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 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
259 struct btrfs_root *root,
260 struct extent_buffer *buf,
261 struct extent_buffer *parent, int parent_slot,
262 struct extent_buffer **cow_ret,
263 u64 search_start, u64 empty_size)
266 struct extent_buffer *cow;
275 btrfs_assert_tree_locked(buf);
278 parent_start = parent->start;
282 WARN_ON(root->ref_cows && trans->transid !=
283 root->fs_info->running_transaction->transid);
284 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
286 level = btrfs_header_level(buf);
287 nritems = btrfs_header_nritems(buf);
289 cow = btrfs_alloc_free_block(trans, root, buf->len,
290 parent_start, root->root_key.objectid,
291 trans->transid, level,
292 search_start, empty_size);
296 /* cow is set to blocking by btrfs_init_new_buffer */
298 copy_extent_buffer(cow, buf, 0, 0, cow->len);
299 btrfs_set_header_bytenr(cow, cow->start);
300 btrfs_set_header_generation(cow, trans->transid);
301 btrfs_set_header_owner(cow, root->root_key.objectid);
302 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN);
304 write_extent_buffer(cow, root->fs_info->fsid,
305 (unsigned long)btrfs_header_fsid(cow),
308 WARN_ON(btrfs_header_generation(buf) > trans->transid);
309 if (btrfs_header_generation(buf) != trans->transid) {
311 ret = btrfs_inc_ref(trans, root, buf, cow, &nr_extents);
315 ret = btrfs_cache_ref(trans, root, buf, nr_extents);
317 } else if (btrfs_header_owner(buf) == BTRFS_TREE_RELOC_OBJECTID) {
319 * There are only two places that can drop reference to
320 * tree blocks owned by living reloc trees, one is here,
321 * the other place is btrfs_drop_subtree. In both places,
322 * we check reference count while tree block is locked.
323 * Furthermore, if reference count is one, it won't get
324 * increased by someone else.
327 ret = btrfs_lookup_extent_ref(trans, root, buf->start,
331 ret = btrfs_update_ref(trans, root, buf, cow,
333 clean_tree_block(trans, root, buf);
335 ret = btrfs_inc_ref(trans, root, buf, cow, NULL);
339 ret = btrfs_update_ref(trans, root, buf, cow, 0, nritems);
342 clean_tree_block(trans, root, buf);
345 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
346 ret = btrfs_reloc_tree_cache_ref(trans, root, cow, buf->start);
350 if (buf == root->node) {
351 WARN_ON(parent && parent != buf);
353 spin_lock(&root->node_lock);
355 extent_buffer_get(cow);
356 spin_unlock(&root->node_lock);
358 if (buf != root->commit_root) {
359 btrfs_free_extent(trans, root, buf->start,
360 buf->len, buf->start,
361 root->root_key.objectid,
362 btrfs_header_generation(buf),
365 free_extent_buffer(buf);
366 add_root_to_dirty_list(root);
368 btrfs_set_node_blockptr(parent, parent_slot,
370 WARN_ON(trans->transid == 0);
371 btrfs_set_node_ptr_generation(parent, parent_slot,
373 btrfs_mark_buffer_dirty(parent);
374 WARN_ON(btrfs_header_generation(parent) != trans->transid);
375 btrfs_free_extent(trans, root, buf->start, buf->len,
376 parent_start, btrfs_header_owner(parent),
377 btrfs_header_generation(parent), level, 1);
380 btrfs_tree_unlock(buf);
381 free_extent_buffer(buf);
382 btrfs_mark_buffer_dirty(cow);
388 * cows a single block, see __btrfs_cow_block for the real work.
389 * This version of it has extra checks so that a block isn't cow'd more than
390 * once per transaction, as long as it hasn't been written yet
392 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
393 struct btrfs_root *root, struct extent_buffer *buf,
394 struct extent_buffer *parent, int parent_slot,
395 struct extent_buffer **cow_ret)
400 if (trans->transaction != root->fs_info->running_transaction) {
401 printk(KERN_CRIT "trans %llu running %llu\n",
402 (unsigned long long)trans->transid,
404 root->fs_info->running_transaction->transid);
407 if (trans->transid != root->fs_info->generation) {
408 printk(KERN_CRIT "trans %llu running %llu\n",
409 (unsigned long long)trans->transid,
410 (unsigned long long)root->fs_info->generation);
414 if (btrfs_header_generation(buf) == trans->transid &&
415 btrfs_header_owner(buf) == root->root_key.objectid &&
416 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN)) {
421 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
424 btrfs_set_lock_blocking(parent);
425 btrfs_set_lock_blocking(buf);
427 ret = __btrfs_cow_block(trans, root, buf, parent,
428 parent_slot, cow_ret, search_start, 0);
433 * helper function for defrag to decide if two blocks pointed to by a
434 * node are actually close by
436 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
438 if (blocknr < other && other - (blocknr + blocksize) < 32768)
440 if (blocknr > other && blocknr - (other + blocksize) < 32768)
446 * compare two keys in a memcmp fashion
448 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
452 btrfs_disk_key_to_cpu(&k1, disk);
454 if (k1.objectid > k2->objectid)
456 if (k1.objectid < k2->objectid)
458 if (k1.type > k2->type)
460 if (k1.type < k2->type)
462 if (k1.offset > k2->offset)
464 if (k1.offset < k2->offset)
470 * same as comp_keys only with two btrfs_key's
472 static int comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
474 if (k1->objectid > k2->objectid)
476 if (k1->objectid < k2->objectid)
478 if (k1->type > k2->type)
480 if (k1->type < k2->type)
482 if (k1->offset > k2->offset)
484 if (k1->offset < k2->offset)
490 * this is used by the defrag code to go through all the
491 * leaves pointed to by a node and reallocate them so that
492 * disk order is close to key order
494 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
495 struct btrfs_root *root, struct extent_buffer *parent,
496 int start_slot, int cache_only, u64 *last_ret,
497 struct btrfs_key *progress)
499 struct extent_buffer *cur;
502 u64 search_start = *last_ret;
512 int progress_passed = 0;
513 struct btrfs_disk_key disk_key;
515 parent_level = btrfs_header_level(parent);
516 if (cache_only && parent_level != 1)
519 if (trans->transaction != root->fs_info->running_transaction)
521 if (trans->transid != root->fs_info->generation)
524 parent_nritems = btrfs_header_nritems(parent);
525 blocksize = btrfs_level_size(root, parent_level - 1);
526 end_slot = parent_nritems;
528 if (parent_nritems == 1)
531 btrfs_set_lock_blocking(parent);
533 for (i = start_slot; i < end_slot; i++) {
536 if (!parent->map_token) {
537 map_extent_buffer(parent,
538 btrfs_node_key_ptr_offset(i),
539 sizeof(struct btrfs_key_ptr),
540 &parent->map_token, &parent->kaddr,
541 &parent->map_start, &parent->map_len,
544 btrfs_node_key(parent, &disk_key, i);
545 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
549 blocknr = btrfs_node_blockptr(parent, i);
550 gen = btrfs_node_ptr_generation(parent, i);
552 last_block = blocknr;
555 other = btrfs_node_blockptr(parent, i - 1);
556 close = close_blocks(blocknr, other, blocksize);
558 if (!close && i < end_slot - 2) {
559 other = btrfs_node_blockptr(parent, i + 1);
560 close = close_blocks(blocknr, other, blocksize);
563 last_block = blocknr;
566 if (parent->map_token) {
567 unmap_extent_buffer(parent, parent->map_token,
569 parent->map_token = NULL;
572 cur = btrfs_find_tree_block(root, blocknr, blocksize);
574 uptodate = btrfs_buffer_uptodate(cur, gen);
577 if (!cur || !uptodate) {
579 free_extent_buffer(cur);
583 cur = read_tree_block(root, blocknr,
585 } else if (!uptodate) {
586 btrfs_read_buffer(cur, gen);
589 if (search_start == 0)
590 search_start = last_block;
592 btrfs_tree_lock(cur);
593 btrfs_set_lock_blocking(cur);
594 err = __btrfs_cow_block(trans, root, cur, parent, i,
597 (end_slot - i) * blocksize));
599 btrfs_tree_unlock(cur);
600 free_extent_buffer(cur);
603 search_start = cur->start;
604 last_block = cur->start;
605 *last_ret = search_start;
606 btrfs_tree_unlock(cur);
607 free_extent_buffer(cur);
609 if (parent->map_token) {
610 unmap_extent_buffer(parent, parent->map_token,
612 parent->map_token = NULL;
618 * The leaf data grows from end-to-front in the node.
619 * this returns the address of the start of the last item,
620 * which is the stop of the leaf data stack
622 static inline unsigned int leaf_data_end(struct btrfs_root *root,
623 struct extent_buffer *leaf)
625 u32 nr = btrfs_header_nritems(leaf);
627 return BTRFS_LEAF_DATA_SIZE(root);
628 return btrfs_item_offset_nr(leaf, nr - 1);
632 * extra debugging checks to make sure all the items in a key are
633 * well formed and in the proper order
635 static int check_node(struct btrfs_root *root, struct btrfs_path *path,
638 struct extent_buffer *parent = NULL;
639 struct extent_buffer *node = path->nodes[level];
640 struct btrfs_disk_key parent_key;
641 struct btrfs_disk_key node_key;
644 struct btrfs_key cpukey;
645 u32 nritems = btrfs_header_nritems(node);
647 if (path->nodes[level + 1])
648 parent = path->nodes[level + 1];
650 slot = path->slots[level];
651 BUG_ON(nritems == 0);
653 parent_slot = path->slots[level + 1];
654 btrfs_node_key(parent, &parent_key, parent_slot);
655 btrfs_node_key(node, &node_key, 0);
656 BUG_ON(memcmp(&parent_key, &node_key,
657 sizeof(struct btrfs_disk_key)));
658 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
659 btrfs_header_bytenr(node));
661 BUG_ON(nritems > BTRFS_NODEPTRS_PER_BLOCK(root));
663 btrfs_node_key_to_cpu(node, &cpukey, slot - 1);
664 btrfs_node_key(node, &node_key, slot);
665 BUG_ON(comp_keys(&node_key, &cpukey) <= 0);
667 if (slot < nritems - 1) {
668 btrfs_node_key_to_cpu(node, &cpukey, slot + 1);
669 btrfs_node_key(node, &node_key, slot);
670 BUG_ON(comp_keys(&node_key, &cpukey) >= 0);
676 * extra checking to make sure all the items in a leaf are
677 * well formed and in the proper order
679 static int check_leaf(struct btrfs_root *root, struct btrfs_path *path,
682 struct extent_buffer *leaf = path->nodes[level];
683 struct extent_buffer *parent = NULL;
685 struct btrfs_key cpukey;
686 struct btrfs_disk_key parent_key;
687 struct btrfs_disk_key leaf_key;
688 int slot = path->slots[0];
690 u32 nritems = btrfs_header_nritems(leaf);
692 if (path->nodes[level + 1])
693 parent = path->nodes[level + 1];
699 parent_slot = path->slots[level + 1];
700 btrfs_node_key(parent, &parent_key, parent_slot);
701 btrfs_item_key(leaf, &leaf_key, 0);
703 BUG_ON(memcmp(&parent_key, &leaf_key,
704 sizeof(struct btrfs_disk_key)));
705 BUG_ON(btrfs_node_blockptr(parent, parent_slot) !=
706 btrfs_header_bytenr(leaf));
708 if (slot != 0 && slot < nritems - 1) {
709 btrfs_item_key(leaf, &leaf_key, slot);
710 btrfs_item_key_to_cpu(leaf, &cpukey, slot - 1);
711 if (comp_keys(&leaf_key, &cpukey) <= 0) {
712 btrfs_print_leaf(root, leaf);
713 printk(KERN_CRIT "slot %d offset bad key\n", slot);
716 if (btrfs_item_offset_nr(leaf, slot - 1) !=
717 btrfs_item_end_nr(leaf, slot)) {
718 btrfs_print_leaf(root, leaf);
719 printk(KERN_CRIT "slot %d offset bad\n", slot);
723 if (slot < nritems - 1) {
724 btrfs_item_key(leaf, &leaf_key, slot);
725 btrfs_item_key_to_cpu(leaf, &cpukey, slot + 1);
726 BUG_ON(comp_keys(&leaf_key, &cpukey) >= 0);
727 if (btrfs_item_offset_nr(leaf, slot) !=
728 btrfs_item_end_nr(leaf, slot + 1)) {
729 btrfs_print_leaf(root, leaf);
730 printk(KERN_CRIT "slot %d offset bad\n", slot);
734 BUG_ON(btrfs_item_offset_nr(leaf, 0) +
735 btrfs_item_size_nr(leaf, 0) != BTRFS_LEAF_DATA_SIZE(root));
739 static noinline int check_block(struct btrfs_root *root,
740 struct btrfs_path *path, int level)
744 return check_leaf(root, path, level);
745 return check_node(root, path, level);
749 * search for key in the extent_buffer. The items start at offset p,
750 * and they are item_size apart. There are 'max' items in p.
752 * the slot in the array is returned via slot, and it points to
753 * the place where you would insert key if it is not found in
756 * slot may point to max if the key is bigger than all of the keys
758 static noinline int generic_bin_search(struct extent_buffer *eb,
760 int item_size, struct btrfs_key *key,
767 struct btrfs_disk_key *tmp = NULL;
768 struct btrfs_disk_key unaligned;
769 unsigned long offset;
770 char *map_token = NULL;
772 unsigned long map_start = 0;
773 unsigned long map_len = 0;
777 mid = (low + high) / 2;
778 offset = p + mid * item_size;
780 if (!map_token || offset < map_start ||
781 (offset + sizeof(struct btrfs_disk_key)) >
782 map_start + map_len) {
784 unmap_extent_buffer(eb, map_token, KM_USER0);
788 err = map_private_extent_buffer(eb, offset,
789 sizeof(struct btrfs_disk_key),
791 &map_start, &map_len, KM_USER0);
794 tmp = (struct btrfs_disk_key *)(kaddr + offset -
797 read_extent_buffer(eb, &unaligned,
798 offset, sizeof(unaligned));
803 tmp = (struct btrfs_disk_key *)(kaddr + offset -
806 ret = comp_keys(tmp, key);
815 unmap_extent_buffer(eb, map_token, KM_USER0);
821 unmap_extent_buffer(eb, map_token, KM_USER0);
826 * simple bin_search frontend that does the right thing for
829 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
830 int level, int *slot)
833 return generic_bin_search(eb,
834 offsetof(struct btrfs_leaf, items),
835 sizeof(struct btrfs_item),
836 key, btrfs_header_nritems(eb),
839 return generic_bin_search(eb,
840 offsetof(struct btrfs_node, ptrs),
841 sizeof(struct btrfs_key_ptr),
842 key, btrfs_header_nritems(eb),
848 /* given a node and slot number, this reads the blocks it points to. The
849 * extent buffer is returned with a reference taken (but unlocked).
850 * NULL is returned on error.
852 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
853 struct extent_buffer *parent, int slot)
855 int level = btrfs_header_level(parent);
858 if (slot >= btrfs_header_nritems(parent))
863 return read_tree_block(root, btrfs_node_blockptr(parent, slot),
864 btrfs_level_size(root, level - 1),
865 btrfs_node_ptr_generation(parent, slot));
869 * node level balancing, used to make sure nodes are in proper order for
870 * item deletion. We balance from the top down, so we have to make sure
871 * that a deletion won't leave an node completely empty later on.
873 static noinline int balance_level(struct btrfs_trans_handle *trans,
874 struct btrfs_root *root,
875 struct btrfs_path *path, int level)
877 struct extent_buffer *right = NULL;
878 struct extent_buffer *mid;
879 struct extent_buffer *left = NULL;
880 struct extent_buffer *parent = NULL;
884 int orig_slot = path->slots[level];
885 int err_on_enospc = 0;
891 mid = path->nodes[level];
893 WARN_ON(!path->locks[level]);
894 WARN_ON(btrfs_header_generation(mid) != trans->transid);
896 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
898 if (level < BTRFS_MAX_LEVEL - 1)
899 parent = path->nodes[level + 1];
900 pslot = path->slots[level + 1];
903 * deal with the case where there is only one pointer in the root
904 * by promoting the node below to a root
907 struct extent_buffer *child;
909 if (btrfs_header_nritems(mid) != 1)
912 /* promote the child to a root */
913 child = read_node_slot(root, mid, 0);
915 btrfs_tree_lock(child);
916 btrfs_set_lock_blocking(child);
917 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
920 spin_lock(&root->node_lock);
922 spin_unlock(&root->node_lock);
924 ret = btrfs_update_extent_ref(trans, root, child->start,
925 mid->start, child->start,
926 root->root_key.objectid,
927 trans->transid, level - 1);
930 add_root_to_dirty_list(root);
931 btrfs_tree_unlock(child);
933 path->locks[level] = 0;
934 path->nodes[level] = NULL;
935 clean_tree_block(trans, root, mid);
936 btrfs_tree_unlock(mid);
937 /* once for the path */
938 free_extent_buffer(mid);
939 ret = btrfs_free_extent(trans, root, mid->start, mid->len,
940 mid->start, root->root_key.objectid,
941 btrfs_header_generation(mid),
943 /* once for the root ptr */
944 free_extent_buffer(mid);
947 if (btrfs_header_nritems(mid) >
948 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
951 if (btrfs_header_nritems(mid) < 2)
954 left = read_node_slot(root, parent, pslot - 1);
956 btrfs_tree_lock(left);
957 btrfs_set_lock_blocking(left);
958 wret = btrfs_cow_block(trans, root, left,
959 parent, pslot - 1, &left);
965 right = read_node_slot(root, parent, pslot + 1);
967 btrfs_tree_lock(right);
968 btrfs_set_lock_blocking(right);
969 wret = btrfs_cow_block(trans, root, right,
970 parent, pslot + 1, &right);
977 /* first, try to make some room in the middle buffer */
979 orig_slot += btrfs_header_nritems(left);
980 wret = push_node_left(trans, root, left, mid, 1);
983 if (btrfs_header_nritems(mid) < 2)
988 * then try to empty the right most buffer into the middle
991 wret = push_node_left(trans, root, mid, right, 1);
992 if (wret < 0 && wret != -ENOSPC)
994 if (btrfs_header_nritems(right) == 0) {
995 u64 bytenr = right->start;
996 u64 generation = btrfs_header_generation(parent);
997 u32 blocksize = right->len;
999 clean_tree_block(trans, root, right);
1000 btrfs_tree_unlock(right);
1001 free_extent_buffer(right);
1003 wret = del_ptr(trans, root, path, level + 1, pslot +
1007 wret = btrfs_free_extent(trans, root, bytenr,
1008 blocksize, parent->start,
1009 btrfs_header_owner(parent),
1010 generation, level, 1);
1014 struct btrfs_disk_key right_key;
1015 btrfs_node_key(right, &right_key, 0);
1016 btrfs_set_node_key(parent, &right_key, pslot + 1);
1017 btrfs_mark_buffer_dirty(parent);
1020 if (btrfs_header_nritems(mid) == 1) {
1022 * we're not allowed to leave a node with one item in the
1023 * tree during a delete. A deletion from lower in the tree
1024 * could try to delete the only pointer in this node.
1025 * So, pull some keys from the left.
1026 * There has to be a left pointer at this point because
1027 * otherwise we would have pulled some pointers from the
1031 wret = balance_node_right(trans, root, mid, left);
1037 wret = push_node_left(trans, root, left, mid, 1);
1043 if (btrfs_header_nritems(mid) == 0) {
1044 /* we've managed to empty the middle node, drop it */
1045 u64 root_gen = btrfs_header_generation(parent);
1046 u64 bytenr = mid->start;
1047 u32 blocksize = mid->len;
1049 clean_tree_block(trans, root, mid);
1050 btrfs_tree_unlock(mid);
1051 free_extent_buffer(mid);
1053 wret = del_ptr(trans, root, path, level + 1, pslot);
1056 wret = btrfs_free_extent(trans, root, bytenr, blocksize,
1058 btrfs_header_owner(parent),
1059 root_gen, level, 1);
1063 /* update the parent key to reflect our changes */
1064 struct btrfs_disk_key mid_key;
1065 btrfs_node_key(mid, &mid_key, 0);
1066 btrfs_set_node_key(parent, &mid_key, pslot);
1067 btrfs_mark_buffer_dirty(parent);
1070 /* update the path */
1072 if (btrfs_header_nritems(left) > orig_slot) {
1073 extent_buffer_get(left);
1074 /* left was locked after cow */
1075 path->nodes[level] = left;
1076 path->slots[level + 1] -= 1;
1077 path->slots[level] = orig_slot;
1079 btrfs_tree_unlock(mid);
1080 free_extent_buffer(mid);
1083 orig_slot -= btrfs_header_nritems(left);
1084 path->slots[level] = orig_slot;
1087 /* double check we haven't messed things up */
1088 check_block(root, path, level);
1090 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1094 btrfs_tree_unlock(right);
1095 free_extent_buffer(right);
1098 if (path->nodes[level] != left)
1099 btrfs_tree_unlock(left);
1100 free_extent_buffer(left);
1105 /* Node balancing for insertion. Here we only split or push nodes around
1106 * when they are completely full. This is also done top down, so we
1107 * have to be pessimistic.
1109 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1110 struct btrfs_root *root,
1111 struct btrfs_path *path, int level)
1113 struct extent_buffer *right = NULL;
1114 struct extent_buffer *mid;
1115 struct extent_buffer *left = NULL;
1116 struct extent_buffer *parent = NULL;
1120 int orig_slot = path->slots[level];
1126 mid = path->nodes[level];
1127 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1128 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1130 if (level < BTRFS_MAX_LEVEL - 1)
1131 parent = path->nodes[level + 1];
1132 pslot = path->slots[level + 1];
1137 left = read_node_slot(root, parent, pslot - 1);
1139 /* first, try to make some room in the middle buffer */
1143 btrfs_tree_lock(left);
1144 btrfs_set_lock_blocking(left);
1146 left_nr = btrfs_header_nritems(left);
1147 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1150 ret = btrfs_cow_block(trans, root, left, parent,
1155 wret = push_node_left(trans, root,
1162 struct btrfs_disk_key disk_key;
1163 orig_slot += left_nr;
1164 btrfs_node_key(mid, &disk_key, 0);
1165 btrfs_set_node_key(parent, &disk_key, pslot);
1166 btrfs_mark_buffer_dirty(parent);
1167 if (btrfs_header_nritems(left) > orig_slot) {
1168 path->nodes[level] = left;
1169 path->slots[level + 1] -= 1;
1170 path->slots[level] = orig_slot;
1171 btrfs_tree_unlock(mid);
1172 free_extent_buffer(mid);
1175 btrfs_header_nritems(left);
1176 path->slots[level] = orig_slot;
1177 btrfs_tree_unlock(left);
1178 free_extent_buffer(left);
1182 btrfs_tree_unlock(left);
1183 free_extent_buffer(left);
1185 right = read_node_slot(root, parent, pslot + 1);
1188 * then try to empty the right most buffer into the middle
1193 btrfs_tree_lock(right);
1194 btrfs_set_lock_blocking(right);
1196 right_nr = btrfs_header_nritems(right);
1197 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1200 ret = btrfs_cow_block(trans, root, right,
1206 wret = balance_node_right(trans, root,
1213 struct btrfs_disk_key disk_key;
1215 btrfs_node_key(right, &disk_key, 0);
1216 btrfs_set_node_key(parent, &disk_key, pslot + 1);
1217 btrfs_mark_buffer_dirty(parent);
1219 if (btrfs_header_nritems(mid) <= orig_slot) {
1220 path->nodes[level] = right;
1221 path->slots[level + 1] += 1;
1222 path->slots[level] = orig_slot -
1223 btrfs_header_nritems(mid);
1224 btrfs_tree_unlock(mid);
1225 free_extent_buffer(mid);
1227 btrfs_tree_unlock(right);
1228 free_extent_buffer(right);
1232 btrfs_tree_unlock(right);
1233 free_extent_buffer(right);
1239 * readahead one full node of leaves, finding things that are close
1240 * to the block in 'slot', and triggering ra on them.
1242 static noinline void reada_for_search(struct btrfs_root *root,
1243 struct btrfs_path *path,
1244 int level, int slot, u64 objectid)
1246 struct extent_buffer *node;
1247 struct btrfs_disk_key disk_key;
1252 int direction = path->reada;
1253 struct extent_buffer *eb;
1261 if (!path->nodes[level])
1264 node = path->nodes[level];
1266 search = btrfs_node_blockptr(node, slot);
1267 blocksize = btrfs_level_size(root, level - 1);
1268 eb = btrfs_find_tree_block(root, search, blocksize);
1270 free_extent_buffer(eb);
1276 nritems = btrfs_header_nritems(node);
1279 if (direction < 0) {
1283 } else if (direction > 0) {
1288 if (path->reada < 0 && objectid) {
1289 btrfs_node_key(node, &disk_key, nr);
1290 if (btrfs_disk_key_objectid(&disk_key) != objectid)
1293 search = btrfs_node_blockptr(node, nr);
1294 if ((search <= target && target - search <= 65536) ||
1295 (search > target && search - target <= 65536)) {
1296 readahead_tree_block(root, search, blocksize,
1297 btrfs_node_ptr_generation(node, nr));
1301 if ((nread > 65536 || nscan > 32))
1307 * returns -EAGAIN if it had to drop the path, or zero if everything was in
1310 static noinline int reada_for_balance(struct btrfs_root *root,
1311 struct btrfs_path *path, int level)
1315 struct extent_buffer *parent;
1316 struct extent_buffer *eb;
1323 parent = path->nodes[level - 1];
1327 nritems = btrfs_header_nritems(parent);
1328 slot = path->slots[level];
1329 blocksize = btrfs_level_size(root, level);
1332 block1 = btrfs_node_blockptr(parent, slot - 1);
1333 gen = btrfs_node_ptr_generation(parent, slot - 1);
1334 eb = btrfs_find_tree_block(root, block1, blocksize);
1335 if (eb && btrfs_buffer_uptodate(eb, gen))
1337 free_extent_buffer(eb);
1339 if (slot < nritems) {
1340 block2 = btrfs_node_blockptr(parent, slot + 1);
1341 gen = btrfs_node_ptr_generation(parent, slot + 1);
1342 eb = btrfs_find_tree_block(root, block2, blocksize);
1343 if (eb && btrfs_buffer_uptodate(eb, gen))
1345 free_extent_buffer(eb);
1347 if (block1 || block2) {
1349 btrfs_release_path(root, path);
1351 readahead_tree_block(root, block1, blocksize, 0);
1353 readahead_tree_block(root, block2, blocksize, 0);
1356 eb = read_tree_block(root, block1, blocksize, 0);
1357 free_extent_buffer(eb);
1360 eb = read_tree_block(root, block2, blocksize, 0);
1361 free_extent_buffer(eb);
1369 * when we walk down the tree, it is usually safe to unlock the higher layers
1370 * in the tree. The exceptions are when our path goes through slot 0, because
1371 * operations on the tree might require changing key pointers higher up in the
1374 * callers might also have set path->keep_locks, which tells this code to keep
1375 * the lock if the path points to the last slot in the block. This is part of
1376 * walking through the tree, and selecting the next slot in the higher block.
1378 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
1379 * if lowest_unlock is 1, level 0 won't be unlocked
1381 static noinline void unlock_up(struct btrfs_path *path, int level,
1385 int skip_level = level;
1387 struct extent_buffer *t;
1389 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1390 if (!path->nodes[i])
1392 if (!path->locks[i])
1394 if (!no_skips && path->slots[i] == 0) {
1398 if (!no_skips && path->keep_locks) {
1401 nritems = btrfs_header_nritems(t);
1402 if (nritems < 1 || path->slots[i] >= nritems - 1) {
1407 if (skip_level < i && i >= lowest_unlock)
1411 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
1412 btrfs_tree_unlock(t);
1419 * This releases any locks held in the path starting at level and
1420 * going all the way up to the root.
1422 * btrfs_search_slot will keep the lock held on higher nodes in a few
1423 * corner cases, such as COW of the block at slot zero in the node. This
1424 * ignores those rules, and it should only be called when there are no
1425 * more updates to be done higher up in the tree.
1427 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
1431 if (path->keep_locks || path->lowest_level)
1434 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1435 if (!path->nodes[i])
1437 if (!path->locks[i])
1439 btrfs_tree_unlock(path->nodes[i]);
1445 * look for key in the tree. path is filled in with nodes along the way
1446 * if key is found, we return zero and you can find the item in the leaf
1447 * level of the path (level 0)
1449 * If the key isn't found, the path points to the slot where it should
1450 * be inserted, and 1 is returned. If there are other errors during the
1451 * search a negative error number is returned.
1453 * if ins_len > 0, nodes and leaves will be split as we walk down the
1454 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1457 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
1458 *root, struct btrfs_key *key, struct btrfs_path *p, int
1461 struct extent_buffer *b;
1462 struct extent_buffer *tmp;
1466 int should_reada = p->reada;
1467 int lowest_unlock = 1;
1469 u8 lowest_level = 0;
1473 lowest_level = p->lowest_level;
1474 WARN_ON(lowest_level && ins_len > 0);
1475 WARN_ON(p->nodes[0] != NULL);
1481 if (p->skip_locking)
1482 b = btrfs_root_node(root);
1484 b = btrfs_lock_root_node(root);
1487 level = btrfs_header_level(b);
1490 * setup the path here so we can release it under lock
1491 * contention with the cow code
1493 p->nodes[level] = b;
1494 if (!p->skip_locking)
1495 p->locks[level] = 1;
1500 /* is a cow on this block not required */
1501 if (btrfs_header_generation(b) == trans->transid &&
1502 btrfs_header_owner(b) == root->root_key.objectid &&
1503 !btrfs_header_flag(b, BTRFS_HEADER_FLAG_WRITTEN)) {
1506 btrfs_set_path_blocking(p);
1508 wret = btrfs_cow_block(trans, root, b,
1509 p->nodes[level + 1],
1510 p->slots[level + 1], &b);
1512 free_extent_buffer(b);
1518 BUG_ON(!cow && ins_len);
1519 if (level != btrfs_header_level(b))
1521 level = btrfs_header_level(b);
1523 p->nodes[level] = b;
1524 if (!p->skip_locking)
1525 p->locks[level] = 1;
1527 btrfs_clear_path_blocking(p, NULL);
1530 * we have a lock on b and as long as we aren't changing
1531 * the tree, there is no way to for the items in b to change.
1532 * It is safe to drop the lock on our parent before we
1533 * go through the expensive btree search on b.
1535 * If cow is true, then we might be changing slot zero,
1536 * which may require changing the parent. So, we can't
1537 * drop the lock until after we know which slot we're
1541 btrfs_unlock_up_safe(p, level + 1);
1543 ret = check_block(root, p, level);
1549 ret = bin_search(b, key, level, &slot);
1552 if (ret && slot > 0)
1554 p->slots[level] = slot;
1555 if ((p->search_for_split || ins_len > 0) &&
1556 btrfs_header_nritems(b) >=
1557 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
1560 sret = reada_for_balance(root, p, level);
1564 btrfs_set_path_blocking(p);
1565 sret = split_node(trans, root, p, level);
1566 btrfs_clear_path_blocking(p, NULL);
1573 b = p->nodes[level];
1574 slot = p->slots[level];
1575 } else if (ins_len < 0 &&
1576 btrfs_header_nritems(b) <
1577 BTRFS_NODEPTRS_PER_BLOCK(root) / 4) {
1580 sret = reada_for_balance(root, p, level);
1584 btrfs_set_path_blocking(p);
1585 sret = balance_level(trans, root, p, level);
1586 btrfs_clear_path_blocking(p, NULL);
1592 b = p->nodes[level];
1594 btrfs_release_path(NULL, p);
1597 slot = p->slots[level];
1598 BUG_ON(btrfs_header_nritems(b) == 1);
1600 unlock_up(p, level, lowest_unlock);
1602 /* this is only true while dropping a snapshot */
1603 if (level == lowest_level) {
1608 blocknr = btrfs_node_blockptr(b, slot);
1609 gen = btrfs_node_ptr_generation(b, slot);
1610 blocksize = btrfs_level_size(root, level - 1);
1612 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
1613 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
1617 * reduce lock contention at high levels
1618 * of the btree by dropping locks before
1622 btrfs_release_path(NULL, p);
1624 free_extent_buffer(tmp);
1626 reada_for_search(root, p,
1630 tmp = read_tree_block(root, blocknr,
1633 free_extent_buffer(tmp);
1636 btrfs_set_path_blocking(p);
1638 free_extent_buffer(tmp);
1640 reada_for_search(root, p,
1643 b = read_node_slot(root, b, slot);
1646 if (!p->skip_locking) {
1649 btrfs_clear_path_blocking(p, NULL);
1650 lret = btrfs_try_spin_lock(b);
1653 btrfs_set_path_blocking(p);
1655 btrfs_clear_path_blocking(p, b);
1659 p->slots[level] = slot;
1661 btrfs_leaf_free_space(root, b) < ins_len) {
1664 btrfs_set_path_blocking(p);
1665 sret = split_leaf(trans, root, key,
1666 p, ins_len, ret == 0);
1667 btrfs_clear_path_blocking(p, NULL);
1675 if (!p->search_for_split)
1676 unlock_up(p, level, lowest_unlock);
1683 * we don't really know what they plan on doing with the path
1684 * from here on, so for now just mark it as blocking
1686 btrfs_set_path_blocking(p);
1690 int btrfs_merge_path(struct btrfs_trans_handle *trans,
1691 struct btrfs_root *root,
1692 struct btrfs_key *node_keys,
1693 u64 *nodes, int lowest_level)
1695 struct extent_buffer *eb;
1696 struct extent_buffer *parent;
1697 struct btrfs_key key;
1706 eb = btrfs_lock_root_node(root);
1707 ret = btrfs_cow_block(trans, root, eb, NULL, 0, &eb);
1710 btrfs_set_lock_blocking(eb);
1714 level = btrfs_header_level(parent);
1715 if (level == 0 || level <= lowest_level)
1718 ret = bin_search(parent, &node_keys[lowest_level], level,
1720 if (ret && slot > 0)
1723 bytenr = btrfs_node_blockptr(parent, slot);
1724 if (nodes[level - 1] == bytenr)
1727 blocksize = btrfs_level_size(root, level - 1);
1728 generation = btrfs_node_ptr_generation(parent, slot);
1729 btrfs_node_key_to_cpu(eb, &key, slot);
1730 key_match = !memcmp(&key, &node_keys[level - 1], sizeof(key));
1732 if (generation == trans->transid) {
1733 eb = read_tree_block(root, bytenr, blocksize,
1735 btrfs_tree_lock(eb);
1736 btrfs_set_lock_blocking(eb);
1740 * if node keys match and node pointer hasn't been modified
1741 * in the running transaction, we can merge the path. for
1742 * blocks owened by reloc trees, the node pointer check is
1743 * skipped, this is because these blocks are fully controlled
1744 * by the space balance code, no one else can modify them.
1746 if (!nodes[level - 1] || !key_match ||
1747 (generation == trans->transid &&
1748 btrfs_header_owner(eb) != BTRFS_TREE_RELOC_OBJECTID)) {
1749 if (level == 1 || level == lowest_level + 1) {
1750 if (generation == trans->transid) {
1751 btrfs_tree_unlock(eb);
1752 free_extent_buffer(eb);
1757 if (generation != trans->transid) {
1758 eb = read_tree_block(root, bytenr, blocksize,
1760 btrfs_tree_lock(eb);
1761 btrfs_set_lock_blocking(eb);
1764 ret = btrfs_cow_block(trans, root, eb, parent, slot,
1768 if (root->root_key.objectid ==
1769 BTRFS_TREE_RELOC_OBJECTID) {
1770 if (!nodes[level - 1]) {
1771 nodes[level - 1] = eb->start;
1772 memcpy(&node_keys[level - 1], &key,
1773 sizeof(node_keys[0]));
1779 btrfs_tree_unlock(parent);
1780 free_extent_buffer(parent);
1785 btrfs_set_node_blockptr(parent, slot, nodes[level - 1]);
1786 btrfs_set_node_ptr_generation(parent, slot, trans->transid);
1787 btrfs_mark_buffer_dirty(parent);
1789 ret = btrfs_inc_extent_ref(trans, root,
1791 blocksize, parent->start,
1792 btrfs_header_owner(parent),
1793 btrfs_header_generation(parent),
1798 * If the block was created in the running transaction,
1799 * it's possible this is the last reference to it, so we
1800 * should drop the subtree.
1802 if (generation == trans->transid) {
1803 ret = btrfs_drop_subtree(trans, root, eb, parent);
1805 btrfs_tree_unlock(eb);
1806 free_extent_buffer(eb);
1808 ret = btrfs_free_extent(trans, root, bytenr,
1809 blocksize, parent->start,
1810 btrfs_header_owner(parent),
1811 btrfs_header_generation(parent),
1817 btrfs_tree_unlock(parent);
1818 free_extent_buffer(parent);
1823 * adjust the pointers going up the tree, starting at level
1824 * making sure the right key of each node is points to 'key'.
1825 * This is used after shifting pointers to the left, so it stops
1826 * fixing up pointers when a given leaf/node is not in slot 0 of the
1829 * If this fails to write a tree block, it returns -1, but continues
1830 * fixing up the blocks in ram so the tree is consistent.
1832 static int fixup_low_keys(struct btrfs_trans_handle *trans,
1833 struct btrfs_root *root, struct btrfs_path *path,
1834 struct btrfs_disk_key *key, int level)
1838 struct extent_buffer *t;
1840 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
1841 int tslot = path->slots[i];
1842 if (!path->nodes[i])
1845 btrfs_set_node_key(t, key, tslot);
1846 btrfs_mark_buffer_dirty(path->nodes[i]);
1856 * This function isn't completely safe. It's the caller's responsibility
1857 * that the new key won't break the order
1859 int btrfs_set_item_key_safe(struct btrfs_trans_handle *trans,
1860 struct btrfs_root *root, struct btrfs_path *path,
1861 struct btrfs_key *new_key)
1863 struct btrfs_disk_key disk_key;
1864 struct extent_buffer *eb;
1867 eb = path->nodes[0];
1868 slot = path->slots[0];
1870 btrfs_item_key(eb, &disk_key, slot - 1);
1871 if (comp_keys(&disk_key, new_key) >= 0)
1874 if (slot < btrfs_header_nritems(eb) - 1) {
1875 btrfs_item_key(eb, &disk_key, slot + 1);
1876 if (comp_keys(&disk_key, new_key) <= 0)
1880 btrfs_cpu_key_to_disk(&disk_key, new_key);
1881 btrfs_set_item_key(eb, &disk_key, slot);
1882 btrfs_mark_buffer_dirty(eb);
1884 fixup_low_keys(trans, root, path, &disk_key, 1);
1889 * try to push data from one node into the next node left in the
1892 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1893 * error, and > 0 if there was no room in the left hand block.
1895 static int push_node_left(struct btrfs_trans_handle *trans,
1896 struct btrfs_root *root, struct extent_buffer *dst,
1897 struct extent_buffer *src, int empty)
1904 src_nritems = btrfs_header_nritems(src);
1905 dst_nritems = btrfs_header_nritems(dst);
1906 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1907 WARN_ON(btrfs_header_generation(src) != trans->transid);
1908 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1910 if (!empty && src_nritems <= 8)
1913 if (push_items <= 0)
1917 push_items = min(src_nritems, push_items);
1918 if (push_items < src_nritems) {
1919 /* leave at least 8 pointers in the node if
1920 * we aren't going to empty it
1922 if (src_nritems - push_items < 8) {
1923 if (push_items <= 8)
1929 push_items = min(src_nritems - 8, push_items);
1931 copy_extent_buffer(dst, src,
1932 btrfs_node_key_ptr_offset(dst_nritems),
1933 btrfs_node_key_ptr_offset(0),
1934 push_items * sizeof(struct btrfs_key_ptr));
1936 if (push_items < src_nritems) {
1937 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
1938 btrfs_node_key_ptr_offset(push_items),
1939 (src_nritems - push_items) *
1940 sizeof(struct btrfs_key_ptr));
1942 btrfs_set_header_nritems(src, src_nritems - push_items);
1943 btrfs_set_header_nritems(dst, dst_nritems + push_items);
1944 btrfs_mark_buffer_dirty(src);
1945 btrfs_mark_buffer_dirty(dst);
1947 ret = btrfs_update_ref(trans, root, src, dst, dst_nritems, push_items);
1954 * try to push data from one node into the next node right in the
1957 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1958 * error, and > 0 if there was no room in the right hand block.
1960 * this will only push up to 1/2 the contents of the left node over
1962 static int balance_node_right(struct btrfs_trans_handle *trans,
1963 struct btrfs_root *root,
1964 struct extent_buffer *dst,
1965 struct extent_buffer *src)
1973 WARN_ON(btrfs_header_generation(src) != trans->transid);
1974 WARN_ON(btrfs_header_generation(dst) != trans->transid);
1976 src_nritems = btrfs_header_nritems(src);
1977 dst_nritems = btrfs_header_nritems(dst);
1978 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
1979 if (push_items <= 0)
1982 if (src_nritems < 4)
1985 max_push = src_nritems / 2 + 1;
1986 /* don't try to empty the node */
1987 if (max_push >= src_nritems)
1990 if (max_push < push_items)
1991 push_items = max_push;
1993 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
1994 btrfs_node_key_ptr_offset(0),
1996 sizeof(struct btrfs_key_ptr));
1998 copy_extent_buffer(dst, src,
1999 btrfs_node_key_ptr_offset(0),
2000 btrfs_node_key_ptr_offset(src_nritems - push_items),
2001 push_items * sizeof(struct btrfs_key_ptr));
2003 btrfs_set_header_nritems(src, src_nritems - push_items);
2004 btrfs_set_header_nritems(dst, dst_nritems + push_items);
2006 btrfs_mark_buffer_dirty(src);
2007 btrfs_mark_buffer_dirty(dst);
2009 ret = btrfs_update_ref(trans, root, src, dst, 0, push_items);
2016 * helper function to insert a new root level in the tree.
2017 * A new node is allocated, and a single item is inserted to
2018 * point to the existing root
2020 * returns zero on success or < 0 on failure.
2022 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
2023 struct btrfs_root *root,
2024 struct btrfs_path *path, int level)
2027 struct extent_buffer *lower;
2028 struct extent_buffer *c;
2029 struct extent_buffer *old;
2030 struct btrfs_disk_key lower_key;
2033 BUG_ON(path->nodes[level]);
2034 BUG_ON(path->nodes[level-1] != root->node);
2036 lower = path->nodes[level-1];
2038 btrfs_item_key(lower, &lower_key, 0);
2040 btrfs_node_key(lower, &lower_key, 0);
2042 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
2043 root->root_key.objectid, trans->transid,
2044 level, root->node->start, 0);
2048 memset_extent_buffer(c, 0, 0, root->nodesize);
2049 btrfs_set_header_nritems(c, 1);
2050 btrfs_set_header_level(c, level);
2051 btrfs_set_header_bytenr(c, c->start);
2052 btrfs_set_header_generation(c, trans->transid);
2053 btrfs_set_header_owner(c, root->root_key.objectid);
2055 write_extent_buffer(c, root->fs_info->fsid,
2056 (unsigned long)btrfs_header_fsid(c),
2059 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
2060 (unsigned long)btrfs_header_chunk_tree_uuid(c),
2063 btrfs_set_node_key(c, &lower_key, 0);
2064 btrfs_set_node_blockptr(c, 0, lower->start);
2065 lower_gen = btrfs_header_generation(lower);
2066 WARN_ON(lower_gen != trans->transid);
2068 btrfs_set_node_ptr_generation(c, 0, lower_gen);
2070 btrfs_mark_buffer_dirty(c);
2072 spin_lock(&root->node_lock);
2075 spin_unlock(&root->node_lock);
2077 ret = btrfs_update_extent_ref(trans, root, lower->start,
2078 lower->start, c->start,
2079 root->root_key.objectid,
2080 trans->transid, level - 1);
2083 /* the super has an extra ref to root->node */
2084 free_extent_buffer(old);
2086 add_root_to_dirty_list(root);
2087 extent_buffer_get(c);
2088 path->nodes[level] = c;
2089 path->locks[level] = 1;
2090 path->slots[level] = 0;
2095 * worker function to insert a single pointer in a node.
2096 * the node should have enough room for the pointer already
2098 * slot and level indicate where you want the key to go, and
2099 * blocknr is the block the key points to.
2101 * returns zero on success and < 0 on any error
2103 static int insert_ptr(struct btrfs_trans_handle *trans, struct btrfs_root
2104 *root, struct btrfs_path *path, struct btrfs_disk_key
2105 *key, u64 bytenr, int slot, int level)
2107 struct extent_buffer *lower;
2110 BUG_ON(!path->nodes[level]);
2111 lower = path->nodes[level];
2112 nritems = btrfs_header_nritems(lower);
2115 if (nritems == BTRFS_NODEPTRS_PER_BLOCK(root))
2117 if (slot != nritems) {
2118 memmove_extent_buffer(lower,
2119 btrfs_node_key_ptr_offset(slot + 1),
2120 btrfs_node_key_ptr_offset(slot),
2121 (nritems - slot) * sizeof(struct btrfs_key_ptr));
2123 btrfs_set_node_key(lower, key, slot);
2124 btrfs_set_node_blockptr(lower, slot, bytenr);
2125 WARN_ON(trans->transid == 0);
2126 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
2127 btrfs_set_header_nritems(lower, nritems + 1);
2128 btrfs_mark_buffer_dirty(lower);
2133 * split the node at the specified level in path in two.
2134 * The path is corrected to point to the appropriate node after the split
2136 * Before splitting this tries to make some room in the node by pushing
2137 * left and right, if either one works, it returns right away.
2139 * returns 0 on success and < 0 on failure
2141 static noinline int split_node(struct btrfs_trans_handle *trans,
2142 struct btrfs_root *root,
2143 struct btrfs_path *path, int level)
2145 struct extent_buffer *c;
2146 struct extent_buffer *split;
2147 struct btrfs_disk_key disk_key;
2153 c = path->nodes[level];
2154 WARN_ON(btrfs_header_generation(c) != trans->transid);
2155 if (c == root->node) {
2156 /* trying to split the root, lets make a new one */
2157 ret = insert_new_root(trans, root, path, level + 1);
2161 ret = push_nodes_for_insert(trans, root, path, level);
2162 c = path->nodes[level];
2163 if (!ret && btrfs_header_nritems(c) <
2164 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
2170 c_nritems = btrfs_header_nritems(c);
2172 split = btrfs_alloc_free_block(trans, root, root->nodesize,
2173 path->nodes[level + 1]->start,
2174 root->root_key.objectid,
2175 trans->transid, level, c->start, 0);
2177 return PTR_ERR(split);
2179 btrfs_set_header_flags(split, btrfs_header_flags(c));
2180 btrfs_set_header_level(split, btrfs_header_level(c));
2181 btrfs_set_header_bytenr(split, split->start);
2182 btrfs_set_header_generation(split, trans->transid);
2183 btrfs_set_header_owner(split, root->root_key.objectid);
2184 btrfs_set_header_flags(split, 0);
2185 write_extent_buffer(split, root->fs_info->fsid,
2186 (unsigned long)btrfs_header_fsid(split),
2188 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
2189 (unsigned long)btrfs_header_chunk_tree_uuid(split),
2192 mid = (c_nritems + 1) / 2;
2194 copy_extent_buffer(split, c,
2195 btrfs_node_key_ptr_offset(0),
2196 btrfs_node_key_ptr_offset(mid),
2197 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
2198 btrfs_set_header_nritems(split, c_nritems - mid);
2199 btrfs_set_header_nritems(c, mid);
2202 btrfs_mark_buffer_dirty(c);
2203 btrfs_mark_buffer_dirty(split);
2205 btrfs_node_key(split, &disk_key, 0);
2206 wret = insert_ptr(trans, root, path, &disk_key, split->start,
2207 path->slots[level + 1] + 1,
2212 ret = btrfs_update_ref(trans, root, c, split, 0, c_nritems - mid);
2215 if (path->slots[level] >= mid) {
2216 path->slots[level] -= mid;
2217 btrfs_tree_unlock(c);
2218 free_extent_buffer(c);
2219 path->nodes[level] = split;
2220 path->slots[level + 1] += 1;
2222 btrfs_tree_unlock(split);
2223 free_extent_buffer(split);
2229 * how many bytes are required to store the items in a leaf. start
2230 * and nr indicate which items in the leaf to check. This totals up the
2231 * space used both by the item structs and the item data
2233 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
2236 int nritems = btrfs_header_nritems(l);
2237 int end = min(nritems, start + nr) - 1;
2241 data_len = btrfs_item_end_nr(l, start);
2242 data_len = data_len - btrfs_item_offset_nr(l, end);
2243 data_len += sizeof(struct btrfs_item) * nr;
2244 WARN_ON(data_len < 0);
2249 * The space between the end of the leaf items and
2250 * the start of the leaf data. IOW, how much room
2251 * the leaf has left for both items and data
2253 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
2254 struct extent_buffer *leaf)
2256 int nritems = btrfs_header_nritems(leaf);
2258 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
2260 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
2261 "used %d nritems %d\n",
2262 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
2263 leaf_space_used(leaf, 0, nritems), nritems);
2269 * push some data in the path leaf to the right, trying to free up at
2270 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2272 * returns 1 if the push failed because the other node didn't have enough
2273 * room, 0 if everything worked out and < 0 if there were major errors.
2275 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
2276 *root, struct btrfs_path *path, int data_size,
2279 struct extent_buffer *left = path->nodes[0];
2280 struct extent_buffer *right;
2281 struct extent_buffer *upper;
2282 struct btrfs_disk_key disk_key;
2288 struct btrfs_item *item;
2296 slot = path->slots[1];
2297 if (!path->nodes[1])
2300 upper = path->nodes[1];
2301 if (slot >= btrfs_header_nritems(upper) - 1)
2304 btrfs_assert_tree_locked(path->nodes[1]);
2306 right = read_node_slot(root, upper, slot + 1);
2307 btrfs_tree_lock(right);
2308 btrfs_set_lock_blocking(right);
2310 free_space = btrfs_leaf_free_space(root, right);
2311 if (free_space < data_size)
2314 /* cow and double check */
2315 ret = btrfs_cow_block(trans, root, right, upper,
2320 free_space = btrfs_leaf_free_space(root, right);
2321 if (free_space < data_size)
2324 left_nritems = btrfs_header_nritems(left);
2325 if (left_nritems == 0)
2333 if (path->slots[0] >= left_nritems)
2334 push_space += data_size;
2336 i = left_nritems - 1;
2338 item = btrfs_item_nr(left, i);
2340 if (!empty && push_items > 0) {
2341 if (path->slots[0] > i)
2343 if (path->slots[0] == i) {
2344 int space = btrfs_leaf_free_space(root, left);
2345 if (space + push_space * 2 > free_space)
2350 if (path->slots[0] == i)
2351 push_space += data_size;
2353 if (!left->map_token) {
2354 map_extent_buffer(left, (unsigned long)item,
2355 sizeof(struct btrfs_item),
2356 &left->map_token, &left->kaddr,
2357 &left->map_start, &left->map_len,
2361 this_item_size = btrfs_item_size(left, item);
2362 if (this_item_size + sizeof(*item) + push_space > free_space)
2366 push_space += this_item_size + sizeof(*item);
2371 if (left->map_token) {
2372 unmap_extent_buffer(left, left->map_token, KM_USER1);
2373 left->map_token = NULL;
2376 if (push_items == 0)
2379 if (!empty && push_items == left_nritems)
2382 /* push left to right */
2383 right_nritems = btrfs_header_nritems(right);
2385 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
2386 push_space -= leaf_data_end(root, left);
2388 /* make room in the right data area */
2389 data_end = leaf_data_end(root, right);
2390 memmove_extent_buffer(right,
2391 btrfs_leaf_data(right) + data_end - push_space,
2392 btrfs_leaf_data(right) + data_end,
2393 BTRFS_LEAF_DATA_SIZE(root) - data_end);
2395 /* copy from the left data area */
2396 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
2397 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2398 btrfs_leaf_data(left) + leaf_data_end(root, left),
2401 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
2402 btrfs_item_nr_offset(0),
2403 right_nritems * sizeof(struct btrfs_item));
2405 /* copy the items from left to right */
2406 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
2407 btrfs_item_nr_offset(left_nritems - push_items),
2408 push_items * sizeof(struct btrfs_item));
2410 /* update the item pointers */
2411 right_nritems += push_items;
2412 btrfs_set_header_nritems(right, right_nritems);
2413 push_space = BTRFS_LEAF_DATA_SIZE(root);
2414 for (i = 0; i < right_nritems; i++) {
2415 item = btrfs_item_nr(right, i);
2416 if (!right->map_token) {
2417 map_extent_buffer(right, (unsigned long)item,
2418 sizeof(struct btrfs_item),
2419 &right->map_token, &right->kaddr,
2420 &right->map_start, &right->map_len,
2423 push_space -= btrfs_item_size(right, item);
2424 btrfs_set_item_offset(right, item, push_space);
2427 if (right->map_token) {
2428 unmap_extent_buffer(right, right->map_token, KM_USER1);
2429 right->map_token = NULL;
2431 left_nritems -= push_items;
2432 btrfs_set_header_nritems(left, left_nritems);
2435 btrfs_mark_buffer_dirty(left);
2436 btrfs_mark_buffer_dirty(right);
2438 ret = btrfs_update_ref(trans, root, left, right, 0, push_items);
2441 btrfs_item_key(right, &disk_key, 0);
2442 btrfs_set_node_key(upper, &disk_key, slot + 1);
2443 btrfs_mark_buffer_dirty(upper);
2445 /* then fixup the leaf pointer in the path */
2446 if (path->slots[0] >= left_nritems) {
2447 path->slots[0] -= left_nritems;
2448 if (btrfs_header_nritems(path->nodes[0]) == 0)
2449 clean_tree_block(trans, root, path->nodes[0]);
2450 btrfs_tree_unlock(path->nodes[0]);
2451 free_extent_buffer(path->nodes[0]);
2452 path->nodes[0] = right;
2453 path->slots[1] += 1;
2455 btrfs_tree_unlock(right);
2456 free_extent_buffer(right);
2461 btrfs_tree_unlock(right);
2462 free_extent_buffer(right);
2467 * push some data in the path leaf to the left, trying to free up at
2468 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2470 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
2471 *root, struct btrfs_path *path, int data_size,
2474 struct btrfs_disk_key disk_key;
2475 struct extent_buffer *right = path->nodes[0];
2476 struct extent_buffer *left;
2482 struct btrfs_item *item;
2483 u32 old_left_nritems;
2489 u32 old_left_item_size;
2491 slot = path->slots[1];
2494 if (!path->nodes[1])
2497 right_nritems = btrfs_header_nritems(right);
2498 if (right_nritems == 0)
2501 btrfs_assert_tree_locked(path->nodes[1]);
2503 left = read_node_slot(root, path->nodes[1], slot - 1);
2504 btrfs_tree_lock(left);
2505 btrfs_set_lock_blocking(left);
2507 free_space = btrfs_leaf_free_space(root, left);
2508 if (free_space < data_size) {
2513 /* cow and double check */
2514 ret = btrfs_cow_block(trans, root, left,
2515 path->nodes[1], slot - 1, &left);
2517 /* we hit -ENOSPC, but it isn't fatal here */
2522 free_space = btrfs_leaf_free_space(root, left);
2523 if (free_space < data_size) {
2531 nr = right_nritems - 1;
2533 for (i = 0; i < nr; i++) {
2534 item = btrfs_item_nr(right, i);
2535 if (!right->map_token) {
2536 map_extent_buffer(right, (unsigned long)item,
2537 sizeof(struct btrfs_item),
2538 &right->map_token, &right->kaddr,
2539 &right->map_start, &right->map_len,
2543 if (!empty && push_items > 0) {
2544 if (path->slots[0] < i)
2546 if (path->slots[0] == i) {
2547 int space = btrfs_leaf_free_space(root, right);
2548 if (space + push_space * 2 > free_space)
2553 if (path->slots[0] == i)
2554 push_space += data_size;
2556 this_item_size = btrfs_item_size(right, item);
2557 if (this_item_size + sizeof(*item) + push_space > free_space)
2561 push_space += this_item_size + sizeof(*item);
2564 if (right->map_token) {
2565 unmap_extent_buffer(right, right->map_token, KM_USER1);
2566 right->map_token = NULL;
2569 if (push_items == 0) {
2573 if (!empty && push_items == btrfs_header_nritems(right))
2576 /* push data from right to left */
2577 copy_extent_buffer(left, right,
2578 btrfs_item_nr_offset(btrfs_header_nritems(left)),
2579 btrfs_item_nr_offset(0),
2580 push_items * sizeof(struct btrfs_item));
2582 push_space = BTRFS_LEAF_DATA_SIZE(root) -
2583 btrfs_item_offset_nr(right, push_items - 1);
2585 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
2586 leaf_data_end(root, left) - push_space,
2587 btrfs_leaf_data(right) +
2588 btrfs_item_offset_nr(right, push_items - 1),
2590 old_left_nritems = btrfs_header_nritems(left);
2591 BUG_ON(old_left_nritems <= 0);
2593 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
2594 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
2597 item = btrfs_item_nr(left, i);
2598 if (!left->map_token) {
2599 map_extent_buffer(left, (unsigned long)item,
2600 sizeof(struct btrfs_item),
2601 &left->map_token, &left->kaddr,
2602 &left->map_start, &left->map_len,
2606 ioff = btrfs_item_offset(left, item);
2607 btrfs_set_item_offset(left, item,
2608 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size));
2610 btrfs_set_header_nritems(left, old_left_nritems + push_items);
2611 if (left->map_token) {
2612 unmap_extent_buffer(left, left->map_token, KM_USER1);
2613 left->map_token = NULL;
2616 /* fixup right node */
2617 if (push_items > right_nritems) {
2618 printk(KERN_CRIT "push items %d nr %u\n", push_items,
2623 if (push_items < right_nritems) {
2624 push_space = btrfs_item_offset_nr(right, push_items - 1) -
2625 leaf_data_end(root, right);
2626 memmove_extent_buffer(right, btrfs_leaf_data(right) +
2627 BTRFS_LEAF_DATA_SIZE(root) - push_space,
2628 btrfs_leaf_data(right) +
2629 leaf_data_end(root, right), push_space);
2631 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
2632 btrfs_item_nr_offset(push_items),
2633 (btrfs_header_nritems(right) - push_items) *
2634 sizeof(struct btrfs_item));
2636 right_nritems -= push_items;
2637 btrfs_set_header_nritems(right, right_nritems);
2638 push_space = BTRFS_LEAF_DATA_SIZE(root);
2639 for (i = 0; i < right_nritems; i++) {
2640 item = btrfs_item_nr(right, i);
2642 if (!right->map_token) {
2643 map_extent_buffer(right, (unsigned long)item,
2644 sizeof(struct btrfs_item),
2645 &right->map_token, &right->kaddr,
2646 &right->map_start, &right->map_len,
2650 push_space = push_space - btrfs_item_size(right, item);
2651 btrfs_set_item_offset(right, item, push_space);
2653 if (right->map_token) {
2654 unmap_extent_buffer(right, right->map_token, KM_USER1);
2655 right->map_token = NULL;
2658 btrfs_mark_buffer_dirty(left);
2660 btrfs_mark_buffer_dirty(right);
2662 ret = btrfs_update_ref(trans, root, right, left,
2663 old_left_nritems, push_items);
2666 btrfs_item_key(right, &disk_key, 0);
2667 wret = fixup_low_keys(trans, root, path, &disk_key, 1);
2671 /* then fixup the leaf pointer in the path */
2672 if (path->slots[0] < push_items) {
2673 path->slots[0] += old_left_nritems;
2674 if (btrfs_header_nritems(path->nodes[0]) == 0)
2675 clean_tree_block(trans, root, path->nodes[0]);
2676 btrfs_tree_unlock(path->nodes[0]);
2677 free_extent_buffer(path->nodes[0]);
2678 path->nodes[0] = left;
2679 path->slots[1] -= 1;
2681 btrfs_tree_unlock(left);
2682 free_extent_buffer(left);
2683 path->slots[0] -= push_items;
2685 BUG_ON(path->slots[0] < 0);
2688 btrfs_tree_unlock(left);
2689 free_extent_buffer(left);
2694 * split the path's leaf in two, making sure there is at least data_size
2695 * available for the resulting leaf level of the path.
2697 * returns 0 if all went well and < 0 on failure.
2699 static noinline int split_leaf(struct btrfs_trans_handle *trans,
2700 struct btrfs_root *root,
2701 struct btrfs_key *ins_key,
2702 struct btrfs_path *path, int data_size,
2705 struct extent_buffer *l;
2709 struct extent_buffer *right;
2716 int num_doubles = 0;
2717 struct btrfs_disk_key disk_key;
2719 /* first try to make some room by pushing left and right */
2720 if (data_size && ins_key->type != BTRFS_DIR_ITEM_KEY) {
2721 wret = push_leaf_right(trans, root, path, data_size, 0);
2725 wret = push_leaf_left(trans, root, path, data_size, 0);
2731 /* did the pushes work? */
2732 if (btrfs_leaf_free_space(root, l) >= data_size)
2736 if (!path->nodes[1]) {
2737 ret = insert_new_root(trans, root, path, 1);
2744 slot = path->slots[0];
2745 nritems = btrfs_header_nritems(l);
2746 mid = (nritems + 1) / 2;
2748 right = btrfs_alloc_free_block(trans, root, root->leafsize,
2749 path->nodes[1]->start,
2750 root->root_key.objectid,
2751 trans->transid, 0, l->start, 0);
2752 if (IS_ERR(right)) {
2754 return PTR_ERR(right);
2757 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
2758 btrfs_set_header_bytenr(right, right->start);
2759 btrfs_set_header_generation(right, trans->transid);
2760 btrfs_set_header_owner(right, root->root_key.objectid);
2761 btrfs_set_header_level(right, 0);
2762 write_extent_buffer(right, root->fs_info->fsid,
2763 (unsigned long)btrfs_header_fsid(right),
2766 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
2767 (unsigned long)btrfs_header_chunk_tree_uuid(right),
2771 leaf_space_used(l, mid, nritems - mid) + data_size >
2772 BTRFS_LEAF_DATA_SIZE(root)) {
2773 if (slot >= nritems) {
2774 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2775 btrfs_set_header_nritems(right, 0);
2776 wret = insert_ptr(trans, root, path,
2777 &disk_key, right->start,
2778 path->slots[1] + 1, 1);
2782 btrfs_tree_unlock(path->nodes[0]);
2783 free_extent_buffer(path->nodes[0]);
2784 path->nodes[0] = right;
2786 path->slots[1] += 1;
2787 btrfs_mark_buffer_dirty(right);
2791 if (mid != nritems &&
2792 leaf_space_used(l, mid, nritems - mid) +
2793 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2798 if (leaf_space_used(l, 0, mid) + data_size >
2799 BTRFS_LEAF_DATA_SIZE(root)) {
2800 if (!extend && data_size && slot == 0) {
2801 btrfs_cpu_key_to_disk(&disk_key, ins_key);
2802 btrfs_set_header_nritems(right, 0);
2803 wret = insert_ptr(trans, root, path,
2809 btrfs_tree_unlock(path->nodes[0]);
2810 free_extent_buffer(path->nodes[0]);
2811 path->nodes[0] = right;
2813 if (path->slots[1] == 0) {
2814 wret = fixup_low_keys(trans, root,
2815 path, &disk_key, 1);
2819 btrfs_mark_buffer_dirty(right);
2821 } else if ((extend || !data_size) && slot == 0) {
2825 if (mid != nritems &&
2826 leaf_space_used(l, mid, nritems - mid) +
2827 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
2833 nritems = nritems - mid;
2834 btrfs_set_header_nritems(right, nritems);
2835 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
2837 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
2838 btrfs_item_nr_offset(mid),
2839 nritems * sizeof(struct btrfs_item));
2841 copy_extent_buffer(right, l,
2842 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
2843 data_copy_size, btrfs_leaf_data(l) +
2844 leaf_data_end(root, l), data_copy_size);
2846 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
2847 btrfs_item_end_nr(l, mid);
2849 for (i = 0; i < nritems; i++) {
2850 struct btrfs_item *item = btrfs_item_nr(right, i);
2853 if (!right->map_token) {
2854 map_extent_buffer(right, (unsigned long)item,
2855 sizeof(struct btrfs_item),
2856 &right->map_token, &right->kaddr,
2857 &right->map_start, &right->map_len,
2861 ioff = btrfs_item_offset(right, item);
2862 btrfs_set_item_offset(right, item, ioff + rt_data_off);
2865 if (right->map_token) {
2866 unmap_extent_buffer(right, right->map_token, KM_USER1);
2867 right->map_token = NULL;
2870 btrfs_set_header_nritems(l, mid);
2872 btrfs_item_key(right, &disk_key, 0);
2873 wret = insert_ptr(trans, root, path, &disk_key, right->start,
2874 path->slots[1] + 1, 1);
2878 btrfs_mark_buffer_dirty(right);
2879 btrfs_mark_buffer_dirty(l);
2880 BUG_ON(path->slots[0] != slot);
2882 ret = btrfs_update_ref(trans, root, l, right, 0, nritems);
2886 btrfs_tree_unlock(path->nodes[0]);
2887 free_extent_buffer(path->nodes[0]);
2888 path->nodes[0] = right;
2889 path->slots[0] -= mid;
2890 path->slots[1] += 1;
2892 btrfs_tree_unlock(right);
2893 free_extent_buffer(right);
2896 BUG_ON(path->slots[0] < 0);
2899 BUG_ON(num_doubles != 0);
2907 * This function splits a single item into two items,
2908 * giving 'new_key' to the new item and splitting the
2909 * old one at split_offset (from the start of the item).
2911 * The path may be released by this operation. After
2912 * the split, the path is pointing to the old item. The
2913 * new item is going to be in the same node as the old one.
2915 * Note, the item being split must be smaller enough to live alone on
2916 * a tree block with room for one extra struct btrfs_item
2918 * This allows us to split the item in place, keeping a lock on the
2919 * leaf the entire time.
2921 int btrfs_split_item(struct btrfs_trans_handle *trans,
2922 struct btrfs_root *root,
2923 struct btrfs_path *path,
2924 struct btrfs_key *new_key,
2925 unsigned long split_offset)
2928 struct extent_buffer *leaf;
2929 struct btrfs_key orig_key;
2930 struct btrfs_item *item;
2931 struct btrfs_item *new_item;
2936 struct btrfs_disk_key disk_key;
2939 leaf = path->nodes[0];
2940 btrfs_item_key_to_cpu(leaf, &orig_key, path->slots[0]);
2941 if (btrfs_leaf_free_space(root, leaf) >= sizeof(struct btrfs_item))
2944 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
2945 btrfs_release_path(root, path);
2947 path->search_for_split = 1;
2948 path->keep_locks = 1;
2950 ret = btrfs_search_slot(trans, root, &orig_key, path, 0, 1);
2951 path->search_for_split = 0;
2953 /* if our item isn't there or got smaller, return now */
2954 if (ret != 0 || item_size != btrfs_item_size_nr(path->nodes[0],
2956 path->keep_locks = 0;
2960 ret = split_leaf(trans, root, &orig_key, path,
2961 sizeof(struct btrfs_item), 1);
2962 path->keep_locks = 0;
2966 * make sure any changes to the path from split_leaf leave it
2967 * in a blocking state
2969 btrfs_set_path_blocking(path);
2971 leaf = path->nodes[0];
2972 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
2975 item = btrfs_item_nr(leaf, path->slots[0]);
2976 orig_offset = btrfs_item_offset(leaf, item);
2977 item_size = btrfs_item_size(leaf, item);
2980 buf = kmalloc(item_size, GFP_NOFS);
2981 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
2982 path->slots[0]), item_size);
2983 slot = path->slots[0] + 1;
2984 leaf = path->nodes[0];
2986 nritems = btrfs_header_nritems(leaf);
2988 if (slot != nritems) {
2989 /* shift the items */
2990 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
2991 btrfs_item_nr_offset(slot),
2992 (nritems - slot) * sizeof(struct btrfs_item));
2996 btrfs_cpu_key_to_disk(&disk_key, new_key);
2997 btrfs_set_item_key(leaf, &disk_key, slot);
2999 new_item = btrfs_item_nr(leaf, slot);
3001 btrfs_set_item_offset(leaf, new_item, orig_offset);
3002 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
3004 btrfs_set_item_offset(leaf, item,
3005 orig_offset + item_size - split_offset);
3006 btrfs_set_item_size(leaf, item, split_offset);
3008 btrfs_set_header_nritems(leaf, nritems + 1);
3010 /* write the data for the start of the original item */
3011 write_extent_buffer(leaf, buf,
3012 btrfs_item_ptr_offset(leaf, path->slots[0]),
3015 /* write the data for the new item */
3016 write_extent_buffer(leaf, buf + split_offset,
3017 btrfs_item_ptr_offset(leaf, slot),
3018 item_size - split_offset);
3019 btrfs_mark_buffer_dirty(leaf);
3022 if (btrfs_leaf_free_space(root, leaf) < 0) {
3023 btrfs_print_leaf(root, leaf);
3031 * make the item pointed to by the path smaller. new_size indicates
3032 * how small to make it, and from_end tells us if we just chop bytes
3033 * off the end of the item or if we shift the item to chop bytes off
3036 int btrfs_truncate_item(struct btrfs_trans_handle *trans,
3037 struct btrfs_root *root,
3038 struct btrfs_path *path,
3039 u32 new_size, int from_end)
3044 struct extent_buffer *leaf;
3045 struct btrfs_item *item;
3047 unsigned int data_end;
3048 unsigned int old_data_start;
3049 unsigned int old_size;
3050 unsigned int size_diff;
3053 slot_orig = path->slots[0];
3054 leaf = path->nodes[0];
3055 slot = path->slots[0];
3057 old_size = btrfs_item_size_nr(leaf, slot);
3058 if (old_size == new_size)
3061 nritems = btrfs_header_nritems(leaf);
3062 data_end = leaf_data_end(root, leaf);
3064 old_data_start = btrfs_item_offset_nr(leaf, slot);
3066 size_diff = old_size - new_size;
3069 BUG_ON(slot >= nritems);
3072 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3074 /* first correct the data pointers */
3075 for (i = slot; i < nritems; i++) {
3077 item = btrfs_item_nr(leaf, i);
3079 if (!leaf->map_token) {
3080 map_extent_buffer(leaf, (unsigned long)item,
3081 sizeof(struct btrfs_item),
3082 &leaf->map_token, &leaf->kaddr,
3083 &leaf->map_start, &leaf->map_len,
3087 ioff = btrfs_item_offset(leaf, item);
3088 btrfs_set_item_offset(leaf, item, ioff + size_diff);
3091 if (leaf->map_token) {
3092 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3093 leaf->map_token = NULL;
3096 /* shift the data */
3098 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3099 data_end + size_diff, btrfs_leaf_data(leaf) +
3100 data_end, old_data_start + new_size - data_end);
3102 struct btrfs_disk_key disk_key;
3105 btrfs_item_key(leaf, &disk_key, slot);
3107 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
3109 struct btrfs_file_extent_item *fi;
3111 fi = btrfs_item_ptr(leaf, slot,
3112 struct btrfs_file_extent_item);
3113 fi = (struct btrfs_file_extent_item *)(
3114 (unsigned long)fi - size_diff);
3116 if (btrfs_file_extent_type(leaf, fi) ==
3117 BTRFS_FILE_EXTENT_INLINE) {
3118 ptr = btrfs_item_ptr_offset(leaf, slot);
3119 memmove_extent_buffer(leaf, ptr,
3121 offsetof(struct btrfs_file_extent_item,
3126 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3127 data_end + size_diff, btrfs_leaf_data(leaf) +
3128 data_end, old_data_start - data_end);
3130 offset = btrfs_disk_key_offset(&disk_key);
3131 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
3132 btrfs_set_item_key(leaf, &disk_key, slot);
3134 fixup_low_keys(trans, root, path, &disk_key, 1);
3137 item = btrfs_item_nr(leaf, slot);
3138 btrfs_set_item_size(leaf, item, new_size);
3139 btrfs_mark_buffer_dirty(leaf);
3142 if (btrfs_leaf_free_space(root, leaf) < 0) {
3143 btrfs_print_leaf(root, leaf);
3150 * make the item pointed to by the path bigger, data_size is the new size.
3152 int btrfs_extend_item(struct btrfs_trans_handle *trans,
3153 struct btrfs_root *root, struct btrfs_path *path,
3159 struct extent_buffer *leaf;
3160 struct btrfs_item *item;
3162 unsigned int data_end;
3163 unsigned int old_data;
3164 unsigned int old_size;
3167 slot_orig = path->slots[0];
3168 leaf = path->nodes[0];
3170 nritems = btrfs_header_nritems(leaf);
3171 data_end = leaf_data_end(root, leaf);
3173 if (btrfs_leaf_free_space(root, leaf) < data_size) {
3174 btrfs_print_leaf(root, leaf);
3177 slot = path->slots[0];
3178 old_data = btrfs_item_end_nr(leaf, slot);
3181 if (slot >= nritems) {
3182 btrfs_print_leaf(root, leaf);
3183 printk(KERN_CRIT "slot %d too large, nritems %d\n",
3189 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3191 /* first correct the data pointers */
3192 for (i = slot; i < nritems; i++) {
3194 item = btrfs_item_nr(leaf, i);
3196 if (!leaf->map_token) {
3197 map_extent_buffer(leaf, (unsigned long)item,
3198 sizeof(struct btrfs_item),
3199 &leaf->map_token, &leaf->kaddr,
3200 &leaf->map_start, &leaf->map_len,
3203 ioff = btrfs_item_offset(leaf, item);
3204 btrfs_set_item_offset(leaf, item, ioff - data_size);
3207 if (leaf->map_token) {
3208 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3209 leaf->map_token = NULL;
3212 /* shift the data */
3213 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3214 data_end - data_size, btrfs_leaf_data(leaf) +
3215 data_end, old_data - data_end);
3217 data_end = old_data;
3218 old_size = btrfs_item_size_nr(leaf, slot);
3219 item = btrfs_item_nr(leaf, slot);
3220 btrfs_set_item_size(leaf, item, old_size + data_size);
3221 btrfs_mark_buffer_dirty(leaf);
3224 if (btrfs_leaf_free_space(root, leaf) < 0) {
3225 btrfs_print_leaf(root, leaf);
3232 * Given a key and some data, insert items into the tree.
3233 * This does all the path init required, making room in the tree if needed.
3234 * Returns the number of keys that were inserted.
3236 int btrfs_insert_some_items(struct btrfs_trans_handle *trans,
3237 struct btrfs_root *root,
3238 struct btrfs_path *path,
3239 struct btrfs_key *cpu_key, u32 *data_size,
3242 struct extent_buffer *leaf;
3243 struct btrfs_item *item;
3250 unsigned int data_end;
3251 struct btrfs_disk_key disk_key;
3252 struct btrfs_key found_key;
3254 for (i = 0; i < nr; i++) {
3255 if (total_size + data_size[i] + sizeof(struct btrfs_item) >
3256 BTRFS_LEAF_DATA_SIZE(root)) {
3260 total_data += data_size[i];
3261 total_size += data_size[i] + sizeof(struct btrfs_item);
3265 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3271 leaf = path->nodes[0];
3273 nritems = btrfs_header_nritems(leaf);
3274 data_end = leaf_data_end(root, leaf);
3276 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3277 for (i = nr; i >= 0; i--) {
3278 total_data -= data_size[i];
3279 total_size -= data_size[i] + sizeof(struct btrfs_item);
3280 if (total_size < btrfs_leaf_free_space(root, leaf))
3286 slot = path->slots[0];
3289 if (slot != nritems) {
3290 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3292 item = btrfs_item_nr(leaf, slot);
3293 btrfs_item_key_to_cpu(leaf, &found_key, slot);
3295 /* figure out how many keys we can insert in here */
3296 total_data = data_size[0];
3297 for (i = 1; i < nr; i++) {
3298 if (comp_cpu_keys(&found_key, cpu_key + i) <= 0)
3300 total_data += data_size[i];
3304 if (old_data < data_end) {
3305 btrfs_print_leaf(root, leaf);
3306 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3307 slot, old_data, data_end);
3311 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3313 /* first correct the data pointers */
3314 WARN_ON(leaf->map_token);
3315 for (i = slot; i < nritems; i++) {
3318 item = btrfs_item_nr(leaf, i);
3319 if (!leaf->map_token) {
3320 map_extent_buffer(leaf, (unsigned long)item,
3321 sizeof(struct btrfs_item),
3322 &leaf->map_token, &leaf->kaddr,
3323 &leaf->map_start, &leaf->map_len,
3327 ioff = btrfs_item_offset(leaf, item);
3328 btrfs_set_item_offset(leaf, item, ioff - total_data);
3330 if (leaf->map_token) {
3331 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3332 leaf->map_token = NULL;
3335 /* shift the items */
3336 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3337 btrfs_item_nr_offset(slot),
3338 (nritems - slot) * sizeof(struct btrfs_item));
3340 /* shift the data */
3341 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3342 data_end - total_data, btrfs_leaf_data(leaf) +
3343 data_end, old_data - data_end);
3344 data_end = old_data;
3347 * this sucks but it has to be done, if we are inserting at
3348 * the end of the leaf only insert 1 of the items, since we
3349 * have no way of knowing whats on the next leaf and we'd have
3350 * to drop our current locks to figure it out
3355 /* setup the item for the new data */
3356 for (i = 0; i < nr; i++) {
3357 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3358 btrfs_set_item_key(leaf, &disk_key, slot + i);
3359 item = btrfs_item_nr(leaf, slot + i);
3360 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3361 data_end -= data_size[i];
3362 btrfs_set_item_size(leaf, item, data_size[i]);
3364 btrfs_set_header_nritems(leaf, nritems + nr);
3365 btrfs_mark_buffer_dirty(leaf);
3369 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3370 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3373 if (btrfs_leaf_free_space(root, leaf) < 0) {
3374 btrfs_print_leaf(root, leaf);
3384 * Given a key and some data, insert items into the tree.
3385 * This does all the path init required, making room in the tree if needed.
3387 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
3388 struct btrfs_root *root,
3389 struct btrfs_path *path,
3390 struct btrfs_key *cpu_key, u32 *data_size,
3393 struct extent_buffer *leaf;
3394 struct btrfs_item *item;
3402 unsigned int data_end;
3403 struct btrfs_disk_key disk_key;
3405 for (i = 0; i < nr; i++)
3406 total_data += data_size[i];
3408 total_size = total_data + (nr * sizeof(struct btrfs_item));
3409 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
3415 slot_orig = path->slots[0];
3416 leaf = path->nodes[0];
3418 nritems = btrfs_header_nritems(leaf);
3419 data_end = leaf_data_end(root, leaf);
3421 if (btrfs_leaf_free_space(root, leaf) < total_size) {
3422 btrfs_print_leaf(root, leaf);
3423 printk(KERN_CRIT "not enough freespace need %u have %d\n",
3424 total_size, btrfs_leaf_free_space(root, leaf));
3428 slot = path->slots[0];
3431 if (slot != nritems) {
3432 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
3434 if (old_data < data_end) {
3435 btrfs_print_leaf(root, leaf);
3436 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
3437 slot, old_data, data_end);
3441 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3443 /* first correct the data pointers */
3444 WARN_ON(leaf->map_token);
3445 for (i = slot; i < nritems; i++) {
3448 item = btrfs_item_nr(leaf, i);
3449 if (!leaf->map_token) {
3450 map_extent_buffer(leaf, (unsigned long)item,
3451 sizeof(struct btrfs_item),
3452 &leaf->map_token, &leaf->kaddr,
3453 &leaf->map_start, &leaf->map_len,
3457 ioff = btrfs_item_offset(leaf, item);
3458 btrfs_set_item_offset(leaf, item, ioff - total_data);
3460 if (leaf->map_token) {
3461 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3462 leaf->map_token = NULL;
3465 /* shift the items */
3466 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
3467 btrfs_item_nr_offset(slot),
3468 (nritems - slot) * sizeof(struct btrfs_item));
3470 /* shift the data */
3471 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3472 data_end - total_data, btrfs_leaf_data(leaf) +
3473 data_end, old_data - data_end);
3474 data_end = old_data;
3477 /* setup the item for the new data */
3478 for (i = 0; i < nr; i++) {
3479 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
3480 btrfs_set_item_key(leaf, &disk_key, slot + i);
3481 item = btrfs_item_nr(leaf, slot + i);
3482 btrfs_set_item_offset(leaf, item, data_end - data_size[i]);
3483 data_end -= data_size[i];
3484 btrfs_set_item_size(leaf, item, data_size[i]);
3486 btrfs_set_header_nritems(leaf, nritems + nr);
3487 btrfs_mark_buffer_dirty(leaf);
3491 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
3492 ret = fixup_low_keys(trans, root, path, &disk_key, 1);
3495 if (btrfs_leaf_free_space(root, leaf) < 0) {
3496 btrfs_print_leaf(root, leaf);
3500 btrfs_unlock_up_safe(path, 1);
3505 * Given a key and some data, insert an item into the tree.
3506 * This does all the path init required, making room in the tree if needed.
3508 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
3509 *root, struct btrfs_key *cpu_key, void *data, u32
3513 struct btrfs_path *path;
3514 struct extent_buffer *leaf;
3517 path = btrfs_alloc_path();
3519 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
3521 leaf = path->nodes[0];
3522 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
3523 write_extent_buffer(leaf, data, ptr, data_size);
3524 btrfs_mark_buffer_dirty(leaf);
3526 btrfs_free_path(path);
3531 * delete the pointer from a given node.
3533 * the tree should have been previously balanced so the deletion does not
3536 static int del_ptr(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3537 struct btrfs_path *path, int level, int slot)
3539 struct extent_buffer *parent = path->nodes[level];
3544 nritems = btrfs_header_nritems(parent);
3545 if (slot != nritems - 1) {
3546 memmove_extent_buffer(parent,
3547 btrfs_node_key_ptr_offset(slot),
3548 btrfs_node_key_ptr_offset(slot + 1),
3549 sizeof(struct btrfs_key_ptr) *
3550 (nritems - slot - 1));
3553 btrfs_set_header_nritems(parent, nritems);
3554 if (nritems == 0 && parent == root->node) {
3555 BUG_ON(btrfs_header_level(root->node) != 1);
3556 /* just turn the root into a leaf and break */
3557 btrfs_set_header_level(root->node, 0);
3558 } else if (slot == 0) {
3559 struct btrfs_disk_key disk_key;
3561 btrfs_node_key(parent, &disk_key, 0);
3562 wret = fixup_low_keys(trans, root, path, &disk_key, level + 1);
3566 btrfs_mark_buffer_dirty(parent);
3571 * a helper function to delete the leaf pointed to by path->slots[1] and
3572 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3573 * already know it, it is faster to have them pass it down than to
3574 * read it out of the node again.
3576 * This deletes the pointer in path->nodes[1] and frees the leaf
3577 * block extent. zero is returned if it all worked out, < 0 otherwise.
3579 * The path must have already been setup for deleting the leaf, including
3580 * all the proper balancing. path->nodes[1] must be locked.
3582 noinline int btrfs_del_leaf(struct btrfs_trans_handle *trans,
3583 struct btrfs_root *root,
3584 struct btrfs_path *path, u64 bytenr)
3587 u64 root_gen = btrfs_header_generation(path->nodes[1]);
3588 u64 parent_start = path->nodes[1]->start;
3589 u64 parent_owner = btrfs_header_owner(path->nodes[1]);
3591 ret = del_ptr(trans, root, path, 1, path->slots[1]);
3596 * btrfs_free_extent is expensive, we want to make sure we
3597 * aren't holding any locks when we call it
3599 btrfs_unlock_up_safe(path, 0);
3601 ret = btrfs_free_extent(trans, root, bytenr,
3602 btrfs_level_size(root, 0),
3603 parent_start, parent_owner,
3608 * delete the item at the leaf level in path. If that empties
3609 * the leaf, remove it from the tree
3611 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
3612 struct btrfs_path *path, int slot, int nr)
3614 struct extent_buffer *leaf;
3615 struct btrfs_item *item;
3623 leaf = path->nodes[0];
3624 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
3626 for (i = 0; i < nr; i++)
3627 dsize += btrfs_item_size_nr(leaf, slot + i);
3629 nritems = btrfs_header_nritems(leaf);
3631 if (slot + nr != nritems) {
3632 int data_end = leaf_data_end(root, leaf);
3634 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
3636 btrfs_leaf_data(leaf) + data_end,
3637 last_off - data_end);
3639 for (i = slot + nr; i < nritems; i++) {
3642 item = btrfs_item_nr(leaf, i);
3643 if (!leaf->map_token) {
3644 map_extent_buffer(leaf, (unsigned long)item,
3645 sizeof(struct btrfs_item),
3646 &leaf->map_token, &leaf->kaddr,
3647 &leaf->map_start, &leaf->map_len,
3650 ioff = btrfs_item_offset(leaf, item);
3651 btrfs_set_item_offset(leaf, item, ioff + dsize);
3654 if (leaf->map_token) {
3655 unmap_extent_buffer(leaf, leaf->map_token, KM_USER1);
3656 leaf->map_token = NULL;
3659 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
3660 btrfs_item_nr_offset(slot + nr),
3661 sizeof(struct btrfs_item) *
3662 (nritems - slot - nr));
3664 btrfs_set_header_nritems(leaf, nritems - nr);
3667 /* delete the leaf if we've emptied it */
3669 if (leaf == root->node) {
3670 btrfs_set_header_level(leaf, 0);
3672 ret = btrfs_del_leaf(trans, root, path, leaf->start);
3676 int used = leaf_space_used(leaf, 0, nritems);
3678 struct btrfs_disk_key disk_key;
3680 btrfs_item_key(leaf, &disk_key, 0);
3681 wret = fixup_low_keys(trans, root, path,
3687 /* delete the leaf if it is mostly empty */
3688 if (used < BTRFS_LEAF_DATA_SIZE(root) / 4) {
3689 /* push_leaf_left fixes the path.
3690 * make sure the path still points to our leaf
3691 * for possible call to del_ptr below
3693 slot = path->slots[1];
3694 extent_buffer_get(leaf);
3696 wret = push_leaf_left(trans, root, path, 1, 1);
3697 if (wret < 0 && wret != -ENOSPC)
3700 if (path->nodes[0] == leaf &&
3701 btrfs_header_nritems(leaf)) {
3702 wret = push_leaf_right(trans, root, path, 1, 1);
3703 if (wret < 0 && wret != -ENOSPC)
3707 if (btrfs_header_nritems(leaf) == 0) {
3708 path->slots[1] = slot;
3709 ret = btrfs_del_leaf(trans, root, path,
3712 free_extent_buffer(leaf);
3714 /* if we're still in the path, make sure
3715 * we're dirty. Otherwise, one of the
3716 * push_leaf functions must have already
3717 * dirtied this buffer
3719 if (path->nodes[0] == leaf)
3720 btrfs_mark_buffer_dirty(leaf);
3721 free_extent_buffer(leaf);
3724 btrfs_mark_buffer_dirty(leaf);
3731 * search the tree again to find a leaf with lesser keys
3732 * returns 0 if it found something or 1 if there are no lesser leaves.
3733 * returns < 0 on io errors.
3735 * This may release the path, and so you may lose any locks held at the
3738 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
3740 struct btrfs_key key;
3741 struct btrfs_disk_key found_key;
3744 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
3748 else if (key.type > 0)
3750 else if (key.objectid > 0)
3755 btrfs_release_path(root, path);
3756 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3759 btrfs_item_key(path->nodes[0], &found_key, 0);
3760 ret = comp_keys(&found_key, &key);
3767 * A helper function to walk down the tree starting at min_key, and looking
3768 * for nodes or leaves that are either in cache or have a minimum
3769 * transaction id. This is used by the btree defrag code, and tree logging
3771 * This does not cow, but it does stuff the starting key it finds back
3772 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3773 * key and get a writable path.
3775 * This does lock as it descends, and path->keep_locks should be set
3776 * to 1 by the caller.
3778 * This honors path->lowest_level to prevent descent past a given level
3781 * min_trans indicates the oldest transaction that you are interested
3782 * in walking through. Any nodes or leaves older than min_trans are
3783 * skipped over (without reading them).
3785 * returns zero if something useful was found, < 0 on error and 1 if there
3786 * was nothing in the tree that matched the search criteria.
3788 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
3789 struct btrfs_key *max_key,
3790 struct btrfs_path *path, int cache_only,
3793 struct extent_buffer *cur;
3794 struct btrfs_key found_key;
3801 WARN_ON(!path->keep_locks);
3803 cur = btrfs_lock_root_node(root);
3804 level = btrfs_header_level(cur);
3805 WARN_ON(path->nodes[level]);
3806 path->nodes[level] = cur;
3807 path->locks[level] = 1;
3809 if (btrfs_header_generation(cur) < min_trans) {
3814 nritems = btrfs_header_nritems(cur);
3815 level = btrfs_header_level(cur);
3816 sret = bin_search(cur, min_key, level, &slot);
3818 /* at the lowest level, we're done, setup the path and exit */
3819 if (level == path->lowest_level) {
3820 if (slot >= nritems)
3823 path->slots[level] = slot;
3824 btrfs_item_key_to_cpu(cur, &found_key, slot);
3827 if (sret && slot > 0)
3830 * check this node pointer against the cache_only and
3831 * min_trans parameters. If it isn't in cache or is too
3832 * old, skip to the next one.
3834 while (slot < nritems) {
3837 struct extent_buffer *tmp;
3838 struct btrfs_disk_key disk_key;
3840 blockptr = btrfs_node_blockptr(cur, slot);
3841 gen = btrfs_node_ptr_generation(cur, slot);
3842 if (gen < min_trans) {
3850 btrfs_node_key(cur, &disk_key, slot);
3851 if (comp_keys(&disk_key, max_key) >= 0) {
3857 tmp = btrfs_find_tree_block(root, blockptr,
3858 btrfs_level_size(root, level - 1));
3860 if (tmp && btrfs_buffer_uptodate(tmp, gen)) {
3861 free_extent_buffer(tmp);
3865 free_extent_buffer(tmp);
3870 * we didn't find a candidate key in this node, walk forward
3871 * and find another one
3873 if (slot >= nritems) {
3874 path->slots[level] = slot;
3875 btrfs_set_path_blocking(path);
3876 sret = btrfs_find_next_key(root, path, min_key, level,
3877 cache_only, min_trans);
3879 btrfs_release_path(root, path);
3885 /* save our key for returning back */
3886 btrfs_node_key_to_cpu(cur, &found_key, slot);
3887 path->slots[level] = slot;
3888 if (level == path->lowest_level) {
3890 unlock_up(path, level, 1);
3893 btrfs_set_path_blocking(path);
3894 cur = read_node_slot(root, cur, slot);
3896 btrfs_tree_lock(cur);
3898 path->locks[level - 1] = 1;
3899 path->nodes[level - 1] = cur;
3900 unlock_up(path, level, 1);
3901 btrfs_clear_path_blocking(path, NULL);
3905 memcpy(min_key, &found_key, sizeof(found_key));
3906 btrfs_set_path_blocking(path);
3911 * this is similar to btrfs_next_leaf, but does not try to preserve
3912 * and fixup the path. It looks for and returns the next key in the
3913 * tree based on the current path and the cache_only and min_trans
3916 * 0 is returned if another key is found, < 0 if there are any errors
3917 * and 1 is returned if there are no higher keys in the tree
3919 * path->keep_locks should be set to 1 on the search made before
3920 * calling this function.
3922 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
3923 struct btrfs_key *key, int lowest_level,
3924 int cache_only, u64 min_trans)
3926 int level = lowest_level;
3928 struct extent_buffer *c;
3930 WARN_ON(!path->keep_locks);
3931 while (level < BTRFS_MAX_LEVEL) {
3932 if (!path->nodes[level])
3935 slot = path->slots[level] + 1;
3936 c = path->nodes[level];
3938 if (slot >= btrfs_header_nritems(c)) {
3940 if (level == BTRFS_MAX_LEVEL)
3945 btrfs_item_key_to_cpu(c, key, slot);
3947 u64 blockptr = btrfs_node_blockptr(c, slot);
3948 u64 gen = btrfs_node_ptr_generation(c, slot);
3951 struct extent_buffer *cur;
3952 cur = btrfs_find_tree_block(root, blockptr,
3953 btrfs_level_size(root, level - 1));
3954 if (!cur || !btrfs_buffer_uptodate(cur, gen)) {
3957 free_extent_buffer(cur);
3960 free_extent_buffer(cur);
3962 if (gen < min_trans) {
3966 btrfs_node_key_to_cpu(c, key, slot);
3974 * search the tree again to find a leaf with greater keys
3975 * returns 0 if it found something or 1 if there are no greater leaves.
3976 * returns < 0 on io errors.
3978 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
3982 struct extent_buffer *c;
3983 struct extent_buffer *next = NULL;
3984 struct btrfs_key key;
3988 nritems = btrfs_header_nritems(path->nodes[0]);
3992 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
3994 btrfs_release_path(root, path);
3995 path->keep_locks = 1;
3996 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
3997 path->keep_locks = 0;
4002 btrfs_set_path_blocking(path);
4003 nritems = btrfs_header_nritems(path->nodes[0]);
4005 * by releasing the path above we dropped all our locks. A balance
4006 * could have added more items next to the key that used to be
4007 * at the very end of the block. So, check again here and
4008 * advance the path if there are now more items available.
4010 if (nritems > 0 && path->slots[0] < nritems - 1) {
4015 while (level < BTRFS_MAX_LEVEL) {
4016 if (!path->nodes[level])
4019 slot = path->slots[level] + 1;
4020 c = path->nodes[level];
4021 if (slot >= btrfs_header_nritems(c)) {
4023 if (level == BTRFS_MAX_LEVEL)
4029 btrfs_tree_unlock(next);
4030 free_extent_buffer(next);
4033 /* the path was set to blocking above */
4034 if (level == 1 && (path->locks[1] || path->skip_locking) &&
4036 reada_for_search(root, path, level, slot, 0);
4038 next = read_node_slot(root, c, slot);
4039 if (!path->skip_locking) {
4040 btrfs_assert_tree_locked(c);
4041 btrfs_tree_lock(next);
4042 btrfs_set_lock_blocking(next);
4046 path->slots[level] = slot;
4049 c = path->nodes[level];
4050 if (path->locks[level])
4051 btrfs_tree_unlock(c);
4052 free_extent_buffer(c);
4053 path->nodes[level] = next;
4054 path->slots[level] = 0;
4055 if (!path->skip_locking)
4056 path->locks[level] = 1;
4060 btrfs_set_path_blocking(path);
4061 if (level == 1 && path->locks[1] && path->reada)
4062 reada_for_search(root, path, level, slot, 0);
4063 next = read_node_slot(root, next, 0);
4064 if (!path->skip_locking) {
4065 btrfs_assert_tree_locked(path->nodes[level]);
4066 btrfs_tree_lock(next);
4067 btrfs_set_lock_blocking(next);
4071 unlock_up(path, 0, 1);
4076 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
4077 * searching until it gets past min_objectid or finds an item of 'type'
4079 * returns 0 if something is found, 1 if nothing was found and < 0 on error
4081 int btrfs_previous_item(struct btrfs_root *root,
4082 struct btrfs_path *path, u64 min_objectid,
4085 struct btrfs_key found_key;
4086 struct extent_buffer *leaf;
4091 if (path->slots[0] == 0) {
4092 btrfs_set_path_blocking(path);
4093 ret = btrfs_prev_leaf(root, path);
4099 leaf = path->nodes[0];
4100 nritems = btrfs_header_nritems(leaf);
4103 if (path->slots[0] == nritems)
4106 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
4107 if (found_key.type == type)
4109 if (found_key.objectid < min_objectid)
4111 if (found_key.objectid == min_objectid &&
4112 found_key.type < type)