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>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle *trans, struct btrfs_root
29 *root, struct btrfs_path *path, int level);
30 static int split_leaf(struct btrfs_trans_handle *trans, struct btrfs_root
31 *root, struct btrfs_key *ins_key,
32 struct btrfs_path *path, int data_size, int extend);
33 static int push_node_left(struct btrfs_trans_handle *trans,
34 struct btrfs_root *root, struct extent_buffer *dst,
35 struct extent_buffer *src, int empty);
36 static int balance_node_right(struct btrfs_trans_handle *trans,
37 struct btrfs_root *root,
38 struct extent_buffer *dst_buf,
39 struct extent_buffer *src_buf);
40 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
42 static void tree_mod_log_free_eb(struct btrfs_fs_info *fs_info,
43 struct extent_buffer *eb);
44 struct extent_buffer *read_old_tree_block(struct btrfs_root *root, u64 bytenr,
45 u32 blocksize, u64 parent_transid,
47 struct extent_buffer *btrfs_find_old_tree_block(struct btrfs_root *root,
48 u64 bytenr, u32 blocksize,
51 struct btrfs_path *btrfs_alloc_path(void)
53 struct btrfs_path *path;
54 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
59 * set all locked nodes in the path to blocking locks. This should
60 * be done before scheduling
62 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
65 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
66 if (!p->nodes[i] || !p->locks[i])
68 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
69 if (p->locks[i] == BTRFS_READ_LOCK)
70 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
71 else if (p->locks[i] == BTRFS_WRITE_LOCK)
72 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
77 * reset all the locked nodes in the patch to spinning locks.
79 * held is used to keep lockdep happy, when lockdep is enabled
80 * we set held to a blocking lock before we go around and
81 * retake all the spinlocks in the path. You can safely use NULL
84 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
85 struct extent_buffer *held, int held_rw)
89 #ifdef CONFIG_DEBUG_LOCK_ALLOC
90 /* lockdep really cares that we take all of these spinlocks
91 * in the right order. If any of the locks in the path are not
92 * currently blocking, it is going to complain. So, make really
93 * really sure by forcing the path to blocking before we clear
97 btrfs_set_lock_blocking_rw(held, held_rw);
98 if (held_rw == BTRFS_WRITE_LOCK)
99 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
100 else if (held_rw == BTRFS_READ_LOCK)
101 held_rw = BTRFS_READ_LOCK_BLOCKING;
103 btrfs_set_path_blocking(p);
106 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
107 if (p->nodes[i] && p->locks[i]) {
108 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
109 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
110 p->locks[i] = BTRFS_WRITE_LOCK;
111 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
112 p->locks[i] = BTRFS_READ_LOCK;
116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 btrfs_clear_lock_blocking_rw(held, held_rw);
122 /* this also releases the path */
123 void btrfs_free_path(struct btrfs_path *p)
127 btrfs_release_path(p);
128 kmem_cache_free(btrfs_path_cachep, p);
132 * path release drops references on the extent buffers in the path
133 * and it drops any locks held by this path
135 * It is safe to call this on paths that no locks or extent buffers held.
137 noinline void btrfs_release_path(struct btrfs_path *p)
141 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
146 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
149 free_extent_buffer(p->nodes[i]);
155 * safely gets a reference on the root node of a tree. A lock
156 * is not taken, so a concurrent writer may put a different node
157 * at the root of the tree. See btrfs_lock_root_node for the
160 * The extent buffer returned by this has a reference taken, so
161 * it won't disappear. It may stop being the root of the tree
162 * at any time because there are no locks held.
164 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
166 struct extent_buffer *eb;
170 eb = rcu_dereference(root->node);
173 * RCU really hurts here, we could free up the root node because
174 * it was cow'ed but we may not get the new root node yet so do
175 * the inc_not_zero dance and if it doesn't work then
176 * synchronize_rcu and try again.
178 if (atomic_inc_not_zero(&eb->refs)) {
188 /* loop around taking references on and locking the root node of the
189 * tree until you end up with a lock on the root. A locked buffer
190 * is returned, with a reference held.
192 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
194 struct extent_buffer *eb;
197 eb = btrfs_root_node(root);
199 if (eb == root->node)
201 btrfs_tree_unlock(eb);
202 free_extent_buffer(eb);
207 /* loop around taking references on and locking the root node of the
208 * tree until you end up with a lock on the root. A locked buffer
209 * is returned, with a reference held.
211 struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
213 struct extent_buffer *eb;
216 eb = btrfs_root_node(root);
217 btrfs_tree_read_lock(eb);
218 if (eb == root->node)
220 btrfs_tree_read_unlock(eb);
221 free_extent_buffer(eb);
226 /* cowonly root (everything not a reference counted cow subvolume), just get
227 * put onto a simple dirty list. transaction.c walks this to make sure they
228 * get properly updated on disk.
230 static void add_root_to_dirty_list(struct btrfs_root *root)
232 spin_lock(&root->fs_info->trans_lock);
233 if (root->track_dirty && list_empty(&root->dirty_list)) {
234 list_add(&root->dirty_list,
235 &root->fs_info->dirty_cowonly_roots);
237 spin_unlock(&root->fs_info->trans_lock);
241 * used by snapshot creation to make a copy of a root for a tree with
242 * a given objectid. The buffer with the new root node is returned in
243 * cow_ret, and this func returns zero on success or a negative error code.
245 int btrfs_copy_root(struct btrfs_trans_handle *trans,
246 struct btrfs_root *root,
247 struct extent_buffer *buf,
248 struct extent_buffer **cow_ret, u64 new_root_objectid)
250 struct extent_buffer *cow;
253 struct btrfs_disk_key disk_key;
255 WARN_ON(root->ref_cows && trans->transid !=
256 root->fs_info->running_transaction->transid);
257 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
259 level = btrfs_header_level(buf);
261 btrfs_item_key(buf, &disk_key, 0);
263 btrfs_node_key(buf, &disk_key, 0);
265 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
266 new_root_objectid, &disk_key, level,
271 copy_extent_buffer(cow, buf, 0, 0, cow->len);
272 btrfs_set_header_bytenr(cow, cow->start);
273 btrfs_set_header_generation(cow, trans->transid);
274 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
275 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
276 BTRFS_HEADER_FLAG_RELOC);
277 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
278 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
280 btrfs_set_header_owner(cow, new_root_objectid);
282 write_extent_buffer(cow, root->fs_info->fsid,
283 (unsigned long)btrfs_header_fsid(cow),
286 WARN_ON(btrfs_header_generation(buf) > trans->transid);
287 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
288 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
290 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
295 btrfs_mark_buffer_dirty(cow);
304 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
305 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
307 MOD_LOG_ROOT_REPLACE,
310 struct tree_mod_move {
315 struct tree_mod_root {
320 struct tree_mod_elem {
322 u64 index; /* shifted logical */
326 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
329 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
332 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
333 struct btrfs_disk_key key;
336 /* this is used for op == MOD_LOG_MOVE_KEYS */
337 struct tree_mod_move move;
339 /* this is used for op == MOD_LOG_ROOT_REPLACE */
340 struct tree_mod_root old_root;
343 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
345 read_lock(&fs_info->tree_mod_log_lock);
348 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
350 read_unlock(&fs_info->tree_mod_log_lock);
353 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
355 write_lock(&fs_info->tree_mod_log_lock);
358 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
360 write_unlock(&fs_info->tree_mod_log_lock);
364 * This adds a new blocker to the tree mod log's blocker list if the @elem
365 * passed does not already have a sequence number set. So when a caller expects
366 * to record tree modifications, it should ensure to set elem->seq to zero
367 * before calling btrfs_get_tree_mod_seq.
368 * Returns a fresh, unused tree log modification sequence number, even if no new
371 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
372 struct seq_list *elem)
376 tree_mod_log_write_lock(fs_info);
377 spin_lock(&fs_info->tree_mod_seq_lock);
379 elem->seq = btrfs_inc_tree_mod_seq(fs_info);
380 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
382 seq = btrfs_inc_tree_mod_seq(fs_info);
383 spin_unlock(&fs_info->tree_mod_seq_lock);
384 tree_mod_log_write_unlock(fs_info);
389 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
390 struct seq_list *elem)
392 struct rb_root *tm_root;
393 struct rb_node *node;
394 struct rb_node *next;
395 struct seq_list *cur_elem;
396 struct tree_mod_elem *tm;
397 u64 min_seq = (u64)-1;
398 u64 seq_putting = elem->seq;
403 spin_lock(&fs_info->tree_mod_seq_lock);
404 list_del(&elem->list);
407 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
408 if (cur_elem->seq < min_seq) {
409 if (seq_putting > cur_elem->seq) {
411 * blocker with lower sequence number exists, we
412 * cannot remove anything from the log
414 spin_unlock(&fs_info->tree_mod_seq_lock);
417 min_seq = cur_elem->seq;
420 spin_unlock(&fs_info->tree_mod_seq_lock);
423 * anything that's lower than the lowest existing (read: blocked)
424 * sequence number can be removed from the tree.
426 tree_mod_log_write_lock(fs_info);
427 tm_root = &fs_info->tree_mod_log;
428 for (node = rb_first(tm_root); node; node = next) {
429 next = rb_next(node);
430 tm = container_of(node, struct tree_mod_elem, node);
431 if (tm->seq > min_seq)
433 rb_erase(node, tm_root);
436 tree_mod_log_write_unlock(fs_info);
440 * key order of the log:
443 * the index is the shifted logical of the *new* root node for root replace
444 * operations, or the shifted logical of the affected block for all other
448 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
450 struct rb_root *tm_root;
451 struct rb_node **new;
452 struct rb_node *parent = NULL;
453 struct tree_mod_elem *cur;
455 BUG_ON(!tm || !tm->seq);
457 tm_root = &fs_info->tree_mod_log;
458 new = &tm_root->rb_node;
460 cur = container_of(*new, struct tree_mod_elem, node);
462 if (cur->index < tm->index)
463 new = &((*new)->rb_left);
464 else if (cur->index > tm->index)
465 new = &((*new)->rb_right);
466 else if (cur->seq < tm->seq)
467 new = &((*new)->rb_left);
468 else if (cur->seq > tm->seq)
469 new = &((*new)->rb_right);
476 rb_link_node(&tm->node, parent, new);
477 rb_insert_color(&tm->node, tm_root);
482 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
483 * returns zero with the tree_mod_log_lock acquired. The caller must hold
484 * this until all tree mod log insertions are recorded in the rb tree and then
485 * call tree_mod_log_write_unlock() to release.
487 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
488 struct extent_buffer *eb) {
490 if (list_empty(&(fs_info)->tree_mod_seq_list))
492 if (eb && btrfs_header_level(eb) == 0)
495 tree_mod_log_write_lock(fs_info);
496 if (list_empty(&fs_info->tree_mod_seq_list)) {
498 * someone emptied the list while we were waiting for the lock.
499 * we must not add to the list when no blocker exists.
501 tree_mod_log_write_unlock(fs_info);
509 * This allocates memory and gets a tree modification sequence number.
511 * Returns <0 on error.
512 * Returns >0 (the added sequence number) on success.
514 static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags,
515 struct tree_mod_elem **tm_ret)
517 struct tree_mod_elem *tm;
520 * once we switch from spin locks to something different, we should
521 * honor the flags parameter here.
523 tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
527 tm->seq = btrfs_inc_tree_mod_seq(fs_info);
532 __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
533 struct extent_buffer *eb, int slot,
534 enum mod_log_op op, gfp_t flags)
537 struct tree_mod_elem *tm;
539 ret = tree_mod_alloc(fs_info, flags, &tm);
543 tm->index = eb->start >> PAGE_CACHE_SHIFT;
544 if (op != MOD_LOG_KEY_ADD) {
545 btrfs_node_key(eb, &tm->key, slot);
546 tm->blockptr = btrfs_node_blockptr(eb, slot);
550 tm->generation = btrfs_node_ptr_generation(eb, slot);
552 return __tree_mod_log_insert(fs_info, tm);
556 tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
557 struct extent_buffer *eb, int slot,
558 enum mod_log_op op, gfp_t flags)
562 if (tree_mod_dont_log(fs_info, eb))
565 ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
567 tree_mod_log_write_unlock(fs_info);
572 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
573 int slot, enum mod_log_op op)
575 return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
579 tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info,
580 struct extent_buffer *eb, int slot,
583 return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
587 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
588 struct extent_buffer *eb, int dst_slot, int src_slot,
589 int nr_items, gfp_t flags)
591 struct tree_mod_elem *tm;
595 if (tree_mod_dont_log(fs_info, eb))
599 * When we override something during the move, we log these removals.
600 * This can only happen when we move towards the beginning of the
601 * buffer, i.e. dst_slot < src_slot.
603 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
604 ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot,
605 MOD_LOG_KEY_REMOVE_WHILE_MOVING);
609 ret = tree_mod_alloc(fs_info, flags, &tm);
613 tm->index = eb->start >> PAGE_CACHE_SHIFT;
615 tm->move.dst_slot = dst_slot;
616 tm->move.nr_items = nr_items;
617 tm->op = MOD_LOG_MOVE_KEYS;
619 ret = __tree_mod_log_insert(fs_info, tm);
621 tree_mod_log_write_unlock(fs_info);
626 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
632 if (btrfs_header_level(eb) == 0)
635 nritems = btrfs_header_nritems(eb);
636 for (i = nritems - 1; i >= 0; i--) {
637 ret = tree_mod_log_insert_key_locked(fs_info, eb, i,
638 MOD_LOG_KEY_REMOVE_WHILE_FREEING);
644 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
645 struct extent_buffer *old_root,
646 struct extent_buffer *new_root, gfp_t flags,
649 struct tree_mod_elem *tm;
652 if (tree_mod_dont_log(fs_info, NULL))
656 __tree_mod_log_free_eb(fs_info, old_root);
658 ret = tree_mod_alloc(fs_info, flags, &tm);
662 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
663 tm->old_root.logical = old_root->start;
664 tm->old_root.level = btrfs_header_level(old_root);
665 tm->generation = btrfs_header_generation(old_root);
666 tm->op = MOD_LOG_ROOT_REPLACE;
668 ret = __tree_mod_log_insert(fs_info, tm);
670 tree_mod_log_write_unlock(fs_info);
674 static struct tree_mod_elem *
675 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
678 struct rb_root *tm_root;
679 struct rb_node *node;
680 struct tree_mod_elem *cur = NULL;
681 struct tree_mod_elem *found = NULL;
682 u64 index = start >> PAGE_CACHE_SHIFT;
684 tree_mod_log_read_lock(fs_info);
685 tm_root = &fs_info->tree_mod_log;
686 node = tm_root->rb_node;
688 cur = container_of(node, struct tree_mod_elem, node);
689 if (cur->index < index) {
690 node = node->rb_left;
691 } else if (cur->index > index) {
692 node = node->rb_right;
693 } else if (cur->seq < min_seq) {
694 node = node->rb_left;
695 } else if (!smallest) {
696 /* we want the node with the highest seq */
698 BUG_ON(found->seq > cur->seq);
700 node = node->rb_left;
701 } else if (cur->seq > min_seq) {
702 /* we want the node with the smallest seq */
704 BUG_ON(found->seq < cur->seq);
706 node = node->rb_right;
712 tree_mod_log_read_unlock(fs_info);
718 * this returns the element from the log with the smallest time sequence
719 * value that's in the log (the oldest log item). any element with a time
720 * sequence lower than min_seq will be ignored.
722 static struct tree_mod_elem *
723 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
726 return __tree_mod_log_search(fs_info, start, min_seq, 1);
730 * this returns the element from the log with the largest time sequence
731 * value that's in the log (the most recent log item). any element with
732 * a time sequence lower than min_seq will be ignored.
734 static struct tree_mod_elem *
735 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
737 return __tree_mod_log_search(fs_info, start, min_seq, 0);
741 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
742 struct extent_buffer *src, unsigned long dst_offset,
743 unsigned long src_offset, int nr_items)
748 if (tree_mod_dont_log(fs_info, NULL))
751 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) {
752 tree_mod_log_write_unlock(fs_info);
756 for (i = 0; i < nr_items; i++) {
757 ret = tree_mod_log_insert_key_locked(fs_info, src,
761 ret = tree_mod_log_insert_key_locked(fs_info, dst,
767 tree_mod_log_write_unlock(fs_info);
771 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
772 int dst_offset, int src_offset, int nr_items)
775 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
781 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
782 struct extent_buffer *eb, int slot, int atomic)
786 ret = tree_mod_log_insert_key_mask(fs_info, eb, slot,
788 atomic ? GFP_ATOMIC : GFP_NOFS);
793 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
795 if (tree_mod_dont_log(fs_info, eb))
798 __tree_mod_log_free_eb(fs_info, eb);
800 tree_mod_log_write_unlock(fs_info);
804 tree_mod_log_set_root_pointer(struct btrfs_root *root,
805 struct extent_buffer *new_root_node,
809 ret = tree_mod_log_insert_root(root->fs_info, root->node,
810 new_root_node, GFP_NOFS, log_removal);
815 * check if the tree block can be shared by multiple trees
817 int btrfs_block_can_be_shared(struct btrfs_root *root,
818 struct extent_buffer *buf)
821 * Tree blocks not in refernece counted trees and tree roots
822 * are never shared. If a block was allocated after the last
823 * snapshot and the block was not allocated by tree relocation,
824 * we know the block is not shared.
826 if (root->ref_cows &&
827 buf != root->node && buf != root->commit_root &&
828 (btrfs_header_generation(buf) <=
829 btrfs_root_last_snapshot(&root->root_item) ||
830 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
832 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
833 if (root->ref_cows &&
834 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
840 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
841 struct btrfs_root *root,
842 struct extent_buffer *buf,
843 struct extent_buffer *cow,
853 * Backrefs update rules:
855 * Always use full backrefs for extent pointers in tree block
856 * allocated by tree relocation.
858 * If a shared tree block is no longer referenced by its owner
859 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
860 * use full backrefs for extent pointers in tree block.
862 * If a tree block is been relocating
863 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
864 * use full backrefs for extent pointers in tree block.
865 * The reason for this is some operations (such as drop tree)
866 * are only allowed for blocks use full backrefs.
869 if (btrfs_block_can_be_shared(root, buf)) {
870 ret = btrfs_lookup_extent_info(trans, root, buf->start,
871 btrfs_header_level(buf), 1,
877 btrfs_std_error(root->fs_info, ret);
882 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
883 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
884 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
889 owner = btrfs_header_owner(buf);
890 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
891 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
894 if ((owner == root->root_key.objectid ||
895 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
896 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
897 ret = btrfs_inc_ref(trans, root, buf, 1, 1);
898 BUG_ON(ret); /* -ENOMEM */
900 if (root->root_key.objectid ==
901 BTRFS_TREE_RELOC_OBJECTID) {
902 ret = btrfs_dec_ref(trans, root, buf, 0, 1);
903 BUG_ON(ret); /* -ENOMEM */
904 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
905 BUG_ON(ret); /* -ENOMEM */
907 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
910 if (root->root_key.objectid ==
911 BTRFS_TREE_RELOC_OBJECTID)
912 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
914 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
915 BUG_ON(ret); /* -ENOMEM */
917 if (new_flags != 0) {
918 ret = btrfs_set_disk_extent_flags(trans, root,
926 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
927 if (root->root_key.objectid ==
928 BTRFS_TREE_RELOC_OBJECTID)
929 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
931 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
932 BUG_ON(ret); /* -ENOMEM */
933 ret = btrfs_dec_ref(trans, root, buf, 1, 1);
934 BUG_ON(ret); /* -ENOMEM */
936 clean_tree_block(trans, root, buf);
943 * does the dirty work in cow of a single block. The parent block (if
944 * supplied) is updated to point to the new cow copy. The new buffer is marked
945 * dirty and returned locked. If you modify the block it needs to be marked
948 * search_start -- an allocation hint for the new block
950 * empty_size -- a hint that you plan on doing more cow. This is the size in
951 * bytes the allocator should try to find free next to the block it returns.
952 * This is just a hint and may be ignored by the allocator.
954 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
955 struct btrfs_root *root,
956 struct extent_buffer *buf,
957 struct extent_buffer *parent, int parent_slot,
958 struct extent_buffer **cow_ret,
959 u64 search_start, u64 empty_size)
961 struct btrfs_disk_key disk_key;
962 struct extent_buffer *cow;
971 btrfs_assert_tree_locked(buf);
973 WARN_ON(root->ref_cows && trans->transid !=
974 root->fs_info->running_transaction->transid);
975 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
977 level = btrfs_header_level(buf);
980 btrfs_item_key(buf, &disk_key, 0);
982 btrfs_node_key(buf, &disk_key, 0);
984 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
986 parent_start = parent->start;
992 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
993 root->root_key.objectid, &disk_key,
994 level, search_start, empty_size);
998 /* cow is set to blocking by btrfs_init_new_buffer */
1000 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1001 btrfs_set_header_bytenr(cow, cow->start);
1002 btrfs_set_header_generation(cow, trans->transid);
1003 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1004 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1005 BTRFS_HEADER_FLAG_RELOC);
1006 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1007 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1009 btrfs_set_header_owner(cow, root->root_key.objectid);
1011 write_extent_buffer(cow, root->fs_info->fsid,
1012 (unsigned long)btrfs_header_fsid(cow),
1015 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1017 btrfs_abort_transaction(trans, root, ret);
1022 btrfs_reloc_cow_block(trans, root, buf, cow);
1024 if (buf == root->node) {
1025 WARN_ON(parent && parent != buf);
1026 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1027 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1028 parent_start = buf->start;
1032 extent_buffer_get(cow);
1033 tree_mod_log_set_root_pointer(root, cow, 1);
1034 rcu_assign_pointer(root->node, cow);
1036 btrfs_free_tree_block(trans, root, buf, parent_start,
1038 free_extent_buffer(buf);
1039 add_root_to_dirty_list(root);
1041 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1042 parent_start = parent->start;
1046 WARN_ON(trans->transid != btrfs_header_generation(parent));
1047 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1048 MOD_LOG_KEY_REPLACE);
1049 btrfs_set_node_blockptr(parent, parent_slot,
1051 btrfs_set_node_ptr_generation(parent, parent_slot,
1053 btrfs_mark_buffer_dirty(parent);
1054 tree_mod_log_free_eb(root->fs_info, buf);
1055 btrfs_free_tree_block(trans, root, buf, parent_start,
1059 btrfs_tree_unlock(buf);
1060 free_extent_buffer_stale(buf);
1061 btrfs_mark_buffer_dirty(cow);
1067 * returns the logical address of the oldest predecessor of the given root.
1068 * entries older than time_seq are ignored.
1070 static struct tree_mod_elem *
1071 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1072 struct extent_buffer *eb_root, u64 time_seq)
1074 struct tree_mod_elem *tm;
1075 struct tree_mod_elem *found = NULL;
1076 u64 root_logical = eb_root->start;
1083 * the very last operation that's logged for a root is the replacement
1084 * operation (if it is replaced at all). this has the index of the *new*
1085 * root, making it the very first operation that's logged for this root.
1088 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1093 * if there are no tree operation for the oldest root, we simply
1094 * return it. this should only happen if that (old) root is at
1101 * if there's an operation that's not a root replacement, we
1102 * found the oldest version of our root. normally, we'll find a
1103 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1105 if (tm->op != MOD_LOG_ROOT_REPLACE)
1109 root_logical = tm->old_root.logical;
1113 /* if there's no old root to return, return what we found instead */
1121 * tm is a pointer to the first operation to rewind within eb. then, all
1122 * previous operations will be rewinded (until we reach something older than
1126 __tree_mod_log_rewind(struct extent_buffer *eb, u64 time_seq,
1127 struct tree_mod_elem *first_tm)
1130 struct rb_node *next;
1131 struct tree_mod_elem *tm = first_tm;
1132 unsigned long o_dst;
1133 unsigned long o_src;
1134 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1136 n = btrfs_header_nritems(eb);
1137 while (tm && tm->seq >= time_seq) {
1139 * all the operations are recorded with the operator used for
1140 * the modification. as we're going backwards, we do the
1141 * opposite of each operation here.
1144 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1145 BUG_ON(tm->slot < n);
1147 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1148 case MOD_LOG_KEY_REMOVE:
1149 btrfs_set_node_key(eb, &tm->key, tm->slot);
1150 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1151 btrfs_set_node_ptr_generation(eb, tm->slot,
1155 case MOD_LOG_KEY_REPLACE:
1156 BUG_ON(tm->slot >= n);
1157 btrfs_set_node_key(eb, &tm->key, tm->slot);
1158 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1159 btrfs_set_node_ptr_generation(eb, tm->slot,
1162 case MOD_LOG_KEY_ADD:
1163 /* if a move operation is needed it's in the log */
1166 case MOD_LOG_MOVE_KEYS:
1167 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1168 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1169 memmove_extent_buffer(eb, o_dst, o_src,
1170 tm->move.nr_items * p_size);
1172 case MOD_LOG_ROOT_REPLACE:
1174 * this operation is special. for roots, this must be
1175 * handled explicitly before rewinding.
1176 * for non-roots, this operation may exist if the node
1177 * was a root: root A -> child B; then A gets empty and
1178 * B is promoted to the new root. in the mod log, we'll
1179 * have a root-replace operation for B, a tree block
1180 * that is no root. we simply ignore that operation.
1184 next = rb_next(&tm->node);
1187 tm = container_of(next, struct tree_mod_elem, node);
1188 if (tm->index != first_tm->index)
1191 btrfs_set_header_nritems(eb, n);
1195 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1196 * is returned. If rewind operations happen, a fresh buffer is returned. The
1197 * returned buffer is always read-locked. If the returned buffer is not the
1198 * input buffer, the lock on the input buffer is released and the input buffer
1199 * is freed (its refcount is decremented).
1201 static struct extent_buffer *
1202 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1205 struct extent_buffer *eb_rewin;
1206 struct tree_mod_elem *tm;
1211 if (btrfs_header_level(eb) == 0)
1214 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1218 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1219 BUG_ON(tm->slot != 0);
1220 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1221 fs_info->tree_root->nodesize);
1223 btrfs_set_header_bytenr(eb_rewin, eb->start);
1224 btrfs_set_header_backref_rev(eb_rewin,
1225 btrfs_header_backref_rev(eb));
1226 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1227 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1229 eb_rewin = btrfs_clone_extent_buffer(eb);
1233 extent_buffer_get(eb_rewin);
1234 btrfs_tree_read_unlock(eb);
1235 free_extent_buffer(eb);
1237 extent_buffer_get(eb_rewin);
1238 btrfs_tree_read_lock(eb_rewin);
1239 __tree_mod_log_rewind(eb_rewin, time_seq, tm);
1240 WARN_ON(btrfs_header_nritems(eb_rewin) >
1241 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1247 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1248 * value. If there are no changes, the current root->root_node is returned. If
1249 * anything changed in between, there's a fresh buffer allocated on which the
1250 * rewind operations are done. In any case, the returned buffer is read locked.
1251 * Returns NULL on error (with no locks held).
1253 static inline struct extent_buffer *
1254 get_old_root(struct btrfs_root *root, u64 time_seq)
1256 struct tree_mod_elem *tm;
1257 struct extent_buffer *eb = NULL;
1258 struct extent_buffer *eb_root;
1259 struct extent_buffer *old;
1260 struct tree_mod_root *old_root = NULL;
1261 u64 old_generation = 0;
1265 eb_root = btrfs_read_lock_root_node(root);
1266 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1270 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1271 old_root = &tm->old_root;
1272 old_generation = tm->generation;
1273 logical = old_root->logical;
1275 logical = eb_root->start;
1278 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1279 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1280 btrfs_tree_read_unlock(eb_root);
1281 free_extent_buffer(eb_root);
1282 blocksize = btrfs_level_size(root, old_root->level);
1283 old = read_tree_block(root, logical, blocksize, 0);
1284 if (!old || !extent_buffer_uptodate(old)) {
1285 free_extent_buffer(old);
1286 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1290 eb = btrfs_clone_extent_buffer(old);
1291 free_extent_buffer(old);
1293 } else if (old_root) {
1294 btrfs_tree_read_unlock(eb_root);
1295 free_extent_buffer(eb_root);
1296 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1298 eb = btrfs_clone_extent_buffer(eb_root);
1299 btrfs_tree_read_unlock(eb_root);
1300 free_extent_buffer(eb_root);
1305 extent_buffer_get(eb);
1306 btrfs_tree_read_lock(eb);
1308 btrfs_set_header_bytenr(eb, eb->start);
1309 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1310 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1311 btrfs_set_header_level(eb, old_root->level);
1312 btrfs_set_header_generation(eb, old_generation);
1315 __tree_mod_log_rewind(eb, time_seq, tm);
1317 WARN_ON(btrfs_header_level(eb) != 0);
1318 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1323 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1325 struct tree_mod_elem *tm;
1327 struct extent_buffer *eb_root = btrfs_root_node(root);
1329 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1330 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1331 level = tm->old_root.level;
1333 level = btrfs_header_level(eb_root);
1335 free_extent_buffer(eb_root);
1340 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1341 struct btrfs_root *root,
1342 struct extent_buffer *buf)
1344 /* ensure we can see the force_cow */
1348 * We do not need to cow a block if
1349 * 1) this block is not created or changed in this transaction;
1350 * 2) this block does not belong to TREE_RELOC tree;
1351 * 3) the root is not forced COW.
1353 * What is forced COW:
1354 * when we create snapshot during commiting the transaction,
1355 * after we've finished coping src root, we must COW the shared
1356 * block to ensure the metadata consistency.
1358 if (btrfs_header_generation(buf) == trans->transid &&
1359 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1360 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1361 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1368 * cows a single block, see __btrfs_cow_block for the real work.
1369 * This version of it has extra checks so that a block isn't cow'd more than
1370 * once per transaction, as long as it hasn't been written yet
1372 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1373 struct btrfs_root *root, struct extent_buffer *buf,
1374 struct extent_buffer *parent, int parent_slot,
1375 struct extent_buffer **cow_ret)
1380 if (trans->transaction != root->fs_info->running_transaction)
1381 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1382 (unsigned long long)trans->transid,
1383 (unsigned long long)
1384 root->fs_info->running_transaction->transid);
1386 if (trans->transid != root->fs_info->generation)
1387 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1388 (unsigned long long)trans->transid,
1389 (unsigned long long)root->fs_info->generation);
1391 if (!should_cow_block(trans, root, buf)) {
1396 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1399 btrfs_set_lock_blocking(parent);
1400 btrfs_set_lock_blocking(buf);
1402 ret = __btrfs_cow_block(trans, root, buf, parent,
1403 parent_slot, cow_ret, search_start, 0);
1405 trace_btrfs_cow_block(root, buf, *cow_ret);
1411 * helper function for defrag to decide if two blocks pointed to by a
1412 * node are actually close by
1414 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1416 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1418 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1424 * compare two keys in a memcmp fashion
1426 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1428 struct btrfs_key k1;
1430 btrfs_disk_key_to_cpu(&k1, disk);
1432 return btrfs_comp_cpu_keys(&k1, k2);
1436 * same as comp_keys only with two btrfs_key's
1438 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1440 if (k1->objectid > k2->objectid)
1442 if (k1->objectid < k2->objectid)
1444 if (k1->type > k2->type)
1446 if (k1->type < k2->type)
1448 if (k1->offset > k2->offset)
1450 if (k1->offset < k2->offset)
1456 * this is used by the defrag code to go through all the
1457 * leaves pointed to by a node and reallocate them so that
1458 * disk order is close to key order
1460 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1461 struct btrfs_root *root, struct extent_buffer *parent,
1462 int start_slot, u64 *last_ret,
1463 struct btrfs_key *progress)
1465 struct extent_buffer *cur;
1468 u64 search_start = *last_ret;
1478 int progress_passed = 0;
1479 struct btrfs_disk_key disk_key;
1481 parent_level = btrfs_header_level(parent);
1483 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1484 WARN_ON(trans->transid != root->fs_info->generation);
1486 parent_nritems = btrfs_header_nritems(parent);
1487 blocksize = btrfs_level_size(root, parent_level - 1);
1488 end_slot = parent_nritems;
1490 if (parent_nritems == 1)
1493 btrfs_set_lock_blocking(parent);
1495 for (i = start_slot; i < end_slot; i++) {
1498 btrfs_node_key(parent, &disk_key, i);
1499 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1502 progress_passed = 1;
1503 blocknr = btrfs_node_blockptr(parent, i);
1504 gen = btrfs_node_ptr_generation(parent, i);
1505 if (last_block == 0)
1506 last_block = blocknr;
1509 other = btrfs_node_blockptr(parent, i - 1);
1510 close = close_blocks(blocknr, other, blocksize);
1512 if (!close && i < end_slot - 2) {
1513 other = btrfs_node_blockptr(parent, i + 1);
1514 close = close_blocks(blocknr, other, blocksize);
1517 last_block = blocknr;
1521 cur = btrfs_find_tree_block(root, blocknr, blocksize);
1523 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1526 if (!cur || !uptodate) {
1528 cur = read_tree_block(root, blocknr,
1530 if (!cur || !extent_buffer_uptodate(cur)) {
1531 free_extent_buffer(cur);
1534 } else if (!uptodate) {
1535 err = btrfs_read_buffer(cur, gen);
1537 free_extent_buffer(cur);
1542 if (search_start == 0)
1543 search_start = last_block;
1545 btrfs_tree_lock(cur);
1546 btrfs_set_lock_blocking(cur);
1547 err = __btrfs_cow_block(trans, root, cur, parent, i,
1550 (end_slot - i) * blocksize));
1552 btrfs_tree_unlock(cur);
1553 free_extent_buffer(cur);
1556 search_start = cur->start;
1557 last_block = cur->start;
1558 *last_ret = search_start;
1559 btrfs_tree_unlock(cur);
1560 free_extent_buffer(cur);
1566 * The leaf data grows from end-to-front in the node.
1567 * this returns the address of the start of the last item,
1568 * which is the stop of the leaf data stack
1570 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1571 struct extent_buffer *leaf)
1573 u32 nr = btrfs_header_nritems(leaf);
1575 return BTRFS_LEAF_DATA_SIZE(root);
1576 return btrfs_item_offset_nr(leaf, nr - 1);
1581 * search for key in the extent_buffer. The items start at offset p,
1582 * and they are item_size apart. There are 'max' items in p.
1584 * the slot in the array is returned via slot, and it points to
1585 * the place where you would insert key if it is not found in
1588 * slot may point to max if the key is bigger than all of the keys
1590 static noinline int generic_bin_search(struct extent_buffer *eb,
1592 int item_size, struct btrfs_key *key,
1599 struct btrfs_disk_key *tmp = NULL;
1600 struct btrfs_disk_key unaligned;
1601 unsigned long offset;
1603 unsigned long map_start = 0;
1604 unsigned long map_len = 0;
1607 while (low < high) {
1608 mid = (low + high) / 2;
1609 offset = p + mid * item_size;
1611 if (!kaddr || offset < map_start ||
1612 (offset + sizeof(struct btrfs_disk_key)) >
1613 map_start + map_len) {
1615 err = map_private_extent_buffer(eb, offset,
1616 sizeof(struct btrfs_disk_key),
1617 &kaddr, &map_start, &map_len);
1620 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1623 read_extent_buffer(eb, &unaligned,
1624 offset, sizeof(unaligned));
1629 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1632 ret = comp_keys(tmp, key);
1648 * simple bin_search frontend that does the right thing for
1651 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1652 int level, int *slot)
1655 return generic_bin_search(eb,
1656 offsetof(struct btrfs_leaf, items),
1657 sizeof(struct btrfs_item),
1658 key, btrfs_header_nritems(eb),
1661 return generic_bin_search(eb,
1662 offsetof(struct btrfs_node, ptrs),
1663 sizeof(struct btrfs_key_ptr),
1664 key, btrfs_header_nritems(eb),
1668 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1669 int level, int *slot)
1671 return bin_search(eb, key, level, slot);
1674 static void root_add_used(struct btrfs_root *root, u32 size)
1676 spin_lock(&root->accounting_lock);
1677 btrfs_set_root_used(&root->root_item,
1678 btrfs_root_used(&root->root_item) + size);
1679 spin_unlock(&root->accounting_lock);
1682 static void root_sub_used(struct btrfs_root *root, u32 size)
1684 spin_lock(&root->accounting_lock);
1685 btrfs_set_root_used(&root->root_item,
1686 btrfs_root_used(&root->root_item) - size);
1687 spin_unlock(&root->accounting_lock);
1690 /* given a node and slot number, this reads the blocks it points to. The
1691 * extent buffer is returned with a reference taken (but unlocked).
1692 * NULL is returned on error.
1694 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1695 struct extent_buffer *parent, int slot)
1697 int level = btrfs_header_level(parent);
1698 struct extent_buffer *eb;
1702 if (slot >= btrfs_header_nritems(parent))
1707 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1708 btrfs_level_size(root, level - 1),
1709 btrfs_node_ptr_generation(parent, slot));
1710 if (eb && !extent_buffer_uptodate(eb)) {
1711 free_extent_buffer(eb);
1719 * node level balancing, used to make sure nodes are in proper order for
1720 * item deletion. We balance from the top down, so we have to make sure
1721 * that a deletion won't leave an node completely empty later on.
1723 static noinline int balance_level(struct btrfs_trans_handle *trans,
1724 struct btrfs_root *root,
1725 struct btrfs_path *path, int level)
1727 struct extent_buffer *right = NULL;
1728 struct extent_buffer *mid;
1729 struct extent_buffer *left = NULL;
1730 struct extent_buffer *parent = NULL;
1734 int orig_slot = path->slots[level];
1740 mid = path->nodes[level];
1742 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1743 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1744 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1746 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1748 if (level < BTRFS_MAX_LEVEL - 1) {
1749 parent = path->nodes[level + 1];
1750 pslot = path->slots[level + 1];
1754 * deal with the case where there is only one pointer in the root
1755 * by promoting the node below to a root
1758 struct extent_buffer *child;
1760 if (btrfs_header_nritems(mid) != 1)
1763 /* promote the child to a root */
1764 child = read_node_slot(root, mid, 0);
1767 btrfs_std_error(root->fs_info, ret);
1771 btrfs_tree_lock(child);
1772 btrfs_set_lock_blocking(child);
1773 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1775 btrfs_tree_unlock(child);
1776 free_extent_buffer(child);
1780 tree_mod_log_set_root_pointer(root, child, 1);
1781 rcu_assign_pointer(root->node, child);
1783 add_root_to_dirty_list(root);
1784 btrfs_tree_unlock(child);
1786 path->locks[level] = 0;
1787 path->nodes[level] = NULL;
1788 clean_tree_block(trans, root, mid);
1789 btrfs_tree_unlock(mid);
1790 /* once for the path */
1791 free_extent_buffer(mid);
1793 root_sub_used(root, mid->len);
1794 btrfs_free_tree_block(trans, root, mid, 0, 1);
1795 /* once for the root ptr */
1796 free_extent_buffer_stale(mid);
1799 if (btrfs_header_nritems(mid) >
1800 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1803 left = read_node_slot(root, parent, pslot - 1);
1805 btrfs_tree_lock(left);
1806 btrfs_set_lock_blocking(left);
1807 wret = btrfs_cow_block(trans, root, left,
1808 parent, pslot - 1, &left);
1814 right = read_node_slot(root, parent, pslot + 1);
1816 btrfs_tree_lock(right);
1817 btrfs_set_lock_blocking(right);
1818 wret = btrfs_cow_block(trans, root, right,
1819 parent, pslot + 1, &right);
1826 /* first, try to make some room in the middle buffer */
1828 orig_slot += btrfs_header_nritems(left);
1829 wret = push_node_left(trans, root, left, mid, 1);
1835 * then try to empty the right most buffer into the middle
1838 wret = push_node_left(trans, root, mid, right, 1);
1839 if (wret < 0 && wret != -ENOSPC)
1841 if (btrfs_header_nritems(right) == 0) {
1842 clean_tree_block(trans, root, right);
1843 btrfs_tree_unlock(right);
1844 del_ptr(root, path, level + 1, pslot + 1);
1845 root_sub_used(root, right->len);
1846 btrfs_free_tree_block(trans, root, right, 0, 1);
1847 free_extent_buffer_stale(right);
1850 struct btrfs_disk_key right_key;
1851 btrfs_node_key(right, &right_key, 0);
1852 tree_mod_log_set_node_key(root->fs_info, parent,
1854 btrfs_set_node_key(parent, &right_key, pslot + 1);
1855 btrfs_mark_buffer_dirty(parent);
1858 if (btrfs_header_nritems(mid) == 1) {
1860 * we're not allowed to leave a node with one item in the
1861 * tree during a delete. A deletion from lower in the tree
1862 * could try to delete the only pointer in this node.
1863 * So, pull some keys from the left.
1864 * There has to be a left pointer at this point because
1865 * otherwise we would have pulled some pointers from the
1870 btrfs_std_error(root->fs_info, ret);
1873 wret = balance_node_right(trans, root, mid, left);
1879 wret = push_node_left(trans, root, left, mid, 1);
1885 if (btrfs_header_nritems(mid) == 0) {
1886 clean_tree_block(trans, root, mid);
1887 btrfs_tree_unlock(mid);
1888 del_ptr(root, path, level + 1, pslot);
1889 root_sub_used(root, mid->len);
1890 btrfs_free_tree_block(trans, root, mid, 0, 1);
1891 free_extent_buffer_stale(mid);
1894 /* update the parent key to reflect our changes */
1895 struct btrfs_disk_key mid_key;
1896 btrfs_node_key(mid, &mid_key, 0);
1897 tree_mod_log_set_node_key(root->fs_info, parent,
1899 btrfs_set_node_key(parent, &mid_key, pslot);
1900 btrfs_mark_buffer_dirty(parent);
1903 /* update the path */
1905 if (btrfs_header_nritems(left) > orig_slot) {
1906 extent_buffer_get(left);
1907 /* left was locked after cow */
1908 path->nodes[level] = left;
1909 path->slots[level + 1] -= 1;
1910 path->slots[level] = orig_slot;
1912 btrfs_tree_unlock(mid);
1913 free_extent_buffer(mid);
1916 orig_slot -= btrfs_header_nritems(left);
1917 path->slots[level] = orig_slot;
1920 /* double check we haven't messed things up */
1922 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1926 btrfs_tree_unlock(right);
1927 free_extent_buffer(right);
1930 if (path->nodes[level] != left)
1931 btrfs_tree_unlock(left);
1932 free_extent_buffer(left);
1937 /* Node balancing for insertion. Here we only split or push nodes around
1938 * when they are completely full. This is also done top down, so we
1939 * have to be pessimistic.
1941 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1942 struct btrfs_root *root,
1943 struct btrfs_path *path, int level)
1945 struct extent_buffer *right = NULL;
1946 struct extent_buffer *mid;
1947 struct extent_buffer *left = NULL;
1948 struct extent_buffer *parent = NULL;
1952 int orig_slot = path->slots[level];
1957 mid = path->nodes[level];
1958 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1960 if (level < BTRFS_MAX_LEVEL - 1) {
1961 parent = path->nodes[level + 1];
1962 pslot = path->slots[level + 1];
1968 left = read_node_slot(root, parent, pslot - 1);
1970 /* first, try to make some room in the middle buffer */
1974 btrfs_tree_lock(left);
1975 btrfs_set_lock_blocking(left);
1977 left_nr = btrfs_header_nritems(left);
1978 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
1981 ret = btrfs_cow_block(trans, root, left, parent,
1986 wret = push_node_left(trans, root,
1993 struct btrfs_disk_key disk_key;
1994 orig_slot += left_nr;
1995 btrfs_node_key(mid, &disk_key, 0);
1996 tree_mod_log_set_node_key(root->fs_info, parent,
1998 btrfs_set_node_key(parent, &disk_key, pslot);
1999 btrfs_mark_buffer_dirty(parent);
2000 if (btrfs_header_nritems(left) > orig_slot) {
2001 path->nodes[level] = left;
2002 path->slots[level + 1] -= 1;
2003 path->slots[level] = orig_slot;
2004 btrfs_tree_unlock(mid);
2005 free_extent_buffer(mid);
2008 btrfs_header_nritems(left);
2009 path->slots[level] = orig_slot;
2010 btrfs_tree_unlock(left);
2011 free_extent_buffer(left);
2015 btrfs_tree_unlock(left);
2016 free_extent_buffer(left);
2018 right = read_node_slot(root, parent, pslot + 1);
2021 * then try to empty the right most buffer into the middle
2026 btrfs_tree_lock(right);
2027 btrfs_set_lock_blocking(right);
2029 right_nr = btrfs_header_nritems(right);
2030 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2033 ret = btrfs_cow_block(trans, root, right,
2039 wret = balance_node_right(trans, root,
2046 struct btrfs_disk_key disk_key;
2048 btrfs_node_key(right, &disk_key, 0);
2049 tree_mod_log_set_node_key(root->fs_info, parent,
2051 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2052 btrfs_mark_buffer_dirty(parent);
2054 if (btrfs_header_nritems(mid) <= orig_slot) {
2055 path->nodes[level] = right;
2056 path->slots[level + 1] += 1;
2057 path->slots[level] = orig_slot -
2058 btrfs_header_nritems(mid);
2059 btrfs_tree_unlock(mid);
2060 free_extent_buffer(mid);
2062 btrfs_tree_unlock(right);
2063 free_extent_buffer(right);
2067 btrfs_tree_unlock(right);
2068 free_extent_buffer(right);
2074 * readahead one full node of leaves, finding things that are close
2075 * to the block in 'slot', and triggering ra on them.
2077 static void reada_for_search(struct btrfs_root *root,
2078 struct btrfs_path *path,
2079 int level, int slot, u64 objectid)
2081 struct extent_buffer *node;
2082 struct btrfs_disk_key disk_key;
2088 int direction = path->reada;
2089 struct extent_buffer *eb;
2097 if (!path->nodes[level])
2100 node = path->nodes[level];
2102 search = btrfs_node_blockptr(node, slot);
2103 blocksize = btrfs_level_size(root, level - 1);
2104 eb = btrfs_find_tree_block(root, search, blocksize);
2106 free_extent_buffer(eb);
2112 nritems = btrfs_header_nritems(node);
2116 if (direction < 0) {
2120 } else if (direction > 0) {
2125 if (path->reada < 0 && objectid) {
2126 btrfs_node_key(node, &disk_key, nr);
2127 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2130 search = btrfs_node_blockptr(node, nr);
2131 if ((search <= target && target - search <= 65536) ||
2132 (search > target && search - target <= 65536)) {
2133 gen = btrfs_node_ptr_generation(node, nr);
2134 readahead_tree_block(root, search, blocksize, gen);
2138 if ((nread > 65536 || nscan > 32))
2144 * returns -EAGAIN if it had to drop the path, or zero if everything was in
2147 static noinline int reada_for_balance(struct btrfs_root *root,
2148 struct btrfs_path *path, int level)
2152 struct extent_buffer *parent;
2153 struct extent_buffer *eb;
2160 parent = path->nodes[level + 1];
2164 nritems = btrfs_header_nritems(parent);
2165 slot = path->slots[level + 1];
2166 blocksize = btrfs_level_size(root, level);
2169 block1 = btrfs_node_blockptr(parent, slot - 1);
2170 gen = btrfs_node_ptr_generation(parent, slot - 1);
2171 eb = btrfs_find_tree_block(root, block1, blocksize);
2173 * if we get -eagain from btrfs_buffer_uptodate, we
2174 * don't want to return eagain here. That will loop
2177 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2179 free_extent_buffer(eb);
2181 if (slot + 1 < nritems) {
2182 block2 = btrfs_node_blockptr(parent, slot + 1);
2183 gen = btrfs_node_ptr_generation(parent, slot + 1);
2184 eb = btrfs_find_tree_block(root, block2, blocksize);
2185 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2187 free_extent_buffer(eb);
2189 if (block1 || block2) {
2192 /* release the whole path */
2193 btrfs_release_path(path);
2195 /* read the blocks */
2197 readahead_tree_block(root, block1, blocksize, 0);
2199 readahead_tree_block(root, block2, blocksize, 0);
2202 eb = read_tree_block(root, block1, blocksize, 0);
2203 free_extent_buffer(eb);
2206 eb = read_tree_block(root, block2, blocksize, 0);
2207 free_extent_buffer(eb);
2215 * when we walk down the tree, it is usually safe to unlock the higher layers
2216 * in the tree. The exceptions are when our path goes through slot 0, because
2217 * operations on the tree might require changing key pointers higher up in the
2220 * callers might also have set path->keep_locks, which tells this code to keep
2221 * the lock if the path points to the last slot in the block. This is part of
2222 * walking through the tree, and selecting the next slot in the higher block.
2224 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2225 * if lowest_unlock is 1, level 0 won't be unlocked
2227 static noinline void unlock_up(struct btrfs_path *path, int level,
2228 int lowest_unlock, int min_write_lock_level,
2229 int *write_lock_level)
2232 int skip_level = level;
2234 struct extent_buffer *t;
2236 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2237 if (!path->nodes[i])
2239 if (!path->locks[i])
2241 if (!no_skips && path->slots[i] == 0) {
2245 if (!no_skips && path->keep_locks) {
2248 nritems = btrfs_header_nritems(t);
2249 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2254 if (skip_level < i && i >= lowest_unlock)
2258 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2259 btrfs_tree_unlock_rw(t, path->locks[i]);
2261 if (write_lock_level &&
2262 i > min_write_lock_level &&
2263 i <= *write_lock_level) {
2264 *write_lock_level = i - 1;
2271 * This releases any locks held in the path starting at level and
2272 * going all the way up to the root.
2274 * btrfs_search_slot will keep the lock held on higher nodes in a few
2275 * corner cases, such as COW of the block at slot zero in the node. This
2276 * ignores those rules, and it should only be called when there are no
2277 * more updates to be done higher up in the tree.
2279 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2283 if (path->keep_locks)
2286 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2287 if (!path->nodes[i])
2289 if (!path->locks[i])
2291 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2297 * helper function for btrfs_search_slot. The goal is to find a block
2298 * in cache without setting the path to blocking. If we find the block
2299 * we return zero and the path is unchanged.
2301 * If we can't find the block, we set the path blocking and do some
2302 * reada. -EAGAIN is returned and the search must be repeated.
2305 read_block_for_search(struct btrfs_trans_handle *trans,
2306 struct btrfs_root *root, struct btrfs_path *p,
2307 struct extent_buffer **eb_ret, int level, int slot,
2308 struct btrfs_key *key, u64 time_seq)
2313 struct extent_buffer *b = *eb_ret;
2314 struct extent_buffer *tmp;
2317 blocknr = btrfs_node_blockptr(b, slot);
2318 gen = btrfs_node_ptr_generation(b, slot);
2319 blocksize = btrfs_level_size(root, level - 1);
2321 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2323 /* first we do an atomic uptodate check */
2324 if (btrfs_buffer_uptodate(tmp, 0, 1) > 0) {
2325 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2327 * we found an up to date block without
2334 /* the pages were up to date, but we failed
2335 * the generation number check. Do a full
2336 * read for the generation number that is correct.
2337 * We must do this without dropping locks so
2338 * we can trust our generation number
2340 free_extent_buffer(tmp);
2341 btrfs_set_path_blocking(p);
2343 /* now we're allowed to do a blocking uptodate check */
2344 tmp = read_tree_block(root, blocknr, blocksize, gen);
2345 if (tmp && btrfs_buffer_uptodate(tmp, gen, 0) > 0) {
2349 free_extent_buffer(tmp);
2350 btrfs_release_path(p);
2356 * reduce lock contention at high levels
2357 * of the btree by dropping locks before
2358 * we read. Don't release the lock on the current
2359 * level because we need to walk this node to figure
2360 * out which blocks to read.
2362 btrfs_unlock_up_safe(p, level + 1);
2363 btrfs_set_path_blocking(p);
2365 free_extent_buffer(tmp);
2367 reada_for_search(root, p, level, slot, key->objectid);
2369 btrfs_release_path(p);
2372 tmp = read_tree_block(root, blocknr, blocksize, 0);
2375 * If the read above didn't mark this buffer up to date,
2376 * it will never end up being up to date. Set ret to EIO now
2377 * and give up so that our caller doesn't loop forever
2380 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2382 free_extent_buffer(tmp);
2388 * helper function for btrfs_search_slot. This does all of the checks
2389 * for node-level blocks and does any balancing required based on
2392 * If no extra work was required, zero is returned. If we had to
2393 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2397 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2398 struct btrfs_root *root, struct btrfs_path *p,
2399 struct extent_buffer *b, int level, int ins_len,
2400 int *write_lock_level)
2403 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2404 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2407 if (*write_lock_level < level + 1) {
2408 *write_lock_level = level + 1;
2409 btrfs_release_path(p);
2413 sret = reada_for_balance(root, p, level);
2417 btrfs_set_path_blocking(p);
2418 sret = split_node(trans, root, p, level);
2419 btrfs_clear_path_blocking(p, NULL, 0);
2426 b = p->nodes[level];
2427 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2428 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2431 if (*write_lock_level < level + 1) {
2432 *write_lock_level = level + 1;
2433 btrfs_release_path(p);
2437 sret = reada_for_balance(root, p, level);
2441 btrfs_set_path_blocking(p);
2442 sret = balance_level(trans, root, p, level);
2443 btrfs_clear_path_blocking(p, NULL, 0);
2449 b = p->nodes[level];
2451 btrfs_release_path(p);
2454 BUG_ON(btrfs_header_nritems(b) == 1);
2465 * look for key in the tree. path is filled in with nodes along the way
2466 * if key is found, we return zero and you can find the item in the leaf
2467 * level of the path (level 0)
2469 * If the key isn't found, the path points to the slot where it should
2470 * be inserted, and 1 is returned. If there are other errors during the
2471 * search a negative error number is returned.
2473 * if ins_len > 0, nodes and leaves will be split as we walk down the
2474 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2477 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2478 *root, struct btrfs_key *key, struct btrfs_path *p, int
2481 struct extent_buffer *b;
2486 int lowest_unlock = 1;
2488 /* everything at write_lock_level or lower must be write locked */
2489 int write_lock_level = 0;
2490 u8 lowest_level = 0;
2491 int min_write_lock_level;
2493 lowest_level = p->lowest_level;
2494 WARN_ON(lowest_level && ins_len > 0);
2495 WARN_ON(p->nodes[0] != NULL);
2500 /* when we are removing items, we might have to go up to level
2501 * two as we update tree pointers Make sure we keep write
2502 * for those levels as well
2504 write_lock_level = 2;
2505 } else if (ins_len > 0) {
2507 * for inserting items, make sure we have a write lock on
2508 * level 1 so we can update keys
2510 write_lock_level = 1;
2514 write_lock_level = -1;
2516 if (cow && (p->keep_locks || p->lowest_level))
2517 write_lock_level = BTRFS_MAX_LEVEL;
2519 min_write_lock_level = write_lock_level;
2523 * we try very hard to do read locks on the root
2525 root_lock = BTRFS_READ_LOCK;
2527 if (p->search_commit_root) {
2529 * the commit roots are read only
2530 * so we always do read locks
2532 b = root->commit_root;
2533 extent_buffer_get(b);
2534 level = btrfs_header_level(b);
2535 if (!p->skip_locking)
2536 btrfs_tree_read_lock(b);
2538 if (p->skip_locking) {
2539 b = btrfs_root_node(root);
2540 level = btrfs_header_level(b);
2542 /* we don't know the level of the root node
2543 * until we actually have it read locked
2545 b = btrfs_read_lock_root_node(root);
2546 level = btrfs_header_level(b);
2547 if (level <= write_lock_level) {
2548 /* whoops, must trade for write lock */
2549 btrfs_tree_read_unlock(b);
2550 free_extent_buffer(b);
2551 b = btrfs_lock_root_node(root);
2552 root_lock = BTRFS_WRITE_LOCK;
2554 /* the level might have changed, check again */
2555 level = btrfs_header_level(b);
2559 p->nodes[level] = b;
2560 if (!p->skip_locking)
2561 p->locks[level] = root_lock;
2564 level = btrfs_header_level(b);
2567 * setup the path here so we can release it under lock
2568 * contention with the cow code
2572 * if we don't really need to cow this block
2573 * then we don't want to set the path blocking,
2574 * so we test it here
2576 if (!should_cow_block(trans, root, b))
2579 btrfs_set_path_blocking(p);
2582 * must have write locks on this node and the
2585 if (level > write_lock_level ||
2586 (level + 1 > write_lock_level &&
2587 level + 1 < BTRFS_MAX_LEVEL &&
2588 p->nodes[level + 1])) {
2589 write_lock_level = level + 1;
2590 btrfs_release_path(p);
2594 err = btrfs_cow_block(trans, root, b,
2595 p->nodes[level + 1],
2596 p->slots[level + 1], &b);
2603 BUG_ON(!cow && ins_len);
2605 p->nodes[level] = b;
2606 btrfs_clear_path_blocking(p, NULL, 0);
2609 * we have a lock on b and as long as we aren't changing
2610 * the tree, there is no way to for the items in b to change.
2611 * It is safe to drop the lock on our parent before we
2612 * go through the expensive btree search on b.
2614 * If cow is true, then we might be changing slot zero,
2615 * which may require changing the parent. So, we can't
2616 * drop the lock until after we know which slot we're
2620 btrfs_unlock_up_safe(p, level + 1);
2622 ret = bin_search(b, key, level, &slot);
2626 if (ret && slot > 0) {
2630 p->slots[level] = slot;
2631 err = setup_nodes_for_search(trans, root, p, b, level,
2632 ins_len, &write_lock_level);
2639 b = p->nodes[level];
2640 slot = p->slots[level];
2643 * slot 0 is special, if we change the key
2644 * we have to update the parent pointer
2645 * which means we must have a write lock
2648 if (slot == 0 && cow &&
2649 write_lock_level < level + 1) {
2650 write_lock_level = level + 1;
2651 btrfs_release_path(p);
2655 unlock_up(p, level, lowest_unlock,
2656 min_write_lock_level, &write_lock_level);
2658 if (level == lowest_level) {
2664 err = read_block_for_search(trans, root, p,
2665 &b, level, slot, key, 0);
2673 if (!p->skip_locking) {
2674 level = btrfs_header_level(b);
2675 if (level <= write_lock_level) {
2676 err = btrfs_try_tree_write_lock(b);
2678 btrfs_set_path_blocking(p);
2680 btrfs_clear_path_blocking(p, b,
2683 p->locks[level] = BTRFS_WRITE_LOCK;
2685 err = btrfs_try_tree_read_lock(b);
2687 btrfs_set_path_blocking(p);
2688 btrfs_tree_read_lock(b);
2689 btrfs_clear_path_blocking(p, b,
2692 p->locks[level] = BTRFS_READ_LOCK;
2694 p->nodes[level] = b;
2697 p->slots[level] = slot;
2699 btrfs_leaf_free_space(root, b) < ins_len) {
2700 if (write_lock_level < 1) {
2701 write_lock_level = 1;
2702 btrfs_release_path(p);
2706 btrfs_set_path_blocking(p);
2707 err = split_leaf(trans, root, key,
2708 p, ins_len, ret == 0);
2709 btrfs_clear_path_blocking(p, NULL, 0);
2717 if (!p->search_for_split)
2718 unlock_up(p, level, lowest_unlock,
2719 min_write_lock_level, &write_lock_level);
2726 * we don't really know what they plan on doing with the path
2727 * from here on, so for now just mark it as blocking
2729 if (!p->leave_spinning)
2730 btrfs_set_path_blocking(p);
2732 btrfs_release_path(p);
2737 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2738 * current state of the tree together with the operations recorded in the tree
2739 * modification log to search for the key in a previous version of this tree, as
2740 * denoted by the time_seq parameter.
2742 * Naturally, there is no support for insert, delete or cow operations.
2744 * The resulting path and return value will be set up as if we called
2745 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2747 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2748 struct btrfs_path *p, u64 time_seq)
2750 struct extent_buffer *b;
2755 int lowest_unlock = 1;
2756 u8 lowest_level = 0;
2758 lowest_level = p->lowest_level;
2759 WARN_ON(p->nodes[0] != NULL);
2761 if (p->search_commit_root) {
2763 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2767 b = get_old_root(root, time_seq);
2768 level = btrfs_header_level(b);
2769 p->locks[level] = BTRFS_READ_LOCK;
2772 level = btrfs_header_level(b);
2773 p->nodes[level] = b;
2774 btrfs_clear_path_blocking(p, NULL, 0);
2777 * we have a lock on b and as long as we aren't changing
2778 * the tree, there is no way to for the items in b to change.
2779 * It is safe to drop the lock on our parent before we
2780 * go through the expensive btree search on b.
2782 btrfs_unlock_up_safe(p, level + 1);
2784 ret = bin_search(b, key, level, &slot);
2788 if (ret && slot > 0) {
2792 p->slots[level] = slot;
2793 unlock_up(p, level, lowest_unlock, 0, NULL);
2795 if (level == lowest_level) {
2801 err = read_block_for_search(NULL, root, p, &b, level,
2802 slot, key, time_seq);
2810 level = btrfs_header_level(b);
2811 err = btrfs_try_tree_read_lock(b);
2813 btrfs_set_path_blocking(p);
2814 btrfs_tree_read_lock(b);
2815 btrfs_clear_path_blocking(p, b,
2818 b = tree_mod_log_rewind(root->fs_info, b, time_seq);
2819 p->locks[level] = BTRFS_READ_LOCK;
2820 p->nodes[level] = b;
2822 p->slots[level] = slot;
2823 unlock_up(p, level, lowest_unlock, 0, NULL);
2829 if (!p->leave_spinning)
2830 btrfs_set_path_blocking(p);
2832 btrfs_release_path(p);
2838 * helper to use instead of search slot if no exact match is needed but
2839 * instead the next or previous item should be returned.
2840 * When find_higher is true, the next higher item is returned, the next lower
2842 * When return_any and find_higher are both true, and no higher item is found,
2843 * return the next lower instead.
2844 * When return_any is true and find_higher is false, and no lower item is found,
2845 * return the next higher instead.
2846 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2849 int btrfs_search_slot_for_read(struct btrfs_root *root,
2850 struct btrfs_key *key, struct btrfs_path *p,
2851 int find_higher, int return_any)
2854 struct extent_buffer *leaf;
2857 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2861 * a return value of 1 means the path is at the position where the
2862 * item should be inserted. Normally this is the next bigger item,
2863 * but in case the previous item is the last in a leaf, path points
2864 * to the first free slot in the previous leaf, i.e. at an invalid
2870 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2871 ret = btrfs_next_leaf(root, p);
2877 * no higher item found, return the next
2882 btrfs_release_path(p);
2886 if (p->slots[0] == 0) {
2887 ret = btrfs_prev_leaf(root, p);
2891 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2897 * no lower item found, return the next
2902 btrfs_release_path(p);
2912 * adjust the pointers going up the tree, starting at level
2913 * making sure the right key of each node is points to 'key'.
2914 * This is used after shifting pointers to the left, so it stops
2915 * fixing up pointers when a given leaf/node is not in slot 0 of the
2919 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
2920 struct btrfs_disk_key *key, int level)
2923 struct extent_buffer *t;
2925 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2926 int tslot = path->slots[i];
2927 if (!path->nodes[i])
2930 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
2931 btrfs_set_node_key(t, key, tslot);
2932 btrfs_mark_buffer_dirty(path->nodes[i]);
2941 * This function isn't completely safe. It's the caller's responsibility
2942 * that the new key won't break the order
2944 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
2945 struct btrfs_key *new_key)
2947 struct btrfs_disk_key disk_key;
2948 struct extent_buffer *eb;
2951 eb = path->nodes[0];
2952 slot = path->slots[0];
2954 btrfs_item_key(eb, &disk_key, slot - 1);
2955 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
2957 if (slot < btrfs_header_nritems(eb) - 1) {
2958 btrfs_item_key(eb, &disk_key, slot + 1);
2959 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
2962 btrfs_cpu_key_to_disk(&disk_key, new_key);
2963 btrfs_set_item_key(eb, &disk_key, slot);
2964 btrfs_mark_buffer_dirty(eb);
2966 fixup_low_keys(root, path, &disk_key, 1);
2970 * try to push data from one node into the next node left in the
2973 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2974 * error, and > 0 if there was no room in the left hand block.
2976 static int push_node_left(struct btrfs_trans_handle *trans,
2977 struct btrfs_root *root, struct extent_buffer *dst,
2978 struct extent_buffer *src, int empty)
2985 src_nritems = btrfs_header_nritems(src);
2986 dst_nritems = btrfs_header_nritems(dst);
2987 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
2988 WARN_ON(btrfs_header_generation(src) != trans->transid);
2989 WARN_ON(btrfs_header_generation(dst) != trans->transid);
2991 if (!empty && src_nritems <= 8)
2994 if (push_items <= 0)
2998 push_items = min(src_nritems, push_items);
2999 if (push_items < src_nritems) {
3000 /* leave at least 8 pointers in the node if
3001 * we aren't going to empty it
3003 if (src_nritems - push_items < 8) {
3004 if (push_items <= 8)
3010 push_items = min(src_nritems - 8, push_items);
3012 tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3014 copy_extent_buffer(dst, src,
3015 btrfs_node_key_ptr_offset(dst_nritems),
3016 btrfs_node_key_ptr_offset(0),
3017 push_items * sizeof(struct btrfs_key_ptr));
3019 if (push_items < src_nritems) {
3021 * don't call tree_mod_log_eb_move here, key removal was already
3022 * fully logged by tree_mod_log_eb_copy above.
3024 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3025 btrfs_node_key_ptr_offset(push_items),
3026 (src_nritems - push_items) *
3027 sizeof(struct btrfs_key_ptr));
3029 btrfs_set_header_nritems(src, src_nritems - push_items);
3030 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3031 btrfs_mark_buffer_dirty(src);
3032 btrfs_mark_buffer_dirty(dst);
3038 * try to push data from one node into the next node right in the
3041 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3042 * error, and > 0 if there was no room in the right hand block.
3044 * this will only push up to 1/2 the contents of the left node over
3046 static int balance_node_right(struct btrfs_trans_handle *trans,
3047 struct btrfs_root *root,
3048 struct extent_buffer *dst,
3049 struct extent_buffer *src)
3057 WARN_ON(btrfs_header_generation(src) != trans->transid);
3058 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3060 src_nritems = btrfs_header_nritems(src);
3061 dst_nritems = btrfs_header_nritems(dst);
3062 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3063 if (push_items <= 0)
3066 if (src_nritems < 4)
3069 max_push = src_nritems / 2 + 1;
3070 /* don't try to empty the node */
3071 if (max_push >= src_nritems)
3074 if (max_push < push_items)
3075 push_items = max_push;
3077 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3078 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3079 btrfs_node_key_ptr_offset(0),
3081 sizeof(struct btrfs_key_ptr));
3083 tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3084 src_nritems - push_items, push_items);
3085 copy_extent_buffer(dst, src,
3086 btrfs_node_key_ptr_offset(0),
3087 btrfs_node_key_ptr_offset(src_nritems - push_items),
3088 push_items * sizeof(struct btrfs_key_ptr));
3090 btrfs_set_header_nritems(src, src_nritems - push_items);
3091 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3093 btrfs_mark_buffer_dirty(src);
3094 btrfs_mark_buffer_dirty(dst);
3100 * helper function to insert a new root level in the tree.
3101 * A new node is allocated, and a single item is inserted to
3102 * point to the existing root
3104 * returns zero on success or < 0 on failure.
3106 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3107 struct btrfs_root *root,
3108 struct btrfs_path *path, int level, int log_removal)
3111 struct extent_buffer *lower;
3112 struct extent_buffer *c;
3113 struct extent_buffer *old;
3114 struct btrfs_disk_key lower_key;
3116 BUG_ON(path->nodes[level]);
3117 BUG_ON(path->nodes[level-1] != root->node);
3119 lower = path->nodes[level-1];
3121 btrfs_item_key(lower, &lower_key, 0);
3123 btrfs_node_key(lower, &lower_key, 0);
3125 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3126 root->root_key.objectid, &lower_key,
3127 level, root->node->start, 0);
3131 root_add_used(root, root->nodesize);
3133 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3134 btrfs_set_header_nritems(c, 1);
3135 btrfs_set_header_level(c, level);
3136 btrfs_set_header_bytenr(c, c->start);
3137 btrfs_set_header_generation(c, trans->transid);
3138 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3139 btrfs_set_header_owner(c, root->root_key.objectid);
3141 write_extent_buffer(c, root->fs_info->fsid,
3142 (unsigned long)btrfs_header_fsid(c),
3145 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3146 (unsigned long)btrfs_header_chunk_tree_uuid(c),
3149 btrfs_set_node_key(c, &lower_key, 0);
3150 btrfs_set_node_blockptr(c, 0, lower->start);
3151 lower_gen = btrfs_header_generation(lower);
3152 WARN_ON(lower_gen != trans->transid);
3154 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3156 btrfs_mark_buffer_dirty(c);
3159 tree_mod_log_set_root_pointer(root, c, log_removal);
3160 rcu_assign_pointer(root->node, c);
3162 /* the super has an extra ref to root->node */
3163 free_extent_buffer(old);
3165 add_root_to_dirty_list(root);
3166 extent_buffer_get(c);
3167 path->nodes[level] = c;
3168 path->locks[level] = BTRFS_WRITE_LOCK;
3169 path->slots[level] = 0;
3174 * worker function to insert a single pointer in a node.
3175 * the node should have enough room for the pointer already
3177 * slot and level indicate where you want the key to go, and
3178 * blocknr is the block the key points to.
3180 static void insert_ptr(struct btrfs_trans_handle *trans,
3181 struct btrfs_root *root, struct btrfs_path *path,
3182 struct btrfs_disk_key *key, u64 bytenr,
3183 int slot, int level)
3185 struct extent_buffer *lower;
3189 BUG_ON(!path->nodes[level]);
3190 btrfs_assert_tree_locked(path->nodes[level]);
3191 lower = path->nodes[level];
3192 nritems = btrfs_header_nritems(lower);
3193 BUG_ON(slot > nritems);
3194 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3195 if (slot != nritems) {
3197 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3198 slot, nritems - slot);
3199 memmove_extent_buffer(lower,
3200 btrfs_node_key_ptr_offset(slot + 1),
3201 btrfs_node_key_ptr_offset(slot),
3202 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3205 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3209 btrfs_set_node_key(lower, key, slot);
3210 btrfs_set_node_blockptr(lower, slot, bytenr);
3211 WARN_ON(trans->transid == 0);
3212 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3213 btrfs_set_header_nritems(lower, nritems + 1);
3214 btrfs_mark_buffer_dirty(lower);
3218 * split the node at the specified level in path in two.
3219 * The path is corrected to point to the appropriate node after the split
3221 * Before splitting this tries to make some room in the node by pushing
3222 * left and right, if either one works, it returns right away.
3224 * returns 0 on success and < 0 on failure
3226 static noinline int split_node(struct btrfs_trans_handle *trans,
3227 struct btrfs_root *root,
3228 struct btrfs_path *path, int level)
3230 struct extent_buffer *c;
3231 struct extent_buffer *split;
3232 struct btrfs_disk_key disk_key;
3237 c = path->nodes[level];
3238 WARN_ON(btrfs_header_generation(c) != trans->transid);
3239 if (c == root->node) {
3241 * trying to split the root, lets make a new one
3243 * tree mod log: We pass 0 as log_removal parameter to
3244 * insert_new_root, because that root buffer will be kept as a
3245 * normal node. We are going to log removal of half of the
3246 * elements below with tree_mod_log_eb_copy. We're holding a
3247 * tree lock on the buffer, which is why we cannot race with
3248 * other tree_mod_log users.
3250 ret = insert_new_root(trans, root, path, level + 1, 0);
3254 ret = push_nodes_for_insert(trans, root, path, level);
3255 c = path->nodes[level];
3256 if (!ret && btrfs_header_nritems(c) <
3257 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3263 c_nritems = btrfs_header_nritems(c);
3264 mid = (c_nritems + 1) / 2;
3265 btrfs_node_key(c, &disk_key, mid);
3267 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3268 root->root_key.objectid,
3269 &disk_key, level, c->start, 0);
3271 return PTR_ERR(split);
3273 root_add_used(root, root->nodesize);
3275 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3276 btrfs_set_header_level(split, btrfs_header_level(c));
3277 btrfs_set_header_bytenr(split, split->start);
3278 btrfs_set_header_generation(split, trans->transid);
3279 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3280 btrfs_set_header_owner(split, root->root_key.objectid);
3281 write_extent_buffer(split, root->fs_info->fsid,
3282 (unsigned long)btrfs_header_fsid(split),
3284 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3285 (unsigned long)btrfs_header_chunk_tree_uuid(split),
3288 tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid);
3289 copy_extent_buffer(split, c,
3290 btrfs_node_key_ptr_offset(0),
3291 btrfs_node_key_ptr_offset(mid),
3292 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3293 btrfs_set_header_nritems(split, c_nritems - mid);
3294 btrfs_set_header_nritems(c, mid);
3297 btrfs_mark_buffer_dirty(c);
3298 btrfs_mark_buffer_dirty(split);
3300 insert_ptr(trans, root, path, &disk_key, split->start,
3301 path->slots[level + 1] + 1, level + 1);
3303 if (path->slots[level] >= mid) {
3304 path->slots[level] -= mid;
3305 btrfs_tree_unlock(c);
3306 free_extent_buffer(c);
3307 path->nodes[level] = split;
3308 path->slots[level + 1] += 1;
3310 btrfs_tree_unlock(split);
3311 free_extent_buffer(split);
3317 * how many bytes are required to store the items in a leaf. start
3318 * and nr indicate which items in the leaf to check. This totals up the
3319 * space used both by the item structs and the item data
3321 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3323 struct btrfs_item *start_item;
3324 struct btrfs_item *end_item;
3325 struct btrfs_map_token token;
3327 int nritems = btrfs_header_nritems(l);
3328 int end = min(nritems, start + nr) - 1;
3332 btrfs_init_map_token(&token);
3333 start_item = btrfs_item_nr(l, start);
3334 end_item = btrfs_item_nr(l, end);
3335 data_len = btrfs_token_item_offset(l, start_item, &token) +
3336 btrfs_token_item_size(l, start_item, &token);
3337 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3338 data_len += sizeof(struct btrfs_item) * nr;
3339 WARN_ON(data_len < 0);
3344 * The space between the end of the leaf items and
3345 * the start of the leaf data. IOW, how much room
3346 * the leaf has left for both items and data
3348 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3349 struct extent_buffer *leaf)
3351 int nritems = btrfs_header_nritems(leaf);
3353 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3355 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
3356 "used %d nritems %d\n",
3357 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3358 leaf_space_used(leaf, 0, nritems), nritems);
3364 * min slot controls the lowest index we're willing to push to the
3365 * right. We'll push up to and including min_slot, but no lower
3367 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3368 struct btrfs_root *root,
3369 struct btrfs_path *path,
3370 int data_size, int empty,
3371 struct extent_buffer *right,
3372 int free_space, u32 left_nritems,
3375 struct extent_buffer *left = path->nodes[0];
3376 struct extent_buffer *upper = path->nodes[1];
3377 struct btrfs_map_token token;
3378 struct btrfs_disk_key disk_key;
3383 struct btrfs_item *item;
3389 btrfs_init_map_token(&token);
3394 nr = max_t(u32, 1, min_slot);
3396 if (path->slots[0] >= left_nritems)
3397 push_space += data_size;
3399 slot = path->slots[1];
3400 i = left_nritems - 1;
3402 item = btrfs_item_nr(left, i);
3404 if (!empty && push_items > 0) {
3405 if (path->slots[0] > i)
3407 if (path->slots[0] == i) {
3408 int space = btrfs_leaf_free_space(root, left);
3409 if (space + push_space * 2 > free_space)
3414 if (path->slots[0] == i)
3415 push_space += data_size;
3417 this_item_size = btrfs_item_size(left, item);
3418 if (this_item_size + sizeof(*item) + push_space > free_space)
3422 push_space += this_item_size + sizeof(*item);
3428 if (push_items == 0)
3431 WARN_ON(!empty && push_items == left_nritems);
3433 /* push left to right */
3434 right_nritems = btrfs_header_nritems(right);
3436 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3437 push_space -= leaf_data_end(root, left);
3439 /* make room in the right data area */
3440 data_end = leaf_data_end(root, right);
3441 memmove_extent_buffer(right,
3442 btrfs_leaf_data(right) + data_end - push_space,
3443 btrfs_leaf_data(right) + data_end,
3444 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3446 /* copy from the left data area */
3447 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3448 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3449 btrfs_leaf_data(left) + leaf_data_end(root, left),
3452 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3453 btrfs_item_nr_offset(0),
3454 right_nritems * sizeof(struct btrfs_item));
3456 /* copy the items from left to right */
3457 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3458 btrfs_item_nr_offset(left_nritems - push_items),
3459 push_items * sizeof(struct btrfs_item));
3461 /* update the item pointers */
3462 right_nritems += push_items;
3463 btrfs_set_header_nritems(right, right_nritems);
3464 push_space = BTRFS_LEAF_DATA_SIZE(root);
3465 for (i = 0; i < right_nritems; i++) {
3466 item = btrfs_item_nr(right, i);
3467 push_space -= btrfs_token_item_size(right, item, &token);
3468 btrfs_set_token_item_offset(right, item, push_space, &token);
3471 left_nritems -= push_items;
3472 btrfs_set_header_nritems(left, left_nritems);
3475 btrfs_mark_buffer_dirty(left);
3477 clean_tree_block(trans, root, left);
3479 btrfs_mark_buffer_dirty(right);
3481 btrfs_item_key(right, &disk_key, 0);
3482 btrfs_set_node_key(upper, &disk_key, slot + 1);
3483 btrfs_mark_buffer_dirty(upper);
3485 /* then fixup the leaf pointer in the path */
3486 if (path->slots[0] >= left_nritems) {
3487 path->slots[0] -= left_nritems;
3488 if (btrfs_header_nritems(path->nodes[0]) == 0)
3489 clean_tree_block(trans, root, path->nodes[0]);
3490 btrfs_tree_unlock(path->nodes[0]);
3491 free_extent_buffer(path->nodes[0]);
3492 path->nodes[0] = right;
3493 path->slots[1] += 1;
3495 btrfs_tree_unlock(right);
3496 free_extent_buffer(right);
3501 btrfs_tree_unlock(right);
3502 free_extent_buffer(right);
3507 * push some data in the path leaf to the right, trying to free up at
3508 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3510 * returns 1 if the push failed because the other node didn't have enough
3511 * room, 0 if everything worked out and < 0 if there were major errors.
3513 * this will push starting from min_slot to the end of the leaf. It won't
3514 * push any slot lower than min_slot
3516 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3517 *root, struct btrfs_path *path,
3518 int min_data_size, int data_size,
3519 int empty, u32 min_slot)
3521 struct extent_buffer *left = path->nodes[0];
3522 struct extent_buffer *right;
3523 struct extent_buffer *upper;
3529 if (!path->nodes[1])
3532 slot = path->slots[1];
3533 upper = path->nodes[1];
3534 if (slot >= btrfs_header_nritems(upper) - 1)
3537 btrfs_assert_tree_locked(path->nodes[1]);
3539 right = read_node_slot(root, upper, slot + 1);
3543 btrfs_tree_lock(right);
3544 btrfs_set_lock_blocking(right);
3546 free_space = btrfs_leaf_free_space(root, right);
3547 if (free_space < data_size)
3550 /* cow and double check */
3551 ret = btrfs_cow_block(trans, root, right, upper,
3556 free_space = btrfs_leaf_free_space(root, right);
3557 if (free_space < data_size)
3560 left_nritems = btrfs_header_nritems(left);
3561 if (left_nritems == 0)
3564 return __push_leaf_right(trans, root, path, min_data_size, empty,
3565 right, free_space, left_nritems, min_slot);
3567 btrfs_tree_unlock(right);
3568 free_extent_buffer(right);
3573 * push some data in the path leaf to the left, trying to free up at
3574 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3576 * max_slot can put a limit on how far into the leaf we'll push items. The
3577 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3580 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3581 struct btrfs_root *root,
3582 struct btrfs_path *path, int data_size,
3583 int empty, struct extent_buffer *left,
3584 int free_space, u32 right_nritems,
3587 struct btrfs_disk_key disk_key;
3588 struct extent_buffer *right = path->nodes[0];
3592 struct btrfs_item *item;
3593 u32 old_left_nritems;
3597 u32 old_left_item_size;
3598 struct btrfs_map_token token;
3600 btrfs_init_map_token(&token);
3603 nr = min(right_nritems, max_slot);
3605 nr = min(right_nritems - 1, max_slot);
3607 for (i = 0; i < nr; i++) {
3608 item = btrfs_item_nr(right, i);
3610 if (!empty && push_items > 0) {
3611 if (path->slots[0] < i)
3613 if (path->slots[0] == i) {
3614 int space = btrfs_leaf_free_space(root, right);
3615 if (space + push_space * 2 > free_space)
3620 if (path->slots[0] == i)
3621 push_space += data_size;
3623 this_item_size = btrfs_item_size(right, item);
3624 if (this_item_size + sizeof(*item) + push_space > free_space)
3628 push_space += this_item_size + sizeof(*item);
3631 if (push_items == 0) {
3635 if (!empty && push_items == btrfs_header_nritems(right))
3638 /* push data from right to left */
3639 copy_extent_buffer(left, right,
3640 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3641 btrfs_item_nr_offset(0),
3642 push_items * sizeof(struct btrfs_item));
3644 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3645 btrfs_item_offset_nr(right, push_items - 1);
3647 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3648 leaf_data_end(root, left) - push_space,
3649 btrfs_leaf_data(right) +
3650 btrfs_item_offset_nr(right, push_items - 1),
3652 old_left_nritems = btrfs_header_nritems(left);
3653 BUG_ON(old_left_nritems <= 0);
3655 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3656 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3659 item = btrfs_item_nr(left, i);
3661 ioff = btrfs_token_item_offset(left, item, &token);
3662 btrfs_set_token_item_offset(left, item,
3663 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3666 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3668 /* fixup right node */
3669 if (push_items > right_nritems)
3670 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3673 if (push_items < right_nritems) {
3674 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3675 leaf_data_end(root, right);
3676 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3677 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3678 btrfs_leaf_data(right) +
3679 leaf_data_end(root, right), push_space);
3681 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3682 btrfs_item_nr_offset(push_items),
3683 (btrfs_header_nritems(right) - push_items) *
3684 sizeof(struct btrfs_item));
3686 right_nritems -= push_items;
3687 btrfs_set_header_nritems(right, right_nritems);
3688 push_space = BTRFS_LEAF_DATA_SIZE(root);
3689 for (i = 0; i < right_nritems; i++) {
3690 item = btrfs_item_nr(right, i);
3692 push_space = push_space - btrfs_token_item_size(right,
3694 btrfs_set_token_item_offset(right, item, push_space, &token);
3697 btrfs_mark_buffer_dirty(left);
3699 btrfs_mark_buffer_dirty(right);
3701 clean_tree_block(trans, root, right);
3703 btrfs_item_key(right, &disk_key, 0);
3704 fixup_low_keys(root, path, &disk_key, 1);
3706 /* then fixup the leaf pointer in the path */
3707 if (path->slots[0] < push_items) {
3708 path->slots[0] += old_left_nritems;
3709 btrfs_tree_unlock(path->nodes[0]);
3710 free_extent_buffer(path->nodes[0]);
3711 path->nodes[0] = left;
3712 path->slots[1] -= 1;
3714 btrfs_tree_unlock(left);
3715 free_extent_buffer(left);
3716 path->slots[0] -= push_items;
3718 BUG_ON(path->slots[0] < 0);
3721 btrfs_tree_unlock(left);
3722 free_extent_buffer(left);
3727 * push some data in the path leaf to the left, trying to free up at
3728 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3730 * max_slot can put a limit on how far into the leaf we'll push items. The
3731 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3734 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3735 *root, struct btrfs_path *path, int min_data_size,
3736 int data_size, int empty, u32 max_slot)
3738 struct extent_buffer *right = path->nodes[0];
3739 struct extent_buffer *left;
3745 slot = path->slots[1];
3748 if (!path->nodes[1])
3751 right_nritems = btrfs_header_nritems(right);
3752 if (right_nritems == 0)
3755 btrfs_assert_tree_locked(path->nodes[1]);
3757 left = read_node_slot(root, path->nodes[1], slot - 1);
3761 btrfs_tree_lock(left);
3762 btrfs_set_lock_blocking(left);
3764 free_space = btrfs_leaf_free_space(root, left);
3765 if (free_space < data_size) {
3770 /* cow and double check */
3771 ret = btrfs_cow_block(trans, root, left,
3772 path->nodes[1], slot - 1, &left);
3774 /* we hit -ENOSPC, but it isn't fatal here */
3780 free_space = btrfs_leaf_free_space(root, left);
3781 if (free_space < data_size) {
3786 return __push_leaf_left(trans, root, path, min_data_size,
3787 empty, left, free_space, right_nritems,
3790 btrfs_tree_unlock(left);
3791 free_extent_buffer(left);
3796 * split the path's leaf in two, making sure there is at least data_size
3797 * available for the resulting leaf level of the path.
3799 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3800 struct btrfs_root *root,
3801 struct btrfs_path *path,
3802 struct extent_buffer *l,
3803 struct extent_buffer *right,
3804 int slot, int mid, int nritems)
3809 struct btrfs_disk_key disk_key;
3810 struct btrfs_map_token token;
3812 btrfs_init_map_token(&token);
3814 nritems = nritems - mid;
3815 btrfs_set_header_nritems(right, nritems);
3816 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
3818 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3819 btrfs_item_nr_offset(mid),
3820 nritems * sizeof(struct btrfs_item));
3822 copy_extent_buffer(right, l,
3823 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
3824 data_copy_size, btrfs_leaf_data(l) +
3825 leaf_data_end(root, l), data_copy_size);
3827 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
3828 btrfs_item_end_nr(l, mid);
3830 for (i = 0; i < nritems; i++) {
3831 struct btrfs_item *item = btrfs_item_nr(right, i);
3834 ioff = btrfs_token_item_offset(right, item, &token);
3835 btrfs_set_token_item_offset(right, item,
3836 ioff + rt_data_off, &token);
3839 btrfs_set_header_nritems(l, mid);
3840 btrfs_item_key(right, &disk_key, 0);
3841 insert_ptr(trans, root, path, &disk_key, right->start,
3842 path->slots[1] + 1, 1);
3844 btrfs_mark_buffer_dirty(right);
3845 btrfs_mark_buffer_dirty(l);
3846 BUG_ON(path->slots[0] != slot);
3849 btrfs_tree_unlock(path->nodes[0]);
3850 free_extent_buffer(path->nodes[0]);
3851 path->nodes[0] = right;
3852 path->slots[0] -= mid;
3853 path->slots[1] += 1;
3855 btrfs_tree_unlock(right);
3856 free_extent_buffer(right);
3859 BUG_ON(path->slots[0] < 0);
3863 * double splits happen when we need to insert a big item in the middle
3864 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3865 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3868 * We avoid this by trying to push the items on either side of our target
3869 * into the adjacent leaves. If all goes well we can avoid the double split
3872 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3873 struct btrfs_root *root,
3874 struct btrfs_path *path,
3882 slot = path->slots[0];
3885 * try to push all the items after our slot into the
3888 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
3895 nritems = btrfs_header_nritems(path->nodes[0]);
3897 * our goal is to get our slot at the start or end of a leaf. If
3898 * we've done so we're done
3900 if (path->slots[0] == 0 || path->slots[0] == nritems)
3903 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3906 /* try to push all the items before our slot into the next leaf */
3907 slot = path->slots[0];
3908 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
3921 * split the path's leaf in two, making sure there is at least data_size
3922 * available for the resulting leaf level of the path.
3924 * returns 0 if all went well and < 0 on failure.
3926 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3927 struct btrfs_root *root,
3928 struct btrfs_key *ins_key,
3929 struct btrfs_path *path, int data_size,
3932 struct btrfs_disk_key disk_key;
3933 struct extent_buffer *l;
3937 struct extent_buffer *right;
3941 int num_doubles = 0;
3942 int tried_avoid_double = 0;
3945 slot = path->slots[0];
3946 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3947 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
3950 /* first try to make some room by pushing left and right */
3952 wret = push_leaf_right(trans, root, path, data_size,
3957 wret = push_leaf_left(trans, root, path, data_size,
3958 data_size, 0, (u32)-1);
3964 /* did the pushes work? */
3965 if (btrfs_leaf_free_space(root, l) >= data_size)
3969 if (!path->nodes[1]) {
3970 ret = insert_new_root(trans, root, path, 1, 1);
3977 slot = path->slots[0];
3978 nritems = btrfs_header_nritems(l);
3979 mid = (nritems + 1) / 2;
3983 leaf_space_used(l, mid, nritems - mid) + data_size >
3984 BTRFS_LEAF_DATA_SIZE(root)) {
3985 if (slot >= nritems) {
3989 if (mid != nritems &&
3990 leaf_space_used(l, mid, nritems - mid) +
3991 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
3992 if (data_size && !tried_avoid_double)
3993 goto push_for_double;
3999 if (leaf_space_used(l, 0, mid) + data_size >
4000 BTRFS_LEAF_DATA_SIZE(root)) {
4001 if (!extend && data_size && slot == 0) {
4003 } else if ((extend || !data_size) && slot == 0) {
4007 if (mid != nritems &&
4008 leaf_space_used(l, mid, nritems - mid) +
4009 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4010 if (data_size && !tried_avoid_double)
4011 goto push_for_double;
4019 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4021 btrfs_item_key(l, &disk_key, mid);
4023 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4024 root->root_key.objectid,
4025 &disk_key, 0, l->start, 0);
4027 return PTR_ERR(right);
4029 root_add_used(root, root->leafsize);
4031 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4032 btrfs_set_header_bytenr(right, right->start);
4033 btrfs_set_header_generation(right, trans->transid);
4034 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4035 btrfs_set_header_owner(right, root->root_key.objectid);
4036 btrfs_set_header_level(right, 0);
4037 write_extent_buffer(right, root->fs_info->fsid,
4038 (unsigned long)btrfs_header_fsid(right),
4041 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4042 (unsigned long)btrfs_header_chunk_tree_uuid(right),
4047 btrfs_set_header_nritems(right, 0);
4048 insert_ptr(trans, root, path, &disk_key, right->start,
4049 path->slots[1] + 1, 1);
4050 btrfs_tree_unlock(path->nodes[0]);
4051 free_extent_buffer(path->nodes[0]);
4052 path->nodes[0] = right;
4054 path->slots[1] += 1;
4056 btrfs_set_header_nritems(right, 0);
4057 insert_ptr(trans, root, path, &disk_key, right->start,
4059 btrfs_tree_unlock(path->nodes[0]);
4060 free_extent_buffer(path->nodes[0]);
4061 path->nodes[0] = right;
4063 if (path->slots[1] == 0)
4064 fixup_low_keys(root, path, &disk_key, 1);
4066 btrfs_mark_buffer_dirty(right);
4070 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4073 BUG_ON(num_doubles != 0);
4081 push_for_double_split(trans, root, path, data_size);
4082 tried_avoid_double = 1;
4083 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4088 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4089 struct btrfs_root *root,
4090 struct btrfs_path *path, int ins_len)
4092 struct btrfs_key key;
4093 struct extent_buffer *leaf;
4094 struct btrfs_file_extent_item *fi;
4099 leaf = path->nodes[0];
4100 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4102 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4103 key.type != BTRFS_EXTENT_CSUM_KEY);
4105 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4108 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4109 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4110 fi = btrfs_item_ptr(leaf, path->slots[0],
4111 struct btrfs_file_extent_item);
4112 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4114 btrfs_release_path(path);
4116 path->keep_locks = 1;
4117 path->search_for_split = 1;
4118 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4119 path->search_for_split = 0;
4124 leaf = path->nodes[0];
4125 /* if our item isn't there or got smaller, return now */
4126 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4129 /* the leaf has changed, it now has room. return now */
4130 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4133 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4134 fi = btrfs_item_ptr(leaf, path->slots[0],
4135 struct btrfs_file_extent_item);
4136 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4140 btrfs_set_path_blocking(path);
4141 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4145 path->keep_locks = 0;
4146 btrfs_unlock_up_safe(path, 1);
4149 path->keep_locks = 0;
4153 static noinline int split_item(struct btrfs_trans_handle *trans,
4154 struct btrfs_root *root,
4155 struct btrfs_path *path,
4156 struct btrfs_key *new_key,
4157 unsigned long split_offset)
4159 struct extent_buffer *leaf;
4160 struct btrfs_item *item;
4161 struct btrfs_item *new_item;
4167 struct btrfs_disk_key disk_key;
4169 leaf = path->nodes[0];
4170 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4172 btrfs_set_path_blocking(path);
4174 item = btrfs_item_nr(leaf, path->slots[0]);
4175 orig_offset = btrfs_item_offset(leaf, item);
4176 item_size = btrfs_item_size(leaf, item);
4178 buf = kmalloc(item_size, GFP_NOFS);
4182 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4183 path->slots[0]), item_size);
4185 slot = path->slots[0] + 1;
4186 nritems = btrfs_header_nritems(leaf);
4187 if (slot != nritems) {
4188 /* shift the items */
4189 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4190 btrfs_item_nr_offset(slot),
4191 (nritems - slot) * sizeof(struct btrfs_item));
4194 btrfs_cpu_key_to_disk(&disk_key, new_key);
4195 btrfs_set_item_key(leaf, &disk_key, slot);
4197 new_item = btrfs_item_nr(leaf, slot);
4199 btrfs_set_item_offset(leaf, new_item, orig_offset);
4200 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4202 btrfs_set_item_offset(leaf, item,
4203 orig_offset + item_size - split_offset);
4204 btrfs_set_item_size(leaf, item, split_offset);
4206 btrfs_set_header_nritems(leaf, nritems + 1);
4208 /* write the data for the start of the original item */
4209 write_extent_buffer(leaf, buf,
4210 btrfs_item_ptr_offset(leaf, path->slots[0]),
4213 /* write the data for the new item */
4214 write_extent_buffer(leaf, buf + split_offset,
4215 btrfs_item_ptr_offset(leaf, slot),
4216 item_size - split_offset);
4217 btrfs_mark_buffer_dirty(leaf);
4219 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4225 * This function splits a single item into two items,
4226 * giving 'new_key' to the new item and splitting the
4227 * old one at split_offset (from the start of the item).
4229 * The path may be released by this operation. After
4230 * the split, the path is pointing to the old item. The
4231 * new item is going to be in the same node as the old one.
4233 * Note, the item being split must be smaller enough to live alone on
4234 * a tree block with room for one extra struct btrfs_item
4236 * This allows us to split the item in place, keeping a lock on the
4237 * leaf the entire time.
4239 int btrfs_split_item(struct btrfs_trans_handle *trans,
4240 struct btrfs_root *root,
4241 struct btrfs_path *path,
4242 struct btrfs_key *new_key,
4243 unsigned long split_offset)
4246 ret = setup_leaf_for_split(trans, root, path,
4247 sizeof(struct btrfs_item));
4251 ret = split_item(trans, root, path, new_key, split_offset);
4256 * This function duplicate a item, giving 'new_key' to the new item.
4257 * It guarantees both items live in the same tree leaf and the new item
4258 * is contiguous with the original item.
4260 * This allows us to split file extent in place, keeping a lock on the
4261 * leaf the entire time.
4263 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4264 struct btrfs_root *root,
4265 struct btrfs_path *path,
4266 struct btrfs_key *new_key)
4268 struct extent_buffer *leaf;
4272 leaf = path->nodes[0];
4273 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4274 ret = setup_leaf_for_split(trans, root, path,
4275 item_size + sizeof(struct btrfs_item));
4280 setup_items_for_insert(root, path, new_key, &item_size,
4281 item_size, item_size +
4282 sizeof(struct btrfs_item), 1);
4283 leaf = path->nodes[0];
4284 memcpy_extent_buffer(leaf,
4285 btrfs_item_ptr_offset(leaf, path->slots[0]),
4286 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4292 * make the item pointed to by the path smaller. new_size indicates
4293 * how small to make it, and from_end tells us if we just chop bytes
4294 * off the end of the item or if we shift the item to chop bytes off
4297 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4298 u32 new_size, int from_end)
4301 struct extent_buffer *leaf;
4302 struct btrfs_item *item;
4304 unsigned int data_end;
4305 unsigned int old_data_start;
4306 unsigned int old_size;
4307 unsigned int size_diff;
4309 struct btrfs_map_token token;
4311 btrfs_init_map_token(&token);
4313 leaf = path->nodes[0];
4314 slot = path->slots[0];
4316 old_size = btrfs_item_size_nr(leaf, slot);
4317 if (old_size == new_size)
4320 nritems = btrfs_header_nritems(leaf);
4321 data_end = leaf_data_end(root, leaf);
4323 old_data_start = btrfs_item_offset_nr(leaf, slot);
4325 size_diff = old_size - new_size;
4328 BUG_ON(slot >= nritems);
4331 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4333 /* first correct the data pointers */
4334 for (i = slot; i < nritems; i++) {
4336 item = btrfs_item_nr(leaf, i);
4338 ioff = btrfs_token_item_offset(leaf, item, &token);
4339 btrfs_set_token_item_offset(leaf, item,
4340 ioff + size_diff, &token);
4343 /* shift the data */
4345 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4346 data_end + size_diff, btrfs_leaf_data(leaf) +
4347 data_end, old_data_start + new_size - data_end);
4349 struct btrfs_disk_key disk_key;
4352 btrfs_item_key(leaf, &disk_key, slot);
4354 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4356 struct btrfs_file_extent_item *fi;
4358 fi = btrfs_item_ptr(leaf, slot,
4359 struct btrfs_file_extent_item);
4360 fi = (struct btrfs_file_extent_item *)(
4361 (unsigned long)fi - size_diff);
4363 if (btrfs_file_extent_type(leaf, fi) ==
4364 BTRFS_FILE_EXTENT_INLINE) {
4365 ptr = btrfs_item_ptr_offset(leaf, slot);
4366 memmove_extent_buffer(leaf, ptr,
4368 offsetof(struct btrfs_file_extent_item,
4373 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4374 data_end + size_diff, btrfs_leaf_data(leaf) +
4375 data_end, old_data_start - data_end);
4377 offset = btrfs_disk_key_offset(&disk_key);
4378 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4379 btrfs_set_item_key(leaf, &disk_key, slot);
4381 fixup_low_keys(root, path, &disk_key, 1);
4384 item = btrfs_item_nr(leaf, slot);
4385 btrfs_set_item_size(leaf, item, new_size);
4386 btrfs_mark_buffer_dirty(leaf);
4388 if (btrfs_leaf_free_space(root, leaf) < 0) {
4389 btrfs_print_leaf(root, leaf);
4395 * make the item pointed to by the path bigger, data_size is the new size.
4397 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4401 struct extent_buffer *leaf;
4402 struct btrfs_item *item;
4404 unsigned int data_end;
4405 unsigned int old_data;
4406 unsigned int old_size;
4408 struct btrfs_map_token token;
4410 btrfs_init_map_token(&token);
4412 leaf = path->nodes[0];
4414 nritems = btrfs_header_nritems(leaf);
4415 data_end = leaf_data_end(root, leaf);
4417 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4418 btrfs_print_leaf(root, leaf);
4421 slot = path->slots[0];
4422 old_data = btrfs_item_end_nr(leaf, slot);
4425 if (slot >= nritems) {
4426 btrfs_print_leaf(root, leaf);
4427 printk(KERN_CRIT "slot %d too large, nritems %d\n",
4433 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4435 /* first correct the data pointers */
4436 for (i = slot; i < nritems; i++) {
4438 item = btrfs_item_nr(leaf, i);
4440 ioff = btrfs_token_item_offset(leaf, item, &token);
4441 btrfs_set_token_item_offset(leaf, item,
4442 ioff - data_size, &token);
4445 /* shift the data */
4446 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4447 data_end - data_size, btrfs_leaf_data(leaf) +
4448 data_end, old_data - data_end);
4450 data_end = old_data;
4451 old_size = btrfs_item_size_nr(leaf, slot);
4452 item = btrfs_item_nr(leaf, slot);
4453 btrfs_set_item_size(leaf, item, old_size + data_size);
4454 btrfs_mark_buffer_dirty(leaf);
4456 if (btrfs_leaf_free_space(root, leaf) < 0) {
4457 btrfs_print_leaf(root, leaf);
4463 * this is a helper for btrfs_insert_empty_items, the main goal here is
4464 * to save stack depth by doing the bulk of the work in a function
4465 * that doesn't call btrfs_search_slot
4467 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4468 struct btrfs_key *cpu_key, u32 *data_size,
4469 u32 total_data, u32 total_size, int nr)
4471 struct btrfs_item *item;
4474 unsigned int data_end;
4475 struct btrfs_disk_key disk_key;
4476 struct extent_buffer *leaf;
4478 struct btrfs_map_token token;
4480 btrfs_init_map_token(&token);
4482 leaf = path->nodes[0];
4483 slot = path->slots[0];
4485 nritems = btrfs_header_nritems(leaf);
4486 data_end = leaf_data_end(root, leaf);
4488 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4489 btrfs_print_leaf(root, leaf);
4490 printk(KERN_CRIT "not enough freespace need %u have %d\n",
4491 total_size, btrfs_leaf_free_space(root, leaf));
4495 if (slot != nritems) {
4496 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4498 if (old_data < data_end) {
4499 btrfs_print_leaf(root, leaf);
4500 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
4501 slot, old_data, data_end);
4505 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4507 /* first correct the data pointers */
4508 for (i = slot; i < nritems; i++) {
4511 item = btrfs_item_nr(leaf, i);
4512 ioff = btrfs_token_item_offset(leaf, item, &token);
4513 btrfs_set_token_item_offset(leaf, item,
4514 ioff - total_data, &token);
4516 /* shift the items */
4517 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4518 btrfs_item_nr_offset(slot),
4519 (nritems - slot) * sizeof(struct btrfs_item));
4521 /* shift the data */
4522 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4523 data_end - total_data, btrfs_leaf_data(leaf) +
4524 data_end, old_data - data_end);
4525 data_end = old_data;
4528 /* setup the item for the new data */
4529 for (i = 0; i < nr; i++) {
4530 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4531 btrfs_set_item_key(leaf, &disk_key, slot + i);
4532 item = btrfs_item_nr(leaf, slot + i);
4533 btrfs_set_token_item_offset(leaf, item,
4534 data_end - data_size[i], &token);
4535 data_end -= data_size[i];
4536 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4539 btrfs_set_header_nritems(leaf, nritems + nr);
4542 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4543 fixup_low_keys(root, path, &disk_key, 1);
4545 btrfs_unlock_up_safe(path, 1);
4546 btrfs_mark_buffer_dirty(leaf);
4548 if (btrfs_leaf_free_space(root, leaf) < 0) {
4549 btrfs_print_leaf(root, leaf);
4555 * Given a key and some data, insert items into the tree.
4556 * This does all the path init required, making room in the tree if needed.
4558 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4559 struct btrfs_root *root,
4560 struct btrfs_path *path,
4561 struct btrfs_key *cpu_key, u32 *data_size,
4570 for (i = 0; i < nr; i++)
4571 total_data += data_size[i];
4573 total_size = total_data + (nr * sizeof(struct btrfs_item));
4574 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4580 slot = path->slots[0];
4583 setup_items_for_insert(root, path, cpu_key, data_size,
4584 total_data, total_size, nr);
4589 * Given a key and some data, insert an item into the tree.
4590 * This does all the path init required, making room in the tree if needed.
4592 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4593 *root, struct btrfs_key *cpu_key, void *data, u32
4597 struct btrfs_path *path;
4598 struct extent_buffer *leaf;
4601 path = btrfs_alloc_path();
4604 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4606 leaf = path->nodes[0];
4607 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4608 write_extent_buffer(leaf, data, ptr, data_size);
4609 btrfs_mark_buffer_dirty(leaf);
4611 btrfs_free_path(path);
4616 * delete the pointer from a given node.
4618 * the tree should have been previously balanced so the deletion does not
4621 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4622 int level, int slot)
4624 struct extent_buffer *parent = path->nodes[level];
4628 nritems = btrfs_header_nritems(parent);
4629 if (slot != nritems - 1) {
4631 tree_mod_log_eb_move(root->fs_info, parent, slot,
4632 slot + 1, nritems - slot - 1);
4633 memmove_extent_buffer(parent,
4634 btrfs_node_key_ptr_offset(slot),
4635 btrfs_node_key_ptr_offset(slot + 1),
4636 sizeof(struct btrfs_key_ptr) *
4637 (nritems - slot - 1));
4639 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4640 MOD_LOG_KEY_REMOVE);
4645 btrfs_set_header_nritems(parent, nritems);
4646 if (nritems == 0 && parent == root->node) {
4647 BUG_ON(btrfs_header_level(root->node) != 1);
4648 /* just turn the root into a leaf and break */
4649 btrfs_set_header_level(root->node, 0);
4650 } else if (slot == 0) {
4651 struct btrfs_disk_key disk_key;
4653 btrfs_node_key(parent, &disk_key, 0);
4654 fixup_low_keys(root, path, &disk_key, level + 1);
4656 btrfs_mark_buffer_dirty(parent);
4660 * a helper function to delete the leaf pointed to by path->slots[1] and
4663 * This deletes the pointer in path->nodes[1] and frees the leaf
4664 * block extent. zero is returned if it all worked out, < 0 otherwise.
4666 * The path must have already been setup for deleting the leaf, including
4667 * all the proper balancing. path->nodes[1] must be locked.
4669 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4670 struct btrfs_root *root,
4671 struct btrfs_path *path,
4672 struct extent_buffer *leaf)
4674 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4675 del_ptr(root, path, 1, path->slots[1]);
4678 * btrfs_free_extent is expensive, we want to make sure we
4679 * aren't holding any locks when we call it
4681 btrfs_unlock_up_safe(path, 0);
4683 root_sub_used(root, leaf->len);
4685 extent_buffer_get(leaf);
4686 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4687 free_extent_buffer_stale(leaf);
4690 * delete the item at the leaf level in path. If that empties
4691 * the leaf, remove it from the tree
4693 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4694 struct btrfs_path *path, int slot, int nr)
4696 struct extent_buffer *leaf;
4697 struct btrfs_item *item;
4704 struct btrfs_map_token token;
4706 btrfs_init_map_token(&token);
4708 leaf = path->nodes[0];
4709 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4711 for (i = 0; i < nr; i++)
4712 dsize += btrfs_item_size_nr(leaf, slot + i);
4714 nritems = btrfs_header_nritems(leaf);
4716 if (slot + nr != nritems) {
4717 int data_end = leaf_data_end(root, leaf);
4719 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4721 btrfs_leaf_data(leaf) + data_end,
4722 last_off - data_end);
4724 for (i = slot + nr; i < nritems; i++) {
4727 item = btrfs_item_nr(leaf, i);
4728 ioff = btrfs_token_item_offset(leaf, item, &token);
4729 btrfs_set_token_item_offset(leaf, item,
4730 ioff + dsize, &token);
4733 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4734 btrfs_item_nr_offset(slot + nr),
4735 sizeof(struct btrfs_item) *
4736 (nritems - slot - nr));
4738 btrfs_set_header_nritems(leaf, nritems - nr);
4741 /* delete the leaf if we've emptied it */
4743 if (leaf == root->node) {
4744 btrfs_set_header_level(leaf, 0);
4746 btrfs_set_path_blocking(path);
4747 clean_tree_block(trans, root, leaf);
4748 btrfs_del_leaf(trans, root, path, leaf);
4751 int used = leaf_space_used(leaf, 0, nritems);
4753 struct btrfs_disk_key disk_key;
4755 btrfs_item_key(leaf, &disk_key, 0);
4756 fixup_low_keys(root, path, &disk_key, 1);
4759 /* delete the leaf if it is mostly empty */
4760 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4761 /* push_leaf_left fixes the path.
4762 * make sure the path still points to our leaf
4763 * for possible call to del_ptr below
4765 slot = path->slots[1];
4766 extent_buffer_get(leaf);
4768 btrfs_set_path_blocking(path);
4769 wret = push_leaf_left(trans, root, path, 1, 1,
4771 if (wret < 0 && wret != -ENOSPC)
4774 if (path->nodes[0] == leaf &&
4775 btrfs_header_nritems(leaf)) {
4776 wret = push_leaf_right(trans, root, path, 1,
4778 if (wret < 0 && wret != -ENOSPC)
4782 if (btrfs_header_nritems(leaf) == 0) {
4783 path->slots[1] = slot;
4784 btrfs_del_leaf(trans, root, path, leaf);
4785 free_extent_buffer(leaf);
4788 /* if we're still in the path, make sure
4789 * we're dirty. Otherwise, one of the
4790 * push_leaf functions must have already
4791 * dirtied this buffer
4793 if (path->nodes[0] == leaf)
4794 btrfs_mark_buffer_dirty(leaf);
4795 free_extent_buffer(leaf);
4798 btrfs_mark_buffer_dirty(leaf);
4805 * search the tree again to find a leaf with lesser keys
4806 * returns 0 if it found something or 1 if there are no lesser leaves.
4807 * returns < 0 on io errors.
4809 * This may release the path, and so you may lose any locks held at the
4812 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4814 struct btrfs_key key;
4815 struct btrfs_disk_key found_key;
4818 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4822 else if (key.type > 0)
4824 else if (key.objectid > 0)
4829 btrfs_release_path(path);
4830 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4833 btrfs_item_key(path->nodes[0], &found_key, 0);
4834 ret = comp_keys(&found_key, &key);
4841 * A helper function to walk down the tree starting at min_key, and looking
4842 * for nodes or leaves that are have a minimum transaction id.
4843 * This is used by the btree defrag code, and tree logging
4845 * This does not cow, but it does stuff the starting key it finds back
4846 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4847 * key and get a writable path.
4849 * This does lock as it descends, and path->keep_locks should be set
4850 * to 1 by the caller.
4852 * This honors path->lowest_level to prevent descent past a given level
4855 * min_trans indicates the oldest transaction that you are interested
4856 * in walking through. Any nodes or leaves older than min_trans are
4857 * skipped over (without reading them).
4859 * returns zero if something useful was found, < 0 on error and 1 if there
4860 * was nothing in the tree that matched the search criteria.
4862 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4863 struct btrfs_key *max_key,
4864 struct btrfs_path *path,
4867 struct extent_buffer *cur;
4868 struct btrfs_key found_key;
4875 WARN_ON(!path->keep_locks);
4877 cur = btrfs_read_lock_root_node(root);
4878 level = btrfs_header_level(cur);
4879 WARN_ON(path->nodes[level]);
4880 path->nodes[level] = cur;
4881 path->locks[level] = BTRFS_READ_LOCK;
4883 if (btrfs_header_generation(cur) < min_trans) {
4888 nritems = btrfs_header_nritems(cur);
4889 level = btrfs_header_level(cur);
4890 sret = bin_search(cur, min_key, level, &slot);
4892 /* at the lowest level, we're done, setup the path and exit */
4893 if (level == path->lowest_level) {
4894 if (slot >= nritems)
4897 path->slots[level] = slot;
4898 btrfs_item_key_to_cpu(cur, &found_key, slot);
4901 if (sret && slot > 0)
4904 * check this node pointer against the min_trans parameters.
4905 * If it is too old, old, skip to the next one.
4907 while (slot < nritems) {
4911 blockptr = btrfs_node_blockptr(cur, slot);
4912 gen = btrfs_node_ptr_generation(cur, slot);
4913 if (gen < min_trans) {
4921 * we didn't find a candidate key in this node, walk forward
4922 * and find another one
4924 if (slot >= nritems) {
4925 path->slots[level] = slot;
4926 btrfs_set_path_blocking(path);
4927 sret = btrfs_find_next_key(root, path, min_key, level,
4930 btrfs_release_path(path);
4936 /* save our key for returning back */
4937 btrfs_node_key_to_cpu(cur, &found_key, slot);
4938 path->slots[level] = slot;
4939 if (level == path->lowest_level) {
4941 unlock_up(path, level, 1, 0, NULL);
4944 btrfs_set_path_blocking(path);
4945 cur = read_node_slot(root, cur, slot);
4946 BUG_ON(!cur); /* -ENOMEM */
4948 btrfs_tree_read_lock(cur);
4950 path->locks[level - 1] = BTRFS_READ_LOCK;
4951 path->nodes[level - 1] = cur;
4952 unlock_up(path, level, 1, 0, NULL);
4953 btrfs_clear_path_blocking(path, NULL, 0);
4957 memcpy(min_key, &found_key, sizeof(found_key));
4958 btrfs_set_path_blocking(path);
4962 static void tree_move_down(struct btrfs_root *root,
4963 struct btrfs_path *path,
4964 int *level, int root_level)
4966 BUG_ON(*level == 0);
4967 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
4968 path->slots[*level]);
4969 path->slots[*level - 1] = 0;
4973 static int tree_move_next_or_upnext(struct btrfs_root *root,
4974 struct btrfs_path *path,
4975 int *level, int root_level)
4979 nritems = btrfs_header_nritems(path->nodes[*level]);
4981 path->slots[*level]++;
4983 while (path->slots[*level] >= nritems) {
4984 if (*level == root_level)
4988 path->slots[*level] = 0;
4989 free_extent_buffer(path->nodes[*level]);
4990 path->nodes[*level] = NULL;
4992 path->slots[*level]++;
4994 nritems = btrfs_header_nritems(path->nodes[*level]);
5001 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5004 static int tree_advance(struct btrfs_root *root,
5005 struct btrfs_path *path,
5006 int *level, int root_level,
5008 struct btrfs_key *key)
5012 if (*level == 0 || !allow_down) {
5013 ret = tree_move_next_or_upnext(root, path, level, root_level);
5015 tree_move_down(root, path, level, root_level);
5020 btrfs_item_key_to_cpu(path->nodes[*level], key,
5021 path->slots[*level]);
5023 btrfs_node_key_to_cpu(path->nodes[*level], key,
5024 path->slots[*level]);
5029 static int tree_compare_item(struct btrfs_root *left_root,
5030 struct btrfs_path *left_path,
5031 struct btrfs_path *right_path,
5036 unsigned long off1, off2;
5038 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5039 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5043 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5044 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5045 right_path->slots[0]);
5047 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5049 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5056 #define ADVANCE_ONLY_NEXT -1
5059 * This function compares two trees and calls the provided callback for
5060 * every changed/new/deleted item it finds.
5061 * If shared tree blocks are encountered, whole subtrees are skipped, making
5062 * the compare pretty fast on snapshotted subvolumes.
5064 * This currently works on commit roots only. As commit roots are read only,
5065 * we don't do any locking. The commit roots are protected with transactions.
5066 * Transactions are ended and rejoined when a commit is tried in between.
5068 * This function checks for modifications done to the trees while comparing.
5069 * If it detects a change, it aborts immediately.
5071 int btrfs_compare_trees(struct btrfs_root *left_root,
5072 struct btrfs_root *right_root,
5073 btrfs_changed_cb_t changed_cb, void *ctx)
5077 struct btrfs_trans_handle *trans = NULL;
5078 struct btrfs_path *left_path = NULL;
5079 struct btrfs_path *right_path = NULL;
5080 struct btrfs_key left_key;
5081 struct btrfs_key right_key;
5082 char *tmp_buf = NULL;
5083 int left_root_level;
5084 int right_root_level;
5087 int left_end_reached;
5088 int right_end_reached;
5093 u64 left_start_ctransid;
5094 u64 right_start_ctransid;
5097 left_path = btrfs_alloc_path();
5102 right_path = btrfs_alloc_path();
5108 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5114 left_path->search_commit_root = 1;
5115 left_path->skip_locking = 1;
5116 right_path->search_commit_root = 1;
5117 right_path->skip_locking = 1;
5119 spin_lock(&left_root->root_item_lock);
5120 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5121 spin_unlock(&left_root->root_item_lock);
5123 spin_lock(&right_root->root_item_lock);
5124 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5125 spin_unlock(&right_root->root_item_lock);
5127 trans = btrfs_join_transaction(left_root);
5128 if (IS_ERR(trans)) {
5129 ret = PTR_ERR(trans);
5135 * Strategy: Go to the first items of both trees. Then do
5137 * If both trees are at level 0
5138 * Compare keys of current items
5139 * If left < right treat left item as new, advance left tree
5141 * If left > right treat right item as deleted, advance right tree
5143 * If left == right do deep compare of items, treat as changed if
5144 * needed, advance both trees and repeat
5145 * If both trees are at the same level but not at level 0
5146 * Compare keys of current nodes/leafs
5147 * If left < right advance left tree and repeat
5148 * If left > right advance right tree and repeat
5149 * If left == right compare blockptrs of the next nodes/leafs
5150 * If they match advance both trees but stay at the same level
5152 * If they don't match advance both trees while allowing to go
5154 * If tree levels are different
5155 * Advance the tree that needs it and repeat
5157 * Advancing a tree means:
5158 * If we are at level 0, try to go to the next slot. If that's not
5159 * possible, go one level up and repeat. Stop when we found a level
5160 * where we could go to the next slot. We may at this point be on a
5163 * If we are not at level 0 and not on shared tree blocks, go one
5166 * If we are not at level 0 and on shared tree blocks, go one slot to
5167 * the right if possible or go up and right.
5170 left_level = btrfs_header_level(left_root->commit_root);
5171 left_root_level = left_level;
5172 left_path->nodes[left_level] = left_root->commit_root;
5173 extent_buffer_get(left_path->nodes[left_level]);
5175 right_level = btrfs_header_level(right_root->commit_root);
5176 right_root_level = right_level;
5177 right_path->nodes[right_level] = right_root->commit_root;
5178 extent_buffer_get(right_path->nodes[right_level]);
5180 if (left_level == 0)
5181 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5182 &left_key, left_path->slots[left_level]);
5184 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5185 &left_key, left_path->slots[left_level]);
5186 if (right_level == 0)
5187 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5188 &right_key, right_path->slots[right_level]);
5190 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5191 &right_key, right_path->slots[right_level]);
5193 left_end_reached = right_end_reached = 0;
5194 advance_left = advance_right = 0;
5198 * We need to make sure the transaction does not get committed
5199 * while we do anything on commit roots. This means, we need to
5200 * join and leave transactions for every item that we process.
5202 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5203 btrfs_release_path(left_path);
5204 btrfs_release_path(right_path);
5206 ret = btrfs_end_transaction(trans, left_root);
5211 /* now rejoin the transaction */
5213 trans = btrfs_join_transaction(left_root);
5214 if (IS_ERR(trans)) {
5215 ret = PTR_ERR(trans);
5220 spin_lock(&left_root->root_item_lock);
5221 ctransid = btrfs_root_ctransid(&left_root->root_item);
5222 spin_unlock(&left_root->root_item_lock);
5223 if (ctransid != left_start_ctransid)
5224 left_start_ctransid = 0;
5226 spin_lock(&right_root->root_item_lock);
5227 ctransid = btrfs_root_ctransid(&right_root->root_item);
5228 spin_unlock(&right_root->root_item_lock);
5229 if (ctransid != right_start_ctransid)
5230 right_start_ctransid = 0;
5232 if (!left_start_ctransid || !right_start_ctransid) {
5233 WARN(1, KERN_WARNING
5234 "btrfs: btrfs_compare_tree detected "
5235 "a change in one of the trees while "
5236 "iterating. This is probably a "
5243 * the commit root may have changed, so start again
5246 left_path->lowest_level = left_level;
5247 right_path->lowest_level = right_level;
5248 ret = btrfs_search_slot(NULL, left_root,
5249 &left_key, left_path, 0, 0);
5252 ret = btrfs_search_slot(NULL, right_root,
5253 &right_key, right_path, 0, 0);
5258 if (advance_left && !left_end_reached) {
5259 ret = tree_advance(left_root, left_path, &left_level,
5261 advance_left != ADVANCE_ONLY_NEXT,
5264 left_end_reached = ADVANCE;
5267 if (advance_right && !right_end_reached) {
5268 ret = tree_advance(right_root, right_path, &right_level,
5270 advance_right != ADVANCE_ONLY_NEXT,
5273 right_end_reached = ADVANCE;
5277 if (left_end_reached && right_end_reached) {
5280 } else if (left_end_reached) {
5281 if (right_level == 0) {
5282 ret = changed_cb(left_root, right_root,
5283 left_path, right_path,
5285 BTRFS_COMPARE_TREE_DELETED,
5290 advance_right = ADVANCE;
5292 } else if (right_end_reached) {
5293 if (left_level == 0) {
5294 ret = changed_cb(left_root, right_root,
5295 left_path, right_path,
5297 BTRFS_COMPARE_TREE_NEW,
5302 advance_left = ADVANCE;
5306 if (left_level == 0 && right_level == 0) {
5307 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5309 ret = changed_cb(left_root, right_root,
5310 left_path, right_path,
5312 BTRFS_COMPARE_TREE_NEW,
5316 advance_left = ADVANCE;
5317 } else if (cmp > 0) {
5318 ret = changed_cb(left_root, right_root,
5319 left_path, right_path,
5321 BTRFS_COMPARE_TREE_DELETED,
5325 advance_right = ADVANCE;
5327 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5328 ret = tree_compare_item(left_root, left_path,
5329 right_path, tmp_buf);
5331 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5332 ret = changed_cb(left_root, right_root,
5333 left_path, right_path,
5335 BTRFS_COMPARE_TREE_CHANGED,
5340 advance_left = ADVANCE;
5341 advance_right = ADVANCE;
5343 } else if (left_level == right_level) {
5344 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5346 advance_left = ADVANCE;
5347 } else if (cmp > 0) {
5348 advance_right = ADVANCE;
5350 left_blockptr = btrfs_node_blockptr(
5351 left_path->nodes[left_level],
5352 left_path->slots[left_level]);
5353 right_blockptr = btrfs_node_blockptr(
5354 right_path->nodes[right_level],
5355 right_path->slots[right_level]);
5356 if (left_blockptr == right_blockptr) {
5358 * As we're on a shared block, don't
5359 * allow to go deeper.
5361 advance_left = ADVANCE_ONLY_NEXT;
5362 advance_right = ADVANCE_ONLY_NEXT;
5364 advance_left = ADVANCE;
5365 advance_right = ADVANCE;
5368 } else if (left_level < right_level) {
5369 advance_right = ADVANCE;
5371 advance_left = ADVANCE;
5376 btrfs_free_path(left_path);
5377 btrfs_free_path(right_path);
5382 ret = btrfs_end_transaction(trans, left_root);
5384 btrfs_end_transaction(trans, left_root);
5391 * this is similar to btrfs_next_leaf, but does not try to preserve
5392 * and fixup the path. It looks for and returns the next key in the
5393 * tree based on the current path and the min_trans parameters.
5395 * 0 is returned if another key is found, < 0 if there are any errors
5396 * and 1 is returned if there are no higher keys in the tree
5398 * path->keep_locks should be set to 1 on the search made before
5399 * calling this function.
5401 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5402 struct btrfs_key *key, int level, u64 min_trans)
5405 struct extent_buffer *c;
5407 WARN_ON(!path->keep_locks);
5408 while (level < BTRFS_MAX_LEVEL) {
5409 if (!path->nodes[level])
5412 slot = path->slots[level] + 1;
5413 c = path->nodes[level];
5415 if (slot >= btrfs_header_nritems(c)) {
5418 struct btrfs_key cur_key;
5419 if (level + 1 >= BTRFS_MAX_LEVEL ||
5420 !path->nodes[level + 1])
5423 if (path->locks[level + 1]) {
5428 slot = btrfs_header_nritems(c) - 1;
5430 btrfs_item_key_to_cpu(c, &cur_key, slot);
5432 btrfs_node_key_to_cpu(c, &cur_key, slot);
5434 orig_lowest = path->lowest_level;
5435 btrfs_release_path(path);
5436 path->lowest_level = level;
5437 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5439 path->lowest_level = orig_lowest;
5443 c = path->nodes[level];
5444 slot = path->slots[level];
5451 btrfs_item_key_to_cpu(c, key, slot);
5453 u64 gen = btrfs_node_ptr_generation(c, slot);
5455 if (gen < min_trans) {
5459 btrfs_node_key_to_cpu(c, key, slot);
5467 * search the tree again to find a leaf with greater keys
5468 * returns 0 if it found something or 1 if there are no greater leaves.
5469 * returns < 0 on io errors.
5471 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5473 return btrfs_next_old_leaf(root, path, 0);
5476 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5481 struct extent_buffer *c;
5482 struct extent_buffer *next;
5483 struct btrfs_key key;
5486 int old_spinning = path->leave_spinning;
5487 int next_rw_lock = 0;
5489 nritems = btrfs_header_nritems(path->nodes[0]);
5493 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5498 btrfs_release_path(path);
5500 path->keep_locks = 1;
5501 path->leave_spinning = 1;
5504 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5506 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5507 path->keep_locks = 0;
5512 nritems = btrfs_header_nritems(path->nodes[0]);
5514 * by releasing the path above we dropped all our locks. A balance
5515 * could have added more items next to the key that used to be
5516 * at the very end of the block. So, check again here and
5517 * advance the path if there are now more items available.
5519 if (nritems > 0 && path->slots[0] < nritems - 1) {
5526 while (level < BTRFS_MAX_LEVEL) {
5527 if (!path->nodes[level]) {
5532 slot = path->slots[level] + 1;
5533 c = path->nodes[level];
5534 if (slot >= btrfs_header_nritems(c)) {
5536 if (level == BTRFS_MAX_LEVEL) {
5544 btrfs_tree_unlock_rw(next, next_rw_lock);
5545 free_extent_buffer(next);
5549 next_rw_lock = path->locks[level];
5550 ret = read_block_for_search(NULL, root, path, &next, level,
5556 btrfs_release_path(path);
5560 if (!path->skip_locking) {
5561 ret = btrfs_try_tree_read_lock(next);
5562 if (!ret && time_seq) {
5564 * If we don't get the lock, we may be racing
5565 * with push_leaf_left, holding that lock while
5566 * itself waiting for the leaf we've currently
5567 * locked. To solve this situation, we give up
5568 * on our lock and cycle.
5570 free_extent_buffer(next);
5571 btrfs_release_path(path);
5576 btrfs_set_path_blocking(path);
5577 btrfs_tree_read_lock(next);
5578 btrfs_clear_path_blocking(path, next,
5581 next_rw_lock = BTRFS_READ_LOCK;
5585 path->slots[level] = slot;
5588 c = path->nodes[level];
5589 if (path->locks[level])
5590 btrfs_tree_unlock_rw(c, path->locks[level]);
5592 free_extent_buffer(c);
5593 path->nodes[level] = next;
5594 path->slots[level] = 0;
5595 if (!path->skip_locking)
5596 path->locks[level] = next_rw_lock;
5600 ret = read_block_for_search(NULL, root, path, &next, level,
5606 btrfs_release_path(path);
5610 if (!path->skip_locking) {
5611 ret = btrfs_try_tree_read_lock(next);
5613 btrfs_set_path_blocking(path);
5614 btrfs_tree_read_lock(next);
5615 btrfs_clear_path_blocking(path, next,
5618 next_rw_lock = BTRFS_READ_LOCK;
5623 unlock_up(path, 0, 1, 0, NULL);
5624 path->leave_spinning = old_spinning;
5626 btrfs_set_path_blocking(path);
5632 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5633 * searching until it gets past min_objectid or finds an item of 'type'
5635 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5637 int btrfs_previous_item(struct btrfs_root *root,
5638 struct btrfs_path *path, u64 min_objectid,
5641 struct btrfs_key found_key;
5642 struct extent_buffer *leaf;
5647 if (path->slots[0] == 0) {
5648 btrfs_set_path_blocking(path);
5649 ret = btrfs_prev_leaf(root, path);
5655 leaf = path->nodes[0];
5656 nritems = btrfs_header_nritems(leaf);
5659 if (path->slots[0] == nritems)
5662 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5663 if (found_key.objectid < min_objectid)
5665 if (found_key.type == type)
5667 if (found_key.objectid == min_objectid &&
5668 found_key.type < type)