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 static int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path);
46 struct btrfs_path *btrfs_alloc_path(void)
48 struct btrfs_path *path;
49 path = kmem_cache_zalloc(btrfs_path_cachep, GFP_NOFS);
54 * set all locked nodes in the path to blocking locks. This should
55 * be done before scheduling
57 noinline void btrfs_set_path_blocking(struct btrfs_path *p)
60 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
61 if (!p->nodes[i] || !p->locks[i])
63 btrfs_set_lock_blocking_rw(p->nodes[i], p->locks[i]);
64 if (p->locks[i] == BTRFS_READ_LOCK)
65 p->locks[i] = BTRFS_READ_LOCK_BLOCKING;
66 else if (p->locks[i] == BTRFS_WRITE_LOCK)
67 p->locks[i] = BTRFS_WRITE_LOCK_BLOCKING;
72 * reset all the locked nodes in the patch to spinning locks.
74 * held is used to keep lockdep happy, when lockdep is enabled
75 * we set held to a blocking lock before we go around and
76 * retake all the spinlocks in the path. You can safely use NULL
79 noinline void btrfs_clear_path_blocking(struct btrfs_path *p,
80 struct extent_buffer *held, int held_rw)
84 #ifdef CONFIG_DEBUG_LOCK_ALLOC
85 /* lockdep really cares that we take all of these spinlocks
86 * in the right order. If any of the locks in the path are not
87 * currently blocking, it is going to complain. So, make really
88 * really sure by forcing the path to blocking before we clear
92 btrfs_set_lock_blocking_rw(held, held_rw);
93 if (held_rw == BTRFS_WRITE_LOCK)
94 held_rw = BTRFS_WRITE_LOCK_BLOCKING;
95 else if (held_rw == BTRFS_READ_LOCK)
96 held_rw = BTRFS_READ_LOCK_BLOCKING;
98 btrfs_set_path_blocking(p);
101 for (i = BTRFS_MAX_LEVEL - 1; i >= 0; i--) {
102 if (p->nodes[i] && p->locks[i]) {
103 btrfs_clear_lock_blocking_rw(p->nodes[i], p->locks[i]);
104 if (p->locks[i] == BTRFS_WRITE_LOCK_BLOCKING)
105 p->locks[i] = BTRFS_WRITE_LOCK;
106 else if (p->locks[i] == BTRFS_READ_LOCK_BLOCKING)
107 p->locks[i] = BTRFS_READ_LOCK;
111 #ifdef CONFIG_DEBUG_LOCK_ALLOC
113 btrfs_clear_lock_blocking_rw(held, held_rw);
117 /* this also releases the path */
118 void btrfs_free_path(struct btrfs_path *p)
122 btrfs_release_path(p);
123 kmem_cache_free(btrfs_path_cachep, p);
127 * path release drops references on the extent buffers in the path
128 * and it drops any locks held by this path
130 * It is safe to call this on paths that no locks or extent buffers held.
132 noinline void btrfs_release_path(struct btrfs_path *p)
136 for (i = 0; i < BTRFS_MAX_LEVEL; i++) {
141 btrfs_tree_unlock_rw(p->nodes[i], p->locks[i]);
144 free_extent_buffer(p->nodes[i]);
150 * safely gets a reference on the root node of a tree. A lock
151 * is not taken, so a concurrent writer may put a different node
152 * at the root of the tree. See btrfs_lock_root_node for the
155 * The extent buffer returned by this has a reference taken, so
156 * it won't disappear. It may stop being the root of the tree
157 * at any time because there are no locks held.
159 struct extent_buffer *btrfs_root_node(struct btrfs_root *root)
161 struct extent_buffer *eb;
165 eb = rcu_dereference(root->node);
168 * RCU really hurts here, we could free up the root node because
169 * it was cow'ed but we may not get the new root node yet so do
170 * the inc_not_zero dance and if it doesn't work then
171 * synchronize_rcu and try again.
173 if (atomic_inc_not_zero(&eb->refs)) {
183 /* loop around taking references on and locking the root node of the
184 * tree until you end up with a lock on the root. A locked buffer
185 * is returned, with a reference held.
187 struct extent_buffer *btrfs_lock_root_node(struct btrfs_root *root)
189 struct extent_buffer *eb;
192 eb = btrfs_root_node(root);
194 if (eb == root->node)
196 btrfs_tree_unlock(eb);
197 free_extent_buffer(eb);
202 /* loop around taking references on and locking the root node of the
203 * tree until you end up with a lock on the root. A locked buffer
204 * is returned, with a reference held.
206 static struct extent_buffer *btrfs_read_lock_root_node(struct btrfs_root *root)
208 struct extent_buffer *eb;
211 eb = btrfs_root_node(root);
212 btrfs_tree_read_lock(eb);
213 if (eb == root->node)
215 btrfs_tree_read_unlock(eb);
216 free_extent_buffer(eb);
221 /* cowonly root (everything not a reference counted cow subvolume), just get
222 * put onto a simple dirty list. transaction.c walks this to make sure they
223 * get properly updated on disk.
225 static void add_root_to_dirty_list(struct btrfs_root *root)
227 spin_lock(&root->fs_info->trans_lock);
228 if (root->track_dirty && list_empty(&root->dirty_list)) {
229 list_add(&root->dirty_list,
230 &root->fs_info->dirty_cowonly_roots);
232 spin_unlock(&root->fs_info->trans_lock);
236 * used by snapshot creation to make a copy of a root for a tree with
237 * a given objectid. The buffer with the new root node is returned in
238 * cow_ret, and this func returns zero on success or a negative error code.
240 int btrfs_copy_root(struct btrfs_trans_handle *trans,
241 struct btrfs_root *root,
242 struct extent_buffer *buf,
243 struct extent_buffer **cow_ret, u64 new_root_objectid)
245 struct extent_buffer *cow;
248 struct btrfs_disk_key disk_key;
250 WARN_ON(root->ref_cows && trans->transid !=
251 root->fs_info->running_transaction->transid);
252 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
254 level = btrfs_header_level(buf);
256 btrfs_item_key(buf, &disk_key, 0);
258 btrfs_node_key(buf, &disk_key, 0);
260 cow = btrfs_alloc_free_block(trans, root, buf->len, 0,
261 new_root_objectid, &disk_key, level,
266 copy_extent_buffer(cow, buf, 0, 0, cow->len);
267 btrfs_set_header_bytenr(cow, cow->start);
268 btrfs_set_header_generation(cow, trans->transid);
269 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
270 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
271 BTRFS_HEADER_FLAG_RELOC);
272 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
273 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
275 btrfs_set_header_owner(cow, new_root_objectid);
277 write_extent_buffer(cow, root->fs_info->fsid,
278 (unsigned long)btrfs_header_fsid(cow),
281 WARN_ON(btrfs_header_generation(buf) > trans->transid);
282 if (new_root_objectid == BTRFS_TREE_RELOC_OBJECTID)
283 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
285 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
290 btrfs_mark_buffer_dirty(cow);
299 MOD_LOG_KEY_REMOVE_WHILE_FREEING,
300 MOD_LOG_KEY_REMOVE_WHILE_MOVING,
302 MOD_LOG_ROOT_REPLACE,
305 struct tree_mod_move {
310 struct tree_mod_root {
315 struct tree_mod_elem {
317 u64 index; /* shifted logical */
321 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
324 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
327 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
328 struct btrfs_disk_key key;
331 /* this is used for op == MOD_LOG_MOVE_KEYS */
332 struct tree_mod_move move;
334 /* this is used for op == MOD_LOG_ROOT_REPLACE */
335 struct tree_mod_root old_root;
338 static inline void tree_mod_log_read_lock(struct btrfs_fs_info *fs_info)
340 read_lock(&fs_info->tree_mod_log_lock);
343 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info *fs_info)
345 read_unlock(&fs_info->tree_mod_log_lock);
348 static inline void tree_mod_log_write_lock(struct btrfs_fs_info *fs_info)
350 write_lock(&fs_info->tree_mod_log_lock);
353 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info *fs_info)
355 write_unlock(&fs_info->tree_mod_log_lock);
359 * Increment the upper half of tree_mod_seq, set lower half zero.
361 * Must be called with fs_info->tree_mod_seq_lock held.
363 static inline u64 btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info *fs_info)
365 u64 seq = atomic64_read(&fs_info->tree_mod_seq);
366 seq &= 0xffffffff00000000ull;
368 atomic64_set(&fs_info->tree_mod_seq, seq);
373 * Increment the lower half of tree_mod_seq.
375 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
376 * are generated should not technically require a spin lock here. (Rationale:
377 * incrementing the minor while incrementing the major seq number is between its
378 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
379 * just returns a unique sequence number as usual.) We have decided to leave
380 * that requirement in here and rethink it once we notice it really imposes a
381 * problem on some workload.
383 static inline u64 btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info *fs_info)
385 return atomic64_inc_return(&fs_info->tree_mod_seq);
389 * return the last minor in the previous major tree_mod_seq number
391 u64 btrfs_tree_mod_seq_prev(u64 seq)
393 return (seq & 0xffffffff00000000ull) - 1ull;
397 * This adds a new blocker to the tree mod log's blocker list if the @elem
398 * passed does not already have a sequence number set. So when a caller expects
399 * to record tree modifications, it should ensure to set elem->seq to zero
400 * before calling btrfs_get_tree_mod_seq.
401 * Returns a fresh, unused tree log modification sequence number, even if no new
404 u64 btrfs_get_tree_mod_seq(struct btrfs_fs_info *fs_info,
405 struct seq_list *elem)
409 tree_mod_log_write_lock(fs_info);
410 spin_lock(&fs_info->tree_mod_seq_lock);
412 elem->seq = btrfs_inc_tree_mod_seq_major(fs_info);
413 list_add_tail(&elem->list, &fs_info->tree_mod_seq_list);
415 seq = btrfs_inc_tree_mod_seq_minor(fs_info);
416 spin_unlock(&fs_info->tree_mod_seq_lock);
417 tree_mod_log_write_unlock(fs_info);
422 void btrfs_put_tree_mod_seq(struct btrfs_fs_info *fs_info,
423 struct seq_list *elem)
425 struct rb_root *tm_root;
426 struct rb_node *node;
427 struct rb_node *next;
428 struct seq_list *cur_elem;
429 struct tree_mod_elem *tm;
430 u64 min_seq = (u64)-1;
431 u64 seq_putting = elem->seq;
436 spin_lock(&fs_info->tree_mod_seq_lock);
437 list_del(&elem->list);
440 list_for_each_entry(cur_elem, &fs_info->tree_mod_seq_list, list) {
441 if (cur_elem->seq < min_seq) {
442 if (seq_putting > cur_elem->seq) {
444 * blocker with lower sequence number exists, we
445 * cannot remove anything from the log
447 spin_unlock(&fs_info->tree_mod_seq_lock);
450 min_seq = cur_elem->seq;
453 spin_unlock(&fs_info->tree_mod_seq_lock);
456 * anything that's lower than the lowest existing (read: blocked)
457 * sequence number can be removed from the tree.
459 tree_mod_log_write_lock(fs_info);
460 tm_root = &fs_info->tree_mod_log;
461 for (node = rb_first(tm_root); node; node = next) {
462 next = rb_next(node);
463 tm = container_of(node, struct tree_mod_elem, node);
464 if (tm->seq > min_seq)
466 rb_erase(node, tm_root);
469 tree_mod_log_write_unlock(fs_info);
473 * key order of the log:
476 * the index is the shifted logical of the *new* root node for root replace
477 * operations, or the shifted logical of the affected block for all other
481 __tree_mod_log_insert(struct btrfs_fs_info *fs_info, struct tree_mod_elem *tm)
483 struct rb_root *tm_root;
484 struct rb_node **new;
485 struct rb_node *parent = NULL;
486 struct tree_mod_elem *cur;
488 BUG_ON(!tm || !tm->seq);
490 tm_root = &fs_info->tree_mod_log;
491 new = &tm_root->rb_node;
493 cur = container_of(*new, struct tree_mod_elem, node);
495 if (cur->index < tm->index)
496 new = &((*new)->rb_left);
497 else if (cur->index > tm->index)
498 new = &((*new)->rb_right);
499 else if (cur->seq < tm->seq)
500 new = &((*new)->rb_left);
501 else if (cur->seq > tm->seq)
502 new = &((*new)->rb_right);
509 rb_link_node(&tm->node, parent, new);
510 rb_insert_color(&tm->node, tm_root);
515 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
516 * returns zero with the tree_mod_log_lock acquired. The caller must hold
517 * this until all tree mod log insertions are recorded in the rb tree and then
518 * call tree_mod_log_write_unlock() to release.
520 static inline int tree_mod_dont_log(struct btrfs_fs_info *fs_info,
521 struct extent_buffer *eb) {
523 if (list_empty(&(fs_info)->tree_mod_seq_list))
525 if (eb && btrfs_header_level(eb) == 0)
528 tree_mod_log_write_lock(fs_info);
529 if (list_empty(&fs_info->tree_mod_seq_list)) {
531 * someone emptied the list while we were waiting for the lock.
532 * we must not add to the list when no blocker exists.
534 tree_mod_log_write_unlock(fs_info);
542 * This allocates memory and gets a tree modification sequence number.
544 * Returns <0 on error.
545 * Returns >0 (the added sequence number) on success.
547 static inline int tree_mod_alloc(struct btrfs_fs_info *fs_info, gfp_t flags,
548 struct tree_mod_elem **tm_ret)
550 struct tree_mod_elem *tm;
553 * once we switch from spin locks to something different, we should
554 * honor the flags parameter here.
556 tm = *tm_ret = kzalloc(sizeof(*tm), GFP_ATOMIC);
560 spin_lock(&fs_info->tree_mod_seq_lock);
561 tm->seq = btrfs_inc_tree_mod_seq_minor(fs_info);
562 spin_unlock(&fs_info->tree_mod_seq_lock);
568 __tree_mod_log_insert_key(struct btrfs_fs_info *fs_info,
569 struct extent_buffer *eb, int slot,
570 enum mod_log_op op, gfp_t flags)
573 struct tree_mod_elem *tm;
575 ret = tree_mod_alloc(fs_info, flags, &tm);
579 tm->index = eb->start >> PAGE_CACHE_SHIFT;
580 if (op != MOD_LOG_KEY_ADD) {
581 btrfs_node_key(eb, &tm->key, slot);
582 tm->blockptr = btrfs_node_blockptr(eb, slot);
586 tm->generation = btrfs_node_ptr_generation(eb, slot);
588 return __tree_mod_log_insert(fs_info, tm);
592 tree_mod_log_insert_key_mask(struct btrfs_fs_info *fs_info,
593 struct extent_buffer *eb, int slot,
594 enum mod_log_op op, gfp_t flags)
598 if (tree_mod_dont_log(fs_info, eb))
601 ret = __tree_mod_log_insert_key(fs_info, eb, slot, op, flags);
603 tree_mod_log_write_unlock(fs_info);
608 tree_mod_log_insert_key(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
609 int slot, enum mod_log_op op)
611 return tree_mod_log_insert_key_mask(fs_info, eb, slot, op, GFP_NOFS);
615 tree_mod_log_insert_key_locked(struct btrfs_fs_info *fs_info,
616 struct extent_buffer *eb, int slot,
619 return __tree_mod_log_insert_key(fs_info, eb, slot, op, GFP_NOFS);
623 tree_mod_log_insert_move(struct btrfs_fs_info *fs_info,
624 struct extent_buffer *eb, int dst_slot, int src_slot,
625 int nr_items, gfp_t flags)
627 struct tree_mod_elem *tm;
631 if (tree_mod_dont_log(fs_info, eb))
635 * When we override something during the move, we log these removals.
636 * This can only happen when we move towards the beginning of the
637 * buffer, i.e. dst_slot < src_slot.
639 for (i = 0; i + dst_slot < src_slot && i < nr_items; i++) {
640 ret = tree_mod_log_insert_key_locked(fs_info, eb, i + dst_slot,
641 MOD_LOG_KEY_REMOVE_WHILE_MOVING);
645 ret = tree_mod_alloc(fs_info, flags, &tm);
649 tm->index = eb->start >> PAGE_CACHE_SHIFT;
651 tm->move.dst_slot = dst_slot;
652 tm->move.nr_items = nr_items;
653 tm->op = MOD_LOG_MOVE_KEYS;
655 ret = __tree_mod_log_insert(fs_info, tm);
657 tree_mod_log_write_unlock(fs_info);
662 __tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
668 if (btrfs_header_level(eb) == 0)
671 nritems = btrfs_header_nritems(eb);
672 for (i = nritems - 1; i >= 0; i--) {
673 ret = tree_mod_log_insert_key_locked(fs_info, eb, i,
674 MOD_LOG_KEY_REMOVE_WHILE_FREEING);
680 tree_mod_log_insert_root(struct btrfs_fs_info *fs_info,
681 struct extent_buffer *old_root,
682 struct extent_buffer *new_root, gfp_t flags,
685 struct tree_mod_elem *tm;
688 if (tree_mod_dont_log(fs_info, NULL))
692 __tree_mod_log_free_eb(fs_info, old_root);
694 ret = tree_mod_alloc(fs_info, flags, &tm);
698 tm->index = new_root->start >> PAGE_CACHE_SHIFT;
699 tm->old_root.logical = old_root->start;
700 tm->old_root.level = btrfs_header_level(old_root);
701 tm->generation = btrfs_header_generation(old_root);
702 tm->op = MOD_LOG_ROOT_REPLACE;
704 ret = __tree_mod_log_insert(fs_info, tm);
706 tree_mod_log_write_unlock(fs_info);
710 static struct tree_mod_elem *
711 __tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq,
714 struct rb_root *tm_root;
715 struct rb_node *node;
716 struct tree_mod_elem *cur = NULL;
717 struct tree_mod_elem *found = NULL;
718 u64 index = start >> PAGE_CACHE_SHIFT;
720 tree_mod_log_read_lock(fs_info);
721 tm_root = &fs_info->tree_mod_log;
722 node = tm_root->rb_node;
724 cur = container_of(node, struct tree_mod_elem, node);
725 if (cur->index < index) {
726 node = node->rb_left;
727 } else if (cur->index > index) {
728 node = node->rb_right;
729 } else if (cur->seq < min_seq) {
730 node = node->rb_left;
731 } else if (!smallest) {
732 /* we want the node with the highest seq */
734 BUG_ON(found->seq > cur->seq);
736 node = node->rb_left;
737 } else if (cur->seq > min_seq) {
738 /* we want the node with the smallest seq */
740 BUG_ON(found->seq < cur->seq);
742 node = node->rb_right;
748 tree_mod_log_read_unlock(fs_info);
754 * this returns the element from the log with the smallest time sequence
755 * value that's in the log (the oldest log item). any element with a time
756 * sequence lower than min_seq will be ignored.
758 static struct tree_mod_elem *
759 tree_mod_log_search_oldest(struct btrfs_fs_info *fs_info, u64 start,
762 return __tree_mod_log_search(fs_info, start, min_seq, 1);
766 * this returns the element from the log with the largest time sequence
767 * value that's in the log (the most recent log item). any element with
768 * a time sequence lower than min_seq will be ignored.
770 static struct tree_mod_elem *
771 tree_mod_log_search(struct btrfs_fs_info *fs_info, u64 start, u64 min_seq)
773 return __tree_mod_log_search(fs_info, start, min_seq, 0);
777 tree_mod_log_eb_copy(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
778 struct extent_buffer *src, unsigned long dst_offset,
779 unsigned long src_offset, int nr_items)
784 if (tree_mod_dont_log(fs_info, NULL))
787 if (btrfs_header_level(dst) == 0 && btrfs_header_level(src) == 0) {
788 tree_mod_log_write_unlock(fs_info);
792 for (i = 0; i < nr_items; i++) {
793 ret = tree_mod_log_insert_key_locked(fs_info, src,
797 ret = tree_mod_log_insert_key_locked(fs_info, dst,
803 tree_mod_log_write_unlock(fs_info);
807 tree_mod_log_eb_move(struct btrfs_fs_info *fs_info, struct extent_buffer *dst,
808 int dst_offset, int src_offset, int nr_items)
811 ret = tree_mod_log_insert_move(fs_info, dst, dst_offset, src_offset,
817 tree_mod_log_set_node_key(struct btrfs_fs_info *fs_info,
818 struct extent_buffer *eb, int slot, int atomic)
822 ret = tree_mod_log_insert_key_mask(fs_info, eb, slot,
824 atomic ? GFP_ATOMIC : GFP_NOFS);
829 tree_mod_log_free_eb(struct btrfs_fs_info *fs_info, struct extent_buffer *eb)
831 if (tree_mod_dont_log(fs_info, eb))
834 __tree_mod_log_free_eb(fs_info, eb);
836 tree_mod_log_write_unlock(fs_info);
840 tree_mod_log_set_root_pointer(struct btrfs_root *root,
841 struct extent_buffer *new_root_node,
845 ret = tree_mod_log_insert_root(root->fs_info, root->node,
846 new_root_node, GFP_NOFS, log_removal);
851 * check if the tree block can be shared by multiple trees
853 int btrfs_block_can_be_shared(struct btrfs_root *root,
854 struct extent_buffer *buf)
857 * Tree blocks not in refernece counted trees and tree roots
858 * are never shared. If a block was allocated after the last
859 * snapshot and the block was not allocated by tree relocation,
860 * we know the block is not shared.
862 if (root->ref_cows &&
863 buf != root->node && buf != root->commit_root &&
864 (btrfs_header_generation(buf) <=
865 btrfs_root_last_snapshot(&root->root_item) ||
866 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)))
868 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
869 if (root->ref_cows &&
870 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
876 static noinline int update_ref_for_cow(struct btrfs_trans_handle *trans,
877 struct btrfs_root *root,
878 struct extent_buffer *buf,
879 struct extent_buffer *cow,
889 * Backrefs update rules:
891 * Always use full backrefs for extent pointers in tree block
892 * allocated by tree relocation.
894 * If a shared tree block is no longer referenced by its owner
895 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
896 * use full backrefs for extent pointers in tree block.
898 * If a tree block is been relocating
899 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
900 * use full backrefs for extent pointers in tree block.
901 * The reason for this is some operations (such as drop tree)
902 * are only allowed for blocks use full backrefs.
905 if (btrfs_block_can_be_shared(root, buf)) {
906 ret = btrfs_lookup_extent_info(trans, root, buf->start,
907 btrfs_header_level(buf), 1,
913 btrfs_std_error(root->fs_info, ret);
918 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
919 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
920 flags = BTRFS_BLOCK_FLAG_FULL_BACKREF;
925 owner = btrfs_header_owner(buf);
926 BUG_ON(owner == BTRFS_TREE_RELOC_OBJECTID &&
927 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF));
930 if ((owner == root->root_key.objectid ||
931 root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) &&
932 !(flags & BTRFS_BLOCK_FLAG_FULL_BACKREF)) {
933 ret = btrfs_inc_ref(trans, root, buf, 1, 1);
934 BUG_ON(ret); /* -ENOMEM */
936 if (root->root_key.objectid ==
937 BTRFS_TREE_RELOC_OBJECTID) {
938 ret = btrfs_dec_ref(trans, root, buf, 0, 1);
939 BUG_ON(ret); /* -ENOMEM */
940 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
941 BUG_ON(ret); /* -ENOMEM */
943 new_flags |= BTRFS_BLOCK_FLAG_FULL_BACKREF;
946 if (root->root_key.objectid ==
947 BTRFS_TREE_RELOC_OBJECTID)
948 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
950 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
951 BUG_ON(ret); /* -ENOMEM */
953 if (new_flags != 0) {
954 int level = btrfs_header_level(buf);
956 ret = btrfs_set_disk_extent_flags(trans, root,
959 new_flags, level, 0);
964 if (flags & BTRFS_BLOCK_FLAG_FULL_BACKREF) {
965 if (root->root_key.objectid ==
966 BTRFS_TREE_RELOC_OBJECTID)
967 ret = btrfs_inc_ref(trans, root, cow, 1, 1);
969 ret = btrfs_inc_ref(trans, root, cow, 0, 1);
970 BUG_ON(ret); /* -ENOMEM */
971 ret = btrfs_dec_ref(trans, root, buf, 1, 1);
972 BUG_ON(ret); /* -ENOMEM */
974 clean_tree_block(trans, root, buf);
981 * does the dirty work in cow of a single block. The parent block (if
982 * supplied) is updated to point to the new cow copy. The new buffer is marked
983 * dirty and returned locked. If you modify the block it needs to be marked
986 * search_start -- an allocation hint for the new block
988 * empty_size -- a hint that you plan on doing more cow. This is the size in
989 * bytes the allocator should try to find free next to the block it returns.
990 * This is just a hint and may be ignored by the allocator.
992 static noinline int __btrfs_cow_block(struct btrfs_trans_handle *trans,
993 struct btrfs_root *root,
994 struct extent_buffer *buf,
995 struct extent_buffer *parent, int parent_slot,
996 struct extent_buffer **cow_ret,
997 u64 search_start, u64 empty_size)
999 struct btrfs_disk_key disk_key;
1000 struct extent_buffer *cow;
1003 int unlock_orig = 0;
1006 if (*cow_ret == buf)
1009 btrfs_assert_tree_locked(buf);
1011 WARN_ON(root->ref_cows && trans->transid !=
1012 root->fs_info->running_transaction->transid);
1013 WARN_ON(root->ref_cows && trans->transid != root->last_trans);
1015 level = btrfs_header_level(buf);
1018 btrfs_item_key(buf, &disk_key, 0);
1020 btrfs_node_key(buf, &disk_key, 0);
1022 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID) {
1024 parent_start = parent->start;
1030 cow = btrfs_alloc_free_block(trans, root, buf->len, parent_start,
1031 root->root_key.objectid, &disk_key,
1032 level, search_start, empty_size);
1034 return PTR_ERR(cow);
1036 /* cow is set to blocking by btrfs_init_new_buffer */
1038 copy_extent_buffer(cow, buf, 0, 0, cow->len);
1039 btrfs_set_header_bytenr(cow, cow->start);
1040 btrfs_set_header_generation(cow, trans->transid);
1041 btrfs_set_header_backref_rev(cow, BTRFS_MIXED_BACKREF_REV);
1042 btrfs_clear_header_flag(cow, BTRFS_HEADER_FLAG_WRITTEN |
1043 BTRFS_HEADER_FLAG_RELOC);
1044 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1045 btrfs_set_header_flag(cow, BTRFS_HEADER_FLAG_RELOC);
1047 btrfs_set_header_owner(cow, root->root_key.objectid);
1049 write_extent_buffer(cow, root->fs_info->fsid,
1050 (unsigned long)btrfs_header_fsid(cow),
1053 ret = update_ref_for_cow(trans, root, buf, cow, &last_ref);
1055 btrfs_abort_transaction(trans, root, ret);
1060 btrfs_reloc_cow_block(trans, root, buf, cow);
1062 if (buf == root->node) {
1063 WARN_ON(parent && parent != buf);
1064 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID ||
1065 btrfs_header_backref_rev(buf) < BTRFS_MIXED_BACKREF_REV)
1066 parent_start = buf->start;
1070 extent_buffer_get(cow);
1071 tree_mod_log_set_root_pointer(root, cow, 1);
1072 rcu_assign_pointer(root->node, cow);
1074 btrfs_free_tree_block(trans, root, buf, parent_start,
1076 free_extent_buffer(buf);
1077 add_root_to_dirty_list(root);
1079 if (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID)
1080 parent_start = parent->start;
1084 WARN_ON(trans->transid != btrfs_header_generation(parent));
1085 tree_mod_log_insert_key(root->fs_info, parent, parent_slot,
1086 MOD_LOG_KEY_REPLACE);
1087 btrfs_set_node_blockptr(parent, parent_slot,
1089 btrfs_set_node_ptr_generation(parent, parent_slot,
1091 btrfs_mark_buffer_dirty(parent);
1092 tree_mod_log_free_eb(root->fs_info, buf);
1093 btrfs_free_tree_block(trans, root, buf, parent_start,
1097 btrfs_tree_unlock(buf);
1098 free_extent_buffer_stale(buf);
1099 btrfs_mark_buffer_dirty(cow);
1105 * returns the logical address of the oldest predecessor of the given root.
1106 * entries older than time_seq are ignored.
1108 static struct tree_mod_elem *
1109 __tree_mod_log_oldest_root(struct btrfs_fs_info *fs_info,
1110 struct extent_buffer *eb_root, u64 time_seq)
1112 struct tree_mod_elem *tm;
1113 struct tree_mod_elem *found = NULL;
1114 u64 root_logical = eb_root->start;
1121 * the very last operation that's logged for a root is the replacement
1122 * operation (if it is replaced at all). this has the index of the *new*
1123 * root, making it the very first operation that's logged for this root.
1126 tm = tree_mod_log_search_oldest(fs_info, root_logical,
1131 * if there are no tree operation for the oldest root, we simply
1132 * return it. this should only happen if that (old) root is at
1139 * if there's an operation that's not a root replacement, we
1140 * found the oldest version of our root. normally, we'll find a
1141 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1143 if (tm->op != MOD_LOG_ROOT_REPLACE)
1147 root_logical = tm->old_root.logical;
1151 /* if there's no old root to return, return what we found instead */
1159 * tm is a pointer to the first operation to rewind within eb. then, all
1160 * previous operations will be rewinded (until we reach something older than
1164 __tree_mod_log_rewind(struct extent_buffer *eb, u64 time_seq,
1165 struct tree_mod_elem *first_tm)
1168 struct rb_node *next;
1169 struct tree_mod_elem *tm = first_tm;
1170 unsigned long o_dst;
1171 unsigned long o_src;
1172 unsigned long p_size = sizeof(struct btrfs_key_ptr);
1174 n = btrfs_header_nritems(eb);
1175 while (tm && tm->seq >= time_seq) {
1177 * all the operations are recorded with the operator used for
1178 * the modification. as we're going backwards, we do the
1179 * opposite of each operation here.
1182 case MOD_LOG_KEY_REMOVE_WHILE_FREEING:
1183 BUG_ON(tm->slot < n);
1185 case MOD_LOG_KEY_REMOVE_WHILE_MOVING:
1186 case MOD_LOG_KEY_REMOVE:
1187 btrfs_set_node_key(eb, &tm->key, tm->slot);
1188 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1189 btrfs_set_node_ptr_generation(eb, tm->slot,
1193 case MOD_LOG_KEY_REPLACE:
1194 BUG_ON(tm->slot >= n);
1195 btrfs_set_node_key(eb, &tm->key, tm->slot);
1196 btrfs_set_node_blockptr(eb, tm->slot, tm->blockptr);
1197 btrfs_set_node_ptr_generation(eb, tm->slot,
1200 case MOD_LOG_KEY_ADD:
1201 /* if a move operation is needed it's in the log */
1204 case MOD_LOG_MOVE_KEYS:
1205 o_dst = btrfs_node_key_ptr_offset(tm->slot);
1206 o_src = btrfs_node_key_ptr_offset(tm->move.dst_slot);
1207 memmove_extent_buffer(eb, o_dst, o_src,
1208 tm->move.nr_items * p_size);
1210 case MOD_LOG_ROOT_REPLACE:
1212 * this operation is special. for roots, this must be
1213 * handled explicitly before rewinding.
1214 * for non-roots, this operation may exist if the node
1215 * was a root: root A -> child B; then A gets empty and
1216 * B is promoted to the new root. in the mod log, we'll
1217 * have a root-replace operation for B, a tree block
1218 * that is no root. we simply ignore that operation.
1222 next = rb_next(&tm->node);
1225 tm = container_of(next, struct tree_mod_elem, node);
1226 if (tm->index != first_tm->index)
1229 btrfs_set_header_nritems(eb, n);
1233 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1234 * is returned. If rewind operations happen, a fresh buffer is returned. The
1235 * returned buffer is always read-locked. If the returned buffer is not the
1236 * input buffer, the lock on the input buffer is released and the input buffer
1237 * is freed (its refcount is decremented).
1239 static struct extent_buffer *
1240 tree_mod_log_rewind(struct btrfs_fs_info *fs_info, struct extent_buffer *eb,
1243 struct extent_buffer *eb_rewin;
1244 struct tree_mod_elem *tm;
1249 if (btrfs_header_level(eb) == 0)
1252 tm = tree_mod_log_search(fs_info, eb->start, time_seq);
1256 if (tm->op == MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1257 BUG_ON(tm->slot != 0);
1258 eb_rewin = alloc_dummy_extent_buffer(eb->start,
1259 fs_info->tree_root->nodesize);
1261 btrfs_set_header_bytenr(eb_rewin, eb->start);
1262 btrfs_set_header_backref_rev(eb_rewin,
1263 btrfs_header_backref_rev(eb));
1264 btrfs_set_header_owner(eb_rewin, btrfs_header_owner(eb));
1265 btrfs_set_header_level(eb_rewin, btrfs_header_level(eb));
1267 eb_rewin = btrfs_clone_extent_buffer(eb);
1271 extent_buffer_get(eb_rewin);
1272 btrfs_tree_read_unlock(eb);
1273 free_extent_buffer(eb);
1275 extent_buffer_get(eb_rewin);
1276 btrfs_tree_read_lock(eb_rewin);
1277 __tree_mod_log_rewind(eb_rewin, time_seq, tm);
1278 WARN_ON(btrfs_header_nritems(eb_rewin) >
1279 BTRFS_NODEPTRS_PER_BLOCK(fs_info->tree_root));
1285 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1286 * value. If there are no changes, the current root->root_node is returned. If
1287 * anything changed in between, there's a fresh buffer allocated on which the
1288 * rewind operations are done. In any case, the returned buffer is read locked.
1289 * Returns NULL on error (with no locks held).
1291 static inline struct extent_buffer *
1292 get_old_root(struct btrfs_root *root, u64 time_seq)
1294 struct tree_mod_elem *tm;
1295 struct extent_buffer *eb = NULL;
1296 struct extent_buffer *eb_root;
1297 struct extent_buffer *old;
1298 struct tree_mod_root *old_root = NULL;
1299 u64 old_generation = 0;
1303 eb_root = btrfs_read_lock_root_node(root);
1304 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1308 if (tm->op == MOD_LOG_ROOT_REPLACE) {
1309 old_root = &tm->old_root;
1310 old_generation = tm->generation;
1311 logical = old_root->logical;
1313 logical = eb_root->start;
1316 tm = tree_mod_log_search(root->fs_info, logical, time_seq);
1317 if (old_root && tm && tm->op != MOD_LOG_KEY_REMOVE_WHILE_FREEING) {
1318 btrfs_tree_read_unlock(eb_root);
1319 free_extent_buffer(eb_root);
1320 blocksize = btrfs_level_size(root, old_root->level);
1321 old = read_tree_block(root, logical, blocksize, 0);
1322 if (!old || !extent_buffer_uptodate(old)) {
1323 free_extent_buffer(old);
1324 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1328 eb = btrfs_clone_extent_buffer(old);
1329 free_extent_buffer(old);
1331 } else if (old_root) {
1332 btrfs_tree_read_unlock(eb_root);
1333 free_extent_buffer(eb_root);
1334 eb = alloc_dummy_extent_buffer(logical, root->nodesize);
1336 eb = btrfs_clone_extent_buffer(eb_root);
1337 btrfs_tree_read_unlock(eb_root);
1338 free_extent_buffer(eb_root);
1343 extent_buffer_get(eb);
1344 btrfs_tree_read_lock(eb);
1346 btrfs_set_header_bytenr(eb, eb->start);
1347 btrfs_set_header_backref_rev(eb, BTRFS_MIXED_BACKREF_REV);
1348 btrfs_set_header_owner(eb, btrfs_header_owner(eb_root));
1349 btrfs_set_header_level(eb, old_root->level);
1350 btrfs_set_header_generation(eb, old_generation);
1353 __tree_mod_log_rewind(eb, time_seq, tm);
1355 WARN_ON(btrfs_header_level(eb) != 0);
1356 WARN_ON(btrfs_header_nritems(eb) > BTRFS_NODEPTRS_PER_BLOCK(root));
1361 int btrfs_old_root_level(struct btrfs_root *root, u64 time_seq)
1363 struct tree_mod_elem *tm;
1365 struct extent_buffer *eb_root = btrfs_root_node(root);
1367 tm = __tree_mod_log_oldest_root(root->fs_info, eb_root, time_seq);
1368 if (tm && tm->op == MOD_LOG_ROOT_REPLACE) {
1369 level = tm->old_root.level;
1371 level = btrfs_header_level(eb_root);
1373 free_extent_buffer(eb_root);
1378 static inline int should_cow_block(struct btrfs_trans_handle *trans,
1379 struct btrfs_root *root,
1380 struct extent_buffer *buf)
1382 /* ensure we can see the force_cow */
1386 * We do not need to cow a block if
1387 * 1) this block is not created or changed in this transaction;
1388 * 2) this block does not belong to TREE_RELOC tree;
1389 * 3) the root is not forced COW.
1391 * What is forced COW:
1392 * when we create snapshot during commiting the transaction,
1393 * after we've finished coping src root, we must COW the shared
1394 * block to ensure the metadata consistency.
1396 if (btrfs_header_generation(buf) == trans->transid &&
1397 !btrfs_header_flag(buf, BTRFS_HEADER_FLAG_WRITTEN) &&
1398 !(root->root_key.objectid != BTRFS_TREE_RELOC_OBJECTID &&
1399 btrfs_header_flag(buf, BTRFS_HEADER_FLAG_RELOC)) &&
1406 * cows a single block, see __btrfs_cow_block for the real work.
1407 * This version of it has extra checks so that a block isn't cow'd more than
1408 * once per transaction, as long as it hasn't been written yet
1410 noinline int btrfs_cow_block(struct btrfs_trans_handle *trans,
1411 struct btrfs_root *root, struct extent_buffer *buf,
1412 struct extent_buffer *parent, int parent_slot,
1413 struct extent_buffer **cow_ret)
1418 if (trans->transaction != root->fs_info->running_transaction)
1419 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1420 (unsigned long long)trans->transid,
1421 (unsigned long long)
1422 root->fs_info->running_transaction->transid);
1424 if (trans->transid != root->fs_info->generation)
1425 WARN(1, KERN_CRIT "trans %llu running %llu\n",
1426 (unsigned long long)trans->transid,
1427 (unsigned long long)root->fs_info->generation);
1429 if (!should_cow_block(trans, root, buf)) {
1434 search_start = buf->start & ~((u64)(1024 * 1024 * 1024) - 1);
1437 btrfs_set_lock_blocking(parent);
1438 btrfs_set_lock_blocking(buf);
1440 ret = __btrfs_cow_block(trans, root, buf, parent,
1441 parent_slot, cow_ret, search_start, 0);
1443 trace_btrfs_cow_block(root, buf, *cow_ret);
1449 * helper function for defrag to decide if two blocks pointed to by a
1450 * node are actually close by
1452 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
1454 if (blocknr < other && other - (blocknr + blocksize) < 32768)
1456 if (blocknr > other && blocknr - (other + blocksize) < 32768)
1462 * compare two keys in a memcmp fashion
1464 static int comp_keys(struct btrfs_disk_key *disk, struct btrfs_key *k2)
1466 struct btrfs_key k1;
1468 btrfs_disk_key_to_cpu(&k1, disk);
1470 return btrfs_comp_cpu_keys(&k1, k2);
1474 * same as comp_keys only with two btrfs_key's
1476 int btrfs_comp_cpu_keys(struct btrfs_key *k1, struct btrfs_key *k2)
1478 if (k1->objectid > k2->objectid)
1480 if (k1->objectid < k2->objectid)
1482 if (k1->type > k2->type)
1484 if (k1->type < k2->type)
1486 if (k1->offset > k2->offset)
1488 if (k1->offset < k2->offset)
1494 * this is used by the defrag code to go through all the
1495 * leaves pointed to by a node and reallocate them so that
1496 * disk order is close to key order
1498 int btrfs_realloc_node(struct btrfs_trans_handle *trans,
1499 struct btrfs_root *root, struct extent_buffer *parent,
1500 int start_slot, u64 *last_ret,
1501 struct btrfs_key *progress)
1503 struct extent_buffer *cur;
1506 u64 search_start = *last_ret;
1516 int progress_passed = 0;
1517 struct btrfs_disk_key disk_key;
1519 parent_level = btrfs_header_level(parent);
1521 WARN_ON(trans->transaction != root->fs_info->running_transaction);
1522 WARN_ON(trans->transid != root->fs_info->generation);
1524 parent_nritems = btrfs_header_nritems(parent);
1525 blocksize = btrfs_level_size(root, parent_level - 1);
1526 end_slot = parent_nritems;
1528 if (parent_nritems == 1)
1531 btrfs_set_lock_blocking(parent);
1533 for (i = start_slot; i < end_slot; i++) {
1536 btrfs_node_key(parent, &disk_key, i);
1537 if (!progress_passed && comp_keys(&disk_key, progress) < 0)
1540 progress_passed = 1;
1541 blocknr = btrfs_node_blockptr(parent, i);
1542 gen = btrfs_node_ptr_generation(parent, i);
1543 if (last_block == 0)
1544 last_block = blocknr;
1547 other = btrfs_node_blockptr(parent, i - 1);
1548 close = close_blocks(blocknr, other, blocksize);
1550 if (!close && i < end_slot - 2) {
1551 other = btrfs_node_blockptr(parent, i + 1);
1552 close = close_blocks(blocknr, other, blocksize);
1555 last_block = blocknr;
1559 cur = btrfs_find_tree_block(root, blocknr, blocksize);
1561 uptodate = btrfs_buffer_uptodate(cur, gen, 0);
1564 if (!cur || !uptodate) {
1566 cur = read_tree_block(root, blocknr,
1568 if (!cur || !extent_buffer_uptodate(cur)) {
1569 free_extent_buffer(cur);
1572 } else if (!uptodate) {
1573 err = btrfs_read_buffer(cur, gen);
1575 free_extent_buffer(cur);
1580 if (search_start == 0)
1581 search_start = last_block;
1583 btrfs_tree_lock(cur);
1584 btrfs_set_lock_blocking(cur);
1585 err = __btrfs_cow_block(trans, root, cur, parent, i,
1588 (end_slot - i) * blocksize));
1590 btrfs_tree_unlock(cur);
1591 free_extent_buffer(cur);
1594 search_start = cur->start;
1595 last_block = cur->start;
1596 *last_ret = search_start;
1597 btrfs_tree_unlock(cur);
1598 free_extent_buffer(cur);
1604 * The leaf data grows from end-to-front in the node.
1605 * this returns the address of the start of the last item,
1606 * which is the stop of the leaf data stack
1608 static inline unsigned int leaf_data_end(struct btrfs_root *root,
1609 struct extent_buffer *leaf)
1611 u32 nr = btrfs_header_nritems(leaf);
1613 return BTRFS_LEAF_DATA_SIZE(root);
1614 return btrfs_item_offset_nr(leaf, nr - 1);
1619 * search for key in the extent_buffer. The items start at offset p,
1620 * and they are item_size apart. There are 'max' items in p.
1622 * the slot in the array is returned via slot, and it points to
1623 * the place where you would insert key if it is not found in
1626 * slot may point to max if the key is bigger than all of the keys
1628 static noinline int generic_bin_search(struct extent_buffer *eb,
1630 int item_size, struct btrfs_key *key,
1637 struct btrfs_disk_key *tmp = NULL;
1638 struct btrfs_disk_key unaligned;
1639 unsigned long offset;
1641 unsigned long map_start = 0;
1642 unsigned long map_len = 0;
1645 while (low < high) {
1646 mid = (low + high) / 2;
1647 offset = p + mid * item_size;
1649 if (!kaddr || offset < map_start ||
1650 (offset + sizeof(struct btrfs_disk_key)) >
1651 map_start + map_len) {
1653 err = map_private_extent_buffer(eb, offset,
1654 sizeof(struct btrfs_disk_key),
1655 &kaddr, &map_start, &map_len);
1658 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1661 read_extent_buffer(eb, &unaligned,
1662 offset, sizeof(unaligned));
1667 tmp = (struct btrfs_disk_key *)(kaddr + offset -
1670 ret = comp_keys(tmp, key);
1686 * simple bin_search frontend that does the right thing for
1689 static int bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1690 int level, int *slot)
1693 return generic_bin_search(eb,
1694 offsetof(struct btrfs_leaf, items),
1695 sizeof(struct btrfs_item),
1696 key, btrfs_header_nritems(eb),
1699 return generic_bin_search(eb,
1700 offsetof(struct btrfs_node, ptrs),
1701 sizeof(struct btrfs_key_ptr),
1702 key, btrfs_header_nritems(eb),
1706 int btrfs_bin_search(struct extent_buffer *eb, struct btrfs_key *key,
1707 int level, int *slot)
1709 return bin_search(eb, key, level, slot);
1712 static void root_add_used(struct btrfs_root *root, u32 size)
1714 spin_lock(&root->accounting_lock);
1715 btrfs_set_root_used(&root->root_item,
1716 btrfs_root_used(&root->root_item) + size);
1717 spin_unlock(&root->accounting_lock);
1720 static void root_sub_used(struct btrfs_root *root, u32 size)
1722 spin_lock(&root->accounting_lock);
1723 btrfs_set_root_used(&root->root_item,
1724 btrfs_root_used(&root->root_item) - size);
1725 spin_unlock(&root->accounting_lock);
1728 /* given a node and slot number, this reads the blocks it points to. The
1729 * extent buffer is returned with a reference taken (but unlocked).
1730 * NULL is returned on error.
1732 static noinline struct extent_buffer *read_node_slot(struct btrfs_root *root,
1733 struct extent_buffer *parent, int slot)
1735 int level = btrfs_header_level(parent);
1736 struct extent_buffer *eb;
1740 if (slot >= btrfs_header_nritems(parent))
1745 eb = read_tree_block(root, btrfs_node_blockptr(parent, slot),
1746 btrfs_level_size(root, level - 1),
1747 btrfs_node_ptr_generation(parent, slot));
1748 if (eb && !extent_buffer_uptodate(eb)) {
1749 free_extent_buffer(eb);
1757 * node level balancing, used to make sure nodes are in proper order for
1758 * item deletion. We balance from the top down, so we have to make sure
1759 * that a deletion won't leave an node completely empty later on.
1761 static noinline int balance_level(struct btrfs_trans_handle *trans,
1762 struct btrfs_root *root,
1763 struct btrfs_path *path, int level)
1765 struct extent_buffer *right = NULL;
1766 struct extent_buffer *mid;
1767 struct extent_buffer *left = NULL;
1768 struct extent_buffer *parent = NULL;
1772 int orig_slot = path->slots[level];
1778 mid = path->nodes[level];
1780 WARN_ON(path->locks[level] != BTRFS_WRITE_LOCK &&
1781 path->locks[level] != BTRFS_WRITE_LOCK_BLOCKING);
1782 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1784 orig_ptr = btrfs_node_blockptr(mid, orig_slot);
1786 if (level < BTRFS_MAX_LEVEL - 1) {
1787 parent = path->nodes[level + 1];
1788 pslot = path->slots[level + 1];
1792 * deal with the case where there is only one pointer in the root
1793 * by promoting the node below to a root
1796 struct extent_buffer *child;
1798 if (btrfs_header_nritems(mid) != 1)
1801 /* promote the child to a root */
1802 child = read_node_slot(root, mid, 0);
1805 btrfs_std_error(root->fs_info, ret);
1809 btrfs_tree_lock(child);
1810 btrfs_set_lock_blocking(child);
1811 ret = btrfs_cow_block(trans, root, child, mid, 0, &child);
1813 btrfs_tree_unlock(child);
1814 free_extent_buffer(child);
1818 tree_mod_log_set_root_pointer(root, child, 1);
1819 rcu_assign_pointer(root->node, child);
1821 add_root_to_dirty_list(root);
1822 btrfs_tree_unlock(child);
1824 path->locks[level] = 0;
1825 path->nodes[level] = NULL;
1826 clean_tree_block(trans, root, mid);
1827 btrfs_tree_unlock(mid);
1828 /* once for the path */
1829 free_extent_buffer(mid);
1831 root_sub_used(root, mid->len);
1832 btrfs_free_tree_block(trans, root, mid, 0, 1);
1833 /* once for the root ptr */
1834 free_extent_buffer_stale(mid);
1837 if (btrfs_header_nritems(mid) >
1838 BTRFS_NODEPTRS_PER_BLOCK(root) / 4)
1841 left = read_node_slot(root, parent, pslot - 1);
1843 btrfs_tree_lock(left);
1844 btrfs_set_lock_blocking(left);
1845 wret = btrfs_cow_block(trans, root, left,
1846 parent, pslot - 1, &left);
1852 right = read_node_slot(root, parent, pslot + 1);
1854 btrfs_tree_lock(right);
1855 btrfs_set_lock_blocking(right);
1856 wret = btrfs_cow_block(trans, root, right,
1857 parent, pslot + 1, &right);
1864 /* first, try to make some room in the middle buffer */
1866 orig_slot += btrfs_header_nritems(left);
1867 wret = push_node_left(trans, root, left, mid, 1);
1873 * then try to empty the right most buffer into the middle
1876 wret = push_node_left(trans, root, mid, right, 1);
1877 if (wret < 0 && wret != -ENOSPC)
1879 if (btrfs_header_nritems(right) == 0) {
1880 clean_tree_block(trans, root, right);
1881 btrfs_tree_unlock(right);
1882 del_ptr(root, path, level + 1, pslot + 1);
1883 root_sub_used(root, right->len);
1884 btrfs_free_tree_block(trans, root, right, 0, 1);
1885 free_extent_buffer_stale(right);
1888 struct btrfs_disk_key right_key;
1889 btrfs_node_key(right, &right_key, 0);
1890 tree_mod_log_set_node_key(root->fs_info, parent,
1892 btrfs_set_node_key(parent, &right_key, pslot + 1);
1893 btrfs_mark_buffer_dirty(parent);
1896 if (btrfs_header_nritems(mid) == 1) {
1898 * we're not allowed to leave a node with one item in the
1899 * tree during a delete. A deletion from lower in the tree
1900 * could try to delete the only pointer in this node.
1901 * So, pull some keys from the left.
1902 * There has to be a left pointer at this point because
1903 * otherwise we would have pulled some pointers from the
1908 btrfs_std_error(root->fs_info, ret);
1911 wret = balance_node_right(trans, root, mid, left);
1917 wret = push_node_left(trans, root, left, mid, 1);
1923 if (btrfs_header_nritems(mid) == 0) {
1924 clean_tree_block(trans, root, mid);
1925 btrfs_tree_unlock(mid);
1926 del_ptr(root, path, level + 1, pslot);
1927 root_sub_used(root, mid->len);
1928 btrfs_free_tree_block(trans, root, mid, 0, 1);
1929 free_extent_buffer_stale(mid);
1932 /* update the parent key to reflect our changes */
1933 struct btrfs_disk_key mid_key;
1934 btrfs_node_key(mid, &mid_key, 0);
1935 tree_mod_log_set_node_key(root->fs_info, parent,
1937 btrfs_set_node_key(parent, &mid_key, pslot);
1938 btrfs_mark_buffer_dirty(parent);
1941 /* update the path */
1943 if (btrfs_header_nritems(left) > orig_slot) {
1944 extent_buffer_get(left);
1945 /* left was locked after cow */
1946 path->nodes[level] = left;
1947 path->slots[level + 1] -= 1;
1948 path->slots[level] = orig_slot;
1950 btrfs_tree_unlock(mid);
1951 free_extent_buffer(mid);
1954 orig_slot -= btrfs_header_nritems(left);
1955 path->slots[level] = orig_slot;
1958 /* double check we haven't messed things up */
1960 btrfs_node_blockptr(path->nodes[level], path->slots[level]))
1964 btrfs_tree_unlock(right);
1965 free_extent_buffer(right);
1968 if (path->nodes[level] != left)
1969 btrfs_tree_unlock(left);
1970 free_extent_buffer(left);
1975 /* Node balancing for insertion. Here we only split or push nodes around
1976 * when they are completely full. This is also done top down, so we
1977 * have to be pessimistic.
1979 static noinline int push_nodes_for_insert(struct btrfs_trans_handle *trans,
1980 struct btrfs_root *root,
1981 struct btrfs_path *path, int level)
1983 struct extent_buffer *right = NULL;
1984 struct extent_buffer *mid;
1985 struct extent_buffer *left = NULL;
1986 struct extent_buffer *parent = NULL;
1990 int orig_slot = path->slots[level];
1995 mid = path->nodes[level];
1996 WARN_ON(btrfs_header_generation(mid) != trans->transid);
1998 if (level < BTRFS_MAX_LEVEL - 1) {
1999 parent = path->nodes[level + 1];
2000 pslot = path->slots[level + 1];
2006 left = read_node_slot(root, parent, pslot - 1);
2008 /* first, try to make some room in the middle buffer */
2012 btrfs_tree_lock(left);
2013 btrfs_set_lock_blocking(left);
2015 left_nr = btrfs_header_nritems(left);
2016 if (left_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2019 ret = btrfs_cow_block(trans, root, left, parent,
2024 wret = push_node_left(trans, root,
2031 struct btrfs_disk_key disk_key;
2032 orig_slot += left_nr;
2033 btrfs_node_key(mid, &disk_key, 0);
2034 tree_mod_log_set_node_key(root->fs_info, parent,
2036 btrfs_set_node_key(parent, &disk_key, pslot);
2037 btrfs_mark_buffer_dirty(parent);
2038 if (btrfs_header_nritems(left) > orig_slot) {
2039 path->nodes[level] = left;
2040 path->slots[level + 1] -= 1;
2041 path->slots[level] = orig_slot;
2042 btrfs_tree_unlock(mid);
2043 free_extent_buffer(mid);
2046 btrfs_header_nritems(left);
2047 path->slots[level] = orig_slot;
2048 btrfs_tree_unlock(left);
2049 free_extent_buffer(left);
2053 btrfs_tree_unlock(left);
2054 free_extent_buffer(left);
2056 right = read_node_slot(root, parent, pslot + 1);
2059 * then try to empty the right most buffer into the middle
2064 btrfs_tree_lock(right);
2065 btrfs_set_lock_blocking(right);
2067 right_nr = btrfs_header_nritems(right);
2068 if (right_nr >= BTRFS_NODEPTRS_PER_BLOCK(root) - 1) {
2071 ret = btrfs_cow_block(trans, root, right,
2077 wret = balance_node_right(trans, root,
2084 struct btrfs_disk_key disk_key;
2086 btrfs_node_key(right, &disk_key, 0);
2087 tree_mod_log_set_node_key(root->fs_info, parent,
2089 btrfs_set_node_key(parent, &disk_key, pslot + 1);
2090 btrfs_mark_buffer_dirty(parent);
2092 if (btrfs_header_nritems(mid) <= orig_slot) {
2093 path->nodes[level] = right;
2094 path->slots[level + 1] += 1;
2095 path->slots[level] = orig_slot -
2096 btrfs_header_nritems(mid);
2097 btrfs_tree_unlock(mid);
2098 free_extent_buffer(mid);
2100 btrfs_tree_unlock(right);
2101 free_extent_buffer(right);
2105 btrfs_tree_unlock(right);
2106 free_extent_buffer(right);
2112 * readahead one full node of leaves, finding things that are close
2113 * to the block in 'slot', and triggering ra on them.
2115 static void reada_for_search(struct btrfs_root *root,
2116 struct btrfs_path *path,
2117 int level, int slot, u64 objectid)
2119 struct extent_buffer *node;
2120 struct btrfs_disk_key disk_key;
2126 int direction = path->reada;
2127 struct extent_buffer *eb;
2135 if (!path->nodes[level])
2138 node = path->nodes[level];
2140 search = btrfs_node_blockptr(node, slot);
2141 blocksize = btrfs_level_size(root, level - 1);
2142 eb = btrfs_find_tree_block(root, search, blocksize);
2144 free_extent_buffer(eb);
2150 nritems = btrfs_header_nritems(node);
2154 if (direction < 0) {
2158 } else if (direction > 0) {
2163 if (path->reada < 0 && objectid) {
2164 btrfs_node_key(node, &disk_key, nr);
2165 if (btrfs_disk_key_objectid(&disk_key) != objectid)
2168 search = btrfs_node_blockptr(node, nr);
2169 if ((search <= target && target - search <= 65536) ||
2170 (search > target && search - target <= 65536)) {
2171 gen = btrfs_node_ptr_generation(node, nr);
2172 readahead_tree_block(root, search, blocksize, gen);
2176 if ((nread > 65536 || nscan > 32))
2182 * returns -EAGAIN if it had to drop the path, or zero if everything was in
2185 static noinline int reada_for_balance(struct btrfs_root *root,
2186 struct btrfs_path *path, int level)
2190 struct extent_buffer *parent;
2191 struct extent_buffer *eb;
2198 parent = path->nodes[level + 1];
2202 nritems = btrfs_header_nritems(parent);
2203 slot = path->slots[level + 1];
2204 blocksize = btrfs_level_size(root, level);
2207 block1 = btrfs_node_blockptr(parent, slot - 1);
2208 gen = btrfs_node_ptr_generation(parent, slot - 1);
2209 eb = btrfs_find_tree_block(root, block1, blocksize);
2211 * if we get -eagain from btrfs_buffer_uptodate, we
2212 * don't want to return eagain here. That will loop
2215 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2217 free_extent_buffer(eb);
2219 if (slot + 1 < nritems) {
2220 block2 = btrfs_node_blockptr(parent, slot + 1);
2221 gen = btrfs_node_ptr_generation(parent, slot + 1);
2222 eb = btrfs_find_tree_block(root, block2, blocksize);
2223 if (eb && btrfs_buffer_uptodate(eb, gen, 1) != 0)
2225 free_extent_buffer(eb);
2227 if (block1 || block2) {
2230 /* release the whole path */
2231 btrfs_release_path(path);
2233 /* read the blocks */
2235 readahead_tree_block(root, block1, blocksize, 0);
2237 readahead_tree_block(root, block2, blocksize, 0);
2240 eb = read_tree_block(root, block1, blocksize, 0);
2241 free_extent_buffer(eb);
2244 eb = read_tree_block(root, block2, blocksize, 0);
2245 free_extent_buffer(eb);
2253 * when we walk down the tree, it is usually safe to unlock the higher layers
2254 * in the tree. The exceptions are when our path goes through slot 0, because
2255 * operations on the tree might require changing key pointers higher up in the
2258 * callers might also have set path->keep_locks, which tells this code to keep
2259 * the lock if the path points to the last slot in the block. This is part of
2260 * walking through the tree, and selecting the next slot in the higher block.
2262 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2263 * if lowest_unlock is 1, level 0 won't be unlocked
2265 static noinline void unlock_up(struct btrfs_path *path, int level,
2266 int lowest_unlock, int min_write_lock_level,
2267 int *write_lock_level)
2270 int skip_level = level;
2272 struct extent_buffer *t;
2274 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2275 if (!path->nodes[i])
2277 if (!path->locks[i])
2279 if (!no_skips && path->slots[i] == 0) {
2283 if (!no_skips && path->keep_locks) {
2286 nritems = btrfs_header_nritems(t);
2287 if (nritems < 1 || path->slots[i] >= nritems - 1) {
2292 if (skip_level < i && i >= lowest_unlock)
2296 if (i >= lowest_unlock && i > skip_level && path->locks[i]) {
2297 btrfs_tree_unlock_rw(t, path->locks[i]);
2299 if (write_lock_level &&
2300 i > min_write_lock_level &&
2301 i <= *write_lock_level) {
2302 *write_lock_level = i - 1;
2309 * This releases any locks held in the path starting at level and
2310 * going all the way up to the root.
2312 * btrfs_search_slot will keep the lock held on higher nodes in a few
2313 * corner cases, such as COW of the block at slot zero in the node. This
2314 * ignores those rules, and it should only be called when there are no
2315 * more updates to be done higher up in the tree.
2317 noinline void btrfs_unlock_up_safe(struct btrfs_path *path, int level)
2321 if (path->keep_locks)
2324 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2325 if (!path->nodes[i])
2327 if (!path->locks[i])
2329 btrfs_tree_unlock_rw(path->nodes[i], path->locks[i]);
2335 * helper function for btrfs_search_slot. The goal is to find a block
2336 * in cache without setting the path to blocking. If we find the block
2337 * we return zero and the path is unchanged.
2339 * If we can't find the block, we set the path blocking and do some
2340 * reada. -EAGAIN is returned and the search must be repeated.
2343 read_block_for_search(struct btrfs_trans_handle *trans,
2344 struct btrfs_root *root, struct btrfs_path *p,
2345 struct extent_buffer **eb_ret, int level, int slot,
2346 struct btrfs_key *key, u64 time_seq)
2351 struct extent_buffer *b = *eb_ret;
2352 struct extent_buffer *tmp;
2355 blocknr = btrfs_node_blockptr(b, slot);
2356 gen = btrfs_node_ptr_generation(b, slot);
2357 blocksize = btrfs_level_size(root, level - 1);
2359 tmp = btrfs_find_tree_block(root, blocknr, blocksize);
2361 /* first we do an atomic uptodate check */
2362 if (btrfs_buffer_uptodate(tmp, gen, 1) > 0) {
2367 /* the pages were up to date, but we failed
2368 * the generation number check. Do a full
2369 * read for the generation number that is correct.
2370 * We must do this without dropping locks so
2371 * we can trust our generation number
2373 btrfs_set_path_blocking(p);
2375 /* now we're allowed to do a blocking uptodate check */
2376 ret = btrfs_read_buffer(tmp, gen);
2381 free_extent_buffer(tmp);
2382 btrfs_release_path(p);
2387 * reduce lock contention at high levels
2388 * of the btree by dropping locks before
2389 * we read. Don't release the lock on the current
2390 * level because we need to walk this node to figure
2391 * out which blocks to read.
2393 btrfs_unlock_up_safe(p, level + 1);
2394 btrfs_set_path_blocking(p);
2396 free_extent_buffer(tmp);
2398 reada_for_search(root, p, level, slot, key->objectid);
2400 btrfs_release_path(p);
2403 tmp = read_tree_block(root, blocknr, blocksize, 0);
2406 * If the read above didn't mark this buffer up to date,
2407 * it will never end up being up to date. Set ret to EIO now
2408 * and give up so that our caller doesn't loop forever
2411 if (!btrfs_buffer_uptodate(tmp, 0, 0))
2413 free_extent_buffer(tmp);
2419 * helper function for btrfs_search_slot. This does all of the checks
2420 * for node-level blocks and does any balancing required based on
2423 * If no extra work was required, zero is returned. If we had to
2424 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2428 setup_nodes_for_search(struct btrfs_trans_handle *trans,
2429 struct btrfs_root *root, struct btrfs_path *p,
2430 struct extent_buffer *b, int level, int ins_len,
2431 int *write_lock_level)
2434 if ((p->search_for_split || ins_len > 0) && btrfs_header_nritems(b) >=
2435 BTRFS_NODEPTRS_PER_BLOCK(root) - 3) {
2438 if (*write_lock_level < level + 1) {
2439 *write_lock_level = level + 1;
2440 btrfs_release_path(p);
2444 sret = reada_for_balance(root, p, level);
2448 btrfs_set_path_blocking(p);
2449 sret = split_node(trans, root, p, level);
2450 btrfs_clear_path_blocking(p, NULL, 0);
2457 b = p->nodes[level];
2458 } else if (ins_len < 0 && btrfs_header_nritems(b) <
2459 BTRFS_NODEPTRS_PER_BLOCK(root) / 2) {
2462 if (*write_lock_level < level + 1) {
2463 *write_lock_level = level + 1;
2464 btrfs_release_path(p);
2468 sret = reada_for_balance(root, p, level);
2472 btrfs_set_path_blocking(p);
2473 sret = balance_level(trans, root, p, level);
2474 btrfs_clear_path_blocking(p, NULL, 0);
2480 b = p->nodes[level];
2482 btrfs_release_path(p);
2485 BUG_ON(btrfs_header_nritems(b) == 1);
2496 * look for key in the tree. path is filled in with nodes along the way
2497 * if key is found, we return zero and you can find the item in the leaf
2498 * level of the path (level 0)
2500 * If the key isn't found, the path points to the slot where it should
2501 * be inserted, and 1 is returned. If there are other errors during the
2502 * search a negative error number is returned.
2504 * if ins_len > 0, nodes and leaves will be split as we walk down the
2505 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2508 int btrfs_search_slot(struct btrfs_trans_handle *trans, struct btrfs_root
2509 *root, struct btrfs_key *key, struct btrfs_path *p, int
2512 struct extent_buffer *b;
2517 int lowest_unlock = 1;
2519 /* everything at write_lock_level or lower must be write locked */
2520 int write_lock_level = 0;
2521 u8 lowest_level = 0;
2522 int min_write_lock_level;
2524 lowest_level = p->lowest_level;
2525 WARN_ON(lowest_level && ins_len > 0);
2526 WARN_ON(p->nodes[0] != NULL);
2531 /* when we are removing items, we might have to go up to level
2532 * two as we update tree pointers Make sure we keep write
2533 * for those levels as well
2535 write_lock_level = 2;
2536 } else if (ins_len > 0) {
2538 * for inserting items, make sure we have a write lock on
2539 * level 1 so we can update keys
2541 write_lock_level = 1;
2545 write_lock_level = -1;
2547 if (cow && (p->keep_locks || p->lowest_level))
2548 write_lock_level = BTRFS_MAX_LEVEL;
2550 min_write_lock_level = write_lock_level;
2554 * we try very hard to do read locks on the root
2556 root_lock = BTRFS_READ_LOCK;
2558 if (p->search_commit_root) {
2560 * the commit roots are read only
2561 * so we always do read locks
2563 b = root->commit_root;
2564 extent_buffer_get(b);
2565 level = btrfs_header_level(b);
2566 if (!p->skip_locking)
2567 btrfs_tree_read_lock(b);
2569 if (p->skip_locking) {
2570 b = btrfs_root_node(root);
2571 level = btrfs_header_level(b);
2573 /* we don't know the level of the root node
2574 * until we actually have it read locked
2576 b = btrfs_read_lock_root_node(root);
2577 level = btrfs_header_level(b);
2578 if (level <= write_lock_level) {
2579 /* whoops, must trade for write lock */
2580 btrfs_tree_read_unlock(b);
2581 free_extent_buffer(b);
2582 b = btrfs_lock_root_node(root);
2583 root_lock = BTRFS_WRITE_LOCK;
2585 /* the level might have changed, check again */
2586 level = btrfs_header_level(b);
2590 p->nodes[level] = b;
2591 if (!p->skip_locking)
2592 p->locks[level] = root_lock;
2595 level = btrfs_header_level(b);
2598 * setup the path here so we can release it under lock
2599 * contention with the cow code
2603 * if we don't really need to cow this block
2604 * then we don't want to set the path blocking,
2605 * so we test it here
2607 if (!should_cow_block(trans, root, b))
2610 btrfs_set_path_blocking(p);
2613 * must have write locks on this node and the
2616 if (level > write_lock_level ||
2617 (level + 1 > write_lock_level &&
2618 level + 1 < BTRFS_MAX_LEVEL &&
2619 p->nodes[level + 1])) {
2620 write_lock_level = level + 1;
2621 btrfs_release_path(p);
2625 err = btrfs_cow_block(trans, root, b,
2626 p->nodes[level + 1],
2627 p->slots[level + 1], &b);
2634 BUG_ON(!cow && ins_len);
2636 p->nodes[level] = b;
2637 btrfs_clear_path_blocking(p, NULL, 0);
2640 * we have a lock on b and as long as we aren't changing
2641 * the tree, there is no way to for the items in b to change.
2642 * It is safe to drop the lock on our parent before we
2643 * go through the expensive btree search on b.
2645 * If cow is true, then we might be changing slot zero,
2646 * which may require changing the parent. So, we can't
2647 * drop the lock until after we know which slot we're
2651 btrfs_unlock_up_safe(p, level + 1);
2653 ret = bin_search(b, key, level, &slot);
2657 if (ret && slot > 0) {
2661 p->slots[level] = slot;
2662 err = setup_nodes_for_search(trans, root, p, b, level,
2663 ins_len, &write_lock_level);
2670 b = p->nodes[level];
2671 slot = p->slots[level];
2674 * slot 0 is special, if we change the key
2675 * we have to update the parent pointer
2676 * which means we must have a write lock
2679 if (slot == 0 && cow &&
2680 write_lock_level < level + 1) {
2681 write_lock_level = level + 1;
2682 btrfs_release_path(p);
2686 unlock_up(p, level, lowest_unlock,
2687 min_write_lock_level, &write_lock_level);
2689 if (level == lowest_level) {
2695 err = read_block_for_search(trans, root, p,
2696 &b, level, slot, key, 0);
2704 if (!p->skip_locking) {
2705 level = btrfs_header_level(b);
2706 if (level <= write_lock_level) {
2707 err = btrfs_try_tree_write_lock(b);
2709 btrfs_set_path_blocking(p);
2711 btrfs_clear_path_blocking(p, b,
2714 p->locks[level] = BTRFS_WRITE_LOCK;
2716 err = btrfs_try_tree_read_lock(b);
2718 btrfs_set_path_blocking(p);
2719 btrfs_tree_read_lock(b);
2720 btrfs_clear_path_blocking(p, b,
2723 p->locks[level] = BTRFS_READ_LOCK;
2725 p->nodes[level] = b;
2728 p->slots[level] = slot;
2730 btrfs_leaf_free_space(root, b) < ins_len) {
2731 if (write_lock_level < 1) {
2732 write_lock_level = 1;
2733 btrfs_release_path(p);
2737 btrfs_set_path_blocking(p);
2738 err = split_leaf(trans, root, key,
2739 p, ins_len, ret == 0);
2740 btrfs_clear_path_blocking(p, NULL, 0);
2748 if (!p->search_for_split)
2749 unlock_up(p, level, lowest_unlock,
2750 min_write_lock_level, &write_lock_level);
2757 * we don't really know what they plan on doing with the path
2758 * from here on, so for now just mark it as blocking
2760 if (!p->leave_spinning)
2761 btrfs_set_path_blocking(p);
2763 btrfs_release_path(p);
2768 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2769 * current state of the tree together with the operations recorded in the tree
2770 * modification log to search for the key in a previous version of this tree, as
2771 * denoted by the time_seq parameter.
2773 * Naturally, there is no support for insert, delete or cow operations.
2775 * The resulting path and return value will be set up as if we called
2776 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2778 int btrfs_search_old_slot(struct btrfs_root *root, struct btrfs_key *key,
2779 struct btrfs_path *p, u64 time_seq)
2781 struct extent_buffer *b;
2786 int lowest_unlock = 1;
2787 u8 lowest_level = 0;
2789 lowest_level = p->lowest_level;
2790 WARN_ON(p->nodes[0] != NULL);
2792 if (p->search_commit_root) {
2794 return btrfs_search_slot(NULL, root, key, p, 0, 0);
2798 b = get_old_root(root, time_seq);
2799 level = btrfs_header_level(b);
2800 p->locks[level] = BTRFS_READ_LOCK;
2803 level = btrfs_header_level(b);
2804 p->nodes[level] = b;
2805 btrfs_clear_path_blocking(p, NULL, 0);
2808 * we have a lock on b and as long as we aren't changing
2809 * the tree, there is no way to for the items in b to change.
2810 * It is safe to drop the lock on our parent before we
2811 * go through the expensive btree search on b.
2813 btrfs_unlock_up_safe(p, level + 1);
2815 ret = bin_search(b, key, level, &slot);
2819 if (ret && slot > 0) {
2823 p->slots[level] = slot;
2824 unlock_up(p, level, lowest_unlock, 0, NULL);
2826 if (level == lowest_level) {
2832 err = read_block_for_search(NULL, root, p, &b, level,
2833 slot, key, time_seq);
2841 level = btrfs_header_level(b);
2842 err = btrfs_try_tree_read_lock(b);
2844 btrfs_set_path_blocking(p);
2845 btrfs_tree_read_lock(b);
2846 btrfs_clear_path_blocking(p, b,
2849 b = tree_mod_log_rewind(root->fs_info, b, time_seq);
2850 p->locks[level] = BTRFS_READ_LOCK;
2851 p->nodes[level] = b;
2853 p->slots[level] = slot;
2854 unlock_up(p, level, lowest_unlock, 0, NULL);
2860 if (!p->leave_spinning)
2861 btrfs_set_path_blocking(p);
2863 btrfs_release_path(p);
2869 * helper to use instead of search slot if no exact match is needed but
2870 * instead the next or previous item should be returned.
2871 * When find_higher is true, the next higher item is returned, the next lower
2873 * When return_any and find_higher are both true, and no higher item is found,
2874 * return the next lower instead.
2875 * When return_any is true and find_higher is false, and no lower item is found,
2876 * return the next higher instead.
2877 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2880 int btrfs_search_slot_for_read(struct btrfs_root *root,
2881 struct btrfs_key *key, struct btrfs_path *p,
2882 int find_higher, int return_any)
2885 struct extent_buffer *leaf;
2888 ret = btrfs_search_slot(NULL, root, key, p, 0, 0);
2892 * a return value of 1 means the path is at the position where the
2893 * item should be inserted. Normally this is the next bigger item,
2894 * but in case the previous item is the last in a leaf, path points
2895 * to the first free slot in the previous leaf, i.e. at an invalid
2901 if (p->slots[0] >= btrfs_header_nritems(leaf)) {
2902 ret = btrfs_next_leaf(root, p);
2908 * no higher item found, return the next
2913 btrfs_release_path(p);
2917 if (p->slots[0] == 0) {
2918 ret = btrfs_prev_leaf(root, p);
2922 p->slots[0] = btrfs_header_nritems(leaf) - 1;
2928 * no lower item found, return the next
2933 btrfs_release_path(p);
2943 * adjust the pointers going up the tree, starting at level
2944 * making sure the right key of each node is points to 'key'.
2945 * This is used after shifting pointers to the left, so it stops
2946 * fixing up pointers when a given leaf/node is not in slot 0 of the
2950 static void fixup_low_keys(struct btrfs_root *root, struct btrfs_path *path,
2951 struct btrfs_disk_key *key, int level)
2954 struct extent_buffer *t;
2956 for (i = level; i < BTRFS_MAX_LEVEL; i++) {
2957 int tslot = path->slots[i];
2958 if (!path->nodes[i])
2961 tree_mod_log_set_node_key(root->fs_info, t, tslot, 1);
2962 btrfs_set_node_key(t, key, tslot);
2963 btrfs_mark_buffer_dirty(path->nodes[i]);
2972 * This function isn't completely safe. It's the caller's responsibility
2973 * that the new key won't break the order
2975 void btrfs_set_item_key_safe(struct btrfs_root *root, struct btrfs_path *path,
2976 struct btrfs_key *new_key)
2978 struct btrfs_disk_key disk_key;
2979 struct extent_buffer *eb;
2982 eb = path->nodes[0];
2983 slot = path->slots[0];
2985 btrfs_item_key(eb, &disk_key, slot - 1);
2986 BUG_ON(comp_keys(&disk_key, new_key) >= 0);
2988 if (slot < btrfs_header_nritems(eb) - 1) {
2989 btrfs_item_key(eb, &disk_key, slot + 1);
2990 BUG_ON(comp_keys(&disk_key, new_key) <= 0);
2993 btrfs_cpu_key_to_disk(&disk_key, new_key);
2994 btrfs_set_item_key(eb, &disk_key, slot);
2995 btrfs_mark_buffer_dirty(eb);
2997 fixup_low_keys(root, path, &disk_key, 1);
3001 * try to push data from one node into the next node left in the
3004 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3005 * error, and > 0 if there was no room in the left hand block.
3007 static int push_node_left(struct btrfs_trans_handle *trans,
3008 struct btrfs_root *root, struct extent_buffer *dst,
3009 struct extent_buffer *src, int empty)
3016 src_nritems = btrfs_header_nritems(src);
3017 dst_nritems = btrfs_header_nritems(dst);
3018 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3019 WARN_ON(btrfs_header_generation(src) != trans->transid);
3020 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3022 if (!empty && src_nritems <= 8)
3025 if (push_items <= 0)
3029 push_items = min(src_nritems, push_items);
3030 if (push_items < src_nritems) {
3031 /* leave at least 8 pointers in the node if
3032 * we aren't going to empty it
3034 if (src_nritems - push_items < 8) {
3035 if (push_items <= 8)
3041 push_items = min(src_nritems - 8, push_items);
3043 tree_mod_log_eb_copy(root->fs_info, dst, src, dst_nritems, 0,
3045 copy_extent_buffer(dst, src,
3046 btrfs_node_key_ptr_offset(dst_nritems),
3047 btrfs_node_key_ptr_offset(0),
3048 push_items * sizeof(struct btrfs_key_ptr));
3050 if (push_items < src_nritems) {
3052 * don't call tree_mod_log_eb_move here, key removal was already
3053 * fully logged by tree_mod_log_eb_copy above.
3055 memmove_extent_buffer(src, btrfs_node_key_ptr_offset(0),
3056 btrfs_node_key_ptr_offset(push_items),
3057 (src_nritems - push_items) *
3058 sizeof(struct btrfs_key_ptr));
3060 btrfs_set_header_nritems(src, src_nritems - push_items);
3061 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3062 btrfs_mark_buffer_dirty(src);
3063 btrfs_mark_buffer_dirty(dst);
3069 * try to push data from one node into the next node right in the
3072 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3073 * error, and > 0 if there was no room in the right hand block.
3075 * this will only push up to 1/2 the contents of the left node over
3077 static int balance_node_right(struct btrfs_trans_handle *trans,
3078 struct btrfs_root *root,
3079 struct extent_buffer *dst,
3080 struct extent_buffer *src)
3088 WARN_ON(btrfs_header_generation(src) != trans->transid);
3089 WARN_ON(btrfs_header_generation(dst) != trans->transid);
3091 src_nritems = btrfs_header_nritems(src);
3092 dst_nritems = btrfs_header_nritems(dst);
3093 push_items = BTRFS_NODEPTRS_PER_BLOCK(root) - dst_nritems;
3094 if (push_items <= 0)
3097 if (src_nritems < 4)
3100 max_push = src_nritems / 2 + 1;
3101 /* don't try to empty the node */
3102 if (max_push >= src_nritems)
3105 if (max_push < push_items)
3106 push_items = max_push;
3108 tree_mod_log_eb_move(root->fs_info, dst, push_items, 0, dst_nritems);
3109 memmove_extent_buffer(dst, btrfs_node_key_ptr_offset(push_items),
3110 btrfs_node_key_ptr_offset(0),
3112 sizeof(struct btrfs_key_ptr));
3114 tree_mod_log_eb_copy(root->fs_info, dst, src, 0,
3115 src_nritems - push_items, push_items);
3116 copy_extent_buffer(dst, src,
3117 btrfs_node_key_ptr_offset(0),
3118 btrfs_node_key_ptr_offset(src_nritems - push_items),
3119 push_items * sizeof(struct btrfs_key_ptr));
3121 btrfs_set_header_nritems(src, src_nritems - push_items);
3122 btrfs_set_header_nritems(dst, dst_nritems + push_items);
3124 btrfs_mark_buffer_dirty(src);
3125 btrfs_mark_buffer_dirty(dst);
3131 * helper function to insert a new root level in the tree.
3132 * A new node is allocated, and a single item is inserted to
3133 * point to the existing root
3135 * returns zero on success or < 0 on failure.
3137 static noinline int insert_new_root(struct btrfs_trans_handle *trans,
3138 struct btrfs_root *root,
3139 struct btrfs_path *path, int level)
3142 struct extent_buffer *lower;
3143 struct extent_buffer *c;
3144 struct extent_buffer *old;
3145 struct btrfs_disk_key lower_key;
3147 BUG_ON(path->nodes[level]);
3148 BUG_ON(path->nodes[level-1] != root->node);
3150 lower = path->nodes[level-1];
3152 btrfs_item_key(lower, &lower_key, 0);
3154 btrfs_node_key(lower, &lower_key, 0);
3156 c = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3157 root->root_key.objectid, &lower_key,
3158 level, root->node->start, 0);
3162 root_add_used(root, root->nodesize);
3164 memset_extent_buffer(c, 0, 0, sizeof(struct btrfs_header));
3165 btrfs_set_header_nritems(c, 1);
3166 btrfs_set_header_level(c, level);
3167 btrfs_set_header_bytenr(c, c->start);
3168 btrfs_set_header_generation(c, trans->transid);
3169 btrfs_set_header_backref_rev(c, BTRFS_MIXED_BACKREF_REV);
3170 btrfs_set_header_owner(c, root->root_key.objectid);
3172 write_extent_buffer(c, root->fs_info->fsid,
3173 (unsigned long)btrfs_header_fsid(c),
3176 write_extent_buffer(c, root->fs_info->chunk_tree_uuid,
3177 (unsigned long)btrfs_header_chunk_tree_uuid(c),
3180 btrfs_set_node_key(c, &lower_key, 0);
3181 btrfs_set_node_blockptr(c, 0, lower->start);
3182 lower_gen = btrfs_header_generation(lower);
3183 WARN_ON(lower_gen != trans->transid);
3185 btrfs_set_node_ptr_generation(c, 0, lower_gen);
3187 btrfs_mark_buffer_dirty(c);
3190 tree_mod_log_set_root_pointer(root, c, 0);
3191 rcu_assign_pointer(root->node, c);
3193 /* the super has an extra ref to root->node */
3194 free_extent_buffer(old);
3196 add_root_to_dirty_list(root);
3197 extent_buffer_get(c);
3198 path->nodes[level] = c;
3199 path->locks[level] = BTRFS_WRITE_LOCK;
3200 path->slots[level] = 0;
3205 * worker function to insert a single pointer in a node.
3206 * the node should have enough room for the pointer already
3208 * slot and level indicate where you want the key to go, and
3209 * blocknr is the block the key points to.
3211 static void insert_ptr(struct btrfs_trans_handle *trans,
3212 struct btrfs_root *root, struct btrfs_path *path,
3213 struct btrfs_disk_key *key, u64 bytenr,
3214 int slot, int level)
3216 struct extent_buffer *lower;
3220 BUG_ON(!path->nodes[level]);
3221 btrfs_assert_tree_locked(path->nodes[level]);
3222 lower = path->nodes[level];
3223 nritems = btrfs_header_nritems(lower);
3224 BUG_ON(slot > nritems);
3225 BUG_ON(nritems == BTRFS_NODEPTRS_PER_BLOCK(root));
3226 if (slot != nritems) {
3228 tree_mod_log_eb_move(root->fs_info, lower, slot + 1,
3229 slot, nritems - slot);
3230 memmove_extent_buffer(lower,
3231 btrfs_node_key_ptr_offset(slot + 1),
3232 btrfs_node_key_ptr_offset(slot),
3233 (nritems - slot) * sizeof(struct btrfs_key_ptr));
3236 ret = tree_mod_log_insert_key(root->fs_info, lower, slot,
3240 btrfs_set_node_key(lower, key, slot);
3241 btrfs_set_node_blockptr(lower, slot, bytenr);
3242 WARN_ON(trans->transid == 0);
3243 btrfs_set_node_ptr_generation(lower, slot, trans->transid);
3244 btrfs_set_header_nritems(lower, nritems + 1);
3245 btrfs_mark_buffer_dirty(lower);
3249 * split the node at the specified level in path in two.
3250 * The path is corrected to point to the appropriate node after the split
3252 * Before splitting this tries to make some room in the node by pushing
3253 * left and right, if either one works, it returns right away.
3255 * returns 0 on success and < 0 on failure
3257 static noinline int split_node(struct btrfs_trans_handle *trans,
3258 struct btrfs_root *root,
3259 struct btrfs_path *path, int level)
3261 struct extent_buffer *c;
3262 struct extent_buffer *split;
3263 struct btrfs_disk_key disk_key;
3268 c = path->nodes[level];
3269 WARN_ON(btrfs_header_generation(c) != trans->transid);
3270 if (c == root->node) {
3272 * trying to split the root, lets make a new one
3274 * tree mod log: We don't log_removal old root in
3275 * insert_new_root, because that root buffer will be kept as a
3276 * normal node. We are going to log removal of half of the
3277 * elements below with tree_mod_log_eb_copy. We're holding a
3278 * tree lock on the buffer, which is why we cannot race with
3279 * other tree_mod_log users.
3281 ret = insert_new_root(trans, root, path, level + 1);
3285 ret = push_nodes_for_insert(trans, root, path, level);
3286 c = path->nodes[level];
3287 if (!ret && btrfs_header_nritems(c) <
3288 BTRFS_NODEPTRS_PER_BLOCK(root) - 3)
3294 c_nritems = btrfs_header_nritems(c);
3295 mid = (c_nritems + 1) / 2;
3296 btrfs_node_key(c, &disk_key, mid);
3298 split = btrfs_alloc_free_block(trans, root, root->nodesize, 0,
3299 root->root_key.objectid,
3300 &disk_key, level, c->start, 0);
3302 return PTR_ERR(split);
3304 root_add_used(root, root->nodesize);
3306 memset_extent_buffer(split, 0, 0, sizeof(struct btrfs_header));
3307 btrfs_set_header_level(split, btrfs_header_level(c));
3308 btrfs_set_header_bytenr(split, split->start);
3309 btrfs_set_header_generation(split, trans->transid);
3310 btrfs_set_header_backref_rev(split, BTRFS_MIXED_BACKREF_REV);
3311 btrfs_set_header_owner(split, root->root_key.objectid);
3312 write_extent_buffer(split, root->fs_info->fsid,
3313 (unsigned long)btrfs_header_fsid(split),
3315 write_extent_buffer(split, root->fs_info->chunk_tree_uuid,
3316 (unsigned long)btrfs_header_chunk_tree_uuid(split),
3319 tree_mod_log_eb_copy(root->fs_info, split, c, 0, mid, c_nritems - mid);
3320 copy_extent_buffer(split, c,
3321 btrfs_node_key_ptr_offset(0),
3322 btrfs_node_key_ptr_offset(mid),
3323 (c_nritems - mid) * sizeof(struct btrfs_key_ptr));
3324 btrfs_set_header_nritems(split, c_nritems - mid);
3325 btrfs_set_header_nritems(c, mid);
3328 btrfs_mark_buffer_dirty(c);
3329 btrfs_mark_buffer_dirty(split);
3331 insert_ptr(trans, root, path, &disk_key, split->start,
3332 path->slots[level + 1] + 1, level + 1);
3334 if (path->slots[level] >= mid) {
3335 path->slots[level] -= mid;
3336 btrfs_tree_unlock(c);
3337 free_extent_buffer(c);
3338 path->nodes[level] = split;
3339 path->slots[level + 1] += 1;
3341 btrfs_tree_unlock(split);
3342 free_extent_buffer(split);
3348 * how many bytes are required to store the items in a leaf. start
3349 * and nr indicate which items in the leaf to check. This totals up the
3350 * space used both by the item structs and the item data
3352 static int leaf_space_used(struct extent_buffer *l, int start, int nr)
3354 struct btrfs_item *start_item;
3355 struct btrfs_item *end_item;
3356 struct btrfs_map_token token;
3358 int nritems = btrfs_header_nritems(l);
3359 int end = min(nritems, start + nr) - 1;
3363 btrfs_init_map_token(&token);
3364 start_item = btrfs_item_nr(l, start);
3365 end_item = btrfs_item_nr(l, end);
3366 data_len = btrfs_token_item_offset(l, start_item, &token) +
3367 btrfs_token_item_size(l, start_item, &token);
3368 data_len = data_len - btrfs_token_item_offset(l, end_item, &token);
3369 data_len += sizeof(struct btrfs_item) * nr;
3370 WARN_ON(data_len < 0);
3375 * The space between the end of the leaf items and
3376 * the start of the leaf data. IOW, how much room
3377 * the leaf has left for both items and data
3379 noinline int btrfs_leaf_free_space(struct btrfs_root *root,
3380 struct extent_buffer *leaf)
3382 int nritems = btrfs_header_nritems(leaf);
3384 ret = BTRFS_LEAF_DATA_SIZE(root) - leaf_space_used(leaf, 0, nritems);
3386 printk(KERN_CRIT "leaf free space ret %d, leaf data size %lu, "
3387 "used %d nritems %d\n",
3388 ret, (unsigned long) BTRFS_LEAF_DATA_SIZE(root),
3389 leaf_space_used(leaf, 0, nritems), nritems);
3395 * min slot controls the lowest index we're willing to push to the
3396 * right. We'll push up to and including min_slot, but no lower
3398 static noinline int __push_leaf_right(struct btrfs_trans_handle *trans,
3399 struct btrfs_root *root,
3400 struct btrfs_path *path,
3401 int data_size, int empty,
3402 struct extent_buffer *right,
3403 int free_space, u32 left_nritems,
3406 struct extent_buffer *left = path->nodes[0];
3407 struct extent_buffer *upper = path->nodes[1];
3408 struct btrfs_map_token token;
3409 struct btrfs_disk_key disk_key;
3414 struct btrfs_item *item;
3420 btrfs_init_map_token(&token);
3425 nr = max_t(u32, 1, min_slot);
3427 if (path->slots[0] >= left_nritems)
3428 push_space += data_size;
3430 slot = path->slots[1];
3431 i = left_nritems - 1;
3433 item = btrfs_item_nr(left, i);
3435 if (!empty && push_items > 0) {
3436 if (path->slots[0] > i)
3438 if (path->slots[0] == i) {
3439 int space = btrfs_leaf_free_space(root, left);
3440 if (space + push_space * 2 > free_space)
3445 if (path->slots[0] == i)
3446 push_space += data_size;
3448 this_item_size = btrfs_item_size(left, item);
3449 if (this_item_size + sizeof(*item) + push_space > free_space)
3453 push_space += this_item_size + sizeof(*item);
3459 if (push_items == 0)
3462 WARN_ON(!empty && push_items == left_nritems);
3464 /* push left to right */
3465 right_nritems = btrfs_header_nritems(right);
3467 push_space = btrfs_item_end_nr(left, left_nritems - push_items);
3468 push_space -= leaf_data_end(root, left);
3470 /* make room in the right data area */
3471 data_end = leaf_data_end(root, right);
3472 memmove_extent_buffer(right,
3473 btrfs_leaf_data(right) + data_end - push_space,
3474 btrfs_leaf_data(right) + data_end,
3475 BTRFS_LEAF_DATA_SIZE(root) - data_end);
3477 /* copy from the left data area */
3478 copy_extent_buffer(right, left, btrfs_leaf_data(right) +
3479 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3480 btrfs_leaf_data(left) + leaf_data_end(root, left),
3483 memmove_extent_buffer(right, btrfs_item_nr_offset(push_items),
3484 btrfs_item_nr_offset(0),
3485 right_nritems * sizeof(struct btrfs_item));
3487 /* copy the items from left to right */
3488 copy_extent_buffer(right, left, btrfs_item_nr_offset(0),
3489 btrfs_item_nr_offset(left_nritems - push_items),
3490 push_items * sizeof(struct btrfs_item));
3492 /* update the item pointers */
3493 right_nritems += push_items;
3494 btrfs_set_header_nritems(right, right_nritems);
3495 push_space = BTRFS_LEAF_DATA_SIZE(root);
3496 for (i = 0; i < right_nritems; i++) {
3497 item = btrfs_item_nr(right, i);
3498 push_space -= btrfs_token_item_size(right, item, &token);
3499 btrfs_set_token_item_offset(right, item, push_space, &token);
3502 left_nritems -= push_items;
3503 btrfs_set_header_nritems(left, left_nritems);
3506 btrfs_mark_buffer_dirty(left);
3508 clean_tree_block(trans, root, left);
3510 btrfs_mark_buffer_dirty(right);
3512 btrfs_item_key(right, &disk_key, 0);
3513 btrfs_set_node_key(upper, &disk_key, slot + 1);
3514 btrfs_mark_buffer_dirty(upper);
3516 /* then fixup the leaf pointer in the path */
3517 if (path->slots[0] >= left_nritems) {
3518 path->slots[0] -= left_nritems;
3519 if (btrfs_header_nritems(path->nodes[0]) == 0)
3520 clean_tree_block(trans, root, path->nodes[0]);
3521 btrfs_tree_unlock(path->nodes[0]);
3522 free_extent_buffer(path->nodes[0]);
3523 path->nodes[0] = right;
3524 path->slots[1] += 1;
3526 btrfs_tree_unlock(right);
3527 free_extent_buffer(right);
3532 btrfs_tree_unlock(right);
3533 free_extent_buffer(right);
3538 * push some data in the path leaf to the right, trying to free up at
3539 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3541 * returns 1 if the push failed because the other node didn't have enough
3542 * room, 0 if everything worked out and < 0 if there were major errors.
3544 * this will push starting from min_slot to the end of the leaf. It won't
3545 * push any slot lower than min_slot
3547 static int push_leaf_right(struct btrfs_trans_handle *trans, struct btrfs_root
3548 *root, struct btrfs_path *path,
3549 int min_data_size, int data_size,
3550 int empty, u32 min_slot)
3552 struct extent_buffer *left = path->nodes[0];
3553 struct extent_buffer *right;
3554 struct extent_buffer *upper;
3560 if (!path->nodes[1])
3563 slot = path->slots[1];
3564 upper = path->nodes[1];
3565 if (slot >= btrfs_header_nritems(upper) - 1)
3568 btrfs_assert_tree_locked(path->nodes[1]);
3570 right = read_node_slot(root, upper, slot + 1);
3574 btrfs_tree_lock(right);
3575 btrfs_set_lock_blocking(right);
3577 free_space = btrfs_leaf_free_space(root, right);
3578 if (free_space < data_size)
3581 /* cow and double check */
3582 ret = btrfs_cow_block(trans, root, right, upper,
3587 free_space = btrfs_leaf_free_space(root, right);
3588 if (free_space < data_size)
3591 left_nritems = btrfs_header_nritems(left);
3592 if (left_nritems == 0)
3595 return __push_leaf_right(trans, root, path, min_data_size, empty,
3596 right, free_space, left_nritems, min_slot);
3598 btrfs_tree_unlock(right);
3599 free_extent_buffer(right);
3604 * push some data in the path leaf to the left, trying to free up at
3605 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3607 * max_slot can put a limit on how far into the leaf we'll push items. The
3608 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3611 static noinline int __push_leaf_left(struct btrfs_trans_handle *trans,
3612 struct btrfs_root *root,
3613 struct btrfs_path *path, int data_size,
3614 int empty, struct extent_buffer *left,
3615 int free_space, u32 right_nritems,
3618 struct btrfs_disk_key disk_key;
3619 struct extent_buffer *right = path->nodes[0];
3623 struct btrfs_item *item;
3624 u32 old_left_nritems;
3628 u32 old_left_item_size;
3629 struct btrfs_map_token token;
3631 btrfs_init_map_token(&token);
3634 nr = min(right_nritems, max_slot);
3636 nr = min(right_nritems - 1, max_slot);
3638 for (i = 0; i < nr; i++) {
3639 item = btrfs_item_nr(right, i);
3641 if (!empty && push_items > 0) {
3642 if (path->slots[0] < i)
3644 if (path->slots[0] == i) {
3645 int space = btrfs_leaf_free_space(root, right);
3646 if (space + push_space * 2 > free_space)
3651 if (path->slots[0] == i)
3652 push_space += data_size;
3654 this_item_size = btrfs_item_size(right, item);
3655 if (this_item_size + sizeof(*item) + push_space > free_space)
3659 push_space += this_item_size + sizeof(*item);
3662 if (push_items == 0) {
3666 if (!empty && push_items == btrfs_header_nritems(right))
3669 /* push data from right to left */
3670 copy_extent_buffer(left, right,
3671 btrfs_item_nr_offset(btrfs_header_nritems(left)),
3672 btrfs_item_nr_offset(0),
3673 push_items * sizeof(struct btrfs_item));
3675 push_space = BTRFS_LEAF_DATA_SIZE(root) -
3676 btrfs_item_offset_nr(right, push_items - 1);
3678 copy_extent_buffer(left, right, btrfs_leaf_data(left) +
3679 leaf_data_end(root, left) - push_space,
3680 btrfs_leaf_data(right) +
3681 btrfs_item_offset_nr(right, push_items - 1),
3683 old_left_nritems = btrfs_header_nritems(left);
3684 BUG_ON(old_left_nritems <= 0);
3686 old_left_item_size = btrfs_item_offset_nr(left, old_left_nritems - 1);
3687 for (i = old_left_nritems; i < old_left_nritems + push_items; i++) {
3690 item = btrfs_item_nr(left, i);
3692 ioff = btrfs_token_item_offset(left, item, &token);
3693 btrfs_set_token_item_offset(left, item,
3694 ioff - (BTRFS_LEAF_DATA_SIZE(root) - old_left_item_size),
3697 btrfs_set_header_nritems(left, old_left_nritems + push_items);
3699 /* fixup right node */
3700 if (push_items > right_nritems)
3701 WARN(1, KERN_CRIT "push items %d nr %u\n", push_items,
3704 if (push_items < right_nritems) {
3705 push_space = btrfs_item_offset_nr(right, push_items - 1) -
3706 leaf_data_end(root, right);
3707 memmove_extent_buffer(right, btrfs_leaf_data(right) +
3708 BTRFS_LEAF_DATA_SIZE(root) - push_space,
3709 btrfs_leaf_data(right) +
3710 leaf_data_end(root, right), push_space);
3712 memmove_extent_buffer(right, btrfs_item_nr_offset(0),
3713 btrfs_item_nr_offset(push_items),
3714 (btrfs_header_nritems(right) - push_items) *
3715 sizeof(struct btrfs_item));
3717 right_nritems -= push_items;
3718 btrfs_set_header_nritems(right, right_nritems);
3719 push_space = BTRFS_LEAF_DATA_SIZE(root);
3720 for (i = 0; i < right_nritems; i++) {
3721 item = btrfs_item_nr(right, i);
3723 push_space = push_space - btrfs_token_item_size(right,
3725 btrfs_set_token_item_offset(right, item, push_space, &token);
3728 btrfs_mark_buffer_dirty(left);
3730 btrfs_mark_buffer_dirty(right);
3732 clean_tree_block(trans, root, right);
3734 btrfs_item_key(right, &disk_key, 0);
3735 fixup_low_keys(root, path, &disk_key, 1);
3737 /* then fixup the leaf pointer in the path */
3738 if (path->slots[0] < push_items) {
3739 path->slots[0] += old_left_nritems;
3740 btrfs_tree_unlock(path->nodes[0]);
3741 free_extent_buffer(path->nodes[0]);
3742 path->nodes[0] = left;
3743 path->slots[1] -= 1;
3745 btrfs_tree_unlock(left);
3746 free_extent_buffer(left);
3747 path->slots[0] -= push_items;
3749 BUG_ON(path->slots[0] < 0);
3752 btrfs_tree_unlock(left);
3753 free_extent_buffer(left);
3758 * push some data in the path leaf to the left, trying to free up at
3759 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3761 * max_slot can put a limit on how far into the leaf we'll push items. The
3762 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3765 static int push_leaf_left(struct btrfs_trans_handle *trans, struct btrfs_root
3766 *root, struct btrfs_path *path, int min_data_size,
3767 int data_size, int empty, u32 max_slot)
3769 struct extent_buffer *right = path->nodes[0];
3770 struct extent_buffer *left;
3776 slot = path->slots[1];
3779 if (!path->nodes[1])
3782 right_nritems = btrfs_header_nritems(right);
3783 if (right_nritems == 0)
3786 btrfs_assert_tree_locked(path->nodes[1]);
3788 left = read_node_slot(root, path->nodes[1], slot - 1);
3792 btrfs_tree_lock(left);
3793 btrfs_set_lock_blocking(left);
3795 free_space = btrfs_leaf_free_space(root, left);
3796 if (free_space < data_size) {
3801 /* cow and double check */
3802 ret = btrfs_cow_block(trans, root, left,
3803 path->nodes[1], slot - 1, &left);
3805 /* we hit -ENOSPC, but it isn't fatal here */
3811 free_space = btrfs_leaf_free_space(root, left);
3812 if (free_space < data_size) {
3817 return __push_leaf_left(trans, root, path, min_data_size,
3818 empty, left, free_space, right_nritems,
3821 btrfs_tree_unlock(left);
3822 free_extent_buffer(left);
3827 * split the path's leaf in two, making sure there is at least data_size
3828 * available for the resulting leaf level of the path.
3830 static noinline void copy_for_split(struct btrfs_trans_handle *trans,
3831 struct btrfs_root *root,
3832 struct btrfs_path *path,
3833 struct extent_buffer *l,
3834 struct extent_buffer *right,
3835 int slot, int mid, int nritems)
3840 struct btrfs_disk_key disk_key;
3841 struct btrfs_map_token token;
3843 btrfs_init_map_token(&token);
3845 nritems = nritems - mid;
3846 btrfs_set_header_nritems(right, nritems);
3847 data_copy_size = btrfs_item_end_nr(l, mid) - leaf_data_end(root, l);
3849 copy_extent_buffer(right, l, btrfs_item_nr_offset(0),
3850 btrfs_item_nr_offset(mid),
3851 nritems * sizeof(struct btrfs_item));
3853 copy_extent_buffer(right, l,
3854 btrfs_leaf_data(right) + BTRFS_LEAF_DATA_SIZE(root) -
3855 data_copy_size, btrfs_leaf_data(l) +
3856 leaf_data_end(root, l), data_copy_size);
3858 rt_data_off = BTRFS_LEAF_DATA_SIZE(root) -
3859 btrfs_item_end_nr(l, mid);
3861 for (i = 0; i < nritems; i++) {
3862 struct btrfs_item *item = btrfs_item_nr(right, i);
3865 ioff = btrfs_token_item_offset(right, item, &token);
3866 btrfs_set_token_item_offset(right, item,
3867 ioff + rt_data_off, &token);
3870 btrfs_set_header_nritems(l, mid);
3871 btrfs_item_key(right, &disk_key, 0);
3872 insert_ptr(trans, root, path, &disk_key, right->start,
3873 path->slots[1] + 1, 1);
3875 btrfs_mark_buffer_dirty(right);
3876 btrfs_mark_buffer_dirty(l);
3877 BUG_ON(path->slots[0] != slot);
3880 btrfs_tree_unlock(path->nodes[0]);
3881 free_extent_buffer(path->nodes[0]);
3882 path->nodes[0] = right;
3883 path->slots[0] -= mid;
3884 path->slots[1] += 1;
3886 btrfs_tree_unlock(right);
3887 free_extent_buffer(right);
3890 BUG_ON(path->slots[0] < 0);
3894 * double splits happen when we need to insert a big item in the middle
3895 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3896 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3899 * We avoid this by trying to push the items on either side of our target
3900 * into the adjacent leaves. If all goes well we can avoid the double split
3903 static noinline int push_for_double_split(struct btrfs_trans_handle *trans,
3904 struct btrfs_root *root,
3905 struct btrfs_path *path,
3913 slot = path->slots[0];
3916 * try to push all the items after our slot into the
3919 ret = push_leaf_right(trans, root, path, 1, data_size, 0, slot);
3926 nritems = btrfs_header_nritems(path->nodes[0]);
3928 * our goal is to get our slot at the start or end of a leaf. If
3929 * we've done so we're done
3931 if (path->slots[0] == 0 || path->slots[0] == nritems)
3934 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
3937 /* try to push all the items before our slot into the next leaf */
3938 slot = path->slots[0];
3939 ret = push_leaf_left(trans, root, path, 1, data_size, 0, slot);
3952 * split the path's leaf in two, making sure there is at least data_size
3953 * available for the resulting leaf level of the path.
3955 * returns 0 if all went well and < 0 on failure.
3957 static noinline int split_leaf(struct btrfs_trans_handle *trans,
3958 struct btrfs_root *root,
3959 struct btrfs_key *ins_key,
3960 struct btrfs_path *path, int data_size,
3963 struct btrfs_disk_key disk_key;
3964 struct extent_buffer *l;
3968 struct extent_buffer *right;
3972 int num_doubles = 0;
3973 int tried_avoid_double = 0;
3976 slot = path->slots[0];
3977 if (extend && data_size + btrfs_item_size_nr(l, slot) +
3978 sizeof(struct btrfs_item) > BTRFS_LEAF_DATA_SIZE(root))
3981 /* first try to make some room by pushing left and right */
3982 if (data_size && path->nodes[1]) {
3983 wret = push_leaf_right(trans, root, path, data_size,
3988 wret = push_leaf_left(trans, root, path, data_size,
3989 data_size, 0, (u32)-1);
3995 /* did the pushes work? */
3996 if (btrfs_leaf_free_space(root, l) >= data_size)
4000 if (!path->nodes[1]) {
4001 ret = insert_new_root(trans, root, path, 1);
4008 slot = path->slots[0];
4009 nritems = btrfs_header_nritems(l);
4010 mid = (nritems + 1) / 2;
4014 leaf_space_used(l, mid, nritems - mid) + data_size >
4015 BTRFS_LEAF_DATA_SIZE(root)) {
4016 if (slot >= nritems) {
4020 if (mid != nritems &&
4021 leaf_space_used(l, mid, nritems - mid) +
4022 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4023 if (data_size && !tried_avoid_double)
4024 goto push_for_double;
4030 if (leaf_space_used(l, 0, mid) + data_size >
4031 BTRFS_LEAF_DATA_SIZE(root)) {
4032 if (!extend && data_size && slot == 0) {
4034 } else if ((extend || !data_size) && slot == 0) {
4038 if (mid != nritems &&
4039 leaf_space_used(l, mid, nritems - mid) +
4040 data_size > BTRFS_LEAF_DATA_SIZE(root)) {
4041 if (data_size && !tried_avoid_double)
4042 goto push_for_double;
4050 btrfs_cpu_key_to_disk(&disk_key, ins_key);
4052 btrfs_item_key(l, &disk_key, mid);
4054 right = btrfs_alloc_free_block(trans, root, root->leafsize, 0,
4055 root->root_key.objectid,
4056 &disk_key, 0, l->start, 0);
4058 return PTR_ERR(right);
4060 root_add_used(root, root->leafsize);
4062 memset_extent_buffer(right, 0, 0, sizeof(struct btrfs_header));
4063 btrfs_set_header_bytenr(right, right->start);
4064 btrfs_set_header_generation(right, trans->transid);
4065 btrfs_set_header_backref_rev(right, BTRFS_MIXED_BACKREF_REV);
4066 btrfs_set_header_owner(right, root->root_key.objectid);
4067 btrfs_set_header_level(right, 0);
4068 write_extent_buffer(right, root->fs_info->fsid,
4069 (unsigned long)btrfs_header_fsid(right),
4072 write_extent_buffer(right, root->fs_info->chunk_tree_uuid,
4073 (unsigned long)btrfs_header_chunk_tree_uuid(right),
4078 btrfs_set_header_nritems(right, 0);
4079 insert_ptr(trans, root, path, &disk_key, right->start,
4080 path->slots[1] + 1, 1);
4081 btrfs_tree_unlock(path->nodes[0]);
4082 free_extent_buffer(path->nodes[0]);
4083 path->nodes[0] = right;
4085 path->slots[1] += 1;
4087 btrfs_set_header_nritems(right, 0);
4088 insert_ptr(trans, root, path, &disk_key, right->start,
4090 btrfs_tree_unlock(path->nodes[0]);
4091 free_extent_buffer(path->nodes[0]);
4092 path->nodes[0] = right;
4094 if (path->slots[1] == 0)
4095 fixup_low_keys(root, path, &disk_key, 1);
4097 btrfs_mark_buffer_dirty(right);
4101 copy_for_split(trans, root, path, l, right, slot, mid, nritems);
4104 BUG_ON(num_doubles != 0);
4112 push_for_double_split(trans, root, path, data_size);
4113 tried_avoid_double = 1;
4114 if (btrfs_leaf_free_space(root, path->nodes[0]) >= data_size)
4119 static noinline int setup_leaf_for_split(struct btrfs_trans_handle *trans,
4120 struct btrfs_root *root,
4121 struct btrfs_path *path, int ins_len)
4123 struct btrfs_key key;
4124 struct extent_buffer *leaf;
4125 struct btrfs_file_extent_item *fi;
4130 leaf = path->nodes[0];
4131 btrfs_item_key_to_cpu(leaf, &key, path->slots[0]);
4133 BUG_ON(key.type != BTRFS_EXTENT_DATA_KEY &&
4134 key.type != BTRFS_EXTENT_CSUM_KEY);
4136 if (btrfs_leaf_free_space(root, leaf) >= ins_len)
4139 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4140 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4141 fi = btrfs_item_ptr(leaf, path->slots[0],
4142 struct btrfs_file_extent_item);
4143 extent_len = btrfs_file_extent_num_bytes(leaf, fi);
4145 btrfs_release_path(path);
4147 path->keep_locks = 1;
4148 path->search_for_split = 1;
4149 ret = btrfs_search_slot(trans, root, &key, path, 0, 1);
4150 path->search_for_split = 0;
4155 leaf = path->nodes[0];
4156 /* if our item isn't there or got smaller, return now */
4157 if (ret > 0 || item_size != btrfs_item_size_nr(leaf, path->slots[0]))
4160 /* the leaf has changed, it now has room. return now */
4161 if (btrfs_leaf_free_space(root, path->nodes[0]) >= ins_len)
4164 if (key.type == BTRFS_EXTENT_DATA_KEY) {
4165 fi = btrfs_item_ptr(leaf, path->slots[0],
4166 struct btrfs_file_extent_item);
4167 if (extent_len != btrfs_file_extent_num_bytes(leaf, fi))
4171 btrfs_set_path_blocking(path);
4172 ret = split_leaf(trans, root, &key, path, ins_len, 1);
4176 path->keep_locks = 0;
4177 btrfs_unlock_up_safe(path, 1);
4180 path->keep_locks = 0;
4184 static noinline int split_item(struct btrfs_trans_handle *trans,
4185 struct btrfs_root *root,
4186 struct btrfs_path *path,
4187 struct btrfs_key *new_key,
4188 unsigned long split_offset)
4190 struct extent_buffer *leaf;
4191 struct btrfs_item *item;
4192 struct btrfs_item *new_item;
4198 struct btrfs_disk_key disk_key;
4200 leaf = path->nodes[0];
4201 BUG_ON(btrfs_leaf_free_space(root, leaf) < sizeof(struct btrfs_item));
4203 btrfs_set_path_blocking(path);
4205 item = btrfs_item_nr(leaf, path->slots[0]);
4206 orig_offset = btrfs_item_offset(leaf, item);
4207 item_size = btrfs_item_size(leaf, item);
4209 buf = kmalloc(item_size, GFP_NOFS);
4213 read_extent_buffer(leaf, buf, btrfs_item_ptr_offset(leaf,
4214 path->slots[0]), item_size);
4216 slot = path->slots[0] + 1;
4217 nritems = btrfs_header_nritems(leaf);
4218 if (slot != nritems) {
4219 /* shift the items */
4220 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + 1),
4221 btrfs_item_nr_offset(slot),
4222 (nritems - slot) * sizeof(struct btrfs_item));
4225 btrfs_cpu_key_to_disk(&disk_key, new_key);
4226 btrfs_set_item_key(leaf, &disk_key, slot);
4228 new_item = btrfs_item_nr(leaf, slot);
4230 btrfs_set_item_offset(leaf, new_item, orig_offset);
4231 btrfs_set_item_size(leaf, new_item, item_size - split_offset);
4233 btrfs_set_item_offset(leaf, item,
4234 orig_offset + item_size - split_offset);
4235 btrfs_set_item_size(leaf, item, split_offset);
4237 btrfs_set_header_nritems(leaf, nritems + 1);
4239 /* write the data for the start of the original item */
4240 write_extent_buffer(leaf, buf,
4241 btrfs_item_ptr_offset(leaf, path->slots[0]),
4244 /* write the data for the new item */
4245 write_extent_buffer(leaf, buf + split_offset,
4246 btrfs_item_ptr_offset(leaf, slot),
4247 item_size - split_offset);
4248 btrfs_mark_buffer_dirty(leaf);
4250 BUG_ON(btrfs_leaf_free_space(root, leaf) < 0);
4256 * This function splits a single item into two items,
4257 * giving 'new_key' to the new item and splitting the
4258 * old one at split_offset (from the start of the item).
4260 * The path may be released by this operation. After
4261 * the split, the path is pointing to the old item. The
4262 * new item is going to be in the same node as the old one.
4264 * Note, the item being split must be smaller enough to live alone on
4265 * a tree block with room for one extra struct btrfs_item
4267 * This allows us to split the item in place, keeping a lock on the
4268 * leaf the entire time.
4270 int btrfs_split_item(struct btrfs_trans_handle *trans,
4271 struct btrfs_root *root,
4272 struct btrfs_path *path,
4273 struct btrfs_key *new_key,
4274 unsigned long split_offset)
4277 ret = setup_leaf_for_split(trans, root, path,
4278 sizeof(struct btrfs_item));
4282 ret = split_item(trans, root, path, new_key, split_offset);
4287 * This function duplicate a item, giving 'new_key' to the new item.
4288 * It guarantees both items live in the same tree leaf and the new item
4289 * is contiguous with the original item.
4291 * This allows us to split file extent in place, keeping a lock on the
4292 * leaf the entire time.
4294 int btrfs_duplicate_item(struct btrfs_trans_handle *trans,
4295 struct btrfs_root *root,
4296 struct btrfs_path *path,
4297 struct btrfs_key *new_key)
4299 struct extent_buffer *leaf;
4303 leaf = path->nodes[0];
4304 item_size = btrfs_item_size_nr(leaf, path->slots[0]);
4305 ret = setup_leaf_for_split(trans, root, path,
4306 item_size + sizeof(struct btrfs_item));
4311 setup_items_for_insert(root, path, new_key, &item_size,
4312 item_size, item_size +
4313 sizeof(struct btrfs_item), 1);
4314 leaf = path->nodes[0];
4315 memcpy_extent_buffer(leaf,
4316 btrfs_item_ptr_offset(leaf, path->slots[0]),
4317 btrfs_item_ptr_offset(leaf, path->slots[0] - 1),
4323 * make the item pointed to by the path smaller. new_size indicates
4324 * how small to make it, and from_end tells us if we just chop bytes
4325 * off the end of the item or if we shift the item to chop bytes off
4328 void btrfs_truncate_item(struct btrfs_root *root, struct btrfs_path *path,
4329 u32 new_size, int from_end)
4332 struct extent_buffer *leaf;
4333 struct btrfs_item *item;
4335 unsigned int data_end;
4336 unsigned int old_data_start;
4337 unsigned int old_size;
4338 unsigned int size_diff;
4340 struct btrfs_map_token token;
4342 btrfs_init_map_token(&token);
4344 leaf = path->nodes[0];
4345 slot = path->slots[0];
4347 old_size = btrfs_item_size_nr(leaf, slot);
4348 if (old_size == new_size)
4351 nritems = btrfs_header_nritems(leaf);
4352 data_end = leaf_data_end(root, leaf);
4354 old_data_start = btrfs_item_offset_nr(leaf, slot);
4356 size_diff = old_size - new_size;
4359 BUG_ON(slot >= nritems);
4362 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4364 /* first correct the data pointers */
4365 for (i = slot; i < nritems; i++) {
4367 item = btrfs_item_nr(leaf, i);
4369 ioff = btrfs_token_item_offset(leaf, item, &token);
4370 btrfs_set_token_item_offset(leaf, item,
4371 ioff + size_diff, &token);
4374 /* shift the data */
4376 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4377 data_end + size_diff, btrfs_leaf_data(leaf) +
4378 data_end, old_data_start + new_size - data_end);
4380 struct btrfs_disk_key disk_key;
4383 btrfs_item_key(leaf, &disk_key, slot);
4385 if (btrfs_disk_key_type(&disk_key) == BTRFS_EXTENT_DATA_KEY) {
4387 struct btrfs_file_extent_item *fi;
4389 fi = btrfs_item_ptr(leaf, slot,
4390 struct btrfs_file_extent_item);
4391 fi = (struct btrfs_file_extent_item *)(
4392 (unsigned long)fi - size_diff);
4394 if (btrfs_file_extent_type(leaf, fi) ==
4395 BTRFS_FILE_EXTENT_INLINE) {
4396 ptr = btrfs_item_ptr_offset(leaf, slot);
4397 memmove_extent_buffer(leaf, ptr,
4399 offsetof(struct btrfs_file_extent_item,
4404 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4405 data_end + size_diff, btrfs_leaf_data(leaf) +
4406 data_end, old_data_start - data_end);
4408 offset = btrfs_disk_key_offset(&disk_key);
4409 btrfs_set_disk_key_offset(&disk_key, offset + size_diff);
4410 btrfs_set_item_key(leaf, &disk_key, slot);
4412 fixup_low_keys(root, path, &disk_key, 1);
4415 item = btrfs_item_nr(leaf, slot);
4416 btrfs_set_item_size(leaf, item, new_size);
4417 btrfs_mark_buffer_dirty(leaf);
4419 if (btrfs_leaf_free_space(root, leaf) < 0) {
4420 btrfs_print_leaf(root, leaf);
4426 * make the item pointed to by the path bigger, data_size is the added size.
4428 void btrfs_extend_item(struct btrfs_root *root, struct btrfs_path *path,
4432 struct extent_buffer *leaf;
4433 struct btrfs_item *item;
4435 unsigned int data_end;
4436 unsigned int old_data;
4437 unsigned int old_size;
4439 struct btrfs_map_token token;
4441 btrfs_init_map_token(&token);
4443 leaf = path->nodes[0];
4445 nritems = btrfs_header_nritems(leaf);
4446 data_end = leaf_data_end(root, leaf);
4448 if (btrfs_leaf_free_space(root, leaf) < data_size) {
4449 btrfs_print_leaf(root, leaf);
4452 slot = path->slots[0];
4453 old_data = btrfs_item_end_nr(leaf, slot);
4456 if (slot >= nritems) {
4457 btrfs_print_leaf(root, leaf);
4458 printk(KERN_CRIT "slot %d too large, nritems %d\n",
4464 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4466 /* first correct the data pointers */
4467 for (i = slot; i < nritems; i++) {
4469 item = btrfs_item_nr(leaf, i);
4471 ioff = btrfs_token_item_offset(leaf, item, &token);
4472 btrfs_set_token_item_offset(leaf, item,
4473 ioff - data_size, &token);
4476 /* shift the data */
4477 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4478 data_end - data_size, btrfs_leaf_data(leaf) +
4479 data_end, old_data - data_end);
4481 data_end = old_data;
4482 old_size = btrfs_item_size_nr(leaf, slot);
4483 item = btrfs_item_nr(leaf, slot);
4484 btrfs_set_item_size(leaf, item, old_size + data_size);
4485 btrfs_mark_buffer_dirty(leaf);
4487 if (btrfs_leaf_free_space(root, leaf) < 0) {
4488 btrfs_print_leaf(root, leaf);
4494 * this is a helper for btrfs_insert_empty_items, the main goal here is
4495 * to save stack depth by doing the bulk of the work in a function
4496 * that doesn't call btrfs_search_slot
4498 void setup_items_for_insert(struct btrfs_root *root, struct btrfs_path *path,
4499 struct btrfs_key *cpu_key, u32 *data_size,
4500 u32 total_data, u32 total_size, int nr)
4502 struct btrfs_item *item;
4505 unsigned int data_end;
4506 struct btrfs_disk_key disk_key;
4507 struct extent_buffer *leaf;
4509 struct btrfs_map_token token;
4511 btrfs_init_map_token(&token);
4513 leaf = path->nodes[0];
4514 slot = path->slots[0];
4516 nritems = btrfs_header_nritems(leaf);
4517 data_end = leaf_data_end(root, leaf);
4519 if (btrfs_leaf_free_space(root, leaf) < total_size) {
4520 btrfs_print_leaf(root, leaf);
4521 printk(KERN_CRIT "not enough freespace need %u have %d\n",
4522 total_size, btrfs_leaf_free_space(root, leaf));
4526 if (slot != nritems) {
4527 unsigned int old_data = btrfs_item_end_nr(leaf, slot);
4529 if (old_data < data_end) {
4530 btrfs_print_leaf(root, leaf);
4531 printk(KERN_CRIT "slot %d old_data %d data_end %d\n",
4532 slot, old_data, data_end);
4536 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4538 /* first correct the data pointers */
4539 for (i = slot; i < nritems; i++) {
4542 item = btrfs_item_nr(leaf, i);
4543 ioff = btrfs_token_item_offset(leaf, item, &token);
4544 btrfs_set_token_item_offset(leaf, item,
4545 ioff - total_data, &token);
4547 /* shift the items */
4548 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot + nr),
4549 btrfs_item_nr_offset(slot),
4550 (nritems - slot) * sizeof(struct btrfs_item));
4552 /* shift the data */
4553 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4554 data_end - total_data, btrfs_leaf_data(leaf) +
4555 data_end, old_data - data_end);
4556 data_end = old_data;
4559 /* setup the item for the new data */
4560 for (i = 0; i < nr; i++) {
4561 btrfs_cpu_key_to_disk(&disk_key, cpu_key + i);
4562 btrfs_set_item_key(leaf, &disk_key, slot + i);
4563 item = btrfs_item_nr(leaf, slot + i);
4564 btrfs_set_token_item_offset(leaf, item,
4565 data_end - data_size[i], &token);
4566 data_end -= data_size[i];
4567 btrfs_set_token_item_size(leaf, item, data_size[i], &token);
4570 btrfs_set_header_nritems(leaf, nritems + nr);
4573 btrfs_cpu_key_to_disk(&disk_key, cpu_key);
4574 fixup_low_keys(root, path, &disk_key, 1);
4576 btrfs_unlock_up_safe(path, 1);
4577 btrfs_mark_buffer_dirty(leaf);
4579 if (btrfs_leaf_free_space(root, leaf) < 0) {
4580 btrfs_print_leaf(root, leaf);
4586 * Given a key and some data, insert items into the tree.
4587 * This does all the path init required, making room in the tree if needed.
4589 int btrfs_insert_empty_items(struct btrfs_trans_handle *trans,
4590 struct btrfs_root *root,
4591 struct btrfs_path *path,
4592 struct btrfs_key *cpu_key, u32 *data_size,
4601 for (i = 0; i < nr; i++)
4602 total_data += data_size[i];
4604 total_size = total_data + (nr * sizeof(struct btrfs_item));
4605 ret = btrfs_search_slot(trans, root, cpu_key, path, total_size, 1);
4611 slot = path->slots[0];
4614 setup_items_for_insert(root, path, cpu_key, data_size,
4615 total_data, total_size, nr);
4620 * Given a key and some data, insert an item into the tree.
4621 * This does all the path init required, making room in the tree if needed.
4623 int btrfs_insert_item(struct btrfs_trans_handle *trans, struct btrfs_root
4624 *root, struct btrfs_key *cpu_key, void *data, u32
4628 struct btrfs_path *path;
4629 struct extent_buffer *leaf;
4632 path = btrfs_alloc_path();
4635 ret = btrfs_insert_empty_item(trans, root, path, cpu_key, data_size);
4637 leaf = path->nodes[0];
4638 ptr = btrfs_item_ptr_offset(leaf, path->slots[0]);
4639 write_extent_buffer(leaf, data, ptr, data_size);
4640 btrfs_mark_buffer_dirty(leaf);
4642 btrfs_free_path(path);
4647 * delete the pointer from a given node.
4649 * the tree should have been previously balanced so the deletion does not
4652 static void del_ptr(struct btrfs_root *root, struct btrfs_path *path,
4653 int level, int slot)
4655 struct extent_buffer *parent = path->nodes[level];
4659 nritems = btrfs_header_nritems(parent);
4660 if (slot != nritems - 1) {
4662 tree_mod_log_eb_move(root->fs_info, parent, slot,
4663 slot + 1, nritems - slot - 1);
4664 memmove_extent_buffer(parent,
4665 btrfs_node_key_ptr_offset(slot),
4666 btrfs_node_key_ptr_offset(slot + 1),
4667 sizeof(struct btrfs_key_ptr) *
4668 (nritems - slot - 1));
4670 ret = tree_mod_log_insert_key(root->fs_info, parent, slot,
4671 MOD_LOG_KEY_REMOVE);
4676 btrfs_set_header_nritems(parent, nritems);
4677 if (nritems == 0 && parent == root->node) {
4678 BUG_ON(btrfs_header_level(root->node) != 1);
4679 /* just turn the root into a leaf and break */
4680 btrfs_set_header_level(root->node, 0);
4681 } else if (slot == 0) {
4682 struct btrfs_disk_key disk_key;
4684 btrfs_node_key(parent, &disk_key, 0);
4685 fixup_low_keys(root, path, &disk_key, level + 1);
4687 btrfs_mark_buffer_dirty(parent);
4691 * a helper function to delete the leaf pointed to by path->slots[1] and
4694 * This deletes the pointer in path->nodes[1] and frees the leaf
4695 * block extent. zero is returned if it all worked out, < 0 otherwise.
4697 * The path must have already been setup for deleting the leaf, including
4698 * all the proper balancing. path->nodes[1] must be locked.
4700 static noinline void btrfs_del_leaf(struct btrfs_trans_handle *trans,
4701 struct btrfs_root *root,
4702 struct btrfs_path *path,
4703 struct extent_buffer *leaf)
4705 WARN_ON(btrfs_header_generation(leaf) != trans->transid);
4706 del_ptr(root, path, 1, path->slots[1]);
4709 * btrfs_free_extent is expensive, we want to make sure we
4710 * aren't holding any locks when we call it
4712 btrfs_unlock_up_safe(path, 0);
4714 root_sub_used(root, leaf->len);
4716 extent_buffer_get(leaf);
4717 btrfs_free_tree_block(trans, root, leaf, 0, 1);
4718 free_extent_buffer_stale(leaf);
4721 * delete the item at the leaf level in path. If that empties
4722 * the leaf, remove it from the tree
4724 int btrfs_del_items(struct btrfs_trans_handle *trans, struct btrfs_root *root,
4725 struct btrfs_path *path, int slot, int nr)
4727 struct extent_buffer *leaf;
4728 struct btrfs_item *item;
4735 struct btrfs_map_token token;
4737 btrfs_init_map_token(&token);
4739 leaf = path->nodes[0];
4740 last_off = btrfs_item_offset_nr(leaf, slot + nr - 1);
4742 for (i = 0; i < nr; i++)
4743 dsize += btrfs_item_size_nr(leaf, slot + i);
4745 nritems = btrfs_header_nritems(leaf);
4747 if (slot + nr != nritems) {
4748 int data_end = leaf_data_end(root, leaf);
4750 memmove_extent_buffer(leaf, btrfs_leaf_data(leaf) +
4752 btrfs_leaf_data(leaf) + data_end,
4753 last_off - data_end);
4755 for (i = slot + nr; i < nritems; i++) {
4758 item = btrfs_item_nr(leaf, i);
4759 ioff = btrfs_token_item_offset(leaf, item, &token);
4760 btrfs_set_token_item_offset(leaf, item,
4761 ioff + dsize, &token);
4764 memmove_extent_buffer(leaf, btrfs_item_nr_offset(slot),
4765 btrfs_item_nr_offset(slot + nr),
4766 sizeof(struct btrfs_item) *
4767 (nritems - slot - nr));
4769 btrfs_set_header_nritems(leaf, nritems - nr);
4772 /* delete the leaf if we've emptied it */
4774 if (leaf == root->node) {
4775 btrfs_set_header_level(leaf, 0);
4777 btrfs_set_path_blocking(path);
4778 clean_tree_block(trans, root, leaf);
4779 btrfs_del_leaf(trans, root, path, leaf);
4782 int used = leaf_space_used(leaf, 0, nritems);
4784 struct btrfs_disk_key disk_key;
4786 btrfs_item_key(leaf, &disk_key, 0);
4787 fixup_low_keys(root, path, &disk_key, 1);
4790 /* delete the leaf if it is mostly empty */
4791 if (used < BTRFS_LEAF_DATA_SIZE(root) / 3) {
4792 /* push_leaf_left fixes the path.
4793 * make sure the path still points to our leaf
4794 * for possible call to del_ptr below
4796 slot = path->slots[1];
4797 extent_buffer_get(leaf);
4799 btrfs_set_path_blocking(path);
4800 wret = push_leaf_left(trans, root, path, 1, 1,
4802 if (wret < 0 && wret != -ENOSPC)
4805 if (path->nodes[0] == leaf &&
4806 btrfs_header_nritems(leaf)) {
4807 wret = push_leaf_right(trans, root, path, 1,
4809 if (wret < 0 && wret != -ENOSPC)
4813 if (btrfs_header_nritems(leaf) == 0) {
4814 path->slots[1] = slot;
4815 btrfs_del_leaf(trans, root, path, leaf);
4816 free_extent_buffer(leaf);
4819 /* if we're still in the path, make sure
4820 * we're dirty. Otherwise, one of the
4821 * push_leaf functions must have already
4822 * dirtied this buffer
4824 if (path->nodes[0] == leaf)
4825 btrfs_mark_buffer_dirty(leaf);
4826 free_extent_buffer(leaf);
4829 btrfs_mark_buffer_dirty(leaf);
4836 * search the tree again to find a leaf with lesser keys
4837 * returns 0 if it found something or 1 if there are no lesser leaves.
4838 * returns < 0 on io errors.
4840 * This may release the path, and so you may lose any locks held at the
4843 int btrfs_prev_leaf(struct btrfs_root *root, struct btrfs_path *path)
4845 struct btrfs_key key;
4846 struct btrfs_disk_key found_key;
4849 btrfs_item_key_to_cpu(path->nodes[0], &key, 0);
4853 else if (key.type > 0)
4855 else if (key.objectid > 0)
4860 btrfs_release_path(path);
4861 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
4864 btrfs_item_key(path->nodes[0], &found_key, 0);
4865 ret = comp_keys(&found_key, &key);
4872 * A helper function to walk down the tree starting at min_key, and looking
4873 * for nodes or leaves that are have a minimum transaction id.
4874 * This is used by the btree defrag code, and tree logging
4876 * This does not cow, but it does stuff the starting key it finds back
4877 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4878 * key and get a writable path.
4880 * This does lock as it descends, and path->keep_locks should be set
4881 * to 1 by the caller.
4883 * This honors path->lowest_level to prevent descent past a given level
4886 * min_trans indicates the oldest transaction that you are interested
4887 * in walking through. Any nodes or leaves older than min_trans are
4888 * skipped over (without reading them).
4890 * returns zero if something useful was found, < 0 on error and 1 if there
4891 * was nothing in the tree that matched the search criteria.
4893 int btrfs_search_forward(struct btrfs_root *root, struct btrfs_key *min_key,
4894 struct btrfs_key *max_key,
4895 struct btrfs_path *path,
4898 struct extent_buffer *cur;
4899 struct btrfs_key found_key;
4906 WARN_ON(!path->keep_locks);
4908 cur = btrfs_read_lock_root_node(root);
4909 level = btrfs_header_level(cur);
4910 WARN_ON(path->nodes[level]);
4911 path->nodes[level] = cur;
4912 path->locks[level] = BTRFS_READ_LOCK;
4914 if (btrfs_header_generation(cur) < min_trans) {
4919 nritems = btrfs_header_nritems(cur);
4920 level = btrfs_header_level(cur);
4921 sret = bin_search(cur, min_key, level, &slot);
4923 /* at the lowest level, we're done, setup the path and exit */
4924 if (level == path->lowest_level) {
4925 if (slot >= nritems)
4928 path->slots[level] = slot;
4929 btrfs_item_key_to_cpu(cur, &found_key, slot);
4932 if (sret && slot > 0)
4935 * check this node pointer against the min_trans parameters.
4936 * If it is too old, old, skip to the next one.
4938 while (slot < nritems) {
4942 blockptr = btrfs_node_blockptr(cur, slot);
4943 gen = btrfs_node_ptr_generation(cur, slot);
4944 if (gen < min_trans) {
4952 * we didn't find a candidate key in this node, walk forward
4953 * and find another one
4955 if (slot >= nritems) {
4956 path->slots[level] = slot;
4957 btrfs_set_path_blocking(path);
4958 sret = btrfs_find_next_key(root, path, min_key, level,
4961 btrfs_release_path(path);
4967 /* save our key for returning back */
4968 btrfs_node_key_to_cpu(cur, &found_key, slot);
4969 path->slots[level] = slot;
4970 if (level == path->lowest_level) {
4972 unlock_up(path, level, 1, 0, NULL);
4975 btrfs_set_path_blocking(path);
4976 cur = read_node_slot(root, cur, slot);
4977 BUG_ON(!cur); /* -ENOMEM */
4979 btrfs_tree_read_lock(cur);
4981 path->locks[level - 1] = BTRFS_READ_LOCK;
4982 path->nodes[level - 1] = cur;
4983 unlock_up(path, level, 1, 0, NULL);
4984 btrfs_clear_path_blocking(path, NULL, 0);
4988 memcpy(min_key, &found_key, sizeof(found_key));
4989 btrfs_set_path_blocking(path);
4993 static void tree_move_down(struct btrfs_root *root,
4994 struct btrfs_path *path,
4995 int *level, int root_level)
4997 BUG_ON(*level == 0);
4998 path->nodes[*level - 1] = read_node_slot(root, path->nodes[*level],
4999 path->slots[*level]);
5000 path->slots[*level - 1] = 0;
5004 static int tree_move_next_or_upnext(struct btrfs_root *root,
5005 struct btrfs_path *path,
5006 int *level, int root_level)
5010 nritems = btrfs_header_nritems(path->nodes[*level]);
5012 path->slots[*level]++;
5014 while (path->slots[*level] >= nritems) {
5015 if (*level == root_level)
5019 path->slots[*level] = 0;
5020 free_extent_buffer(path->nodes[*level]);
5021 path->nodes[*level] = NULL;
5023 path->slots[*level]++;
5025 nritems = btrfs_header_nritems(path->nodes[*level]);
5032 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5035 static int tree_advance(struct btrfs_root *root,
5036 struct btrfs_path *path,
5037 int *level, int root_level,
5039 struct btrfs_key *key)
5043 if (*level == 0 || !allow_down) {
5044 ret = tree_move_next_or_upnext(root, path, level, root_level);
5046 tree_move_down(root, path, level, root_level);
5051 btrfs_item_key_to_cpu(path->nodes[*level], key,
5052 path->slots[*level]);
5054 btrfs_node_key_to_cpu(path->nodes[*level], key,
5055 path->slots[*level]);
5060 static int tree_compare_item(struct btrfs_root *left_root,
5061 struct btrfs_path *left_path,
5062 struct btrfs_path *right_path,
5067 unsigned long off1, off2;
5069 len1 = btrfs_item_size_nr(left_path->nodes[0], left_path->slots[0]);
5070 len2 = btrfs_item_size_nr(right_path->nodes[0], right_path->slots[0]);
5074 off1 = btrfs_item_ptr_offset(left_path->nodes[0], left_path->slots[0]);
5075 off2 = btrfs_item_ptr_offset(right_path->nodes[0],
5076 right_path->slots[0]);
5078 read_extent_buffer(left_path->nodes[0], tmp_buf, off1, len1);
5080 cmp = memcmp_extent_buffer(right_path->nodes[0], tmp_buf, off2, len1);
5087 #define ADVANCE_ONLY_NEXT -1
5090 * This function compares two trees and calls the provided callback for
5091 * every changed/new/deleted item it finds.
5092 * If shared tree blocks are encountered, whole subtrees are skipped, making
5093 * the compare pretty fast on snapshotted subvolumes.
5095 * This currently works on commit roots only. As commit roots are read only,
5096 * we don't do any locking. The commit roots are protected with transactions.
5097 * Transactions are ended and rejoined when a commit is tried in between.
5099 * This function checks for modifications done to the trees while comparing.
5100 * If it detects a change, it aborts immediately.
5102 int btrfs_compare_trees(struct btrfs_root *left_root,
5103 struct btrfs_root *right_root,
5104 btrfs_changed_cb_t changed_cb, void *ctx)
5108 struct btrfs_trans_handle *trans = NULL;
5109 struct btrfs_path *left_path = NULL;
5110 struct btrfs_path *right_path = NULL;
5111 struct btrfs_key left_key;
5112 struct btrfs_key right_key;
5113 char *tmp_buf = NULL;
5114 int left_root_level;
5115 int right_root_level;
5118 int left_end_reached;
5119 int right_end_reached;
5124 u64 left_start_ctransid;
5125 u64 right_start_ctransid;
5128 left_path = btrfs_alloc_path();
5133 right_path = btrfs_alloc_path();
5139 tmp_buf = kmalloc(left_root->leafsize, GFP_NOFS);
5145 left_path->search_commit_root = 1;
5146 left_path->skip_locking = 1;
5147 right_path->search_commit_root = 1;
5148 right_path->skip_locking = 1;
5150 spin_lock(&left_root->root_item_lock);
5151 left_start_ctransid = btrfs_root_ctransid(&left_root->root_item);
5152 spin_unlock(&left_root->root_item_lock);
5154 spin_lock(&right_root->root_item_lock);
5155 right_start_ctransid = btrfs_root_ctransid(&right_root->root_item);
5156 spin_unlock(&right_root->root_item_lock);
5158 trans = btrfs_join_transaction(left_root);
5159 if (IS_ERR(trans)) {
5160 ret = PTR_ERR(trans);
5166 * Strategy: Go to the first items of both trees. Then do
5168 * If both trees are at level 0
5169 * Compare keys of current items
5170 * If left < right treat left item as new, advance left tree
5172 * If left > right treat right item as deleted, advance right tree
5174 * If left == right do deep compare of items, treat as changed if
5175 * needed, advance both trees and repeat
5176 * If both trees are at the same level but not at level 0
5177 * Compare keys of current nodes/leafs
5178 * If left < right advance left tree and repeat
5179 * If left > right advance right tree and repeat
5180 * If left == right compare blockptrs of the next nodes/leafs
5181 * If they match advance both trees but stay at the same level
5183 * If they don't match advance both trees while allowing to go
5185 * If tree levels are different
5186 * Advance the tree that needs it and repeat
5188 * Advancing a tree means:
5189 * If we are at level 0, try to go to the next slot. If that's not
5190 * possible, go one level up and repeat. Stop when we found a level
5191 * where we could go to the next slot. We may at this point be on a
5194 * If we are not at level 0 and not on shared tree blocks, go one
5197 * If we are not at level 0 and on shared tree blocks, go one slot to
5198 * the right if possible or go up and right.
5201 left_level = btrfs_header_level(left_root->commit_root);
5202 left_root_level = left_level;
5203 left_path->nodes[left_level] = left_root->commit_root;
5204 extent_buffer_get(left_path->nodes[left_level]);
5206 right_level = btrfs_header_level(right_root->commit_root);
5207 right_root_level = right_level;
5208 right_path->nodes[right_level] = right_root->commit_root;
5209 extent_buffer_get(right_path->nodes[right_level]);
5211 if (left_level == 0)
5212 btrfs_item_key_to_cpu(left_path->nodes[left_level],
5213 &left_key, left_path->slots[left_level]);
5215 btrfs_node_key_to_cpu(left_path->nodes[left_level],
5216 &left_key, left_path->slots[left_level]);
5217 if (right_level == 0)
5218 btrfs_item_key_to_cpu(right_path->nodes[right_level],
5219 &right_key, right_path->slots[right_level]);
5221 btrfs_node_key_to_cpu(right_path->nodes[right_level],
5222 &right_key, right_path->slots[right_level]);
5224 left_end_reached = right_end_reached = 0;
5225 advance_left = advance_right = 0;
5229 * We need to make sure the transaction does not get committed
5230 * while we do anything on commit roots. This means, we need to
5231 * join and leave transactions for every item that we process.
5233 if (trans && btrfs_should_end_transaction(trans, left_root)) {
5234 btrfs_release_path(left_path);
5235 btrfs_release_path(right_path);
5237 ret = btrfs_end_transaction(trans, left_root);
5242 /* now rejoin the transaction */
5244 trans = btrfs_join_transaction(left_root);
5245 if (IS_ERR(trans)) {
5246 ret = PTR_ERR(trans);
5251 spin_lock(&left_root->root_item_lock);
5252 ctransid = btrfs_root_ctransid(&left_root->root_item);
5253 spin_unlock(&left_root->root_item_lock);
5254 if (ctransid != left_start_ctransid)
5255 left_start_ctransid = 0;
5257 spin_lock(&right_root->root_item_lock);
5258 ctransid = btrfs_root_ctransid(&right_root->root_item);
5259 spin_unlock(&right_root->root_item_lock);
5260 if (ctransid != right_start_ctransid)
5261 right_start_ctransid = 0;
5263 if (!left_start_ctransid || !right_start_ctransid) {
5264 WARN(1, KERN_WARNING
5265 "btrfs: btrfs_compare_tree detected "
5266 "a change in one of the trees while "
5267 "iterating. This is probably a "
5274 * the commit root may have changed, so start again
5277 left_path->lowest_level = left_level;
5278 right_path->lowest_level = right_level;
5279 ret = btrfs_search_slot(NULL, left_root,
5280 &left_key, left_path, 0, 0);
5283 ret = btrfs_search_slot(NULL, right_root,
5284 &right_key, right_path, 0, 0);
5289 if (advance_left && !left_end_reached) {
5290 ret = tree_advance(left_root, left_path, &left_level,
5292 advance_left != ADVANCE_ONLY_NEXT,
5295 left_end_reached = ADVANCE;
5298 if (advance_right && !right_end_reached) {
5299 ret = tree_advance(right_root, right_path, &right_level,
5301 advance_right != ADVANCE_ONLY_NEXT,
5304 right_end_reached = ADVANCE;
5308 if (left_end_reached && right_end_reached) {
5311 } else if (left_end_reached) {
5312 if (right_level == 0) {
5313 ret = changed_cb(left_root, right_root,
5314 left_path, right_path,
5316 BTRFS_COMPARE_TREE_DELETED,
5321 advance_right = ADVANCE;
5323 } else if (right_end_reached) {
5324 if (left_level == 0) {
5325 ret = changed_cb(left_root, right_root,
5326 left_path, right_path,
5328 BTRFS_COMPARE_TREE_NEW,
5333 advance_left = ADVANCE;
5337 if (left_level == 0 && right_level == 0) {
5338 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5340 ret = changed_cb(left_root, right_root,
5341 left_path, right_path,
5343 BTRFS_COMPARE_TREE_NEW,
5347 advance_left = ADVANCE;
5348 } else if (cmp > 0) {
5349 ret = changed_cb(left_root, right_root,
5350 left_path, right_path,
5352 BTRFS_COMPARE_TREE_DELETED,
5356 advance_right = ADVANCE;
5358 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5359 ret = tree_compare_item(left_root, left_path,
5360 right_path, tmp_buf);
5362 WARN_ON(!extent_buffer_uptodate(left_path->nodes[0]));
5363 ret = changed_cb(left_root, right_root,
5364 left_path, right_path,
5366 BTRFS_COMPARE_TREE_CHANGED,
5371 advance_left = ADVANCE;
5372 advance_right = ADVANCE;
5374 } else if (left_level == right_level) {
5375 cmp = btrfs_comp_cpu_keys(&left_key, &right_key);
5377 advance_left = ADVANCE;
5378 } else if (cmp > 0) {
5379 advance_right = ADVANCE;
5381 left_blockptr = btrfs_node_blockptr(
5382 left_path->nodes[left_level],
5383 left_path->slots[left_level]);
5384 right_blockptr = btrfs_node_blockptr(
5385 right_path->nodes[right_level],
5386 right_path->slots[right_level]);
5387 if (left_blockptr == right_blockptr) {
5389 * As we're on a shared block, don't
5390 * allow to go deeper.
5392 advance_left = ADVANCE_ONLY_NEXT;
5393 advance_right = ADVANCE_ONLY_NEXT;
5395 advance_left = ADVANCE;
5396 advance_right = ADVANCE;
5399 } else if (left_level < right_level) {
5400 advance_right = ADVANCE;
5402 advance_left = ADVANCE;
5407 btrfs_free_path(left_path);
5408 btrfs_free_path(right_path);
5413 ret = btrfs_end_transaction(trans, left_root);
5415 btrfs_end_transaction(trans, left_root);
5422 * this is similar to btrfs_next_leaf, but does not try to preserve
5423 * and fixup the path. It looks for and returns the next key in the
5424 * tree based on the current path and the min_trans parameters.
5426 * 0 is returned if another key is found, < 0 if there are any errors
5427 * and 1 is returned if there are no higher keys in the tree
5429 * path->keep_locks should be set to 1 on the search made before
5430 * calling this function.
5432 int btrfs_find_next_key(struct btrfs_root *root, struct btrfs_path *path,
5433 struct btrfs_key *key, int level, u64 min_trans)
5436 struct extent_buffer *c;
5438 WARN_ON(!path->keep_locks);
5439 while (level < BTRFS_MAX_LEVEL) {
5440 if (!path->nodes[level])
5443 slot = path->slots[level] + 1;
5444 c = path->nodes[level];
5446 if (slot >= btrfs_header_nritems(c)) {
5449 struct btrfs_key cur_key;
5450 if (level + 1 >= BTRFS_MAX_LEVEL ||
5451 !path->nodes[level + 1])
5454 if (path->locks[level + 1]) {
5459 slot = btrfs_header_nritems(c) - 1;
5461 btrfs_item_key_to_cpu(c, &cur_key, slot);
5463 btrfs_node_key_to_cpu(c, &cur_key, slot);
5465 orig_lowest = path->lowest_level;
5466 btrfs_release_path(path);
5467 path->lowest_level = level;
5468 ret = btrfs_search_slot(NULL, root, &cur_key, path,
5470 path->lowest_level = orig_lowest;
5474 c = path->nodes[level];
5475 slot = path->slots[level];
5482 btrfs_item_key_to_cpu(c, key, slot);
5484 u64 gen = btrfs_node_ptr_generation(c, slot);
5486 if (gen < min_trans) {
5490 btrfs_node_key_to_cpu(c, key, slot);
5498 * search the tree again to find a leaf with greater keys
5499 * returns 0 if it found something or 1 if there are no greater leaves.
5500 * returns < 0 on io errors.
5502 int btrfs_next_leaf(struct btrfs_root *root, struct btrfs_path *path)
5504 return btrfs_next_old_leaf(root, path, 0);
5507 int btrfs_next_old_leaf(struct btrfs_root *root, struct btrfs_path *path,
5512 struct extent_buffer *c;
5513 struct extent_buffer *next;
5514 struct btrfs_key key;
5517 int old_spinning = path->leave_spinning;
5518 int next_rw_lock = 0;
5520 nritems = btrfs_header_nritems(path->nodes[0]);
5524 btrfs_item_key_to_cpu(path->nodes[0], &key, nritems - 1);
5529 btrfs_release_path(path);
5531 path->keep_locks = 1;
5532 path->leave_spinning = 1;
5535 ret = btrfs_search_old_slot(root, &key, path, time_seq);
5537 ret = btrfs_search_slot(NULL, root, &key, path, 0, 0);
5538 path->keep_locks = 0;
5543 nritems = btrfs_header_nritems(path->nodes[0]);
5545 * by releasing the path above we dropped all our locks. A balance
5546 * could have added more items next to the key that used to be
5547 * at the very end of the block. So, check again here and
5548 * advance the path if there are now more items available.
5550 if (nritems > 0 && path->slots[0] < nritems - 1) {
5557 while (level < BTRFS_MAX_LEVEL) {
5558 if (!path->nodes[level]) {
5563 slot = path->slots[level] + 1;
5564 c = path->nodes[level];
5565 if (slot >= btrfs_header_nritems(c)) {
5567 if (level == BTRFS_MAX_LEVEL) {
5575 btrfs_tree_unlock_rw(next, next_rw_lock);
5576 free_extent_buffer(next);
5580 next_rw_lock = path->locks[level];
5581 ret = read_block_for_search(NULL, root, path, &next, level,
5587 btrfs_release_path(path);
5591 if (!path->skip_locking) {
5592 ret = btrfs_try_tree_read_lock(next);
5593 if (!ret && time_seq) {
5595 * If we don't get the lock, we may be racing
5596 * with push_leaf_left, holding that lock while
5597 * itself waiting for the leaf we've currently
5598 * locked. To solve this situation, we give up
5599 * on our lock and cycle.
5601 free_extent_buffer(next);
5602 btrfs_release_path(path);
5607 btrfs_set_path_blocking(path);
5608 btrfs_tree_read_lock(next);
5609 btrfs_clear_path_blocking(path, next,
5612 next_rw_lock = BTRFS_READ_LOCK;
5616 path->slots[level] = slot;
5619 c = path->nodes[level];
5620 if (path->locks[level])
5621 btrfs_tree_unlock_rw(c, path->locks[level]);
5623 free_extent_buffer(c);
5624 path->nodes[level] = next;
5625 path->slots[level] = 0;
5626 if (!path->skip_locking)
5627 path->locks[level] = next_rw_lock;
5631 ret = read_block_for_search(NULL, root, path, &next, level,
5637 btrfs_release_path(path);
5641 if (!path->skip_locking) {
5642 ret = btrfs_try_tree_read_lock(next);
5644 btrfs_set_path_blocking(path);
5645 btrfs_tree_read_lock(next);
5646 btrfs_clear_path_blocking(path, next,
5649 next_rw_lock = BTRFS_READ_LOCK;
5654 unlock_up(path, 0, 1, 0, NULL);
5655 path->leave_spinning = old_spinning;
5657 btrfs_set_path_blocking(path);
5663 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5664 * searching until it gets past min_objectid or finds an item of 'type'
5666 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5668 int btrfs_previous_item(struct btrfs_root *root,
5669 struct btrfs_path *path, u64 min_objectid,
5672 struct btrfs_key found_key;
5673 struct extent_buffer *leaf;
5678 if (path->slots[0] == 0) {
5679 btrfs_set_path_blocking(path);
5680 ret = btrfs_prev_leaf(root, path);
5686 leaf = path->nodes[0];
5687 nritems = btrfs_header_nritems(leaf);
5690 if (path->slots[0] == nritems)
5693 btrfs_item_key_to_cpu(leaf, &found_key, path->slots[0]);
5694 if (found_key.objectid < min_objectid)
5696 if (found_key.type == type)
5698 if (found_key.objectid == min_objectid &&
5699 found_key.type < type)