, parent_slot,
391 WARN_ON(trans->transid == 0);
392 btrfs_set_node_ptr_generation(parent, parent_slot,
394 btrfs_mark_buffer_dirty(parent);
395 WARN_ON(btrfs_header_generation(parent) != trans->transid);
396 btrfs_free_extent(trans, root, buf->start, buf->len,
397 parent_start, btrfs_header_owner(parent),
398 btrfs_header_generation(parent), level, 1);
401 btrfs_tree_unlock(buf);
402 free_extent_buffer(buf);
403 btrfs_mark_buffer_dirty(cow);
409 * cows a single block, see __btrfs_cow_block for the real work.
410 * This version of it has extra checks so that a block isn't cow'd more than
411 * once per transaction, as long as it hasn't been written yet
413 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
414 struct btrfs_root *root, struct extent_buffer *buf,
415 struct extent_buffer *parent, int parent_slot,
416 struct extent_buffer **cow_ret, u64 prealloc_dest)
421 if (trans->transaction != root->fs_info->running_transaction) {
422 printk(KERN_CRIT "trans %llu running %llu\n",
423 (unsigned long long)trans->transid,
425 root->fs_info->running_transaction->transid);
428 if (trans->transid != root->fs_info->generation) {
429 printk(KERN_CRIT "trans %llu running %llu\n",
430 (unsigned long long)trans->transid,
431 (unsigned long long)root->fs_info->generation);
435 if (btrfs_header_generation(buf) == trans->transid &&
436 btrfs_header_owner(buf) == root->root_key.objectid &&
437 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
439 WARN_ON(prealloc_dest);
443 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
446 btrfs_set_lock_blocking(parent);
447 btrfs_set_lock_blocking(buf);
449 ret = __btrfs_cow_block(trans, root, buf, parent,
450 parent_slot, cow_ret, search_start, 0,
456 * helper function for defrag to decide if two blocks pointed to by a
457 * node are actually close by
459 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
461 if (blocknr < other && other - (blocknr + blocksize) < 32768)
463 if (blocknr > other && blocknr - (other + blocksize) < 32768)
469 * compare two keys in a memcmp fashion
471 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
475 btrfs_disk_key_to_cpu(&k1, disk);
477 if (k1.objectid > k2->objectid)
479 if (k1.objectid < k2->objectid)
481 if (k1.type > k2->type)
483 if (k1.type < k2->type)
485 if (k1.offset > k2->offset)
487 if (k1.offset < k2->offset)
493 * same as comp_keys only with two btrfs_key's
495 static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
497 if (k1->objectid > k2->objectid)
499 if (k1->objectid < k2->objectid)
501 if (k1->type > k2->type)
503 if (k1->type < k2->type)
505 if (k1->offset > k2->offset)
507 if (k1->offset < k2->offset)
513 * this is used by the defrag code to go through all the
514 * leaves pointed to by a node and reallocate them so that
515 * disk order is close to key order
517 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
518 struct btrfs_root *root, struct extent_buffer *parent,
519 int start_slot, int cache_only, u64 *last_ret,
520 struct btrfs_key *progress)
522 struct extent_buffer *cur;
525 u64 search_start = *last_ret;
535 int progress_passed = 0;
536 struct btrfs_disk_key disk_key;
538 parent_level = btrfs_header_level(parent);
539 if (cache_only && parent_level != 1)
542 if (trans->transaction != root->fs_info->running_transaction)
544 if (trans->transid != root->fs_info->generation)
547 parent_nritems = btrfs_header_nritems(parent);
548 blocksize = btrfs_level_size(root, parent_level - 1);
549 end_slot = parent_nritems;
551 if (parent_nritems == 1)
554 btrfs_set_lock_blocking(parent);
556 for (i = start_slot; i < end_slot; i++) {
559 if (!parent->map_token) {
560 map_extent_buffer(parent,
561 btrfs_node_key_ptr_offset(i),
562 sizeof(struct btrfs_key_ptr),
563 &parent->map_token, &parent->kaddr,
564 &parent->map_start, &parent->map_len,
567 btrfs_node_key(parent, &disk_key, i);
568 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
572 blocknr = btrfs_node_blockptr(parent, i);
573 gen = btrfs_node_ptr_generation(parent, i);
575 last_block = blocknr;
578 other = btrfs_node_blockptr(parent, i - 1);
579 close = close_blocks(blocknr, other, blocksize);
581 if (!close && i < end_slot - 2) {
582 other = btrfs_node_blockptr(parent, i + 1);
583 close = close_blocks(blocknr, other, blocksize);
586 last_block = blocknr;
589 if (parent->map_token) {
590 unmap_extent_buffer(parent, parent->map_token,
592 parent->map_token = NULL;
595 cur = btrfs_find_tree_block(root, blocknr, blocksize);
597 uptodate = btrfs_buffer_uptodate(cur, gen);
600 if (!cur || !uptodate) {
602 free_extent_buffer(cur);
606 cur = read_tree_block(root, blocknr,
608 } else if (!uptodate) {
609 btrfs_read_buffer(cur, gen);
612 if (search_start == 0)
613 search_start = last_block;
615 btrfs_tree_lock(cur);
616 btrfs_set_lock_blocking(cur);
617 err = __btrfs_cow_block(trans, root, cur, parent, i,
620 (end_slot - i) * blocksize), 0);
622 btrfs_tree_unlock(cur);
623 free_extent_buffer(cur);
626 search_start = cur->start;
627 last_block = cur->start;
628 *last_ret = search_start;
629 btrfs_tree_unlock(cur);
630 free_extent_buffer(cur);
632 if (parent->map_token) {
633 unmap_extent_buffer(parent, parent->map_token,
635 parent->map_token = NULL;
641 * The leaf data grows from end-to-front in the node.
642 * this returns the address of the start of the last item,
643 * which is the stop of the leaf data stack
645 static inline unsigned int leaf_data_end(struct btrfs_root *root,
646 struct extent_buffer *leaf)
648 u32 nr = btrfs_header_nritems(leaf);
650 return BTRFS_LEAF_DATA_SIZE(root);
651 return btrfs_item_offset_nr(leaf, nr - 1);
655 * extra debugging checks to make sure all the items in a key are
656 * well formed and in the proper order
658 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
661 struct extent_buffer *parent = NULL;
662 struct extent_buffer *node = path->nodes[level];
663 struct btrfs_disk_key parent_key;
664 struct btrfs_disk_key node_key;
667 struct btrfs_key cpukey;
668 u32 nritems = btrfs_header_nritems(node);
670 if (path->nodes[level + 1])
671 parent = path->nodes[level + 1];
673 slot = path->slots[level];
674 BUG_ON(nritems == 0);
676 parent_slot = path->slots[level + 1];
677 btrfs_node_key(parent, &parent_key, parent_slot);
678 btrfs_node_key(node, &node_key, 0);
679 BUG_ON(memcmp(&parent_key, &node_key,
680 sizeof(struct btrfs_disk_key)));
681 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
682 btrfs_header_bytenr(node));
684 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
686 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
687 btrfs_node_key(node, &node_key, slot);
688 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
690 if (slot < nritems - 1) {
691 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
692 btrfs_node_key(node, &node_key, slot);
693 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
699 * extra checking to make sure all the items in a leaf are
700 * well formed and in the proper order
702 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
705 struct extent_buffer *leaf = path->nodes[level];
706 struct extent_buffer *parent = NULL;
708 struct btrfs_key cpukey;
709 struct btrfs_disk_key parent_key;
710 struct btrfs_disk_key leaf_key;
711 int slot = path->slots[0];
713 u32 nritems = btrfs_header_nritems(leaf);
715 if (path->nodes[level + 1])
716 parent = path->nodes[level + 1];
722 parent_slot = path->slots[level + 1];
723 btrfs_node_key(parent, &parent_key, parent_slot);
724 btrfs_item_key(leaf, &leaf_key, 0);
726 BUG_ON(memcmp(&parent_key, &leaf_key,
727 sizeof(struct btrfs_disk_key)));
728 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
729 btrfs_header_bytenr(leaf));
731 if (slot != 0 && slot < nritems - 1) {
732 btrfs_item_key(leaf, &leaf_key, slot);
733 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
734 if (comp_keys(&leaf_key, &cpukey) <= 0) {
735 btrfs_print_leaf(root, leaf);
736 printk(KERN_CRIT "slot %d offset bad key\n", slot);
739 if (btrfs_item_offset_nr(leaf, slot - 1) !=
740 btrfs_item_end_nr(leaf, slot)) {
741 btrfs_print_leaf(root, leaf);
742 printk(KERN_CRIT "slot %d offset bad\n", slot);
746 if (slot < nritems - 1) {
747 btrfs_item_key(leaf, &leaf_key, slot);
748 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
749 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
750 if (btrfs_item_offset_nr(leaf, slot) !=
751 btrfs_item_end_nr(leaf, slot + 1)) {
752 btrfs_print_leaf(root, leaf);
753 printk(KERN_CRIT "slot %d offset bad\n", slot);
757 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
758 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
762 static noinline int check_block(struct btrfs_root *root,
763 struct btrfs_path *path, int level)
767 return check_leaf(root, path, level);
768 return check_node(root, path, level);
772 * search for key in the extent_buffer. The items start at offset p,
773 * and they are item_size apart. There are 'max' items in p.
775 * the slot in the array is returned via slot, and it points to
776 * the place where you would insert key if it is not found in
779 * slot may point to max if the key is bigger than all of the keys
781 static noinline int generic_bin_search(struct extent_buffer *eb,
783 int item_size, struct btrfs_key *key,
790 struct btrfs_disk_key *tmp = NULL;
791 struct btrfs_disk_key unaligned;
792 unsigned long offset;
793 char *map_token = NULL;
795 unsigned long map_start = 0;
796 unsigned long map_len = 0;
800 mid = (low + high) / 2;
801 offset = p + mid * item_size;
803 if (!map_token || offset < map_start ||
804 (offset + sizeof(struct btrfs_disk_key)) >
805 map_start + map_len) {
807 unmap_extent_buffer(eb, map_token, KM_USER0);
811 err = map_private_extent_buffer(eb, offset,
812 sizeof(struct btrfs_disk_key),
814 &map_start, &map_len, KM_USER0);
817 tmp = (struct btrfs_disk_key *)(kaddr + offset -
820 read_extent_buffer(eb, &unaligned,
821 offset, sizeof(unaligned));
826 tmp = (struct btrfs_disk_key *)(kaddr + offset -
829 ret = comp_keys(tmp, key);
838 unmap_extent_buffer(eb, map_token, KM_USER0);
844 unmap_extent_buffer(eb, map_token, KM_USER0);
849 * simple bin_search frontend that does the right thing for
852 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
853 int level, int *slot)
856 return generic_bin_search(eb,
857 offsetof(struct btrfs_leaf, items),
858 sizeof(struct btrfs_item),
859 key, btrfs_header_nritems(eb),
862 return generic_bin_search(eb,
863 offsetof(struct btrfs_node, ptrs),
864 sizeof(struct btrfs_key_ptr),
865 key, btrfs_header_nritems(eb),
871 /* given a node and slot number, this reads the blocks it points to. The
872 * extent buffer is returned with a reference taken (but unlocked).
873 * NULL is returned on error.
875 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
876 struct extent_buffer *parent, int slot)
878 int level = btrfs_header_level(parent);
881 if (slot >= btrfs_header_nritems(parent))
886 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
887 btrfs_level_size(root, level - 1),
888 btrfs_node_ptr_generation(parent, slot));
892 * node level balancing, used to make sure nodes are in proper order for
893 * item deletion. We balance from the top down, so we have to make sure
894 * that a deletion won't leave an node completely empty later on.
896 static noinline int balance_level(struct btrfs_trans_handle *trans,
897 struct btrfs_root *root,
898 struct btrfs_path *path, int level)
900 struct extent_buffer *right = NULL;
901 struct extent_buffer *mid;
902 struct extent_buffer *left = NULL;
903 struct extent_buffer *parent = NULL;
907 int orig_slot = path->slots[level];
908 int err_on_enospc = 0;
914 mid = path->nodes[level];
916 WARN_ON(!path->locks[level]);
917 WARN_ON(btrfs_header_generation(mid) != trans->transid);
919 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
921 if (level < BTRFS_MAX_LEVEL - 1)
922 parent = path->nodes[level + 1];
923 pslot = path->slots[level + 1];
926 * deal with the case where there is only one pointer in the root
927 * by promoting the node below to a root
930 struct extent_buffer *child;
932 if (btrfs_header_nritems(mid) != 1)
935 /* promote the child to a root */
936 child = read_node_slot(root, mid, 0);
938 btrfs_tree_lock(child);
939 btrfs_set_lock_blocking(child);
940 ret = btrfs_cow_block(trans, root, child, mid, 0, &child, 0);
943 spin_lock(&root->node_lock);
945 spin_unlock(&root->node_lock);
947 ret = btrfs_update_extent_ref(trans, root, child->start,
948 mid->start, child->start,
949 root->root_key.objectid,
950 trans->transid, level - 1);
953 add_root_to_dirty_list(root);
954 btrfs_tree_unlock(child);
956 path->locks[level] = 0;
957 path->nodes[level] = NULL;
958 clean_tree_block(trans, root, mid);
959 btrfs_tree_unlock(mid);
960 /* once for the path */
961 free_extent_buffer(mid);
962 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
963 mid->start, root->root_key.objectid,
964 btrfs_header_generation(mid),
966 /* once for the root ptr */
967 free_extent_buffer(mid);
970 if (btrfs_header_nritems(mid) >
971 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
974 if (btrfs_header_nritems(mid) < 2)
977 left = read_node_slot(root, parent, pslot - 1);
979 btrfs_tree_lock(left);
980 btrfs_set_lock_blocking(left);
981 wret = btrfs_cow_block(trans, root, left,
982 parent, pslot - 1, &left, 0);
988 right = read_node_slot(root, parent, pslot + 1);
990 btrfs_tree_lock(right);
991 btrfs_set_lock_blocking(right);
992 wret = btrfs_cow_block(trans, root, right,
993 parent, pslot + 1, &right, 0);
1000 /* first, try to make some room in the middle buffer */
1002 orig_slot += btrfs_header_nritems(left);
1003 wret = push_node_left(trans, root, left, mid, 1);
1006 if (btrfs_header_nritems(mid) < 2)
1011 * then try to empty the right most buffer into the middle
1014 wret = push_node_left(trans, root, mid, right, 1);
1015 if (wret < 0 && wret != -ENOSPC)
1017 if (btrfs_header_nritems(right) == 0) {
1018 u64 bytenr = right->start;
1019 u64 generation = btrfs_header_generation(parent);
1020 u32 blocksize = right->len;
1022 clean_tree_block(trans, root, right);
1023 btrfs_tree_unlock(right);
1024 free_extent_buffer(right);
1026 wret = del_ptr(trans, root, path, level + 1, pslot +
1030 wret = btrfs_free_extent(trans, root, bytenr,
1031 blocksize, parent->start,
1032 btrfs_header_owner(parent),
1033 generation, level, 1);
1037 struct btrfs_disk_key right_key;
1038 btrfs_node_key(right, &right_key, 0);
1039 btrfs_set_node_key(parent, &right_key, pslot + 1);
1040 btrfs_mark_buffer_dirty(parent);
1043 if (btrfs_header_nritems(mid) == 1) {
1045 * we're not allowed to leave a node with one item in the
1046 * tree during a delete. A deletion from lower in the tree
1047 * could try to delete the only pointer in this node.
1048 * So, pull some keys from the left.
1049 * There has to be a left pointer at this point because
1050 * otherwise we would have pulled some pointers from the
1054 wret = balance_node_right(trans, root, mid, left);
1060 wret = push_node_left(trans, root, left, mid, 1);
1066 if (btrfs_header_nritems(mid) == 0) {
1067 /* we've managed to empty the middle node, drop it */
1068 u64 root_gen = btrfs_header_generation(parent);
1069 u64 bytenr = mid->start;
1070 u32 blocksize = mid->len;
1072 clean_tree_block(trans, root, mid);
1073 btrfs_tree_unlock(mid);
1074 free_extent_buffer(mid);
1076 wret = del_ptr(trans, root, path, level + 1, pslot);
1079 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1081 btrfs_header_owner(parent),
1082 root_gen, level, 1);
1086 /* update the parent key to reflect our changes */
1087 struct btrfs_disk_key mid_key;
1088 btrfs_node_key(mid, &mid_key, 0);
1089 btrfs_set_node_key(parent, &mid_key, pslot);
1090 btrfs_mark_buffer_dirty(parent);
1093 /* update the path */
1095 if (btrfs_header_nritems(left) > orig_slot) {
1096 extent_buffer_get(left);
1097 /* left was locked after cow */
1098 path->nodes[level] = left;
1099 path->slots[level + 1] -= 1;
1100 path->slots[level] = orig_slot;
1102 btrfs_tree_unlock(mid);
1103 free_extent_buffer(mid);
1106 orig_slot -= btrfs_header_nritems(left);
1107 path->slots[level] = orig_slot;
1110 /* double check we haven't messed things up */
1111 check_block(root, path, level);
1113 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1117 btrfs_tree_unlock(right);
1118 free_extent_buffer(right);
1121 if (path->nodes[level] != left)
1122 btrfs_tree_unlock(left);
1123 free_extent_buffer(left);
1128 /* Node balancing for insertion. Here we only split or push nodes around
1129 * when they are completely full. This is also done top down, so we
1130 * have to be pessimistic.
1132 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1133 struct btrfs_root *root,
1134 struct btrfs_path *path, int level)
1136 struct extent_buffer *right = NULL;
1137 struct extent_buffer *mid;
1138 struct extent_buffer *left = NULL;
1139 struct extent_buffer *parent = NULL;
1143 int orig_slot = path->slots[level];
1149 mid = path->nodes[level];
1150 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1151 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1153 if (level < BTRFS_MAX_LEVEL - 1)
1154 parent = path->nodes[level + 1];
1155 pslot = path->slots[level + 1];
1160 left = read_node_slot(root, parent, pslot - 1);
1162 /* first, try to make some room in the middle buffer */
1166 btrfs_tree_lock(left);
1167 btrfs_set_lock_blocking(left);
1169 left_nr = btrfs_header_nritems(left);
1170 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1173 ret = btrfs_cow_block(trans, root, left, parent,
1174 pslot - 1, &left, 0);
1178 wret = push_node_left(trans, root,
1185 struct btrfs_disk_key disk_key;
1186 orig_slot += left_nr;
1187 btrfs_node_key(mid, &disk_key, 0);
1188 btrfs_set_node_key(parent, &disk_key, pslot);
1189 btrfs_mark_buffer_dirty(parent);
1190 if (btrfs_header_nritems(left) > orig_slot) {
1191 path->nodes[level] = left;
1192 path->slots[level + 1] -= 1;
1193 path->slots[level] = orig_slot;
1194 btrfs_tree_unlock(mid);
1195 free_extent_buffer(mid);
1198 btrfs_header_nritems(left);
1199 path->slots[level] = orig_slot;
1200 btrfs_tree_unlock(left);
1201 free_extent_buffer(left);
1205 btrfs_tree_unlock(left);
1206 free_extent_buffer(left);
1208 right = read_node_slot(root, parent, pslot + 1);
1211 * then try to empty the right most buffer into the middle
1216 btrfs_tree_lock(right);
1217 btrfs_set_lock_blocking(right);
1219 right_nr = btrfs_header_nritems(right);
1220 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1223 ret = btrfs_cow_block(trans, root, right,
1229 wret = balance_node_right(trans, root,
1236 struct btrfs_disk_key disk_key;
1238 btrfs_node_key(right, &disk_key, 0);
1239 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1240 btrfs_mark_buffer_dirty(parent);
1242 if (btrfs_header_nritems(mid) <= orig_slot) {
1243 path->nodes[level] = right;
1244 path->slots[level + 1] += 1;
1245 path->slots[level] = orig_slot -
1246 btrfs_header_nritems(mid);
1247 btrfs_tree_unlock(mid);
1248 free_extent_buffer(mid);
1250 btrfs_tree_unlock(right);
1251 free_extent_buffer(right);
1255 btrfs_tree_unlock(right);
1256 free_extent_buffer(right);
1262 * readahead one full node of leaves, finding things that are close
1263 * to the block in 'slot', and triggering ra on them.
1265 static noinline void reada_for_search(struct btrfs_root *root,
1266 struct btrfs_path *path,
1267 int level, int slot, u64 objectid)
1269 struct extent_buffer *node;
1270 struct btrfs_disk_key disk_key;
1275 int direction = path->reada;
1276 struct extent_buffer *eb;
1284 if (!path->nodes[level])
1287 node = path->nodes[level];
1289 search = btrfs_node_blockptr(node, slot);
1290 blocksize = btrfs_level_size(root, level - 1);
1291 eb = btrfs_find_tree_block(root, search, blocksize);
1293 free_extent_buffer(eb);
1299 nritems = btrfs_header_nritems(node);
1302 if (direction < 0) {
1306 } else if (direction > 0) {
1311 if (path->reada < 0 && objectid) {
1312 btrfs_node_key(node, &disk_key, nr);
1313 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1316 search = btrfs_node_blockptr(node, nr);
1317 if ((search <= target && target - search <= 65536) ||
1318 (search > target && search - target <= 65536)) {
1319 readahead_tree_block(root, search, blocksize,
1320 btrfs_node_ptr_generation(node, nr));
1324 if ((nread > 65536 || nscan > 32))
1330 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1333 static noinline int reada_for_balance(struct btrfs_root *root,
1334 struct btrfs_path *path, int level)
1338 struct extent_buffer *parent;
1339 struct extent_buffer *eb;
1346 parent = path->nodes[level - 1];
1350 nritems = btrfs_header_nritems(parent);
1351 slot = path->slots[level];
1352 blocksize = btrfs_level_size(root, level);
1355 block1 = btrfs_node_blockptr(parent, slot - 1);
1356 gen = btrfs_node_ptr_generation(parent, slot - 1);
1357 eb = btrfs_find_tree_block(root, block1, blocksize);
1358 if (eb && btrfs_buffer_uptodate(eb, gen))
1360 free_extent_buffer(eb);
1362 if (slot < nritems) {
1363 block2 = btrfs_node_blockptr(parent, slot + 1);
1364 gen = btrfs_node_ptr_generation(parent, slot + 1);
1365 eb = btrfs_find_tree_block(root, block2, blocksize);
1366 if (eb && btrfs_buffer_uptodate(eb, gen))
1368 free_extent_buffer(eb);
1370 if (block1 || block2) {
1372 btrfs_release_path(root, path);
1374 readahead_tree_block(root, block1, blocksize, 0);
1376 readahead_tree_block(root, block2, blocksize, 0);
1379 eb = read_tree_block(root, block1, blocksize, 0);
1380 free_extent_buffer(eb);
1383 eb = read_tree_block(root, block2, blocksize, 0);
1384 free_extent_buffer(eb);
1392 * when we walk down the tree, it is usually safe to unlock the higher layers
1393 * in the tree. The exceptions are when our path goes through slot 0, because
1394 * operations on the tree might require changing key pointers higher up in the
1397 * callers might also have set path->keep_locks, which tells this code to keep
1398 * the lock if the path points to the last slot in the block. This is part of
1399 * walking through the tree, and selecting the next slot in the higher block.
1401 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1402 * if lowest_unlock is 1, level 0 won't be unlocked
1404 static noinline void unlock_up(struct btrfs_path *path, int level,
1408 int skip_level = level;
1410 struct extent_buffer *t;
1412 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1413 if (!path->nodes[i])
1415 if (!path->locks[i])
1417 if (!no_skips && path->slots[i] == 0) {
1421 if (!no_skips && path->keep_locks) {
1424 nritems = btrfs_header_nritems(t);
1425 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1430 if (skip_level < i && i >= lowest_unlock)
1434 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1435 btrfs_tree_unlock(t);
1442 * This releases any locks held in the path starting at level and
1443 * going all the way up to the root.
1445 * btrfs_search_slot will keep the lock held on higher nodes in a few
1446 * corner cases, such as COW of the block at slot zero in the node. This
1447 * ignores those rules, and it should only be called when there are no
1448 * more updates to be done higher up in the tree.
1450 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1454 if (path->keep_locks || path->lowest_level)
1457 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1458 if (!path->nodes[i])
1460 if (!path->locks[i])
1462 btrfs_tree_unlock(path->nodes[i]);
1468 * look for key in the tree. path is filled in with nodes along the way
1469 * if key is found, we return zero and you can find the item in the leaf
1470 * level of the path (level 0)
1472 * If the key isn't found, the path points to the slot where it should
1473 * be inserted, and 1 is returned. If there are other errors during the
1474 * search a negative error number is returned.
1476 * if ins_len > 0, nodes and leaves will be split as we walk down the
1477 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1480 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1481 *root, struct btrfs_key *key, struct btrfs_path *p, int
1484 struct extent_buffer *b;
1485 struct extent_buffer *tmp;
1489 int should_reada = p->reada;
1490 int lowest_unlock = 1;
1492 u8 lowest_level = 0;
1495 struct btrfs_key prealloc_block;
1497 lowest_level = p->lowest_level;
1498 WARN_ON(lowest_level && ins_len > 0);
1499 WARN_ON(p->nodes[0] != NULL);
1504 prealloc_block.objectid = 0;
1507 if (p->skip_locking)
1508 b = btrfs_root_node(root);
1510 b = btrfs_lock_root_node(root);
1513 level = btrfs_header_level(b);
1516 * setup the path here so we can release it under lock
1517 * contention with the cow code
1519 p->nodes[level] = b;
1520 if (!p->skip_locking)
1521 p->locks[level] = 1;
1526 /* is a cow on this block not required */
1527 if (btrfs_header_generation(b) == trans->transid &&
1528 btrfs_header_owner(b) == root->root_key.objectid &&
1529 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1533 /* ok, we have to cow, is our old prealloc the right
1536 if (prealloc_block.objectid &&
1537 prealloc_block.offset != b->len) {
1538 btrfs_release_path(root, p);
1539 btrfs_free_reserved_extent(root,
1540 prealloc_block.objectid,
1541 prealloc_block.offset);
1542 prealloc_block.objectid = 0;
1547 * for higher level blocks, try not to allocate blocks
1548 * with the block and the parent locks held.
1550 if (level > 0 && !prealloc_block.objectid) {
1552 u64 hint = b->start;
1554 btrfs_release_path(root, p);
1555 ret = btrfs_reserve_extent(trans, root,
1558 &prealloc_block, 0);
1563 btrfs_set_path_blocking(p);
1565 wret = btrfs_cow_block(trans, root, b,
1566 p->nodes[level + 1],
1567 p->slots[level + 1],
1568 &b, prealloc_block.objectid);
1569 prealloc_block.objectid = 0;
1571 free_extent_buffer(b);
1577 BUG_ON(!cow && ins_len);
1578 if (level != btrfs_header_level(b))
1580 level = btrfs_header_level(b);
1582 p->nodes[level] = b;
1583 if (!p->skip_locking)
1584 p->locks[level] = 1;
1586 btrfs_clear_path_blocking(p, NULL);
1589 * we have a lock on b and as long as we aren't changing
1590 * the tree, there is no way to for the items in b to change.
1591 * It is safe to drop the lock on our parent before we
1592 * go through the expensive btree search on b.
1594 * If cow is true, then we might be changing slot zero,
1595 * which may require changing the parent. So, we can't
1596 * drop the lock until after we know which slot we're
1600 btrfs_unlock_up_safe(p, level + 1);
1602 ret = check_block(root, p, level);
1608 ret = bin_search(b, key, level, &slot);
1611 if (ret && slot > 0)
1613 p->slots[level] = slot;
1614 if ((p->search_for_split || ins_len > 0) &&
1615 btrfs_header_nritems(b) >=
1616 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1619 sret = reada_for_balance(root, p, level);
1623 btrfs_set_path_blocking(p);
1624 sret = split_node(trans, root, p, level);
1625 btrfs_clear_path_blocking(p, NULL);
1632 b = p->nodes[level];
1633 slot = p->slots[level];
1634 } else if (ins_len < 0 &&
1635 btrfs_header_nritems(b) <
1636 BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
1639 sret = reada_for_balance(root, p, level);
1643 btrfs_set_path_blocking(p);
1644 sret = balance_level(trans, root, p, level);
1645 btrfs_clear_path_blocking(p, NULL);
1651 b = p->nodes[level];
1653 btrfs_release_path(NULL, p);
1656 slot = p->slots[level];
1657 BUG_ON(btrfs_header_nritems(b) == 1);
1659 unlock_up(p, level, lowest_unlock);
1661 /* this is only true while dropping a snapshot */
1662 if (level == lowest_level) {
1667 blocknr = btrfs_node_blockptr(b, slot);
1668 gen = btrfs_node_ptr_generation(b, slot);
1669 blocksize = btrfs_level_size(root, level - 1);
1671 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1672 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1676 * reduce lock contention at high levels
1677 * of the btree by dropping locks before
1681 btrfs_release_path(NULL, p);
1683 free_extent_buffer(tmp);
1685 reada_for_search(root, p,
1689 tmp = read_tree_block(root, blocknr,
1692 free_extent_buffer(tmp);
1695 btrfs_set_path_blocking(p);
1697 free_extent_buffer(tmp);
1699 reada_for_search(root, p,
1702 b = read_node_slot(root, b, slot);
1705 if (!p->skip_locking) {
1708 btrfs_clear_path_blocking(p, NULL);
1709 lret = btrfs_try_spin_lock(b);
1712 btrfs_set_path_blocking(p);
1714 btrfs_clear_path_blocking(p, b);
1718 p->slots[level] = slot;
1720 btrfs_leaf_free_space(root, b) < ins_len) {
1723 btrfs_set_path_blocking(p);
1724 sret = split_leaf(trans, root, key,
1725 p, ins_len, ret == 0);
1726 btrfs_clear_path_blocking(p, NULL);
1734 if (!p->search_for_split)
1735 unlock_up(p, level, lowest_unlock);
1742 * we don't really know what they plan on doing with the path
1743 * from here on, so for now just mark it as blocking
1745 btrfs_set_path_blocking(p);
1746 if (prealloc_block.objectid) {
1747 btrfs_free_reserved_extent(root,
1748 prealloc_block.objectid,
1749 prealloc_block.offset);
1754 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1755 struct btrfs_root *root,
1756 struct btrfs_key *node_keys,
1757 u64 *nodes, int lowest_level)
1759 struct extent_buffer *eb;
1760 struct extent_buffer *parent;
1761 struct btrfs_key key;
1770 eb = btrfs_lock_root_node(root);
1771 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb, 0);
1774 btrfs_set_lock_blocking(eb);
1778 level = btrfs_header_level(parent);
1779 if (level == 0 || level <= lowest_level)
1782 ret = bin_search(parent, &node_keys[lowest_level], level,
1784 if (ret && slot > 0)
1787 bytenr = btrfs_node_blockptr(parent, slot);
1788 if (nodes[level - 1] == bytenr)
1791 blocksize = btrfs_level_size(root, level - 1);
1792 generation = btrfs_node_ptr_generation(parent, slot);
1793 btrfs_node_key_to_cpu(eb, &key, slot);
1794 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1796 if (generation == trans->transid) {
1797 eb = read_tree_block(root, bytenr, blocksize,
1799 btrfs_tree_lock(eb);
1800 btrfs_set_lock_blocking(eb);
1804 * if node keys match and node pointer hasn't been modified
1805 * in the running transaction, we can merge the path. for
1806 * blocks owened by reloc trees, the node pointer check is
1807 * skipped, this is because these blocks are fully controlled
1808 * by the space balance code, no one else can modify them.
1810 if (!nodes[level - 1] || !key_match ||
1811 (generation == trans->transid &&
1812 btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1813 if (level == 1 || level == lowest_level + 1) {
1814 if (generation == trans->transid) {
1815 btrfs_tree_unlock(eb);
1816 free_extent_buffer(eb);
1821 if (generation != trans->transid) {
1822 eb = read_tree_block(root, bytenr, blocksize,
1824 btrfs_tree_lock(eb);
1825 btrfs_set_lock_blocking(eb);
1828 ret = btrfs_cow_block(trans, root, eb, parent, slot,
1832 if (root->root_key.objectid ==
1833 BTRFS_TREE_RELOC_OBJECTID) {
1834 if (!nodes[level - 1]) {
1835 nodes[level - 1] = eb->start;
1836 memcpy(&node_keys[level - 1], &key,
1837 sizeof(node_keys[0]));
1843 btrfs_tree_unlock(parent);
1844 free_extent_buffer(parent);
1849 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1850 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1851 btrfs_mark_buffer_dirty(parent);
1853 ret = btrfs_inc_extent_ref(trans, root,
1855 blocksize, parent->start,
1856 btrfs_header_owner(parent),
1857 btrfs_header_generation(parent),
1862 * If the block was created in the running transaction,
1863 * it's possible this is the last reference to it, so we
1864 * should drop the subtree.
1866 if (generation == trans->transid) {
1867 ret = btrfs_drop_subtree(trans, root, eb, parent);
1869 btrfs_tree_unlock(eb);
1870 free_extent_buffer(eb);
1872 ret = btrfs_free_extent(trans, root, bytenr,
1873 blocksize, parent->start,
1874 btrfs_header_owner(parent),
1875 btrfs_header_generation(parent),
1881 btrfs_tree_unlock(parent);
1882 free_extent_buffer(parent);
1887 * adjust the pointers going up the tree, starting at level
1888 * making sure the right key of each node is points to 'key'.
1889 * This is used after shifting pointers to the left, so it stops
1890 * fixing up pointers when a given leaf/node is not in slot 0 of the
1893 * If this fails to write a tree block, it returns -1, but continues
1894 * fixing up the blocks in ram so the tree is consistent.
1896 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1897 struct btrfs_root *root, struct btrfs_path *path,
1898 struct btrfs_disk_key *key, int level)
1902 struct extent_buffer *t;
1904 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1905 int tslot = path->slots[i];
1906 if (!path->nodes[i])
1909 btrfs_set_node_key(t, key, tslot);
1910 btrfs_mark_buffer_dirty(path->nodes[i]);
1920 * This function isn't completely safe. It's the caller's responsibility
1921 * that the new key won't break the order
1923 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1924 struct btrfs_root *root, struct btrfs_path *path,
1925 struct btrfs_key *new_key)
1927 struct btrfs_disk_key disk_key;
1928 struct extent_buffer *eb;
1931 eb = path->nodes[0];
1932 slot = path->slots[0];
1934 btrfs_item_key(eb, &disk_key, slot - 1);
1935 if (comp_keys(&disk_key, new_key) >= 0)
1938 if (slot < btrfs_header_nritems(eb) - 1) {
1939 btrfs_item_key(eb, &disk_key, slot + 1);
1940 if (comp_keys(&disk_key, new_key) <= 0)
1944 btrfs_cpu_key_to_disk(&disk_key, new_key);
1945 btrfs_set_item_key(eb, &disk_key, slot);
1946 btrfs_mark_buffer_dirty(eb);
1948 fixup_low_keys(trans, root, path, &disk_key, 1);
1953 * try to push data from one node into the next node left in the
1956 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1957 * error, and > 0 if there was no room in the left hand block.
1959 static int push_node_left(struct btrfs_trans_handle *trans,
1960 struct btrfs_root *root, struct extent_buffer *dst,
1961 struct extent_buffer *src, int empty)
1968 src_nritems = btrfs_header_nritems(src);
1969 dst_nritems = btrfs_header_nritems(dst);
1970 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1971 WARN_ON(btrfs_header_generation(src) != trans->transid);
1972 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1974 if (!empty && src_nritems <= 8)
1977 if (push_items <= 0)
1981 push_items = min(src_nritems, push_items);
1982 if (push_items < src_nritems) {
1983 /* leave at least 8 pointers in the node if
1984 * we aren't going to empty it
1986 if (src_nritems - push_items < 8) {
1987 if (push_items <= 8)
1993 push_items = min(src_nritems - 8, push_items);
1995 copy_extent_buffer(dst, src,
1996 btrfs_node_key_ptr_offset(dst_nritems),
1997 btrfs_node_key_ptr_offset(0),
1998 push_items * sizeof(struct btrfs_key_ptr));
2000 if (push_items < src_nritems) {
2001 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
2002 btrfs_node_key_ptr_offset(push_items),
2003 (src_nritems - push_items) *
2004 sizeof(struct btrfs_key_ptr));
2006 btrfs_set_header_nritems(src, src_nritems - push_items);
2007 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2008 btrfs_mark_buffer_dirty(src);
2009 btrfs_mark_buffer_dirty(dst);
2011 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
2018 * try to push data from one node into the next node right in the
2021 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2022 * error, and > 0 if there was no room in the right hand block.
2024 * this will only push up to 1/2 the contents of the left node over
2026 static int balance_node_right(struct btrfs_trans_handle *trans,
2027 struct btrfs_root *root,
2028 struct extent_buffer *dst,
2029 struct extent_buffer *src)
2037 WARN_ON(btrfs_header_generation(src) != trans->transid);
2038 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2040 src_nritems = btrfs_header_nritems(src);
2041 dst_nritems = btrfs_header_nritems(dst);
2042 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2043 if (push_items <= 0)
2046 if (src_nritems < 4)
2049 max_push = src_nritems / 2 + 1;
2050 /* don't try to empty the node */
2051 if (max_push >= src_nritems)
2054 if (max_push < push_items)
2055 push_items = max_push;
2057 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
2058 btrfs_node_key_ptr_offset(0),
2060 sizeof(struct btrfs_key_ptr));
2062 copy_extent_buffer(dst, src,
2063 btrfs_node_key_ptr_offset(0),
2064 btrfs_node_key_ptr_offset(src_nritems - push_items),
2065 push_items * sizeof(struct btrfs_key_ptr));
2067 btrfs_set_header_nritems(src, src_nritems - push_items);
2068 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2070 btrfs_mark_buffer_dirty(src);
2071 btrfs_mark_buffer_dirty(dst);
2073 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2080 * helper function to insert a new root level in the tree.
2081 * A new node is allocated, and a single item is inserted to
2082 * point to the existing root
2084 * returns zero on success or < 0 on failure.
2086 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2087 struct btrfs_root *root,
2088 struct btrfs_path *path, int level)
2091 struct extent_buffer *lower;
2092 struct extent_buffer *c;
2093 struct extent_buffer *old;
2094 struct btrfs_disk_key lower_key;
2097 BUG_ON(path->nodes[level]);
2098 BUG_ON(path->nodes[level-1] != root->node);
2100 lower = path->nodes[level-1];
2102 btrfs_item_key(lower, &lower_key, 0);
2104 btrfs_node_key(lower, &lower_key, 0);
2106 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2107 root->root_key.objectid, trans->transid,
2108 level, root->node->start, 0);
2112 memset_extent_buffer(c, 0, 0, root->nodesize);
2113 btrfs_set_header_nritems(c, 1);
2114 btrfs_set_header_level(c, level);
2115 btrfs_set_header_bytenr(c, c->start);
2116 btrfs_set_header_generation(c, trans->transid);
2117 btrfs_set_header_owner(c, root->root_key.objectid);
2119 write_extent_buffer(c, root->fs_info->fsid,
2120 (unsigned long)btrfs_header_fsid(c),
2123 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2124 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2127 btrfs_set_node_key(c, &lower_key, 0);
2128 btrfs_set_node_blockptr(c, 0, lower->start);
2129 lower_gen = btrfs_header_generation(lower);
2130 WARN_ON(lower_gen != trans->transid);
2132 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2134 btrfs_mark_buffer_dirty(c);
2136 spin_lock(&root->node_lock);
2139 spin_unlock(&root->node_lock);
2141 ret = btrfs_update_extent_ref(trans, root, lower->start,
2142 lower->start, c->start,
2143 root->root_key.objectid,
2144 trans->transid, level - 1);
2147 /* the super has an extra ref to root->node */
2148 free_extent_buffer(old);
2150 add_root_to_dirty_list(root);
2151 extent_buffer_get(c);
2152 path->nodes[level] = c;
2153 path->locks[level] = 1;
2154 path->slots[level] = 0;
2159 * worker function to insert a single pointer in a node.
2160 * the node should have enough room for the pointer already
2162 * slot and level indicate where you want the key to go, and
2163 * blocknr is the block the key points to.
2165 * returns zero on success and < 0 on any error
2167 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2168 *root, struct btrfs_path *path, struct btrfs_disk_key
2169 *key, u64 bytenr, int slot, int level)
2171 struct extent_buffer *lower;
2174 BUG_ON(!path->nodes[level]);
2175 lower = path->nodes[level];
2176 nritems = btrfs_header_nritems(lower);
2179 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2181 if (slot != nritems) {
2182 memmove_extent_buffer(lower,
2183 btrfs_node_key_ptr_offset(slot + 1),
2184 btrfs_node_key_ptr_offset(slot),
2185 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2187 btrfs_set_node_key(lower, key, slot);
2188 btrfs_set_node_blockptr(lower, slot, bytenr);
2189 WARN_ON(trans->transid == 0);
2190 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2191 btrfs_set_header_nritems(lower, nritems + 1);
2192 btrfs_mark_buffer_dirty(lower);
2197 * split the node at the specified level in path in two.
2198 * The path is corrected to point to the appropriate node after the split
2200 * Before splitting this tries to make some room in the node by pushing
2201 * left and right, if either one works, it returns right away.
2203 * returns 0 on success and < 0 on failure
2205 static noinline int split_node(struct btrfs_trans_handle *trans,
2206 struct btrfs_root *root,
2207 struct btrfs_path *path, int level)
2209 struct extent_buffer *c;
2210 struct extent_buffer *split;
2211 struct btrfs_disk_key disk_key;
2217 c = path->nodes[level];
2218 WARN_ON(btrfs_header_generation(c) != trans->transid);
2219 if (c == root->node) {
2220 /* trying to split the root, lets make a new one */
2221 ret = insert_new_root(trans, root, path, level + 1);
2225 ret = push_nodes_for_insert(trans, root, path, level);
2226 c = path->nodes[level];
2227 if (!ret && btrfs_header_nritems(c) <
2228 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2234 c_nritems = btrfs_header_nritems(c);
2236 split = btrfs_alloc_free_block(trans, root, root->nodesize,
2237 path->nodes[level + 1]->start,
2238 root->root_key.objectid,
2239 trans->transid, level, c->start, 0);
2241 return PTR_ERR(split);
2243 btrfs_set_header_flags(split, btrfs_header_flags(c));
2244 btrfs_set_header_level(split, btrfs_header_level(c));
2245 btrfs_set_header_bytenr(split, split->start);
2246 btrfs_set_header_generation(split, trans->transid);
2247 btrfs_set_header_owner(split, root->root_key.objectid);
2248 btrfs_set_header_flags(split, 0);
2249 write_extent_buffer(split, root->fs_info->fsid,
2250 (unsigned long)btrfs_header_fsid(split),
2252 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2253 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2256 mid = (c_nritems + 1) / 2;
2258 copy_extent_buffer(split, c,
2259 btrfs_node_key_ptr_offset(0),
2260 btrfs_node_key_ptr_offset(mid),
2261 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2262 btrfs_set_header_nritems(split, c_nritems - mid);
2263 btrfs_set_header_nritems(c, mid);
2266 btrfs_mark_buffer_dirty(c);
2267 btrfs_mark_buffer_dirty(split);
2269 btrfs_node_key(split, &disk_key, 0);
2270 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2271 path->slots[level + 1] + 1,
2276 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2279 if (path->slots[level] >= mid) {
2280 path->slots[level] -= mid;
2281 btrfs_tree_unlock(c);
2282 free_extent_buffer(c);
2283 path->nodes[level] = split;
2284 path->slots[level + 1] += 1;
2286 btrfs_tree_unlock(split);
2287 free_extent_buffer(split);
2293 * how many bytes are required to store the items in a leaf. start
2294 * and nr indicate which items in the leaf to check. This totals up the
2295 * space used both by the item structs and the item data
2297 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2300 int nritems = btrfs_header_nritems(l);
2301 int end = min(nritems, start + nr) - 1;
2305 data_len = btrfs_item_end_nr(l, start);
2306 data_len = data_len - btrfs_item_offset_nr(l, end);
2307 data_len += sizeof(struct btrfs_item) * nr;
2308 WARN_ON(data_len < 0);
2313 * The space between the end of the leaf items and
2314 * the start of the leaf data. IOW, how much room
2315 * the leaf has left for both items and data
2317 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2318 struct extent_buffer *leaf)
2320 int nritems = btrfs_header_nritems(leaf);
2322 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2324 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2325 "used %d nritems %d\n",
2326 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2327 leaf_space_used(leaf, 0, nritems), nritems);
2333 * push some data in the path leaf to the right, trying to free up at
2334 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2336 * returns 1 if the push failed because the other node didn't have enough
2337 * room, 0 if everything worked out and < 0 if there were major errors.
2339 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2340 *root, struct btrfs_path *path, int data_size,
2343 struct extent_buffer *left = path->nodes[0];
2344 struct extent_buffer *right;
2345 struct extent_buffer *upper;
2346 struct btrfs_disk_key disk_key;
2352 struct btrfs_item *item;
2360 slot = path->slots[1];
2361 if (!path->nodes[1])
2364 upper = path->nodes[1];
2365 if (slot >= btrfs_header_nritems(upper) - 1)
2368 btrfs_assert_tree_locked(path->nodes[1]);
2370 right = read_node_slot(root, upper, slot + 1);
2371 btrfs_tree_lock(right);
2372 btrfs_set_lock_blocking(right);
2374 free_space = btrfs_leaf_free_space(root, right);
2375 if (free_space < data_size)
2378 /* cow and double check */
2379 ret = btrfs_cow_block(trans, root, right, upper,
2380 slot + 1, &right, 0);
2384 free_space = btrfs_leaf_free_space(root, right);
2385 if (free_space < data_size)
2388 left_nritems = btrfs_header_nritems(left);
2389 if (left_nritems == 0)
2397 if (path->slots[0] >= left_nritems)
2398 push_space += data_size;
2400 i = left_nritems - 1;
2402 item = btrfs_item_nr(left, i);
2404 if (!empty && push_items > 0) {
2405 if (path->slots[0] > i)
2407 if (path->slots[0] == i) {
2408 int space = btrfs_leaf_free_space(root, left);
2409 if (space + push_space * 2 > free_space)
2414 if (path->slots[0] == i)
2415 push_space += data_size;
2417 if (!left->map_token) {
2418 map_extent_buffer(left, (unsigned long)item,
2419 sizeof(struct btrfs_item),
2420 &left->map_token, &left->kaddr,
2421 &left->map_start, &left->map_len,
2425 this_item_size = btrfs_item_size(left, item);
2426 if (this_item_size + sizeof(*item) + push_space > free_space)
2430 push_space += this_item_size + sizeof(*item);
2435 if (left->map_token) {
2436 unmap_extent_buffer(left, left->map_token, KM_USER1);
2437 left->map_token = NULL;
2440 if (push_items == 0)
2443 if (!empty && push_items == left_nritems)
2446 /* push left to right */
2447 right_nritems = btrfs_header_nritems(right);
2449 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2450 push_space -= leaf_data_end(root, left);
2452 /* make room in the right data area */
2453 data_end = leaf_data_end(root, right);
2454 memmove_extent_buffer(right,
2455 btrfs_leaf_data(right) + data_end - push_space,
2456 btrfs_leaf_data(right) + data_end,
2457 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2459 /* copy from the left data area */
2460 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2461 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2462 btrfs_leaf_data(left) + leaf_data_end(root, left),
2465 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2466 btrfs_item_nr_offset(0),
2467 right_nritems * sizeof(struct btrfs_item));
2469 /* copy the items from left to right */
2470 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2471 btrfs_item_nr_offset(left_nritems - push_items),
2472 push_items * sizeof(struct btrfs_item));
2474 /* update the item pointers */
2475 right_nritems += push_items;
2476 btrfs_set_header_nritems(right, right_nritems);
2477 push_space = BTRFS_LEAF_DATA_SIZE(root);
2478 for (i = 0; i < right_nritems; i++) {
2479 item = btrfs_item_nr(right, i);
2480 if (!right->map_token) {
2481 map_extent_buffer(right, (unsigned long)item,
2482 sizeof(struct btrfs_item),
2483 &right->map_token, &right->kaddr,
2484 &right->map_start, &right->map_len,
2487 push_space -= btrfs_item_size(right, item);
2488 btrfs_set_item_offset(right, item, push_space);
2491 if (right->map_token) {
2492 unmap_extent_buffer(right, right->map_token, KM_USER1);
2493 right->map_token = NULL;
2495 left_nritems -= push_items;
2496 btrfs_set_header_nritems(left, left_nritems);
2499 btrfs_mark_buffer_dirty(left);
2500 btrfs_mark_buffer_dirty(right);
2502 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2505 btrfs_item_key(right, &disk_key, 0);
2506 btrfs_set_node_key(upper, &disk_key, slot + 1);
2507 btrfs_mark_buffer_dirty(upper);
2509 /* then fixup the leaf pointer in the path */
2510 if (path->slots[0] >= left_nritems) {
2511 path->slots[0] -= left_nritems;
2512 if (btrfs_header_nritems(path->nodes[0]) == 0)
2513 clean_tree_block(trans, root, path->nodes[0]);
2514 btrfs_tree_unlock(path->nodes[0]);
2515 free_extent_buffer(path->nodes[0]);
2516 path->nodes[0] = right;
2517 path->slots[1] += 1;
2519 btrfs_tree_unlock(right);
2520 free_extent_buffer(right);
2525 btrfs_tree_unlock(right);
2526 free_extent_buffer(right);
2531 * push some data in the path leaf to the left, trying to free up at
2532 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2534 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2535 *root, struct btrfs_path *path, int data_size,
2538 struct btrfs_disk_key disk_key;
2539 struct extent_buffer *right = path->nodes[0];
2540 struct extent_buffer *left;
2546 struct btrfs_item *item;
2547 u32 old_left_nritems;
2553 u32 old_left_item_size;
2555 slot = path->slots[1];
2558 if (!path->nodes[1])
2561 right_nritems = btrfs_header_nritems(right);
2562 if (right_nritems == 0)
2565 btrfs_assert_tree_locked(path->nodes[1]);
2567 left = read_node_slot(root, path->nodes[1], slot - 1);
2568 btrfs_tree_lock(left);
2569 btrfs_set_lock_blocking(left);
2571 free_space = btrfs_leaf_free_space(root, left);
2572 if (free_space < data_size) {
2577 /* cow and double check */
2578 ret = btrfs_cow_block(trans, root, left,
2579 path->nodes[1], slot - 1, &left, 0);
2581 /* we hit -ENOSPC, but it isn't fatal here */
2586 free_space = btrfs_leaf_free_space(root, left);
2587 if (free_space < data_size) {
2595 nr = right_nritems - 1;
2597 for (i = 0; i < nr; i++) {
2598 item = btrfs_item_nr(right, i);
2599 if (!right->map_token) {
2600 map_extent_buffer(right, (unsigned long)item,
2601 sizeof(struct btrfs_item),
2602 &right->map_token, &right->kaddr,
2603 &right->map_start, &right->map_len,
2607 if (!empty && push_items > 0) {
2608 if (path->slots[0] < i)
2610 if (path->slots[0] == i) {
2611 int space = btrfs_leaf_free_space(root, right);
2612 if (space + push_space * 2 > free_space)
2617 if (path->slots[0] == i)
2618 push_space += data_size;
2620 this_item_size = btrfs_item_size(right, item);
2621 if (this_item_size + sizeof(*item) + push_space > free_space)
2625 push_space += this_item_size + sizeof(*item);
2628 if (right->map_token) {
2629 unmap_extent_buffer(right, right->map_token, KM_USER1);
2630 right->map_token = NULL;
2633 if (push_items == 0) {
2637 if (!empty && push_items == btrfs_header_nritems(right))
2640 /* push data from right to left */
2641 copy_extent_buffer(left, right,
2642 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2643 btrfs_item_nr_offset(0),
2644 push_items * sizeof(struct btrfs_item));
2646 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2647 btrfs_item_offset_nr(right, push_items - 1);
2649 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2650 leaf_data_end(root, left) - push_space,
2651 btrfs_leaf_data(right) +
2652 btrfs_item_offset_nr(right, push_items - 1),
2654 old_left_nritems = btrfs_header_nritems(left);
2655 BUG_ON(old_left_nritems <= 0);
2657 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2658 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2661 item = btrfs_item_nr(left, i);
2662 if (!left->map_token) {
2663 map_extent_buffer(left, (unsigned long)item,
2664 sizeof(struct btrfs_item),
2665 &left->map_token, &left->kaddr,
2666 &left->map_start, &left->map_len,
2670 ioff = btrfs_item_offset(left, item);
2671 btrfs_set_item_offset(left, item,
2672 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2674 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2675 if (left->map_token) {
2676 unmap_extent_buffer(left, left->map_token, KM_USER1);
2677 left->map_token = NULL;
2680 /* fixup right node */
2681 if (push_items > right_nritems) {
2682 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2687 if (push_items < right_nritems) {
2688 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2689 leaf_data_end(root, right);
2690 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2691 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2692 btrfs_leaf_data(right) +
2693 leaf_data_end(root, right), push_space);
2695 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2696 btrfs_item_nr_offset(push_items),
2697 (btrfs_header_nritems(right) - push_items) *
2698 sizeof(struct btrfs_item));
2700 right_nritems -= push_items;
2701 btrfs_set_header_nritems(right, right_nritems);
2702 push_space = BTRFS_LEAF_DATA_SIZE(root);
2703 for (i = 0; i < right_nritems; i++) {
2704 item = btrfs_item_nr(right, i);
2706 if (!right->map_token) {
2707 map_extent_buffer(right, (unsigned long)item,
2708 sizeof(struct btrfs_item),
2709 &right->map_token, &right->kaddr,
2710 &right->map_start, &right->map_len,
2714 push_space = push_space - btrfs_item_size(right, item);
2715 btrfs_set_item_offset(right, item, push_space);
2717 if (right->map_token) {
2718 unmap_extent_buffer(right, right->map_token, KM_USER1);
2719 right->map_token = NULL;
2722 btrfs_mark_buffer_dirty(left);
2724 btrfs_mark_buffer_dirty(right);
2726 ret = btrfs_update_ref(trans, root, right, left,
2727 old_left_nritems, push_items);
2730 btrfs_item_key(right, &disk_key, 0);
2731 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2735 /* then fixup the leaf pointer in the path */
2736 if (path->slots[0] < push_items) {
2737 path->slots[0] += old_left_nritems;
2738 if (btrfs_header_nritems(path->nodes[0]) == 0)
2739 clean_tree_block(trans, root, path->nodes[0]);
2740 btrfs_tree_unlock(path->nodes[0]);
2741 free_extent_buffer(path->nodes[0]);
2742 path->nodes[0] = left;
2743 path->slots[1] -= 1;
2745 btrfs_tree_unlock(left);
2746 free_extent_buffer(left);
2747 path->slots[0] -= push_items;
2749 BUG_ON(path->slots[0] < 0);
2752 btrfs_tree_unlock(left);
2753 free_extent_buffer(left);
2758 * split the path's leaf in two, making sure there is at least data_size
2759 * available for the resulting leaf level of the path.
2761 * returns 0 if all went well and < 0 on failure.
2763 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2764 struct btrfs_root *root,
2765 struct btrfs_key *ins_key,
2766 struct btrfs_path *path, int data_size,
2769 struct extent_buffer *l;
2773 struct extent_buffer *right;
2780 int num_doubles = 0;
2781 struct btrfs_disk_key disk_key;
2783 /* first try to make some room by pushing left and right */
2784 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2785 wret = push_leaf_right(trans, root, path, data_size, 0);
2789 wret = push_leaf_left(trans, root, path, data_size, 0);
2795 /* did the pushes work? */
2796 if (btrfs_leaf_free_space(root, l) >= data_size)
2800 if (!path->nodes[1]) {
2801 ret = insert_new_root(trans, root, path, 1);
2808 slot = path->slots[0];
2809 nritems = btrfs_header_nritems(l);
2810 mid = (nritems + 1) / 2;
2812 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2813 path->nodes[1]->start,
2814 root->root_key.objectid,
2815 trans->transid, 0, l->start, 0);
2816 if (IS_ERR(right)) {
2818 return PTR_ERR(right);
2821 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2822 btrfs_set_header_bytenr(right, right->start);
2823 btrfs_set_header_generation(right, trans->transid);
2824 btrfs_set_header_owner(right, root->root_key.objectid);
2825 btrfs_set_header_level(right, 0);
2826 write_extent_buffer(right, root->fs_info->fsid,
2827 (unsigned long)btrfs_header_fsid(right),
2830 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2831 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2835 leaf_space_used(l, mid, nritems - mid) + data_size >
2836 BTRFS_LEAF_DATA_SIZE(root)) {
2837 if (slot >= nritems) {
2838 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2839 btrfs_set_header_nritems(right, 0);
2840 wret = insert_ptr(trans, root, path,
2841 &disk_key, right->start,
2842 path->slots[1] + 1, 1);
2846 btrfs_tree_unlock(path->nodes[0]);
2847 free_extent_buffer(path->nodes[0]);
2848 path->nodes[0] = right;
2850 path->slots[1] += 1;
2851 btrfs_mark_buffer_dirty(right);
2855 if (mid != nritems &&
2856 leaf_space_used(l, mid, nritems - mid) +
2857 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2862 if (leaf_space_used(l, 0, mid) + data_size >
2863 BTRFS_LEAF_DATA_SIZE(root)) {
2864 if (!extend && data_size && slot == 0) {
2865 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2866 btrfs_set_header_nritems(right, 0);
2867 wret = insert_ptr(trans, root, path,
2873 btrfs_tree_unlock(path->nodes[0]);
2874 free_extent_buffer(path->nodes[0]);
2875 path->nodes[0] = right;
2877 if (path->slots[1] == 0) {
2878 wret = fixup_low_keys(trans, root,
2879 path, &disk_key, 1);
2883 btrfs_mark_buffer_dirty(right);
2885 } else if ((extend || !data_size) && slot == 0) {
2889 if (mid != nritems &&
2890 leaf_space_used(l, mid, nritems - mid) +
2891 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2897 nritems = nritems - mid;
2898 btrfs_set_header_nritems(right, nritems);
2899 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2901 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2902 btrfs_item_nr_offset(mid),
2903 nritems * sizeof(struct btrfs_item));
2905 copy_extent_buffer(right, l,
2906 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2907 data_copy_size, btrfs_leaf_data(l) +
2908 leaf_data_end(root, l), data_copy_size);
2910 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2911 btrfs_item_end_nr(l, mid);
2913 for (i = 0; i < nritems; i++) {
2914 struct btrfs_item *item = btrfs_item_nr(right, i);
2917 if (!right->map_token) {
2918 map_extent_buffer(right, (unsigned long)item,
2919 sizeof(struct btrfs_item),
2920 &right->map_token, &right->kaddr,
2921 &right->map_start, &right->map_len,
2925 ioff = btrfs_item_offset(right, item);
2926 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2929 if (right->map_token) {
2930 unmap_extent_buffer(right, right->map_token, KM_USER1);
2931 right->map_token = NULL;
2934 btrfs_set_header_nritems(l, mid);
2936 btrfs_item_key(right, &disk_key, 0);
2937 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2938 path->slots[1] + 1, 1);
2942 btrfs_mark_buffer_dirty(right);
2943 btrfs_mark_buffer_dirty(l);
2944 BUG_ON(path->slots[0] != slot);
2946 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2950 btrfs_tree_unlock(path->nodes[0]);
2951 free_extent_buffer(path->nodes[0]);
2952 path->nodes[0] = right;
2953 path->slots[0] -= mid;
2954 path->slots[1] += 1;
2956 btrfs_tree_unlock(right);
2957 free_extent_buffer(right);
2960 BUG_ON(path->slots[0] < 0);
2963 BUG_ON(num_doubles != 0);
2971 * This function splits a single item into two items,
2972 * giving 'new_key' to the new item and splitting the
2973 * old one at split_offset (from the start of the item).
2975 * The path may be released by this operation. After
2976 * the split, the path is pointing to the old item. The
2977 * new item is going to be in the same node as the old one.
2979 * Note, the item being split must be smaller enough to live alone on
2980 * a tree block with room for one extra struct btrfs_item
2982 * This allows us to split the item in place, keeping a lock on the
2983 * leaf the entire time.
2985 int btrfs_split_item(struct btrfs_trans_handle *trans,
2986 struct btrfs_root *root,
2987 struct btrfs_path *path,
2988 struct btrfs_key *new_key,
2989 unsigned long split_offset)
2992 struct extent_buffer *leaf;
2993 struct btrfs_key orig_key;
2994 struct btrfs_item *item;
2995 struct btrfs_item *new_item;
3000 struct btrfs_disk_key disk_key;
3003 leaf = path->nodes[0];
3004 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
3005 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
3008 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
3009 btrfs_release_path(root, path);
3011 path->search_for_split = 1;
3012 path->keep_locks = 1;
3014 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
3015 path->search_for_split = 0;
3017 /* if our item isn't there or got smaller, return now */
3018 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
3020 path->keep_locks = 0;
3024 ret = split_leaf(trans, root, &orig_key, path,
3025 sizeof(struct btrfs_item), 1);
3026 path->keep_locks = 0;
3030 * make sure any changes to the path from split_leaf leave it
3031 * in a blocking state
3033 btrfs_set_path_blocking(path);
3035 leaf = path->nodes[0];
3036 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
3039 item = btrfs_item_nr(leaf, path->slots[0]);
3040 orig_offset = btrfs_item_offset(leaf, item);
3041 item_size = btrfs_item_size(leaf, item);
3044 buf = kmalloc(item_size, GFP_NOFS);
3045 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
3046 path->slots[0]), item_size);
3047 slot = path->slots[0] + 1;
3048 leaf = path->nodes[0];
3050 nritems = btrfs_header_nritems(leaf);
3052 if (slot != nritems) {
3053 /* shift the items */
3054 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
3055 btrfs_item_nr_offset(slot),
3056 (nritems - slot) * sizeof(struct btrfs_item));
3060 btrfs_cpu_key_to_disk(&disk_key, new_key);
3061 btrfs_set_item_key(leaf, &disk_key, slot);
3063 new_item = btrfs_item_nr(leaf, slot);
3065 btrfs_set_item_offset(leaf, new_item, orig_offset);
3066 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3068 btrfs_set_item_offset(leaf, item,
3069 orig_offset + item_size - split_offset);
3070 btrfs_set_item_size(leaf, item, split_offset);
3072 btrfs_set_header_nritems(leaf, nritems + 1);
3074 /* write the data for the start of the original item */
3075 write_extent_buffer(leaf, buf,
3076 btrfs_item_ptr_offset(leaf, path->slots[0]),
3079 /* write the data for the new item */
3080 write_extent_buffer(leaf, buf + split_offset,
3081 btrfs_item_ptr_offset(leaf, slot),
3082 item_size - split_offset);
3083 btrfs_mark_buffer_dirty(leaf);
3086 if (btrfs_leaf_free_space(root, leaf) < 0) {
3087 btrfs_print_leaf(root, leaf);
3095 * make the item pointed to by the path smaller. new_size indicates
3096 * how small to make it, and from_end tells us if we just chop bytes
3097 * off the end of the item or if we shift the item to chop bytes off
3100 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3101 struct btrfs_root *root,
3102 struct btrfs_path *path,
3103 u32 new_size, int from_end)
3108 struct extent_buffer *leaf;
3109 struct btrfs_item *item;
3111 unsigned int data_end;
3112 unsigned int old_data_start;
3113 unsigned int old_size;
3114 unsigned int size_diff;
3117 slot_orig = path->slots[0];
3118 leaf = path->nodes[0];
3119 slot = path->slots[0];
3121 old_size = btrfs_item_size_nr(leaf, slot);
3122 if (old_size == new_size)
3125 nritems = btrfs_header_nritems(leaf);
3126 data_end = leaf_data_end(root, leaf);
3128 old_data_start = btrfs_item_offset_nr(leaf, slot);
3130 size_diff = old_size - new_size;
3133 BUG_ON(slot >= nritems);
3136 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3138 /* first correct the data pointers */
3139 for (i = slot; i < nritems; i++) {
3141 item = btrfs_item_nr(leaf, i);
3143 if (!leaf->map_token) {
3144 map_extent_buffer(leaf, (unsigned long)item,
3145 sizeof(struct btrfs_item),
3146 &leaf->map_token, &leaf->kaddr,
3147 &leaf->map_start, &leaf->map_len,
3151 ioff = btrfs_item_offset(leaf, item);
3152 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3155 if (leaf->map_token) {
3156 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3157 leaf->map_token = NULL;
3160 /* shift the data */
3162 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3163 data_end + size_diff, btrfs_leaf_data(leaf) +
3164 data_end, old_data_start + new_size - data_end);
3166 struct btrfs_disk_key disk_key;
3169 btrfs_item_key(leaf, &disk_key, slot);
3171 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3173 struct btrfs_file_extent_item *fi;
3175 fi = btrfs_item_ptr(leaf, slot,
3176 struct btrfs_file_extent_item);
3177 fi = (struct btrfs_file_extent_item *)(
3178 (unsigned long)fi - size_diff);
3180 if (btrfs_file_extent_type(leaf, fi) ==
3181 BTRFS_FILE_EXTENT_INLINE) {
3182 ptr = btrfs_item_ptr_offset(leaf, slot);
3183 memmove_extent_buffer(leaf, ptr,
3185 offsetof(struct btrfs_file_extent_item,
3190 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3191 data_end + size_diff, btrfs_leaf_data(leaf) +
3192 data_end, old_data_start - data_end);
3194 offset = btrfs_disk_key_offset(&disk_key);
3195 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3196 btrfs_set_item_key(leaf, &disk_key, slot);
3198 fixup_low_keys(trans, root, path, &disk_key, 1);
3201 item = btrfs_item_nr(leaf, slot);
3202 btrfs_set_item_size(leaf, item, new_size);
3203 btrfs_mark_buffer_dirty(leaf);
3206 if (btrfs_leaf_free_space(root, leaf) < 0) {
3207 btrfs_print_leaf(root, leaf);
3214 * make the item pointed to by the path bigger, data_size is the new size.
3216 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3217 struct btrfs_root *root, struct btrfs_path *path,
3223 struct extent_buffer *leaf;
3224 struct btrfs_item *item;
3226 unsigned int data_end;
3227 unsigned int old_data;
3228 unsigned int old_size;
3231 slot_orig = path->slots[0];
3232 leaf = path->nodes[0];
3234 nritems = btrfs_header_nritems(leaf);
3235 data_end = leaf_data_end(root, leaf);
3237 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3238 btrfs_print_leaf(root, leaf);
3241 slot = path->slots[0];
3242 old_data = btrfs_item_end_nr(leaf, slot);
3245 if (slot >= nritems) {
3246 btrfs_print_leaf(root, leaf);
3247 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3253 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3255 /* first correct the data pointers */
3256 for (i = slot; i < nritems; i++) {
3258 item = btrfs_item_nr(leaf, i);
3260 if (!leaf->map_token) {
3261 map_extent_buffer(leaf, (unsigned long)item,
3262 sizeof(struct btrfs_item),
3263 &leaf->map_token, &leaf->kaddr,
3264 &leaf->map_start, &leaf->map_len,
3267 ioff = btrfs_item_offset(leaf, item);
3268 btrfs_set_item_offset(leaf, item, ioff - data_size);
3271 if (leaf->map_token) {
3272 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3273 leaf->map_token = NULL;
3276 /* shift the data */
3277 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3278 data_end - data_size, btrfs_leaf_data(leaf) +
3279 data_end, old_data - data_end);
3281 data_end = old_data;
3282 old_size = btrfs_item_size_nr(leaf, slot);
3283 item = btrfs_item_nr(leaf, slot);
3284 btrfs_set_item_size(leaf, item, old_size + data_size);
3285 btrfs_mark_buffer_dirty(leaf);
3288 if (btrfs_leaf_free_space(root, leaf) < 0) {
3289 btrfs_print_leaf(root, leaf);
3296 * Given a key and some data, insert items into the tree.
3297 * This does all the path init required, making room in the tree if needed.
3298 * Returns the number of keys that were inserted.
3300 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3301 struct btrfs_root *root,
3302 struct btrfs_path *path,
3303 struct btrfs_key *cpu_key, u32 *data_size,
3306 struct extent_buffer *leaf;
3307 struct btrfs_item *item;
3314 unsigned int data_end;
3315 struct btrfs_disk_key disk_key;
3316 struct btrfs_key found_key;
3318 for (i = 0; i < nr; i++) {
3319 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3320 BTRFS_LEAF_DATA_SIZE(root)) {
3324 total_data += data_size[i];
3325 total_size += data_size[i] + sizeof(struct btrfs_item);
3329 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3335 leaf = path->nodes[0];
3337 nritems = btrfs_header_nritems(leaf);
3338 data_end = leaf_data_end(root, leaf);
3340 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3341 for (i = nr; i >= 0; i--) {
3342 total_data -= data_size[i];
3343 total_size -= data_size[i] + sizeof(struct btrfs_item);
3344 if (total_size < btrfs_leaf_free_space(root, leaf))
3350 slot = path->slots[0];
3353 if (slot != nritems) {
3354 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3356 item = btrfs_item_nr(leaf, slot);
3357 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3359 /* figure out how many keys we can insert in here */
3360 total_data = data_size[0];
3361 for (i = 1; i < nr; i++) {
3362 if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3364 total_data += data_size[i];
3368 if (old_data < data_end) {
3369 btrfs_print_leaf(root, leaf);
3370 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3371 slot, old_data, data_end);
3375 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3377 /* first correct the data pointers */
3378 WARN_ON(leaf->map_token);
3379 for (i = slot; i < nritems; i++) {
3382 item = btrfs_item_nr(leaf, i);
3383 if (!leaf->map_token) {
3384 map_extent_buffer(leaf, (unsigned long)item,
3385 sizeof(struct btrfs_item),
3386 &leaf->map_token, &leaf->kaddr,
3387 &leaf->map_start, &leaf->map_len,
3391 ioff = btrfs_item_offset(leaf, item);
3392 btrfs_set_item_offset(leaf, item, ioff - total_data);
3394 if (leaf->map_token) {
3395 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3396 leaf->map_token = NULL;
3399 /* shift the items */
3400 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3401 btrfs_item_nr_offset(slot),
3402 (nritems - slot) * sizeof(struct btrfs_item));
3404 /* shift the data */
3405 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3406 data_end - total_data, btrfs_leaf_data(leaf) +
3407 data_end, old_data - data_end);
3408 data_end = old_data;
3411 * this sucks but it has to be done, if we are inserting at
3412 * the end of the leaf only insert 1 of the items, since we
3413 * have no way of knowing whats on the next leaf and we'd have
3414 * to drop our current locks to figure it out
3419 /* setup the item for the new data */
3420 for (i = 0; i < nr; i++) {
3421 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3422 btrfs_set_item_key(leaf, &disk_key, slot + i);
3423 item = btrfs_item_nr(leaf, slot + i);
3424 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3425 data_end -= data_size[i];
3426 btrfs_set_item_size(leaf, item, data_size[i]);
3428 btrfs_set_header_nritems(leaf, nritems + nr);
3429 btrfs_mark_buffer_dirty(leaf);
3433 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3434 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3437 if (btrfs_leaf_free_space(root, leaf) < 0) {
3438 btrfs_print_leaf(root, leaf);
3448 * Given a key and some data, insert items into the tree.
3449 * This does all the path init required, making room in the tree if needed.
3451 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3452 struct btrfs_root *root,
3453 struct btrfs_path *path,
3454 struct btrfs_key *cpu_key, u32 *data_size,
3457 struct extent_buffer *leaf;
3458 struct btrfs_item *item;
3466 unsigned int data_end;
3467 struct btrfs_disk_key disk_key;
3469 for (i = 0; i < nr; i++)
3470 total_data += data_size[i];
3472 total_size = total_data + (nr * sizeof(struct btrfs_item));
3473 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3479 slot_orig = path->slots[0];
3480 leaf = path->nodes[0];
3482 nritems = btrfs_header_nritems(leaf);
3483 data_end = leaf_data_end(root, leaf);
3485 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3486 btrfs_print_leaf(root, leaf);
3487 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3488 total_size, btrfs_leaf_free_space(root, leaf));
3492 slot = path->slots[0];
3495 if (slot != nritems) {
3496 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3498 if (old_data < data_end) {
3499 btrfs_print_leaf(root, leaf);
3500 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3501 slot, old_data, data_end);
3505 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3507 /* first correct the data pointers */
3508 WARN_ON(leaf->map_token);
3509 for (i = slot; i < nritems; i++) {
3512 item = btrfs_item_nr(leaf, i);
3513 if (!leaf->map_token) {
3514 map_extent_buffer(leaf, (unsigned long)item,
3515 sizeof(struct btrfs_item),
3516 &leaf->map_token, &leaf->kaddr,
3517 &leaf->map_start, &leaf->map_len,
3521 ioff = btrfs_item_offset(leaf, item);
3522 btrfs_set_item_offset(leaf, item, ioff - total_data);
3524 if (leaf->map_token) {
3525 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3526 leaf->map_token = NULL;
3529 /* shift the items */
3530 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3531 btrfs_item_nr_offset(slot),
3532 (nritems - slot) * sizeof(struct btrfs_item));
3534 /* shift the data */
3535 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3536 data_end - total_data, btrfs_leaf_data(leaf) +
3537 data_end, old_data - data_end);
3538 data_end = old_data;
3541 /* setup the item for the new data */
3542 for (i = 0; i < nr; i++) {
3543 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3544 btrfs_set_item_key(leaf, &disk_key, slot + i);
3545 item = btrfs_item_nr(leaf, slot + i);
3546 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3547 data_end -= data_size[i];
3548 btrfs_set_item_size(leaf, item, data_size[i]);
3550 btrfs_set_header_nritems(leaf, nritems + nr);
3551 btrfs_mark_buffer_dirty(leaf);
3555 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3556 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3559 if (btrfs_leaf_free_space(root, leaf) < 0) {
3560 btrfs_print_leaf(root, leaf);
3564 btrfs_unlock_up_safe(path, 1);
3569 * Given a key and some data, insert an item into the tree.
3570 * This does all the path init required, making room in the tree if needed.
3572 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3573 *root, struct btrfs_key *cpu_key, void *data, u32
3577 struct btrfs_path *path;
3578 struct extent_buffer *leaf;
3581 path = btrfs_alloc_path();
3583 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3585 leaf = path->nodes[0];
3586 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3587 write_extent_buffer(leaf, data, ptr, data_size);
3588 btrfs_mark_buffer_dirty(leaf);
3590 btrfs_free_path(path);
3595 * delete the pointer from a given node.
3597 * the tree should have been previously balanced so the deletion does not
3600 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3601 struct btrfs_path *path, int level, int slot)
3603 struct extent_buffer *parent = path->nodes[level];
3608 nritems = btrfs_header_nritems(parent);
3609 if (slot != nritems - 1) {
3610 memmove_extent_buffer(parent,
3611 btrfs_node_key_ptr_offset(slot),
3612 btrfs_node_key_ptr_offset(slot + 1),
3613 sizeof(struct btrfs_key_ptr) *
3614 (nritems - slot - 1));
3617 btrfs_set_header_nritems(parent, nritems);
3618 if (nritems == 0 && parent == root->node) {
3619 BUG_ON(btrfs_header_level(root->node) != 1);
3620 /* just turn the root into a leaf and break */
3621 btrfs_set_header_level(root->node, 0);
3622 } else if (slot == 0) {
3623 struct btrfs_disk_key disk_key;
3625 btrfs_node_key(parent, &disk_key, 0);
3626 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3630 btrfs_mark_buffer_dirty(parent);
3635 * a helper function to delete the leaf pointed to by path->slots[1] and
3636 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3637 * already know it, it is faster to have them pass it down than to
3638 * read it out of the node again.
3640 * This deletes the pointer in path->nodes[1] and frees the leaf
3641 * block extent. zero is returned if it all worked out, < 0 otherwise.
3643 * The path must have already been setup for deleting the leaf, including
3644 * all the proper balancing. path->nodes[1] must be locked.
3646 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3647 struct btrfs_root *root,
3648 struct btrfs_path *path, u64 bytenr)
3651 u64 root_gen = btrfs_header_generation(path->nodes[1]);
3652 u64 parent_start = path->nodes[1]->start;
3653 u64 parent_owner = btrfs_header_owner(path->nodes[1]);
3655 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3660 * btrfs_free_extent is expensive, we want to make sure we
3661 * aren't holding any locks when we call it
3663 btrfs_unlock_up_safe(path, 0);
3665 ret = btrfs_free_extent(trans, root, bytenr,
3666 btrfs_level_size(root, 0),
3667 parent_start, parent_owner,
3672 * delete the item at the leaf level in path. If that empties
3673 * the leaf, remove it from the tree
3675 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3676 struct btrfs_path *path, int slot, int nr)
3678 struct extent_buffer *leaf;
3679 struct btrfs_item *item;
3687 leaf = path->nodes[0];
3688 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3690 for (i = 0; i < nr; i++)
3691 dsize += btrfs_item_size_nr(leaf, slot + i);
3693 nritems = btrfs_header_nritems(leaf);
3695 if (slot + nr != nritems) {
3696 int data_end = leaf_data_end(root, leaf);
3698 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3700 btrfs_leaf_data(leaf) + data_end,
3701 last_off - data_end);
3703 for (i = slot + nr; i < nritems; i++) {
3706 item = btrfs_item_nr(leaf, i);
3707 if (!leaf->map_token) {
3708 map_extent_buffer(leaf, (unsigned long)item,
3709 sizeof(struct btrfs_item),
3710 &leaf->map_token, &leaf->kaddr,
3711 &leaf->map_start, &leaf->map_len,
3714 ioff = btrfs_item_offset(leaf, item);
3715 btrfs_set_item_offset(leaf, item, ioff + dsize);
3718 if (leaf->map_token) {
3719 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3720 leaf->map_token = NULL;
3723 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3724 btrfs_item_nr_offset(slot + nr),
3725 sizeof(struct btrfs_item) *
3726 (nritems - slot - nr));
3728 btrfs_set_header_nritems(leaf, nritems - nr);
3731 /* delete the leaf if we've emptied it */
3733 if (leaf == root->node) {
3734 btrfs_set_header_level(leaf, 0);
3736 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3740 int used = leaf_space_used(leaf, 0, nritems);
3742 struct btrfs_disk_key disk_key;
3744 btrfs_item_key(leaf, &disk_key, 0);
3745 wret = fixup_low_keys(trans, root, path,
3751 /* delete the leaf if it is mostly empty */
3752 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3753 /* push_leaf_left fixes the path.
3754 * make sure the path still points to our leaf
3755 * for possible call to del_ptr below
3757 slot = path->slots[1];
3758 extent_buffer_get(leaf);
3760 wret = push_leaf_left(trans, root, path, 1, 1);
3761 if (wret < 0 && wret != -ENOSPC)
3764 if (path->nodes[0] == leaf &&
3765 btrfs_header_nritems(leaf)) {
3766 wret = push_leaf_right(trans, root, path, 1, 1);
3767 if (wret < 0 && wret != -ENOSPC)
3771 if (btrfs_header_nritems(leaf) == 0) {
3772 path->slots[1] = slot;
3773 ret = btrfs_del_leaf(trans, root, path,
3776 free_extent_buffer(leaf);
3778 /* if we're still in the path, make sure
3779 * we're dirty. Otherwise, one of the
3780 * push_leaf functions must have already
3781 * dirtied this buffer
3783 if (path->nodes[0] == leaf)
3784 btrfs_mark_buffer_dirty(leaf);
3785 free_extent_buffer(leaf);
3788 btrfs_mark_buffer_dirty(leaf);
3795 * search the tree again to find a leaf with lesser keys
3796 * returns 0 if it found something or 1 if there are no lesser leaves.
3797 * returns < 0 on io errors.
3799 * This may release the path, and so you may lose any locks held at the
3802 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3804 struct btrfs_key key;
3805 struct btrfs_disk_key found_key;
3808 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3812 else if (key.type > 0)
3814 else if (key.objectid > 0)
3819 btrfs_release_path(root, path);
3820 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3823 btrfs_item_key(path->nodes[0], &found_key, 0);
3824 ret = comp_keys(&found_key, &key);
3831 * A helper function to walk down the tree starting at min_key, and looking
3832 * for nodes or leaves that are either in cache or have a minimum
3833 * transaction id. This is used by the btree defrag code, and tree logging
3835 * This does not cow, but it does stuff the starting key it finds back
3836 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3837 * key and get a writable path.
3839 * This does lock as it descends, and path->keep_locks should be set
3840 * to 1 by the caller.
3842 * This honors path->lowest_level to prevent descent past a given level
3845 * min_trans indicates the oldest transaction that you are interested
3846 * in walking through. Any nodes or leaves older than min_trans are
3847 * skipped over (without reading them).
3849 * returns zero if something useful was found, < 0 on error and 1 if there
3850 * was nothing in the tree that matched the search criteria.
3852 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3853 struct btrfs_key *max_key,
3854 struct btrfs_path *path, int cache_only,
3857 struct extent_buffer *cur;
3858 struct btrfs_key found_key;
3865 WARN_ON(!path->keep_locks);
3867 cur = btrfs_lock_root_node(root);
3868 level = btrfs_header_level(cur);
3869 WARN_ON(path->nodes[level]);
3870 path->nodes[level] = cur;
3871 path->locks[level] = 1;
3873 if (btrfs_header_generation(cur) < min_trans) {
3878 nritems = btrfs_header_nritems(cur);
3879 level = btrfs_header_level(cur);
3880 sret = bin_search(cur, min_key, level, &slot);
3882 /* at the lowest level, we're done, setup the path and exit */
3883 if (level == path->lowest_level) {
3884 if (slot >= nritems)
3887 path->slots[level] = slot;
3888 btrfs_item_key_to_cpu(cur, &found_key, slot);
3891 if (sret && slot > 0)
3894 * check this node pointer against the cache_only and
3895 * min_trans parameters. If it isn't in cache or is too
3896 * old, skip to the next one.
3898 while (slot < nritems) {
3901 struct extent_buffer *tmp;
3902 struct btrfs_disk_key disk_key;
3904 blockptr = btrfs_node_blockptr(cur, slot);
3905 gen = btrfs_node_ptr_generation(cur, slot);
3906 if (gen < min_trans) {
3914 btrfs_node_key(cur, &disk_key, slot);
3915 if (comp_keys(&disk_key, max_key) >= 0) {
3921 tmp = btrfs_find_tree_block(root, blockptr,
3922 btrfs_level_size(root, level - 1));
3924 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3925 free_extent_buffer(tmp);
3929 free_extent_buffer(tmp);
3934 * we didn't find a candidate key in this node, walk forward
3935 * and find another one
3937 if (slot >= nritems) {
3938 path->slots[level] = slot;
3939 btrfs_set_path_blocking(path);
3940 sret = btrfs_find_next_key(root, path, min_key, level,
3941 cache_only, min_trans);
3943 btrfs_release_path(root, path);
3949 /* save our key for returning back */
3950 btrfs_node_key_to_cpu(cur, &found_key, slot);
3951 path->slots[level] = slot;
3952 if (level == path->lowest_level) {
3954 unlock_up(path, level, 1);
3957 btrfs_set_path_blocking(path);
3958 cur = read_node_slot(root, cur, slot);
3960 btrfs_tree_lock(cur);
3962 path->locks[level - 1] = 1;
3963 path->nodes[level - 1] = cur;
3964 unlock_up(path, level, 1);
3965 btrfs_clear_path_blocking(path, NULL);
3969 memcpy(min_key, &found_key, sizeof(found_key));
3970 btrfs_set_path_blocking(path);
3975 * this is similar to btrfs_next_leaf, but does not try to preserve
3976 * and fixup the path. It looks for and returns the next key in the
3977 * tree based on the current path and the cache_only and min_trans
3980 * 0 is returned if another key is found, < 0 if there are any errors
3981 * and 1 is returned if there are no higher keys in the tree
3983 * path->keep_locks should be set to 1 on the search made before
3984 * calling this function.
3986 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3987 struct btrfs_key *key, int lowest_level,
3988 int cache_only, u64 min_trans)
3990 int level = lowest_level;
3992 struct extent_buffer *c;
3994 WARN_ON(!path->keep_locks);
3995 while (level < BTRFS_MAX_LEVEL) {
3996 if (!path->nodes[level])
3999 slot = path->slots[level] + 1;
4000 c = path->nodes[level];
4002 if (slot >= btrfs_header_nritems(c)) {
4004 if (level == BTRFS_MAX_LEVEL)
4009 btrfs_item_key_to_cpu(c, key, slot);
4011 u64 blockptr = btrfs_node_blockptr(c, slot);
4012 u64 gen = btrfs_node_ptr_generation(c, slot);
4015 struct extent_buffer *cur;
4016 cur = btrfs_find_tree_block(root, blockptr,
4017 btrfs_level_size(root, level - 1));
4018 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
4021 free_extent_buffer(cur);
4024 free_extent_buffer(cur);
4026 if (gen < min_trans) {
4030 btrfs_node_key_to_cpu(c, key, slot);
4038 * search the tree again to find a leaf with greater keys
4039 * returns 0 if it found something or 1 if there are no greater leaves.
4040 * returns < 0 on io errors.
4042 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
4046 struct extent_buffer *c;
4047 struct extent_buffer *next = NULL;
4048 struct btrfs_key key;
4052 nritems = btrfs_header_nritems(path->nodes[0]);
4056 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
4058 btrfs_release_path(root, path);
4059 path->keep_locks = 1;
4060 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4061 path->keep_locks = 0;
4066 btrfs_set_path_blocking(path);
4067 nritems = btrfs_header_nritems(path->nodes[0]);
4069 * by releasing the path above we dropped all our locks. A balance
4070 * could have added more items next to the key that used to be
4071 * at the very end of the block. So, check again here and
4072 * advance the path if there are now more items available.
4074 if (nritems > 0 && path->slots[0] < nritems - 1) {
4079 while (level < BTRFS_MAX_LEVEL) {
4080 if (!path->nodes[level])
4083 slot = path->slots[level] + 1;
4084 c = path->nodes[level];
4085 if (slot >= btrfs_header_nritems(c)) {
4087 if (level == BTRFS_MAX_LEVEL)
4093 btrfs_tree_unlock(next);
4094 free_extent_buffer(next);
4097 /* the path was set to blocking above */
4098 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4100 reada_for_search(root, path, level, slot, 0);
4102 next = read_node_slot(root, c, slot);
4103 if (!path->skip_locking) {
4104 btrfs_assert_tree_locked(c);
4105 btrfs_tree_lock(next);
4106 btrfs_set_lock_blocking(next);
4110 path->slots[level] = slot;
4113 c = path->nodes[level];
4114 if (path->locks[level])
4115 btrfs_tree_unlock(c);
4116 free_extent_buffer(c);
4117 path->nodes[level] = next;
4118 path->slots[level] = 0;
4119 if (!path->skip_locking)
4120 path->locks[level] = 1;
4124 btrfs_set_path_blocking(path);
4125 if (level == 1 && path->locks[1] && path->reada)
4126 reada_for_search(root, path, level, slot, 0);
4127 next = read_node_slot(root, next, 0);
4128 if (!path->skip_locking) {
4129 btrfs_assert_tree_locked(path->nodes[level]);
4130 btrfs_tree_lock(next);
4131 btrfs_set_lock_blocking(next);
4135 unlock_up(path, 0, 1);
4140 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4141 * searching until it gets past min_objectid or finds an item of 'type'
4143 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4145 int btrfs_previous_item(struct btrfs_root *root,
4146 struct btrfs_path *path, u64 min_objectid,
4149 struct btrfs_key found_key;
4150 struct extent_buffer *leaf;
4155 if (path->slots[0] == 0) {
4156 btrfs_set_path_blocking(path);
4157 ret = btrfs_prev_leaf(root, path);
4163 leaf = path->nodes[0];
4164 nritems = btrfs_header_nritems(leaf);
4167 if (path->slots[0] == nritems)
4170 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4171 if (found_key.type == type)
4173 if (found_key.objectid < min_objectid)
4175 if (found_key.objectid == min_objectid &&
4176 found_key.type < type)