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
, btrfs_header_fsid(),
280 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
281 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
282 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
284 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
289 btrfs_mark_buffer_dirty(cow
);
298 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
299 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
301 MOD_LOG_ROOT_REPLACE
,
304 struct tree_mod_move
{
309 struct tree_mod_root
{
314 struct tree_mod_elem
{
316 u64 index
; /* shifted logical */
320 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
323 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
326 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
327 struct btrfs_disk_key key
;
330 /* this is used for op == MOD_LOG_MOVE_KEYS */
331 struct tree_mod_move move
;
333 /* this is used for op == MOD_LOG_ROOT_REPLACE */
334 struct tree_mod_root old_root
;
337 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
339 read_lock(&fs_info
->tree_mod_log_lock
);
342 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
344 read_unlock(&fs_info
->tree_mod_log_lock
);
347 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
349 write_lock(&fs_info
->tree_mod_log_lock
);
352 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
354 write_unlock(&fs_info
->tree_mod_log_lock
);
358 * Increment the upper half of tree_mod_seq, set lower half zero.
360 * Must be called with fs_info->tree_mod_seq_lock held.
362 static inline u64
btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info
*fs_info
)
364 u64 seq
= atomic64_read(&fs_info
->tree_mod_seq
);
365 seq
&= 0xffffffff00000000ull
;
367 atomic64_set(&fs_info
->tree_mod_seq
, seq
);
372 * Increment the lower half of tree_mod_seq.
374 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
375 * are generated should not technically require a spin lock here. (Rationale:
376 * incrementing the minor while incrementing the major seq number is between its
377 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
378 * just returns a unique sequence number as usual.) We have decided to leave
379 * that requirement in here and rethink it once we notice it really imposes a
380 * problem on some workload.
382 static inline u64
btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info
*fs_info
)
384 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
388 * return the last minor in the previous major tree_mod_seq number
390 u64
btrfs_tree_mod_seq_prev(u64 seq
)
392 return (seq
& 0xffffffff00000000ull
) - 1ull;
396 * This adds a new blocker to the tree mod log's blocker list if the @elem
397 * passed does not already have a sequence number set. So when a caller expects
398 * to record tree modifications, it should ensure to set elem->seq to zero
399 * before calling btrfs_get_tree_mod_seq.
400 * Returns a fresh, unused tree log modification sequence number, even if no new
403 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
404 struct seq_list
*elem
)
408 tree_mod_log_write_lock(fs_info
);
409 spin_lock(&fs_info
->tree_mod_seq_lock
);
411 elem
->seq
= btrfs_inc_tree_mod_seq_major(fs_info
);
412 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
414 seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
415 spin_unlock(&fs_info
->tree_mod_seq_lock
);
416 tree_mod_log_write_unlock(fs_info
);
421 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
422 struct seq_list
*elem
)
424 struct rb_root
*tm_root
;
425 struct rb_node
*node
;
426 struct rb_node
*next
;
427 struct seq_list
*cur_elem
;
428 struct tree_mod_elem
*tm
;
429 u64 min_seq
= (u64
)-1;
430 u64 seq_putting
= elem
->seq
;
435 spin_lock(&fs_info
->tree_mod_seq_lock
);
436 list_del(&elem
->list
);
439 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
440 if (cur_elem
->seq
< min_seq
) {
441 if (seq_putting
> cur_elem
->seq
) {
443 * blocker with lower sequence number exists, we
444 * cannot remove anything from the log
446 spin_unlock(&fs_info
->tree_mod_seq_lock
);
449 min_seq
= cur_elem
->seq
;
452 spin_unlock(&fs_info
->tree_mod_seq_lock
);
455 * anything that's lower than the lowest existing (read: blocked)
456 * sequence number can be removed from the tree.
458 tree_mod_log_write_lock(fs_info
);
459 tm_root
= &fs_info
->tree_mod_log
;
460 for (node
= rb_first(tm_root
); node
; node
= next
) {
461 next
= rb_next(node
);
462 tm
= container_of(node
, struct tree_mod_elem
, node
);
463 if (tm
->seq
> min_seq
)
465 rb_erase(node
, tm_root
);
468 tree_mod_log_write_unlock(fs_info
);
472 * key order of the log:
475 * the index is the shifted logical of the *new* root node for root replace
476 * operations, or the shifted logical of the affected block for all other
480 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
482 struct rb_root
*tm_root
;
483 struct rb_node
**new;
484 struct rb_node
*parent
= NULL
;
485 struct tree_mod_elem
*cur
;
490 tree_mod_log_write_lock(fs_info
);
491 if (list_empty(&fs_info
->tree_mod_seq_list
)) {
492 tree_mod_log_write_unlock(fs_info
);
494 * Ok we no longer care about logging modifications, free up tm
495 * and return 0. Any callers shouldn't be using tm after
496 * calling tree_mod_log_insert, but if they do we can just
497 * change this to return a special error code to let the callers
498 * do their own thing.
504 spin_lock(&fs_info
->tree_mod_seq_lock
);
505 tm
->seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
506 spin_unlock(&fs_info
->tree_mod_seq_lock
);
508 tm_root
= &fs_info
->tree_mod_log
;
509 new = &tm_root
->rb_node
;
511 cur
= container_of(*new, struct tree_mod_elem
, node
);
513 if (cur
->index
< tm
->index
)
514 new = &((*new)->rb_left
);
515 else if (cur
->index
> tm
->index
)
516 new = &((*new)->rb_right
);
517 else if (cur
->seq
< tm
->seq
)
518 new = &((*new)->rb_left
);
519 else if (cur
->seq
> tm
->seq
)
520 new = &((*new)->rb_right
);
528 rb_link_node(&tm
->node
, parent
, new);
529 rb_insert_color(&tm
->node
, tm_root
);
531 tree_mod_log_write_unlock(fs_info
);
536 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
537 * returns zero with the tree_mod_log_lock acquired. The caller must hold
538 * this until all tree mod log insertions are recorded in the rb tree and then
539 * call tree_mod_log_write_unlock() to release.
541 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
542 struct extent_buffer
*eb
) {
544 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
546 if (eb
&& btrfs_header_level(eb
) == 0)
552 __tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
553 struct extent_buffer
*eb
, int slot
,
554 enum mod_log_op op
, gfp_t flags
)
556 struct tree_mod_elem
*tm
;
558 tm
= kzalloc(sizeof(*tm
), flags
);
562 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
563 if (op
!= MOD_LOG_KEY_ADD
) {
564 btrfs_node_key(eb
, &tm
->key
, slot
);
565 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
569 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
571 return __tree_mod_log_insert(fs_info
, tm
);
575 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
576 struct extent_buffer
*eb
, int slot
,
577 enum mod_log_op op
, gfp_t flags
)
579 if (tree_mod_dont_log(fs_info
, eb
))
582 return __tree_mod_log_insert_key(fs_info
, eb
, slot
, op
, flags
);
586 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
587 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
588 int nr_items
, gfp_t flags
)
590 struct tree_mod_elem
*tm
;
594 if (tree_mod_dont_log(fs_info
, eb
))
598 * When we override something during the move, we log these removals.
599 * This can only happen when we move towards the beginning of the
600 * buffer, i.e. dst_slot < src_slot.
602 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
603 ret
= __tree_mod_log_insert_key(fs_info
, eb
, i
+ dst_slot
,
604 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, GFP_NOFS
);
608 tm
= kzalloc(sizeof(*tm
), flags
);
612 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
614 tm
->move
.dst_slot
= dst_slot
;
615 tm
->move
.nr_items
= nr_items
;
616 tm
->op
= MOD_LOG_MOVE_KEYS
;
618 return __tree_mod_log_insert(fs_info
, tm
);
622 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
628 if (btrfs_header_level(eb
) == 0)
631 nritems
= btrfs_header_nritems(eb
);
632 for (i
= nritems
- 1; i
>= 0; i
--) {
633 ret
= __tree_mod_log_insert_key(fs_info
, eb
, i
,
634 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
640 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
641 struct extent_buffer
*old_root
,
642 struct extent_buffer
*new_root
, gfp_t flags
,
645 struct tree_mod_elem
*tm
;
647 if (tree_mod_dont_log(fs_info
, NULL
))
651 __tree_mod_log_free_eb(fs_info
, old_root
);
653 tm
= kzalloc(sizeof(*tm
), flags
);
657 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
658 tm
->old_root
.logical
= old_root
->start
;
659 tm
->old_root
.level
= btrfs_header_level(old_root
);
660 tm
->generation
= btrfs_header_generation(old_root
);
661 tm
->op
= MOD_LOG_ROOT_REPLACE
;
663 return __tree_mod_log_insert(fs_info
, tm
);
666 static struct tree_mod_elem
*
667 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
670 struct rb_root
*tm_root
;
671 struct rb_node
*node
;
672 struct tree_mod_elem
*cur
= NULL
;
673 struct tree_mod_elem
*found
= NULL
;
674 u64 index
= start
>> PAGE_CACHE_SHIFT
;
676 tree_mod_log_read_lock(fs_info
);
677 tm_root
= &fs_info
->tree_mod_log
;
678 node
= tm_root
->rb_node
;
680 cur
= container_of(node
, struct tree_mod_elem
, node
);
681 if (cur
->index
< index
) {
682 node
= node
->rb_left
;
683 } else if (cur
->index
> index
) {
684 node
= node
->rb_right
;
685 } else if (cur
->seq
< min_seq
) {
686 node
= node
->rb_left
;
687 } else if (!smallest
) {
688 /* we want the node with the highest seq */
690 BUG_ON(found
->seq
> cur
->seq
);
692 node
= node
->rb_left
;
693 } else if (cur
->seq
> min_seq
) {
694 /* we want the node with the smallest seq */
696 BUG_ON(found
->seq
< cur
->seq
);
698 node
= node
->rb_right
;
704 tree_mod_log_read_unlock(fs_info
);
710 * this returns the element from the log with the smallest time sequence
711 * value that's in the log (the oldest log item). any element with a time
712 * sequence lower than min_seq will be ignored.
714 static struct tree_mod_elem
*
715 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
718 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
722 * this returns the element from the log with the largest time sequence
723 * value that's in the log (the most recent log item). any element with
724 * a time sequence lower than min_seq will be ignored.
726 static struct tree_mod_elem
*
727 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
729 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
733 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
734 struct extent_buffer
*src
, unsigned long dst_offset
,
735 unsigned long src_offset
, int nr_items
)
740 if (tree_mod_dont_log(fs_info
, NULL
))
743 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
746 for (i
= 0; i
< nr_items
; i
++) {
747 ret
= __tree_mod_log_insert_key(fs_info
, src
,
749 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
751 ret
= __tree_mod_log_insert_key(fs_info
, dst
,
760 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
761 int dst_offset
, int src_offset
, int nr_items
)
764 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
770 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
771 struct extent_buffer
*eb
, int slot
, int atomic
)
775 ret
= __tree_mod_log_insert_key(fs_info
, eb
, slot
,
777 atomic
? GFP_ATOMIC
: GFP_NOFS
);
782 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
784 if (tree_mod_dont_log(fs_info
, eb
))
786 __tree_mod_log_free_eb(fs_info
, eb
);
790 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
791 struct extent_buffer
*new_root_node
,
795 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
796 new_root_node
, GFP_NOFS
, log_removal
);
801 * check if the tree block can be shared by multiple trees
803 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
804 struct extent_buffer
*buf
)
807 * Tree blocks not in refernece counted trees and tree roots
808 * are never shared. If a block was allocated after the last
809 * snapshot and the block was not allocated by tree relocation,
810 * we know the block is not shared.
812 if (root
->ref_cows
&&
813 buf
!= root
->node
&& buf
!= root
->commit_root
&&
814 (btrfs_header_generation(buf
) <=
815 btrfs_root_last_snapshot(&root
->root_item
) ||
816 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
818 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
819 if (root
->ref_cows
&&
820 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
826 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
827 struct btrfs_root
*root
,
828 struct extent_buffer
*buf
,
829 struct extent_buffer
*cow
,
839 * Backrefs update rules:
841 * Always use full backrefs for extent pointers in tree block
842 * allocated by tree relocation.
844 * If a shared tree block is no longer referenced by its owner
845 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
846 * use full backrefs for extent pointers in tree block.
848 * If a tree block is been relocating
849 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
850 * use full backrefs for extent pointers in tree block.
851 * The reason for this is some operations (such as drop tree)
852 * are only allowed for blocks use full backrefs.
855 if (btrfs_block_can_be_shared(root
, buf
)) {
856 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
857 btrfs_header_level(buf
), 1,
863 btrfs_std_error(root
->fs_info
, ret
);
868 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
869 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
870 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
875 owner
= btrfs_header_owner(buf
);
876 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
877 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
880 if ((owner
== root
->root_key
.objectid
||
881 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
882 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
883 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
884 BUG_ON(ret
); /* -ENOMEM */
886 if (root
->root_key
.objectid
==
887 BTRFS_TREE_RELOC_OBJECTID
) {
888 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
889 BUG_ON(ret
); /* -ENOMEM */
890 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
891 BUG_ON(ret
); /* -ENOMEM */
893 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
896 if (root
->root_key
.objectid
==
897 BTRFS_TREE_RELOC_OBJECTID
)
898 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
900 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
901 BUG_ON(ret
); /* -ENOMEM */
903 if (new_flags
!= 0) {
904 int level
= btrfs_header_level(buf
);
906 ret
= btrfs_set_disk_extent_flags(trans
, root
,
909 new_flags
, level
, 0);
914 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
915 if (root
->root_key
.objectid
==
916 BTRFS_TREE_RELOC_OBJECTID
)
917 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
919 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
920 BUG_ON(ret
); /* -ENOMEM */
921 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
922 BUG_ON(ret
); /* -ENOMEM */
924 clean_tree_block(trans
, root
, buf
);
931 * does the dirty work in cow of a single block. The parent block (if
932 * supplied) is updated to point to the new cow copy. The new buffer is marked
933 * dirty and returned locked. If you modify the block it needs to be marked
936 * search_start -- an allocation hint for the new block
938 * empty_size -- a hint that you plan on doing more cow. This is the size in
939 * bytes the allocator should try to find free next to the block it returns.
940 * This is just a hint and may be ignored by the allocator.
942 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
943 struct btrfs_root
*root
,
944 struct extent_buffer
*buf
,
945 struct extent_buffer
*parent
, int parent_slot
,
946 struct extent_buffer
**cow_ret
,
947 u64 search_start
, u64 empty_size
)
949 struct btrfs_disk_key disk_key
;
950 struct extent_buffer
*cow
;
959 btrfs_assert_tree_locked(buf
);
961 WARN_ON(root
->ref_cows
&& trans
->transid
!=
962 root
->fs_info
->running_transaction
->transid
);
963 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
965 level
= btrfs_header_level(buf
);
968 btrfs_item_key(buf
, &disk_key
, 0);
970 btrfs_node_key(buf
, &disk_key
, 0);
972 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
974 parent_start
= parent
->start
;
980 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
981 root
->root_key
.objectid
, &disk_key
,
982 level
, search_start
, empty_size
);
986 /* cow is set to blocking by btrfs_init_new_buffer */
988 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
989 btrfs_set_header_bytenr(cow
, cow
->start
);
990 btrfs_set_header_generation(cow
, trans
->transid
);
991 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
992 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
993 BTRFS_HEADER_FLAG_RELOC
);
994 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
995 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
997 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
999 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
1002 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1004 btrfs_abort_transaction(trans
, root
, ret
);
1008 if (root
->ref_cows
) {
1009 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1014 if (buf
== root
->node
) {
1015 WARN_ON(parent
&& parent
!= buf
);
1016 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1017 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1018 parent_start
= buf
->start
;
1022 extent_buffer_get(cow
);
1023 tree_mod_log_set_root_pointer(root
, cow
, 1);
1024 rcu_assign_pointer(root
->node
, cow
);
1026 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1028 free_extent_buffer(buf
);
1029 add_root_to_dirty_list(root
);
1031 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1032 parent_start
= parent
->start
;
1036 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1037 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1038 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1039 btrfs_set_node_blockptr(parent
, parent_slot
,
1041 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1043 btrfs_mark_buffer_dirty(parent
);
1045 tree_mod_log_free_eb(root
->fs_info
, buf
);
1046 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1050 btrfs_tree_unlock(buf
);
1051 free_extent_buffer_stale(buf
);
1052 btrfs_mark_buffer_dirty(cow
);
1058 * returns the logical address of the oldest predecessor of the given root.
1059 * entries older than time_seq are ignored.
1061 static struct tree_mod_elem
*
1062 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1063 struct extent_buffer
*eb_root
, u64 time_seq
)
1065 struct tree_mod_elem
*tm
;
1066 struct tree_mod_elem
*found
= NULL
;
1067 u64 root_logical
= eb_root
->start
;
1074 * the very last operation that's logged for a root is the replacement
1075 * operation (if it is replaced at all). this has the index of the *new*
1076 * root, making it the very first operation that's logged for this root.
1079 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1084 * if there are no tree operation for the oldest root, we simply
1085 * return it. this should only happen if that (old) root is at
1092 * if there's an operation that's not a root replacement, we
1093 * found the oldest version of our root. normally, we'll find a
1094 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1096 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1100 root_logical
= tm
->old_root
.logical
;
1104 /* if there's no old root to return, return what we found instead */
1112 * tm is a pointer to the first operation to rewind within eb. then, all
1113 * previous operations will be rewinded (until we reach something older than
1117 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1118 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1121 struct rb_node
*next
;
1122 struct tree_mod_elem
*tm
= first_tm
;
1123 unsigned long o_dst
;
1124 unsigned long o_src
;
1125 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1127 n
= btrfs_header_nritems(eb
);
1128 tree_mod_log_read_lock(fs_info
);
1129 while (tm
&& tm
->seq
>= time_seq
) {
1131 * all the operations are recorded with the operator used for
1132 * the modification. as we're going backwards, we do the
1133 * opposite of each operation here.
1136 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1137 BUG_ON(tm
->slot
< n
);
1139 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1140 case MOD_LOG_KEY_REMOVE
:
1141 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1142 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1143 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1147 case MOD_LOG_KEY_REPLACE
:
1148 BUG_ON(tm
->slot
>= n
);
1149 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1150 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1151 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1154 case MOD_LOG_KEY_ADD
:
1155 /* if a move operation is needed it's in the log */
1158 case MOD_LOG_MOVE_KEYS
:
1159 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1160 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1161 memmove_extent_buffer(eb
, o_dst
, o_src
,
1162 tm
->move
.nr_items
* p_size
);
1164 case MOD_LOG_ROOT_REPLACE
:
1166 * this operation is special. for roots, this must be
1167 * handled explicitly before rewinding.
1168 * for non-roots, this operation may exist if the node
1169 * was a root: root A -> child B; then A gets empty and
1170 * B is promoted to the new root. in the mod log, we'll
1171 * have a root-replace operation for B, a tree block
1172 * that is no root. we simply ignore that operation.
1176 next
= rb_next(&tm
->node
);
1179 tm
= container_of(next
, struct tree_mod_elem
, node
);
1180 if (tm
->index
!= first_tm
->index
)
1183 tree_mod_log_read_unlock(fs_info
);
1184 btrfs_set_header_nritems(eb
, n
);
1188 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1189 * is returned. If rewind operations happen, a fresh buffer is returned. The
1190 * returned buffer is always read-locked. If the returned buffer is not the
1191 * input buffer, the lock on the input buffer is released and the input buffer
1192 * is freed (its refcount is decremented).
1194 static struct extent_buffer
*
1195 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1196 struct extent_buffer
*eb
, u64 time_seq
)
1198 struct extent_buffer
*eb_rewin
;
1199 struct tree_mod_elem
*tm
;
1204 if (btrfs_header_level(eb
) == 0)
1207 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1211 btrfs_set_path_blocking(path
);
1212 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1214 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1215 BUG_ON(tm
->slot
!= 0);
1216 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1217 fs_info
->tree_root
->nodesize
);
1219 btrfs_tree_read_unlock_blocking(eb
);
1220 free_extent_buffer(eb
);
1223 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1224 btrfs_set_header_backref_rev(eb_rewin
,
1225 btrfs_header_backref_rev(eb
));
1226 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1227 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1229 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1231 btrfs_tree_read_unlock_blocking(eb
);
1232 free_extent_buffer(eb
);
1237 btrfs_clear_path_blocking(path
, NULL
, BTRFS_READ_LOCK
);
1238 btrfs_tree_read_unlock_blocking(eb
);
1239 free_extent_buffer(eb
);
1241 extent_buffer_get(eb_rewin
);
1242 btrfs_tree_read_lock(eb_rewin
);
1243 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1244 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1245 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1251 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1252 * value. If there are no changes, the current root->root_node is returned. If
1253 * anything changed in between, there's a fresh buffer allocated on which the
1254 * rewind operations are done. In any case, the returned buffer is read locked.
1255 * Returns NULL on error (with no locks held).
1257 static inline struct extent_buffer
*
1258 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1260 struct tree_mod_elem
*tm
;
1261 struct extent_buffer
*eb
= NULL
;
1262 struct extent_buffer
*eb_root
;
1263 struct extent_buffer
*old
;
1264 struct tree_mod_root
*old_root
= NULL
;
1265 u64 old_generation
= 0;
1269 eb_root
= btrfs_read_lock_root_node(root
);
1270 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1274 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1275 old_root
= &tm
->old_root
;
1276 old_generation
= tm
->generation
;
1277 logical
= old_root
->logical
;
1279 logical
= eb_root
->start
;
1282 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1283 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1284 btrfs_tree_read_unlock(eb_root
);
1285 free_extent_buffer(eb_root
);
1286 blocksize
= btrfs_level_size(root
, old_root
->level
);
1287 old
= read_tree_block(root
, logical
, blocksize
, 0);
1288 if (WARN_ON(!old
|| !extent_buffer_uptodate(old
))) {
1289 free_extent_buffer(old
);
1290 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1293 eb
= btrfs_clone_extent_buffer(old
);
1294 free_extent_buffer(old
);
1296 } else if (old_root
) {
1297 btrfs_tree_read_unlock(eb_root
);
1298 free_extent_buffer(eb_root
);
1299 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1301 btrfs_set_lock_blocking_rw(eb_root
, BTRFS_READ_LOCK
);
1302 eb
= btrfs_clone_extent_buffer(eb_root
);
1303 btrfs_tree_read_unlock_blocking(eb_root
);
1304 free_extent_buffer(eb_root
);
1309 extent_buffer_get(eb
);
1310 btrfs_tree_read_lock(eb
);
1312 btrfs_set_header_bytenr(eb
, eb
->start
);
1313 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1314 btrfs_set_header_owner(eb
, btrfs_header_owner(eb_root
));
1315 btrfs_set_header_level(eb
, old_root
->level
);
1316 btrfs_set_header_generation(eb
, old_generation
);
1319 __tree_mod_log_rewind(root
->fs_info
, eb
, time_seq
, tm
);
1321 WARN_ON(btrfs_header_level(eb
) != 0);
1322 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1327 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1329 struct tree_mod_elem
*tm
;
1331 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1333 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1334 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1335 level
= tm
->old_root
.level
;
1337 level
= btrfs_header_level(eb_root
);
1339 free_extent_buffer(eb_root
);
1344 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1345 struct btrfs_root
*root
,
1346 struct extent_buffer
*buf
)
1348 /* ensure we can see the force_cow */
1352 * We do not need to cow a block if
1353 * 1) this block is not created or changed in this transaction;
1354 * 2) this block does not belong to TREE_RELOC tree;
1355 * 3) the root is not forced COW.
1357 * What is forced COW:
1358 * when we create snapshot during commiting the transaction,
1359 * after we've finished coping src root, we must COW the shared
1360 * block to ensure the metadata consistency.
1362 if (btrfs_header_generation(buf
) == trans
->transid
&&
1363 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1364 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1365 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1372 * cows a single block, see __btrfs_cow_block for the real work.
1373 * This version of it has extra checks so that a block isn't cow'd more than
1374 * once per transaction, as long as it hasn't been written yet
1376 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1377 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1378 struct extent_buffer
*parent
, int parent_slot
,
1379 struct extent_buffer
**cow_ret
)
1384 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1385 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1387 root
->fs_info
->running_transaction
->transid
);
1389 if (trans
->transid
!= root
->fs_info
->generation
)
1390 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1391 trans
->transid
, root
->fs_info
->generation
);
1393 if (!should_cow_block(trans
, root
, buf
)) {
1398 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1401 btrfs_set_lock_blocking(parent
);
1402 btrfs_set_lock_blocking(buf
);
1404 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1405 parent_slot
, cow_ret
, search_start
, 0);
1407 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1413 * helper function for defrag to decide if two blocks pointed to by a
1414 * node are actually close by
1416 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1418 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1420 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1426 * compare two keys in a memcmp fashion
1428 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1430 struct btrfs_key k1
;
1432 btrfs_disk_key_to_cpu(&k1
, disk
);
1434 return btrfs_comp_cpu_keys(&k1
, k2
);
1438 * same as comp_keys only with two btrfs_key's
1440 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1442 if (k1
->objectid
> k2
->objectid
)
1444 if (k1
->objectid
< k2
->objectid
)
1446 if (k1
->type
> k2
->type
)
1448 if (k1
->type
< k2
->type
)
1450 if (k1
->offset
> k2
->offset
)
1452 if (k1
->offset
< k2
->offset
)
1458 * this is used by the defrag code to go through all the
1459 * leaves pointed to by a node and reallocate them so that
1460 * disk order is close to key order
1462 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1463 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1464 int start_slot
, u64
*last_ret
,
1465 struct btrfs_key
*progress
)
1467 struct extent_buffer
*cur
;
1470 u64 search_start
= *last_ret
;
1480 int progress_passed
= 0;
1481 struct btrfs_disk_key disk_key
;
1483 parent_level
= btrfs_header_level(parent
);
1485 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1486 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1488 parent_nritems
= btrfs_header_nritems(parent
);
1489 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1490 end_slot
= parent_nritems
;
1492 if (parent_nritems
== 1)
1495 btrfs_set_lock_blocking(parent
);
1497 for (i
= start_slot
; i
< end_slot
; i
++) {
1500 btrfs_node_key(parent
, &disk_key
, i
);
1501 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1504 progress_passed
= 1;
1505 blocknr
= btrfs_node_blockptr(parent
, i
);
1506 gen
= btrfs_node_ptr_generation(parent
, i
);
1507 if (last_block
== 0)
1508 last_block
= blocknr
;
1511 other
= btrfs_node_blockptr(parent
, i
- 1);
1512 close
= close_blocks(blocknr
, other
, blocksize
);
1514 if (!close
&& i
< end_slot
- 2) {
1515 other
= btrfs_node_blockptr(parent
, i
+ 1);
1516 close
= close_blocks(blocknr
, other
, blocksize
);
1519 last_block
= blocknr
;
1523 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1525 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1528 if (!cur
|| !uptodate
) {
1530 cur
= read_tree_block(root
, blocknr
,
1532 if (!cur
|| !extent_buffer_uptodate(cur
)) {
1533 free_extent_buffer(cur
);
1536 } else if (!uptodate
) {
1537 err
= btrfs_read_buffer(cur
, gen
);
1539 free_extent_buffer(cur
);
1544 if (search_start
== 0)
1545 search_start
= last_block
;
1547 btrfs_tree_lock(cur
);
1548 btrfs_set_lock_blocking(cur
);
1549 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1552 (end_slot
- i
) * blocksize
));
1554 btrfs_tree_unlock(cur
);
1555 free_extent_buffer(cur
);
1558 search_start
= cur
->start
;
1559 last_block
= cur
->start
;
1560 *last_ret
= search_start
;
1561 btrfs_tree_unlock(cur
);
1562 free_extent_buffer(cur
);
1568 * The leaf data grows from end-to-front in the node.
1569 * this returns the address of the start of the last item,
1570 * which is the stop of the leaf data stack
1572 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1573 struct extent_buffer
*leaf
)
1575 u32 nr
= btrfs_header_nritems(leaf
);
1577 return BTRFS_LEAF_DATA_SIZE(root
);
1578 return btrfs_item_offset_nr(leaf
, nr
- 1);
1583 * search for key in the extent_buffer. The items start at offset p,
1584 * and they are item_size apart. There are 'max' items in p.
1586 * the slot in the array is returned via slot, and it points to
1587 * the place where you would insert key if it is not found in
1590 * slot may point to max if the key is bigger than all of the keys
1592 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1594 int item_size
, struct btrfs_key
*key
,
1601 struct btrfs_disk_key
*tmp
= NULL
;
1602 struct btrfs_disk_key unaligned
;
1603 unsigned long offset
;
1605 unsigned long map_start
= 0;
1606 unsigned long map_len
= 0;
1609 while (low
< high
) {
1610 mid
= (low
+ high
) / 2;
1611 offset
= p
+ mid
* item_size
;
1613 if (!kaddr
|| offset
< map_start
||
1614 (offset
+ sizeof(struct btrfs_disk_key
)) >
1615 map_start
+ map_len
) {
1617 err
= map_private_extent_buffer(eb
, offset
,
1618 sizeof(struct btrfs_disk_key
),
1619 &kaddr
, &map_start
, &map_len
);
1622 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1625 read_extent_buffer(eb
, &unaligned
,
1626 offset
, sizeof(unaligned
));
1631 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1634 ret
= comp_keys(tmp
, key
);
1650 * simple bin_search frontend that does the right thing for
1653 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1654 int level
, int *slot
)
1657 return generic_bin_search(eb
,
1658 offsetof(struct btrfs_leaf
, items
),
1659 sizeof(struct btrfs_item
),
1660 key
, btrfs_header_nritems(eb
),
1663 return generic_bin_search(eb
,
1664 offsetof(struct btrfs_node
, ptrs
),
1665 sizeof(struct btrfs_key_ptr
),
1666 key
, btrfs_header_nritems(eb
),
1670 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1671 int level
, int *slot
)
1673 return bin_search(eb
, key
, level
, slot
);
1676 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1678 spin_lock(&root
->accounting_lock
);
1679 btrfs_set_root_used(&root
->root_item
,
1680 btrfs_root_used(&root
->root_item
) + size
);
1681 spin_unlock(&root
->accounting_lock
);
1684 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1686 spin_lock(&root
->accounting_lock
);
1687 btrfs_set_root_used(&root
->root_item
,
1688 btrfs_root_used(&root
->root_item
) - size
);
1689 spin_unlock(&root
->accounting_lock
);
1692 /* given a node and slot number, this reads the blocks it points to. The
1693 * extent buffer is returned with a reference taken (but unlocked).
1694 * NULL is returned on error.
1696 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1697 struct extent_buffer
*parent
, int slot
)
1699 int level
= btrfs_header_level(parent
);
1700 struct extent_buffer
*eb
;
1704 if (slot
>= btrfs_header_nritems(parent
))
1709 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1710 btrfs_level_size(root
, level
- 1),
1711 btrfs_node_ptr_generation(parent
, slot
));
1712 if (eb
&& !extent_buffer_uptodate(eb
)) {
1713 free_extent_buffer(eb
);
1721 * node level balancing, used to make sure nodes are in proper order for
1722 * item deletion. We balance from the top down, so we have to make sure
1723 * that a deletion won't leave an node completely empty later on.
1725 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1726 struct btrfs_root
*root
,
1727 struct btrfs_path
*path
, int level
)
1729 struct extent_buffer
*right
= NULL
;
1730 struct extent_buffer
*mid
;
1731 struct extent_buffer
*left
= NULL
;
1732 struct extent_buffer
*parent
= NULL
;
1736 int orig_slot
= path
->slots
[level
];
1742 mid
= path
->nodes
[level
];
1744 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1745 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1746 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1748 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1750 if (level
< BTRFS_MAX_LEVEL
- 1) {
1751 parent
= path
->nodes
[level
+ 1];
1752 pslot
= path
->slots
[level
+ 1];
1756 * deal with the case where there is only one pointer in the root
1757 * by promoting the node below to a root
1760 struct extent_buffer
*child
;
1762 if (btrfs_header_nritems(mid
) != 1)
1765 /* promote the child to a root */
1766 child
= read_node_slot(root
, mid
, 0);
1769 btrfs_std_error(root
->fs_info
, ret
);
1773 btrfs_tree_lock(child
);
1774 btrfs_set_lock_blocking(child
);
1775 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1777 btrfs_tree_unlock(child
);
1778 free_extent_buffer(child
);
1782 tree_mod_log_set_root_pointer(root
, child
, 1);
1783 rcu_assign_pointer(root
->node
, child
);
1785 add_root_to_dirty_list(root
);
1786 btrfs_tree_unlock(child
);
1788 path
->locks
[level
] = 0;
1789 path
->nodes
[level
] = NULL
;
1790 clean_tree_block(trans
, root
, mid
);
1791 btrfs_tree_unlock(mid
);
1792 /* once for the path */
1793 free_extent_buffer(mid
);
1795 root_sub_used(root
, mid
->len
);
1796 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1797 /* once for the root ptr */
1798 free_extent_buffer_stale(mid
);
1801 if (btrfs_header_nritems(mid
) >
1802 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1805 left
= read_node_slot(root
, parent
, pslot
- 1);
1807 btrfs_tree_lock(left
);
1808 btrfs_set_lock_blocking(left
);
1809 wret
= btrfs_cow_block(trans
, root
, left
,
1810 parent
, pslot
- 1, &left
);
1816 right
= read_node_slot(root
, parent
, pslot
+ 1);
1818 btrfs_tree_lock(right
);
1819 btrfs_set_lock_blocking(right
);
1820 wret
= btrfs_cow_block(trans
, root
, right
,
1821 parent
, pslot
+ 1, &right
);
1828 /* first, try to make some room in the middle buffer */
1830 orig_slot
+= btrfs_header_nritems(left
);
1831 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1837 * then try to empty the right most buffer into the middle
1840 wret
= push_node_left(trans
, root
, mid
, right
, 1);
1841 if (wret
< 0 && wret
!= -ENOSPC
)
1843 if (btrfs_header_nritems(right
) == 0) {
1844 clean_tree_block(trans
, root
, right
);
1845 btrfs_tree_unlock(right
);
1846 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
1847 root_sub_used(root
, right
->len
);
1848 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
1849 free_extent_buffer_stale(right
);
1852 struct btrfs_disk_key right_key
;
1853 btrfs_node_key(right
, &right_key
, 0);
1854 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1856 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1857 btrfs_mark_buffer_dirty(parent
);
1860 if (btrfs_header_nritems(mid
) == 1) {
1862 * we're not allowed to leave a node with one item in the
1863 * tree during a delete. A deletion from lower in the tree
1864 * could try to delete the only pointer in this node.
1865 * So, pull some keys from the left.
1866 * There has to be a left pointer at this point because
1867 * otherwise we would have pulled some pointers from the
1872 btrfs_std_error(root
->fs_info
, ret
);
1875 wret
= balance_node_right(trans
, root
, mid
, left
);
1881 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1887 if (btrfs_header_nritems(mid
) == 0) {
1888 clean_tree_block(trans
, root
, mid
);
1889 btrfs_tree_unlock(mid
);
1890 del_ptr(root
, path
, level
+ 1, pslot
);
1891 root_sub_used(root
, mid
->len
);
1892 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1893 free_extent_buffer_stale(mid
);
1896 /* update the parent key to reflect our changes */
1897 struct btrfs_disk_key mid_key
;
1898 btrfs_node_key(mid
, &mid_key
, 0);
1899 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1901 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1902 btrfs_mark_buffer_dirty(parent
);
1905 /* update the path */
1907 if (btrfs_header_nritems(left
) > orig_slot
) {
1908 extent_buffer_get(left
);
1909 /* left was locked after cow */
1910 path
->nodes
[level
] = left
;
1911 path
->slots
[level
+ 1] -= 1;
1912 path
->slots
[level
] = orig_slot
;
1914 btrfs_tree_unlock(mid
);
1915 free_extent_buffer(mid
);
1918 orig_slot
-= btrfs_header_nritems(left
);
1919 path
->slots
[level
] = orig_slot
;
1922 /* double check we haven't messed things up */
1924 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
1928 btrfs_tree_unlock(right
);
1929 free_extent_buffer(right
);
1932 if (path
->nodes
[level
] != left
)
1933 btrfs_tree_unlock(left
);
1934 free_extent_buffer(left
);
1939 /* Node balancing for insertion. Here we only split or push nodes around
1940 * when they are completely full. This is also done top down, so we
1941 * have to be pessimistic.
1943 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
1944 struct btrfs_root
*root
,
1945 struct btrfs_path
*path
, int level
)
1947 struct extent_buffer
*right
= NULL
;
1948 struct extent_buffer
*mid
;
1949 struct extent_buffer
*left
= NULL
;
1950 struct extent_buffer
*parent
= NULL
;
1954 int orig_slot
= path
->slots
[level
];
1959 mid
= path
->nodes
[level
];
1960 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1962 if (level
< BTRFS_MAX_LEVEL
- 1) {
1963 parent
= path
->nodes
[level
+ 1];
1964 pslot
= path
->slots
[level
+ 1];
1970 left
= read_node_slot(root
, parent
, pslot
- 1);
1972 /* first, try to make some room in the middle buffer */
1976 btrfs_tree_lock(left
);
1977 btrfs_set_lock_blocking(left
);
1979 left_nr
= btrfs_header_nritems(left
);
1980 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1983 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1988 wret
= push_node_left(trans
, root
,
1995 struct btrfs_disk_key disk_key
;
1996 orig_slot
+= left_nr
;
1997 btrfs_node_key(mid
, &disk_key
, 0);
1998 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2000 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2001 btrfs_mark_buffer_dirty(parent
);
2002 if (btrfs_header_nritems(left
) > orig_slot
) {
2003 path
->nodes
[level
] = left
;
2004 path
->slots
[level
+ 1] -= 1;
2005 path
->slots
[level
] = orig_slot
;
2006 btrfs_tree_unlock(mid
);
2007 free_extent_buffer(mid
);
2010 btrfs_header_nritems(left
);
2011 path
->slots
[level
] = orig_slot
;
2012 btrfs_tree_unlock(left
);
2013 free_extent_buffer(left
);
2017 btrfs_tree_unlock(left
);
2018 free_extent_buffer(left
);
2020 right
= read_node_slot(root
, parent
, pslot
+ 1);
2023 * then try to empty the right most buffer into the middle
2028 btrfs_tree_lock(right
);
2029 btrfs_set_lock_blocking(right
);
2031 right_nr
= btrfs_header_nritems(right
);
2032 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2035 ret
= btrfs_cow_block(trans
, root
, right
,
2041 wret
= balance_node_right(trans
, root
,
2048 struct btrfs_disk_key disk_key
;
2050 btrfs_node_key(right
, &disk_key
, 0);
2051 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2053 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2054 btrfs_mark_buffer_dirty(parent
);
2056 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2057 path
->nodes
[level
] = right
;
2058 path
->slots
[level
+ 1] += 1;
2059 path
->slots
[level
] = orig_slot
-
2060 btrfs_header_nritems(mid
);
2061 btrfs_tree_unlock(mid
);
2062 free_extent_buffer(mid
);
2064 btrfs_tree_unlock(right
);
2065 free_extent_buffer(right
);
2069 btrfs_tree_unlock(right
);
2070 free_extent_buffer(right
);
2076 * readahead one full node of leaves, finding things that are close
2077 * to the block in 'slot', and triggering ra on them.
2079 static void reada_for_search(struct btrfs_root
*root
,
2080 struct btrfs_path
*path
,
2081 int level
, int slot
, u64 objectid
)
2083 struct extent_buffer
*node
;
2084 struct btrfs_disk_key disk_key
;
2090 int direction
= path
->reada
;
2091 struct extent_buffer
*eb
;
2099 if (!path
->nodes
[level
])
2102 node
= path
->nodes
[level
];
2104 search
= btrfs_node_blockptr(node
, slot
);
2105 blocksize
= btrfs_level_size(root
, level
- 1);
2106 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
2108 free_extent_buffer(eb
);
2114 nritems
= btrfs_header_nritems(node
);
2118 if (direction
< 0) {
2122 } else if (direction
> 0) {
2127 if (path
->reada
< 0 && objectid
) {
2128 btrfs_node_key(node
, &disk_key
, nr
);
2129 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2132 search
= btrfs_node_blockptr(node
, nr
);
2133 if ((search
<= target
&& target
- search
<= 65536) ||
2134 (search
> target
&& search
- target
<= 65536)) {
2135 gen
= btrfs_node_ptr_generation(node
, nr
);
2136 readahead_tree_block(root
, search
, blocksize
, gen
);
2140 if ((nread
> 65536 || nscan
> 32))
2145 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2146 struct btrfs_path
*path
, int level
)
2150 struct extent_buffer
*parent
;
2151 struct extent_buffer
*eb
;
2157 parent
= path
->nodes
[level
+ 1];
2161 nritems
= btrfs_header_nritems(parent
);
2162 slot
= path
->slots
[level
+ 1];
2163 blocksize
= btrfs_level_size(root
, level
);
2166 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2167 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2168 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2170 * if we get -eagain from btrfs_buffer_uptodate, we
2171 * don't want to return eagain here. That will loop
2174 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2176 free_extent_buffer(eb
);
2178 if (slot
+ 1 < nritems
) {
2179 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2180 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2181 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2182 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2184 free_extent_buffer(eb
);
2188 readahead_tree_block(root
, block1
, blocksize
, 0);
2190 readahead_tree_block(root
, block2
, blocksize
, 0);
2195 * when we walk down the tree, it is usually safe to unlock the higher layers
2196 * in the tree. The exceptions are when our path goes through slot 0, because
2197 * operations on the tree might require changing key pointers higher up in the
2200 * callers might also have set path->keep_locks, which tells this code to keep
2201 * the lock if the path points to the last slot in the block. This is part of
2202 * walking through the tree, and selecting the next slot in the higher block.
2204 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2205 * if lowest_unlock is 1, level 0 won't be unlocked
2207 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2208 int lowest_unlock
, int min_write_lock_level
,
2209 int *write_lock_level
)
2212 int skip_level
= level
;
2214 struct extent_buffer
*t
;
2216 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2217 if (!path
->nodes
[i
])
2219 if (!path
->locks
[i
])
2221 if (!no_skips
&& path
->slots
[i
] == 0) {
2225 if (!no_skips
&& path
->keep_locks
) {
2228 nritems
= btrfs_header_nritems(t
);
2229 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2234 if (skip_level
< i
&& i
>= lowest_unlock
)
2238 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2239 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2241 if (write_lock_level
&&
2242 i
> min_write_lock_level
&&
2243 i
<= *write_lock_level
) {
2244 *write_lock_level
= i
- 1;
2251 * This releases any locks held in the path starting at level and
2252 * going all the way up to the root.
2254 * btrfs_search_slot will keep the lock held on higher nodes in a few
2255 * corner cases, such as COW of the block at slot zero in the node. This
2256 * ignores those rules, and it should only be called when there are no
2257 * more updates to be done higher up in the tree.
2259 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2263 if (path
->keep_locks
)
2266 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2267 if (!path
->nodes
[i
])
2269 if (!path
->locks
[i
])
2271 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2277 * helper function for btrfs_search_slot. The goal is to find a block
2278 * in cache without setting the path to blocking. If we find the block
2279 * we return zero and the path is unchanged.
2281 * If we can't find the block, we set the path blocking and do some
2282 * reada. -EAGAIN is returned and the search must be repeated.
2285 read_block_for_search(struct btrfs_trans_handle
*trans
,
2286 struct btrfs_root
*root
, struct btrfs_path
*p
,
2287 struct extent_buffer
**eb_ret
, int level
, int slot
,
2288 struct btrfs_key
*key
, u64 time_seq
)
2293 struct extent_buffer
*b
= *eb_ret
;
2294 struct extent_buffer
*tmp
;
2297 blocknr
= btrfs_node_blockptr(b
, slot
);
2298 gen
= btrfs_node_ptr_generation(b
, slot
);
2299 blocksize
= btrfs_level_size(root
, level
- 1);
2301 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2303 /* first we do an atomic uptodate check */
2304 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2309 /* the pages were up to date, but we failed
2310 * the generation number check. Do a full
2311 * read for the generation number that is correct.
2312 * We must do this without dropping locks so
2313 * we can trust our generation number
2315 btrfs_set_path_blocking(p
);
2317 /* now we're allowed to do a blocking uptodate check */
2318 ret
= btrfs_read_buffer(tmp
, gen
);
2323 free_extent_buffer(tmp
);
2324 btrfs_release_path(p
);
2329 * reduce lock contention at high levels
2330 * of the btree by dropping locks before
2331 * we read. Don't release the lock on the current
2332 * level because we need to walk this node to figure
2333 * out which blocks to read.
2335 btrfs_unlock_up_safe(p
, level
+ 1);
2336 btrfs_set_path_blocking(p
);
2338 free_extent_buffer(tmp
);
2340 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2342 btrfs_release_path(p
);
2345 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2348 * If the read above didn't mark this buffer up to date,
2349 * it will never end up being up to date. Set ret to EIO now
2350 * and give up so that our caller doesn't loop forever
2353 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2355 free_extent_buffer(tmp
);
2361 * helper function for btrfs_search_slot. This does all of the checks
2362 * for node-level blocks and does any balancing required based on
2365 * If no extra work was required, zero is returned. If we had to
2366 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2370 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2371 struct btrfs_root
*root
, struct btrfs_path
*p
,
2372 struct extent_buffer
*b
, int level
, int ins_len
,
2373 int *write_lock_level
)
2376 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2377 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2380 if (*write_lock_level
< level
+ 1) {
2381 *write_lock_level
= level
+ 1;
2382 btrfs_release_path(p
);
2386 btrfs_set_path_blocking(p
);
2387 reada_for_balance(root
, p
, level
);
2388 sret
= split_node(trans
, root
, p
, level
);
2389 btrfs_clear_path_blocking(p
, NULL
, 0);
2396 b
= p
->nodes
[level
];
2397 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2398 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2401 if (*write_lock_level
< level
+ 1) {
2402 *write_lock_level
= level
+ 1;
2403 btrfs_release_path(p
);
2407 btrfs_set_path_blocking(p
);
2408 reada_for_balance(root
, p
, level
);
2409 sret
= balance_level(trans
, root
, p
, level
);
2410 btrfs_clear_path_blocking(p
, NULL
, 0);
2416 b
= p
->nodes
[level
];
2418 btrfs_release_path(p
);
2421 BUG_ON(btrfs_header_nritems(b
) == 1);
2431 static void key_search_validate(struct extent_buffer
*b
,
2432 struct btrfs_key
*key
,
2435 #ifdef CONFIG_BTRFS_ASSERT
2436 struct btrfs_disk_key disk_key
;
2438 btrfs_cpu_key_to_disk(&disk_key
, key
);
2441 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2442 offsetof(struct btrfs_leaf
, items
[0].key
),
2445 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2446 offsetof(struct btrfs_node
, ptrs
[0].key
),
2451 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2452 int level
, int *prev_cmp
, int *slot
)
2454 if (*prev_cmp
!= 0) {
2455 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2459 key_search_validate(b
, key
, level
);
2466 * look for key in the tree. path is filled in with nodes along the way
2467 * if key is found, we return zero and you can find the item in the leaf
2468 * level of the path (level 0)
2470 * If the key isn't found, the path points to the slot where it should
2471 * be inserted, and 1 is returned. If there are other errors during the
2472 * search a negative error number is returned.
2474 * if ins_len > 0, nodes and leaves will be split as we walk down the
2475 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2478 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2479 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2482 struct extent_buffer
*b
;
2487 int lowest_unlock
= 1;
2489 /* everything at write_lock_level or lower must be write locked */
2490 int write_lock_level
= 0;
2491 u8 lowest_level
= 0;
2492 int min_write_lock_level
;
2495 lowest_level
= p
->lowest_level
;
2496 WARN_ON(lowest_level
&& ins_len
> 0);
2497 WARN_ON(p
->nodes
[0] != NULL
);
2502 /* when we are removing items, we might have to go up to level
2503 * two as we update tree pointers Make sure we keep write
2504 * for those levels as well
2506 write_lock_level
= 2;
2507 } else if (ins_len
> 0) {
2509 * for inserting items, make sure we have a write lock on
2510 * level 1 so we can update keys
2512 write_lock_level
= 1;
2516 write_lock_level
= -1;
2518 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2519 write_lock_level
= BTRFS_MAX_LEVEL
;
2521 min_write_lock_level
= write_lock_level
;
2526 * we try very hard to do read locks on the root
2528 root_lock
= BTRFS_READ_LOCK
;
2530 if (p
->search_commit_root
) {
2532 * the commit roots are read only
2533 * so we always do read locks
2535 b
= root
->commit_root
;
2536 extent_buffer_get(b
);
2537 level
= btrfs_header_level(b
);
2538 if (!p
->skip_locking
)
2539 btrfs_tree_read_lock(b
);
2541 if (p
->skip_locking
) {
2542 b
= btrfs_root_node(root
);
2543 level
= btrfs_header_level(b
);
2545 /* we don't know the level of the root node
2546 * until we actually have it read locked
2548 b
= btrfs_read_lock_root_node(root
);
2549 level
= btrfs_header_level(b
);
2550 if (level
<= write_lock_level
) {
2551 /* whoops, must trade for write lock */
2552 btrfs_tree_read_unlock(b
);
2553 free_extent_buffer(b
);
2554 b
= btrfs_lock_root_node(root
);
2555 root_lock
= BTRFS_WRITE_LOCK
;
2557 /* the level might have changed, check again */
2558 level
= btrfs_header_level(b
);
2562 p
->nodes
[level
] = b
;
2563 if (!p
->skip_locking
)
2564 p
->locks
[level
] = root_lock
;
2567 level
= btrfs_header_level(b
);
2570 * setup the path here so we can release it under lock
2571 * contention with the cow code
2575 * if we don't really need to cow this block
2576 * then we don't want to set the path blocking,
2577 * so we test it here
2579 if (!should_cow_block(trans
, root
, b
))
2582 btrfs_set_path_blocking(p
);
2585 * must have write locks on this node and the
2588 if (level
> write_lock_level
||
2589 (level
+ 1 > write_lock_level
&&
2590 level
+ 1 < BTRFS_MAX_LEVEL
&&
2591 p
->nodes
[level
+ 1])) {
2592 write_lock_level
= level
+ 1;
2593 btrfs_release_path(p
);
2597 err
= btrfs_cow_block(trans
, root
, b
,
2598 p
->nodes
[level
+ 1],
2599 p
->slots
[level
+ 1], &b
);
2606 BUG_ON(!cow
&& ins_len
);
2608 p
->nodes
[level
] = b
;
2609 btrfs_clear_path_blocking(p
, NULL
, 0);
2612 * we have a lock on b and as long as we aren't changing
2613 * the tree, there is no way to for the items in b to change.
2614 * It is safe to drop the lock on our parent before we
2615 * go through the expensive btree search on b.
2617 * If cow is true, then we might be changing slot zero,
2618 * which may require changing the parent. So, we can't
2619 * drop the lock until after we know which slot we're
2623 btrfs_unlock_up_safe(p
, level
+ 1);
2625 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2629 if (ret
&& slot
> 0) {
2633 p
->slots
[level
] = slot
;
2634 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2635 ins_len
, &write_lock_level
);
2642 b
= p
->nodes
[level
];
2643 slot
= p
->slots
[level
];
2646 * slot 0 is special, if we change the key
2647 * we have to update the parent pointer
2648 * which means we must have a write lock
2651 if (slot
== 0 && cow
&&
2652 write_lock_level
< level
+ 1) {
2653 write_lock_level
= level
+ 1;
2654 btrfs_release_path(p
);
2658 unlock_up(p
, level
, lowest_unlock
,
2659 min_write_lock_level
, &write_lock_level
);
2661 if (level
== lowest_level
) {
2667 err
= read_block_for_search(trans
, root
, p
,
2668 &b
, level
, slot
, key
, 0);
2676 if (!p
->skip_locking
) {
2677 level
= btrfs_header_level(b
);
2678 if (level
<= write_lock_level
) {
2679 err
= btrfs_try_tree_write_lock(b
);
2681 btrfs_set_path_blocking(p
);
2683 btrfs_clear_path_blocking(p
, b
,
2686 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2688 err
= btrfs_try_tree_read_lock(b
);
2690 btrfs_set_path_blocking(p
);
2691 btrfs_tree_read_lock(b
);
2692 btrfs_clear_path_blocking(p
, b
,
2695 p
->locks
[level
] = BTRFS_READ_LOCK
;
2697 p
->nodes
[level
] = b
;
2700 p
->slots
[level
] = slot
;
2702 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2703 if (write_lock_level
< 1) {
2704 write_lock_level
= 1;
2705 btrfs_release_path(p
);
2709 btrfs_set_path_blocking(p
);
2710 err
= split_leaf(trans
, root
, key
,
2711 p
, ins_len
, ret
== 0);
2712 btrfs_clear_path_blocking(p
, NULL
, 0);
2720 if (!p
->search_for_split
)
2721 unlock_up(p
, level
, lowest_unlock
,
2722 min_write_lock_level
, &write_lock_level
);
2729 * we don't really know what they plan on doing with the path
2730 * from here on, so for now just mark it as blocking
2732 if (!p
->leave_spinning
)
2733 btrfs_set_path_blocking(p
);
2735 btrfs_release_path(p
);
2740 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2741 * current state of the tree together with the operations recorded in the tree
2742 * modification log to search for the key in a previous version of this tree, as
2743 * denoted by the time_seq parameter.
2745 * Naturally, there is no support for insert, delete or cow operations.
2747 * The resulting path and return value will be set up as if we called
2748 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2750 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2751 struct btrfs_path
*p
, u64 time_seq
)
2753 struct extent_buffer
*b
;
2758 int lowest_unlock
= 1;
2759 u8 lowest_level
= 0;
2762 lowest_level
= p
->lowest_level
;
2763 WARN_ON(p
->nodes
[0] != NULL
);
2765 if (p
->search_commit_root
) {
2767 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2771 b
= get_old_root(root
, time_seq
);
2772 level
= btrfs_header_level(b
);
2773 p
->locks
[level
] = BTRFS_READ_LOCK
;
2776 level
= btrfs_header_level(b
);
2777 p
->nodes
[level
] = b
;
2778 btrfs_clear_path_blocking(p
, NULL
, 0);
2781 * we have a lock on b and as long as we aren't changing
2782 * the tree, there is no way to for the items in b to change.
2783 * It is safe to drop the lock on our parent before we
2784 * go through the expensive btree search on b.
2786 btrfs_unlock_up_safe(p
, level
+ 1);
2789 * Since we can unwind eb's we want to do a real search every
2793 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2797 if (ret
&& slot
> 0) {
2801 p
->slots
[level
] = slot
;
2802 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2804 if (level
== lowest_level
) {
2810 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
2811 slot
, key
, time_seq
);
2819 level
= btrfs_header_level(b
);
2820 err
= btrfs_try_tree_read_lock(b
);
2822 btrfs_set_path_blocking(p
);
2823 btrfs_tree_read_lock(b
);
2824 btrfs_clear_path_blocking(p
, b
,
2827 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
2832 p
->locks
[level
] = BTRFS_READ_LOCK
;
2833 p
->nodes
[level
] = b
;
2835 p
->slots
[level
] = slot
;
2836 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2842 if (!p
->leave_spinning
)
2843 btrfs_set_path_blocking(p
);
2845 btrfs_release_path(p
);
2851 * helper to use instead of search slot if no exact match is needed but
2852 * instead the next or previous item should be returned.
2853 * When find_higher is true, the next higher item is returned, the next lower
2855 * When return_any and find_higher are both true, and no higher item is found,
2856 * return the next lower instead.
2857 * When return_any is true and find_higher is false, and no lower item is found,
2858 * return the next higher instead.
2859 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2862 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
2863 struct btrfs_key
*key
, struct btrfs_path
*p
,
2864 int find_higher
, int return_any
)
2867 struct extent_buffer
*leaf
;
2870 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2874 * a return value of 1 means the path is at the position where the
2875 * item should be inserted. Normally this is the next bigger item,
2876 * but in case the previous item is the last in a leaf, path points
2877 * to the first free slot in the previous leaf, i.e. at an invalid
2883 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2884 ret
= btrfs_next_leaf(root
, p
);
2890 * no higher item found, return the next
2895 btrfs_release_path(p
);
2899 if (p
->slots
[0] == 0) {
2900 ret
= btrfs_prev_leaf(root
, p
);
2904 p
->slots
[0] = btrfs_header_nritems(leaf
) - 1;
2910 * no lower item found, return the next
2915 btrfs_release_path(p
);
2925 * adjust the pointers going up the tree, starting at level
2926 * making sure the right key of each node is points to 'key'.
2927 * This is used after shifting pointers to the left, so it stops
2928 * fixing up pointers when a given leaf/node is not in slot 0 of the
2932 static void fixup_low_keys(struct btrfs_root
*root
, struct btrfs_path
*path
,
2933 struct btrfs_disk_key
*key
, int level
)
2936 struct extent_buffer
*t
;
2938 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2939 int tslot
= path
->slots
[i
];
2940 if (!path
->nodes
[i
])
2943 tree_mod_log_set_node_key(root
->fs_info
, t
, tslot
, 1);
2944 btrfs_set_node_key(t
, key
, tslot
);
2945 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
2954 * This function isn't completely safe. It's the caller's responsibility
2955 * that the new key won't break the order
2957 void btrfs_set_item_key_safe(struct btrfs_root
*root
, struct btrfs_path
*path
,
2958 struct btrfs_key
*new_key
)
2960 struct btrfs_disk_key disk_key
;
2961 struct extent_buffer
*eb
;
2964 eb
= path
->nodes
[0];
2965 slot
= path
->slots
[0];
2967 btrfs_item_key(eb
, &disk_key
, slot
- 1);
2968 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
2970 if (slot
< btrfs_header_nritems(eb
) - 1) {
2971 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
2972 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
2975 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2976 btrfs_set_item_key(eb
, &disk_key
, slot
);
2977 btrfs_mark_buffer_dirty(eb
);
2979 fixup_low_keys(root
, path
, &disk_key
, 1);
2983 * try to push data from one node into the next node left in the
2986 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2987 * error, and > 0 if there was no room in the left hand block.
2989 static int push_node_left(struct btrfs_trans_handle
*trans
,
2990 struct btrfs_root
*root
, struct extent_buffer
*dst
,
2991 struct extent_buffer
*src
, int empty
)
2998 src_nritems
= btrfs_header_nritems(src
);
2999 dst_nritems
= btrfs_header_nritems(dst
);
3000 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3001 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3002 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3004 if (!empty
&& src_nritems
<= 8)
3007 if (push_items
<= 0)
3011 push_items
= min(src_nritems
, push_items
);
3012 if (push_items
< src_nritems
) {
3013 /* leave at least 8 pointers in the node if
3014 * we aren't going to empty it
3016 if (src_nritems
- push_items
< 8) {
3017 if (push_items
<= 8)
3023 push_items
= min(src_nritems
- 8, push_items
);
3025 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3027 copy_extent_buffer(dst
, src
,
3028 btrfs_node_key_ptr_offset(dst_nritems
),
3029 btrfs_node_key_ptr_offset(0),
3030 push_items
* sizeof(struct btrfs_key_ptr
));
3032 if (push_items
< src_nritems
) {
3034 * don't call tree_mod_log_eb_move here, key removal was already
3035 * fully logged by tree_mod_log_eb_copy above.
3037 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3038 btrfs_node_key_ptr_offset(push_items
),
3039 (src_nritems
- push_items
) *
3040 sizeof(struct btrfs_key_ptr
));
3042 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3043 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3044 btrfs_mark_buffer_dirty(src
);
3045 btrfs_mark_buffer_dirty(dst
);
3051 * try to push data from one node into the next node right in the
3054 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3055 * error, and > 0 if there was no room in the right hand block.
3057 * this will only push up to 1/2 the contents of the left node over
3059 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3060 struct btrfs_root
*root
,
3061 struct extent_buffer
*dst
,
3062 struct extent_buffer
*src
)
3070 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3071 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3073 src_nritems
= btrfs_header_nritems(src
);
3074 dst_nritems
= btrfs_header_nritems(dst
);
3075 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3076 if (push_items
<= 0)
3079 if (src_nritems
< 4)
3082 max_push
= src_nritems
/ 2 + 1;
3083 /* don't try to empty the node */
3084 if (max_push
>= src_nritems
)
3087 if (max_push
< push_items
)
3088 push_items
= max_push
;
3090 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3091 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3092 btrfs_node_key_ptr_offset(0),
3094 sizeof(struct btrfs_key_ptr
));
3096 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3097 src_nritems
- push_items
, push_items
);
3098 copy_extent_buffer(dst
, src
,
3099 btrfs_node_key_ptr_offset(0),
3100 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3101 push_items
* sizeof(struct btrfs_key_ptr
));
3103 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3104 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3106 btrfs_mark_buffer_dirty(src
);
3107 btrfs_mark_buffer_dirty(dst
);
3113 * helper function to insert a new root level in the tree.
3114 * A new node is allocated, and a single item is inserted to
3115 * point to the existing root
3117 * returns zero on success or < 0 on failure.
3119 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3120 struct btrfs_root
*root
,
3121 struct btrfs_path
*path
, int level
)
3124 struct extent_buffer
*lower
;
3125 struct extent_buffer
*c
;
3126 struct extent_buffer
*old
;
3127 struct btrfs_disk_key lower_key
;
3129 BUG_ON(path
->nodes
[level
]);
3130 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3132 lower
= path
->nodes
[level
-1];
3134 btrfs_item_key(lower
, &lower_key
, 0);
3136 btrfs_node_key(lower
, &lower_key
, 0);
3138 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3139 root
->root_key
.objectid
, &lower_key
,
3140 level
, root
->node
->start
, 0);
3144 root_add_used(root
, root
->nodesize
);
3146 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3147 btrfs_set_header_nritems(c
, 1);
3148 btrfs_set_header_level(c
, level
);
3149 btrfs_set_header_bytenr(c
, c
->start
);
3150 btrfs_set_header_generation(c
, trans
->transid
);
3151 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3152 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3154 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3157 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3158 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3160 btrfs_set_node_key(c
, &lower_key
, 0);
3161 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3162 lower_gen
= btrfs_header_generation(lower
);
3163 WARN_ON(lower_gen
!= trans
->transid
);
3165 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3167 btrfs_mark_buffer_dirty(c
);
3170 tree_mod_log_set_root_pointer(root
, c
, 0);
3171 rcu_assign_pointer(root
->node
, c
);
3173 /* the super has an extra ref to root->node */
3174 free_extent_buffer(old
);
3176 add_root_to_dirty_list(root
);
3177 extent_buffer_get(c
);
3178 path
->nodes
[level
] = c
;
3179 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
3180 path
->slots
[level
] = 0;
3185 * worker function to insert a single pointer in a node.
3186 * the node should have enough room for the pointer already
3188 * slot and level indicate where you want the key to go, and
3189 * blocknr is the block the key points to.
3191 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3192 struct btrfs_root
*root
, struct btrfs_path
*path
,
3193 struct btrfs_disk_key
*key
, u64 bytenr
,
3194 int slot
, int level
)
3196 struct extent_buffer
*lower
;
3200 BUG_ON(!path
->nodes
[level
]);
3201 btrfs_assert_tree_locked(path
->nodes
[level
]);
3202 lower
= path
->nodes
[level
];
3203 nritems
= btrfs_header_nritems(lower
);
3204 BUG_ON(slot
> nritems
);
3205 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3206 if (slot
!= nritems
) {
3208 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3209 slot
, nritems
- slot
);
3210 memmove_extent_buffer(lower
,
3211 btrfs_node_key_ptr_offset(slot
+ 1),
3212 btrfs_node_key_ptr_offset(slot
),
3213 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3216 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3217 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3220 btrfs_set_node_key(lower
, key
, slot
);
3221 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3222 WARN_ON(trans
->transid
== 0);
3223 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3224 btrfs_set_header_nritems(lower
, nritems
+ 1);
3225 btrfs_mark_buffer_dirty(lower
);
3229 * split the node at the specified level in path in two.
3230 * The path is corrected to point to the appropriate node after the split
3232 * Before splitting this tries to make some room in the node by pushing
3233 * left and right, if either one works, it returns right away.
3235 * returns 0 on success and < 0 on failure
3237 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3238 struct btrfs_root
*root
,
3239 struct btrfs_path
*path
, int level
)
3241 struct extent_buffer
*c
;
3242 struct extent_buffer
*split
;
3243 struct btrfs_disk_key disk_key
;
3248 c
= path
->nodes
[level
];
3249 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3250 if (c
== root
->node
) {
3252 * trying to split the root, lets make a new one
3254 * tree mod log: We don't log_removal old root in
3255 * insert_new_root, because that root buffer will be kept as a
3256 * normal node. We are going to log removal of half of the
3257 * elements below with tree_mod_log_eb_copy. We're holding a
3258 * tree lock on the buffer, which is why we cannot race with
3259 * other tree_mod_log users.
3261 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3265 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3266 c
= path
->nodes
[level
];
3267 if (!ret
&& btrfs_header_nritems(c
) <
3268 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3274 c_nritems
= btrfs_header_nritems(c
);
3275 mid
= (c_nritems
+ 1) / 2;
3276 btrfs_node_key(c
, &disk_key
, mid
);
3278 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3279 root
->root_key
.objectid
,
3280 &disk_key
, level
, c
->start
, 0);
3282 return PTR_ERR(split
);
3284 root_add_used(root
, root
->nodesize
);
3286 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3287 btrfs_set_header_level(split
, btrfs_header_level(c
));
3288 btrfs_set_header_bytenr(split
, split
->start
);
3289 btrfs_set_header_generation(split
, trans
->transid
);
3290 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3291 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3292 write_extent_buffer(split
, root
->fs_info
->fsid
,
3293 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3294 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3295 btrfs_header_chunk_tree_uuid(split
),
3298 tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0, mid
, c_nritems
- mid
);
3299 copy_extent_buffer(split
, c
,
3300 btrfs_node_key_ptr_offset(0),
3301 btrfs_node_key_ptr_offset(mid
),
3302 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3303 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3304 btrfs_set_header_nritems(c
, mid
);
3307 btrfs_mark_buffer_dirty(c
);
3308 btrfs_mark_buffer_dirty(split
);
3310 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3311 path
->slots
[level
+ 1] + 1, level
+ 1);
3313 if (path
->slots
[level
] >= mid
) {
3314 path
->slots
[level
] -= mid
;
3315 btrfs_tree_unlock(c
);
3316 free_extent_buffer(c
);
3317 path
->nodes
[level
] = split
;
3318 path
->slots
[level
+ 1] += 1;
3320 btrfs_tree_unlock(split
);
3321 free_extent_buffer(split
);
3327 * how many bytes are required to store the items in a leaf. start
3328 * and nr indicate which items in the leaf to check. This totals up the
3329 * space used both by the item structs and the item data
3331 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3333 struct btrfs_item
*start_item
;
3334 struct btrfs_item
*end_item
;
3335 struct btrfs_map_token token
;
3337 int nritems
= btrfs_header_nritems(l
);
3338 int end
= min(nritems
, start
+ nr
) - 1;
3342 btrfs_init_map_token(&token
);
3343 start_item
= btrfs_item_nr(start
);
3344 end_item
= btrfs_item_nr(end
);
3345 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3346 btrfs_token_item_size(l
, start_item
, &token
);
3347 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3348 data_len
+= sizeof(struct btrfs_item
) * nr
;
3349 WARN_ON(data_len
< 0);
3354 * The space between the end of the leaf items and
3355 * the start of the leaf data. IOW, how much room
3356 * the leaf has left for both items and data
3358 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3359 struct extent_buffer
*leaf
)
3361 int nritems
= btrfs_header_nritems(leaf
);
3363 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3365 printk(KERN_CRIT
"leaf free space ret %d, leaf data size %lu, "
3366 "used %d nritems %d\n",
3367 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3368 leaf_space_used(leaf
, 0, nritems
), nritems
);
3374 * min slot controls the lowest index we're willing to push to the
3375 * right. We'll push up to and including min_slot, but no lower
3377 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3378 struct btrfs_root
*root
,
3379 struct btrfs_path
*path
,
3380 int data_size
, int empty
,
3381 struct extent_buffer
*right
,
3382 int free_space
, u32 left_nritems
,
3385 struct extent_buffer
*left
= path
->nodes
[0];
3386 struct extent_buffer
*upper
= path
->nodes
[1];
3387 struct btrfs_map_token token
;
3388 struct btrfs_disk_key disk_key
;
3393 struct btrfs_item
*item
;
3399 btrfs_init_map_token(&token
);
3404 nr
= max_t(u32
, 1, min_slot
);
3406 if (path
->slots
[0] >= left_nritems
)
3407 push_space
+= data_size
;
3409 slot
= path
->slots
[1];
3410 i
= left_nritems
- 1;
3412 item
= btrfs_item_nr(i
);
3414 if (!empty
&& push_items
> 0) {
3415 if (path
->slots
[0] > i
)
3417 if (path
->slots
[0] == i
) {
3418 int space
= btrfs_leaf_free_space(root
, left
);
3419 if (space
+ push_space
* 2 > free_space
)
3424 if (path
->slots
[0] == i
)
3425 push_space
+= data_size
;
3427 this_item_size
= btrfs_item_size(left
, item
);
3428 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3432 push_space
+= this_item_size
+ sizeof(*item
);
3438 if (push_items
== 0)
3441 WARN_ON(!empty
&& push_items
== left_nritems
);
3443 /* push left to right */
3444 right_nritems
= btrfs_header_nritems(right
);
3446 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3447 push_space
-= leaf_data_end(root
, left
);
3449 /* make room in the right data area */
3450 data_end
= leaf_data_end(root
, right
);
3451 memmove_extent_buffer(right
,
3452 btrfs_leaf_data(right
) + data_end
- push_space
,
3453 btrfs_leaf_data(right
) + data_end
,
3454 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3456 /* copy from the left data area */
3457 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3458 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3459 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3462 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3463 btrfs_item_nr_offset(0),
3464 right_nritems
* sizeof(struct btrfs_item
));
3466 /* copy the items from left to right */
3467 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3468 btrfs_item_nr_offset(left_nritems
- push_items
),
3469 push_items
* sizeof(struct btrfs_item
));
3471 /* update the item pointers */
3472 right_nritems
+= push_items
;
3473 btrfs_set_header_nritems(right
, right_nritems
);
3474 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3475 for (i
= 0; i
< right_nritems
; i
++) {
3476 item
= btrfs_item_nr(i
);
3477 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3478 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3481 left_nritems
-= push_items
;
3482 btrfs_set_header_nritems(left
, left_nritems
);
3485 btrfs_mark_buffer_dirty(left
);
3487 clean_tree_block(trans
, root
, left
);
3489 btrfs_mark_buffer_dirty(right
);
3491 btrfs_item_key(right
, &disk_key
, 0);
3492 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3493 btrfs_mark_buffer_dirty(upper
);
3495 /* then fixup the leaf pointer in the path */
3496 if (path
->slots
[0] >= left_nritems
) {
3497 path
->slots
[0] -= left_nritems
;
3498 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3499 clean_tree_block(trans
, root
, path
->nodes
[0]);
3500 btrfs_tree_unlock(path
->nodes
[0]);
3501 free_extent_buffer(path
->nodes
[0]);
3502 path
->nodes
[0] = right
;
3503 path
->slots
[1] += 1;
3505 btrfs_tree_unlock(right
);
3506 free_extent_buffer(right
);
3511 btrfs_tree_unlock(right
);
3512 free_extent_buffer(right
);
3517 * push some data in the path leaf to the right, trying to free up at
3518 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3520 * returns 1 if the push failed because the other node didn't have enough
3521 * room, 0 if everything worked out and < 0 if there were major errors.
3523 * this will push starting from min_slot to the end of the leaf. It won't
3524 * push any slot lower than min_slot
3526 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3527 *root
, struct btrfs_path
*path
,
3528 int min_data_size
, int data_size
,
3529 int empty
, u32 min_slot
)
3531 struct extent_buffer
*left
= path
->nodes
[0];
3532 struct extent_buffer
*right
;
3533 struct extent_buffer
*upper
;
3539 if (!path
->nodes
[1])
3542 slot
= path
->slots
[1];
3543 upper
= path
->nodes
[1];
3544 if (slot
>= btrfs_header_nritems(upper
) - 1)
3547 btrfs_assert_tree_locked(path
->nodes
[1]);
3549 right
= read_node_slot(root
, upper
, slot
+ 1);
3553 btrfs_tree_lock(right
);
3554 btrfs_set_lock_blocking(right
);
3556 free_space
= btrfs_leaf_free_space(root
, right
);
3557 if (free_space
< data_size
)
3560 /* cow and double check */
3561 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3566 free_space
= btrfs_leaf_free_space(root
, right
);
3567 if (free_space
< data_size
)
3570 left_nritems
= btrfs_header_nritems(left
);
3571 if (left_nritems
== 0)
3574 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3575 right
, free_space
, left_nritems
, min_slot
);
3577 btrfs_tree_unlock(right
);
3578 free_extent_buffer(right
);
3583 * push some data in the path leaf to the left, trying to free up at
3584 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3586 * max_slot can put a limit on how far into the leaf we'll push items. The
3587 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3590 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3591 struct btrfs_root
*root
,
3592 struct btrfs_path
*path
, int data_size
,
3593 int empty
, struct extent_buffer
*left
,
3594 int free_space
, u32 right_nritems
,
3597 struct btrfs_disk_key disk_key
;
3598 struct extent_buffer
*right
= path
->nodes
[0];
3602 struct btrfs_item
*item
;
3603 u32 old_left_nritems
;
3607 u32 old_left_item_size
;
3608 struct btrfs_map_token token
;
3610 btrfs_init_map_token(&token
);
3613 nr
= min(right_nritems
, max_slot
);
3615 nr
= min(right_nritems
- 1, max_slot
);
3617 for (i
= 0; i
< nr
; i
++) {
3618 item
= btrfs_item_nr(i
);
3620 if (!empty
&& push_items
> 0) {
3621 if (path
->slots
[0] < i
)
3623 if (path
->slots
[0] == i
) {
3624 int space
= btrfs_leaf_free_space(root
, right
);
3625 if (space
+ push_space
* 2 > free_space
)
3630 if (path
->slots
[0] == i
)
3631 push_space
+= data_size
;
3633 this_item_size
= btrfs_item_size(right
, item
);
3634 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3638 push_space
+= this_item_size
+ sizeof(*item
);
3641 if (push_items
== 0) {
3645 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3647 /* push data from right to left */
3648 copy_extent_buffer(left
, right
,
3649 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3650 btrfs_item_nr_offset(0),
3651 push_items
* sizeof(struct btrfs_item
));
3653 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3654 btrfs_item_offset_nr(right
, push_items
- 1);
3656 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3657 leaf_data_end(root
, left
) - push_space
,
3658 btrfs_leaf_data(right
) +
3659 btrfs_item_offset_nr(right
, push_items
- 1),
3661 old_left_nritems
= btrfs_header_nritems(left
);
3662 BUG_ON(old_left_nritems
<= 0);
3664 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3665 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3668 item
= btrfs_item_nr(i
);
3670 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3671 btrfs_set_token_item_offset(left
, item
,
3672 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3675 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3677 /* fixup right node */
3678 if (push_items
> right_nritems
)
3679 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3682 if (push_items
< right_nritems
) {
3683 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3684 leaf_data_end(root
, right
);
3685 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3686 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3687 btrfs_leaf_data(right
) +
3688 leaf_data_end(root
, right
), push_space
);
3690 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3691 btrfs_item_nr_offset(push_items
),
3692 (btrfs_header_nritems(right
) - push_items
) *
3693 sizeof(struct btrfs_item
));
3695 right_nritems
-= push_items
;
3696 btrfs_set_header_nritems(right
, right_nritems
);
3697 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3698 for (i
= 0; i
< right_nritems
; i
++) {
3699 item
= btrfs_item_nr(i
);
3701 push_space
= push_space
- btrfs_token_item_size(right
,
3703 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3706 btrfs_mark_buffer_dirty(left
);
3708 btrfs_mark_buffer_dirty(right
);
3710 clean_tree_block(trans
, root
, right
);
3712 btrfs_item_key(right
, &disk_key
, 0);
3713 fixup_low_keys(root
, path
, &disk_key
, 1);
3715 /* then fixup the leaf pointer in the path */
3716 if (path
->slots
[0] < push_items
) {
3717 path
->slots
[0] += old_left_nritems
;
3718 btrfs_tree_unlock(path
->nodes
[0]);
3719 free_extent_buffer(path
->nodes
[0]);
3720 path
->nodes
[0] = left
;
3721 path
->slots
[1] -= 1;
3723 btrfs_tree_unlock(left
);
3724 free_extent_buffer(left
);
3725 path
->slots
[0] -= push_items
;
3727 BUG_ON(path
->slots
[0] < 0);
3730 btrfs_tree_unlock(left
);
3731 free_extent_buffer(left
);
3736 * push some data in the path leaf to the left, trying to free up at
3737 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3739 * max_slot can put a limit on how far into the leaf we'll push items. The
3740 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3743 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3744 *root
, struct btrfs_path
*path
, int min_data_size
,
3745 int data_size
, int empty
, u32 max_slot
)
3747 struct extent_buffer
*right
= path
->nodes
[0];
3748 struct extent_buffer
*left
;
3754 slot
= path
->slots
[1];
3757 if (!path
->nodes
[1])
3760 right_nritems
= btrfs_header_nritems(right
);
3761 if (right_nritems
== 0)
3764 btrfs_assert_tree_locked(path
->nodes
[1]);
3766 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
3770 btrfs_tree_lock(left
);
3771 btrfs_set_lock_blocking(left
);
3773 free_space
= btrfs_leaf_free_space(root
, left
);
3774 if (free_space
< data_size
) {
3779 /* cow and double check */
3780 ret
= btrfs_cow_block(trans
, root
, left
,
3781 path
->nodes
[1], slot
- 1, &left
);
3783 /* we hit -ENOSPC, but it isn't fatal here */
3789 free_space
= btrfs_leaf_free_space(root
, left
);
3790 if (free_space
< data_size
) {
3795 return __push_leaf_left(trans
, root
, path
, min_data_size
,
3796 empty
, left
, free_space
, right_nritems
,
3799 btrfs_tree_unlock(left
);
3800 free_extent_buffer(left
);
3805 * split the path's leaf in two, making sure there is at least data_size
3806 * available for the resulting leaf level of the path.
3808 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
3809 struct btrfs_root
*root
,
3810 struct btrfs_path
*path
,
3811 struct extent_buffer
*l
,
3812 struct extent_buffer
*right
,
3813 int slot
, int mid
, int nritems
)
3818 struct btrfs_disk_key disk_key
;
3819 struct btrfs_map_token token
;
3821 btrfs_init_map_token(&token
);
3823 nritems
= nritems
- mid
;
3824 btrfs_set_header_nritems(right
, nritems
);
3825 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
3827 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
3828 btrfs_item_nr_offset(mid
),
3829 nritems
* sizeof(struct btrfs_item
));
3831 copy_extent_buffer(right
, l
,
3832 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
3833 data_copy_size
, btrfs_leaf_data(l
) +
3834 leaf_data_end(root
, l
), data_copy_size
);
3836 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
3837 btrfs_item_end_nr(l
, mid
);
3839 for (i
= 0; i
< nritems
; i
++) {
3840 struct btrfs_item
*item
= btrfs_item_nr(i
);
3843 ioff
= btrfs_token_item_offset(right
, item
, &token
);
3844 btrfs_set_token_item_offset(right
, item
,
3845 ioff
+ rt_data_off
, &token
);
3848 btrfs_set_header_nritems(l
, mid
);
3849 btrfs_item_key(right
, &disk_key
, 0);
3850 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3851 path
->slots
[1] + 1, 1);
3853 btrfs_mark_buffer_dirty(right
);
3854 btrfs_mark_buffer_dirty(l
);
3855 BUG_ON(path
->slots
[0] != slot
);
3858 btrfs_tree_unlock(path
->nodes
[0]);
3859 free_extent_buffer(path
->nodes
[0]);
3860 path
->nodes
[0] = right
;
3861 path
->slots
[0] -= mid
;
3862 path
->slots
[1] += 1;
3864 btrfs_tree_unlock(right
);
3865 free_extent_buffer(right
);
3868 BUG_ON(path
->slots
[0] < 0);
3872 * double splits happen when we need to insert a big item in the middle
3873 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3874 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3877 * We avoid this by trying to push the items on either side of our target
3878 * into the adjacent leaves. If all goes well we can avoid the double split
3881 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
3882 struct btrfs_root
*root
,
3883 struct btrfs_path
*path
,
3891 slot
= path
->slots
[0];
3894 * try to push all the items after our slot into the
3897 ret
= push_leaf_right(trans
, root
, path
, 1, data_size
, 0, slot
);
3904 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3906 * our goal is to get our slot at the start or end of a leaf. If
3907 * we've done so we're done
3909 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
3912 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
3915 /* try to push all the items before our slot into the next leaf */
3916 slot
= path
->slots
[0];
3917 ret
= push_leaf_left(trans
, root
, path
, 1, data_size
, 0, slot
);
3930 * split the path's leaf in two, making sure there is at least data_size
3931 * available for the resulting leaf level of the path.
3933 * returns 0 if all went well and < 0 on failure.
3935 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
3936 struct btrfs_root
*root
,
3937 struct btrfs_key
*ins_key
,
3938 struct btrfs_path
*path
, int data_size
,
3941 struct btrfs_disk_key disk_key
;
3942 struct extent_buffer
*l
;
3946 struct extent_buffer
*right
;
3950 int num_doubles
= 0;
3951 int tried_avoid_double
= 0;
3954 slot
= path
->slots
[0];
3955 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
3956 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
3959 /* first try to make some room by pushing left and right */
3960 if (data_size
&& path
->nodes
[1]) {
3961 wret
= push_leaf_right(trans
, root
, path
, data_size
,
3966 wret
= push_leaf_left(trans
, root
, path
, data_size
,
3967 data_size
, 0, (u32
)-1);
3973 /* did the pushes work? */
3974 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
3978 if (!path
->nodes
[1]) {
3979 ret
= insert_new_root(trans
, root
, path
, 1);
3986 slot
= path
->slots
[0];
3987 nritems
= btrfs_header_nritems(l
);
3988 mid
= (nritems
+ 1) / 2;
3992 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
3993 BTRFS_LEAF_DATA_SIZE(root
)) {
3994 if (slot
>= nritems
) {
3998 if (mid
!= nritems
&&
3999 leaf_space_used(l
, mid
, nritems
- mid
) +
4000 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4001 if (data_size
&& !tried_avoid_double
)
4002 goto push_for_double
;
4008 if (leaf_space_used(l
, 0, mid
) + data_size
>
4009 BTRFS_LEAF_DATA_SIZE(root
)) {
4010 if (!extend
&& data_size
&& slot
== 0) {
4012 } else if ((extend
|| !data_size
) && slot
== 0) {
4016 if (mid
!= nritems
&&
4017 leaf_space_used(l
, mid
, nritems
- mid
) +
4018 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4019 if (data_size
&& !tried_avoid_double
)
4020 goto push_for_double
;
4028 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4030 btrfs_item_key(l
, &disk_key
, mid
);
4032 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
4033 root
->root_key
.objectid
,
4034 &disk_key
, 0, l
->start
, 0);
4036 return PTR_ERR(right
);
4038 root_add_used(root
, root
->leafsize
);
4040 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4041 btrfs_set_header_bytenr(right
, right
->start
);
4042 btrfs_set_header_generation(right
, trans
->transid
);
4043 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4044 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4045 btrfs_set_header_level(right
, 0);
4046 write_extent_buffer(right
, root
->fs_info
->fsid
,
4047 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4049 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
4050 btrfs_header_chunk_tree_uuid(right
),
4055 btrfs_set_header_nritems(right
, 0);
4056 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4057 path
->slots
[1] + 1, 1);
4058 btrfs_tree_unlock(path
->nodes
[0]);
4059 free_extent_buffer(path
->nodes
[0]);
4060 path
->nodes
[0] = right
;
4062 path
->slots
[1] += 1;
4064 btrfs_set_header_nritems(right
, 0);
4065 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4067 btrfs_tree_unlock(path
->nodes
[0]);
4068 free_extent_buffer(path
->nodes
[0]);
4069 path
->nodes
[0] = right
;
4071 if (path
->slots
[1] == 0)
4072 fixup_low_keys(root
, path
, &disk_key
, 1);
4074 btrfs_mark_buffer_dirty(right
);
4078 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4081 BUG_ON(num_doubles
!= 0);
4089 push_for_double_split(trans
, root
, path
, data_size
);
4090 tried_avoid_double
= 1;
4091 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4096 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4097 struct btrfs_root
*root
,
4098 struct btrfs_path
*path
, int ins_len
)
4100 struct btrfs_key key
;
4101 struct extent_buffer
*leaf
;
4102 struct btrfs_file_extent_item
*fi
;
4107 leaf
= path
->nodes
[0];
4108 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4110 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4111 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4113 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4116 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4117 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4118 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4119 struct btrfs_file_extent_item
);
4120 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4122 btrfs_release_path(path
);
4124 path
->keep_locks
= 1;
4125 path
->search_for_split
= 1;
4126 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4127 path
->search_for_split
= 0;
4132 leaf
= path
->nodes
[0];
4133 /* if our item isn't there or got smaller, return now */
4134 if (ret
> 0 || item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4137 /* the leaf has changed, it now has room. return now */
4138 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4141 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4142 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4143 struct btrfs_file_extent_item
);
4144 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4148 btrfs_set_path_blocking(path
);
4149 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4153 path
->keep_locks
= 0;
4154 btrfs_unlock_up_safe(path
, 1);
4157 path
->keep_locks
= 0;
4161 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4162 struct btrfs_root
*root
,
4163 struct btrfs_path
*path
,
4164 struct btrfs_key
*new_key
,
4165 unsigned long split_offset
)
4167 struct extent_buffer
*leaf
;
4168 struct btrfs_item
*item
;
4169 struct btrfs_item
*new_item
;
4175 struct btrfs_disk_key disk_key
;
4177 leaf
= path
->nodes
[0];
4178 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4180 btrfs_set_path_blocking(path
);
4182 item
= btrfs_item_nr(path
->slots
[0]);
4183 orig_offset
= btrfs_item_offset(leaf
, item
);
4184 item_size
= btrfs_item_size(leaf
, item
);
4186 buf
= kmalloc(item_size
, GFP_NOFS
);
4190 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4191 path
->slots
[0]), item_size
);
4193 slot
= path
->slots
[0] + 1;
4194 nritems
= btrfs_header_nritems(leaf
);
4195 if (slot
!= nritems
) {
4196 /* shift the items */
4197 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4198 btrfs_item_nr_offset(slot
),
4199 (nritems
- slot
) * sizeof(struct btrfs_item
));
4202 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4203 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4205 new_item
= btrfs_item_nr(slot
);
4207 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4208 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4210 btrfs_set_item_offset(leaf
, item
,
4211 orig_offset
+ item_size
- split_offset
);
4212 btrfs_set_item_size(leaf
, item
, split_offset
);
4214 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4216 /* write the data for the start of the original item */
4217 write_extent_buffer(leaf
, buf
,
4218 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4221 /* write the data for the new item */
4222 write_extent_buffer(leaf
, buf
+ split_offset
,
4223 btrfs_item_ptr_offset(leaf
, slot
),
4224 item_size
- split_offset
);
4225 btrfs_mark_buffer_dirty(leaf
);
4227 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4233 * This function splits a single item into two items,
4234 * giving 'new_key' to the new item and splitting the
4235 * old one at split_offset (from the start of the item).
4237 * The path may be released by this operation. After
4238 * the split, the path is pointing to the old item. The
4239 * new item is going to be in the same node as the old one.
4241 * Note, the item being split must be smaller enough to live alone on
4242 * a tree block with room for one extra struct btrfs_item
4244 * This allows us to split the item in place, keeping a lock on the
4245 * leaf the entire time.
4247 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4248 struct btrfs_root
*root
,
4249 struct btrfs_path
*path
,
4250 struct btrfs_key
*new_key
,
4251 unsigned long split_offset
)
4254 ret
= setup_leaf_for_split(trans
, root
, path
,
4255 sizeof(struct btrfs_item
));
4259 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4264 * This function duplicate a item, giving 'new_key' to the new item.
4265 * It guarantees both items live in the same tree leaf and the new item
4266 * is contiguous with the original item.
4268 * This allows us to split file extent in place, keeping a lock on the
4269 * leaf the entire time.
4271 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4272 struct btrfs_root
*root
,
4273 struct btrfs_path
*path
,
4274 struct btrfs_key
*new_key
)
4276 struct extent_buffer
*leaf
;
4280 leaf
= path
->nodes
[0];
4281 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4282 ret
= setup_leaf_for_split(trans
, root
, path
,
4283 item_size
+ sizeof(struct btrfs_item
));
4288 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4289 item_size
, item_size
+
4290 sizeof(struct btrfs_item
), 1);
4291 leaf
= path
->nodes
[0];
4292 memcpy_extent_buffer(leaf
,
4293 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4294 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4300 * make the item pointed to by the path smaller. new_size indicates
4301 * how small to make it, and from_end tells us if we just chop bytes
4302 * off the end of the item or if we shift the item to chop bytes off
4305 void btrfs_truncate_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4306 u32 new_size
, int from_end
)
4309 struct extent_buffer
*leaf
;
4310 struct btrfs_item
*item
;
4312 unsigned int data_end
;
4313 unsigned int old_data_start
;
4314 unsigned int old_size
;
4315 unsigned int size_diff
;
4317 struct btrfs_map_token token
;
4319 btrfs_init_map_token(&token
);
4321 leaf
= path
->nodes
[0];
4322 slot
= path
->slots
[0];
4324 old_size
= btrfs_item_size_nr(leaf
, slot
);
4325 if (old_size
== new_size
)
4328 nritems
= btrfs_header_nritems(leaf
);
4329 data_end
= leaf_data_end(root
, leaf
);
4331 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4333 size_diff
= old_size
- new_size
;
4336 BUG_ON(slot
>= nritems
);
4339 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4341 /* first correct the data pointers */
4342 for (i
= slot
; i
< nritems
; i
++) {
4344 item
= btrfs_item_nr(i
);
4346 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4347 btrfs_set_token_item_offset(leaf
, item
,
4348 ioff
+ size_diff
, &token
);
4351 /* shift the data */
4353 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4354 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4355 data_end
, old_data_start
+ new_size
- data_end
);
4357 struct btrfs_disk_key disk_key
;
4360 btrfs_item_key(leaf
, &disk_key
, slot
);
4362 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4364 struct btrfs_file_extent_item
*fi
;
4366 fi
= btrfs_item_ptr(leaf
, slot
,
4367 struct btrfs_file_extent_item
);
4368 fi
= (struct btrfs_file_extent_item
*)(
4369 (unsigned long)fi
- size_diff
);
4371 if (btrfs_file_extent_type(leaf
, fi
) ==
4372 BTRFS_FILE_EXTENT_INLINE
) {
4373 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4374 memmove_extent_buffer(leaf
, ptr
,
4376 offsetof(struct btrfs_file_extent_item
,
4381 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4382 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4383 data_end
, old_data_start
- data_end
);
4385 offset
= btrfs_disk_key_offset(&disk_key
);
4386 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4387 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4389 fixup_low_keys(root
, path
, &disk_key
, 1);
4392 item
= btrfs_item_nr(slot
);
4393 btrfs_set_item_size(leaf
, item
, new_size
);
4394 btrfs_mark_buffer_dirty(leaf
);
4396 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4397 btrfs_print_leaf(root
, leaf
);
4403 * make the item pointed to by the path bigger, data_size is the added size.
4405 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4409 struct extent_buffer
*leaf
;
4410 struct btrfs_item
*item
;
4412 unsigned int data_end
;
4413 unsigned int old_data
;
4414 unsigned int old_size
;
4416 struct btrfs_map_token token
;
4418 btrfs_init_map_token(&token
);
4420 leaf
= path
->nodes
[0];
4422 nritems
= btrfs_header_nritems(leaf
);
4423 data_end
= leaf_data_end(root
, leaf
);
4425 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4426 btrfs_print_leaf(root
, leaf
);
4429 slot
= path
->slots
[0];
4430 old_data
= btrfs_item_end_nr(leaf
, slot
);
4433 if (slot
>= nritems
) {
4434 btrfs_print_leaf(root
, leaf
);
4435 printk(KERN_CRIT
"slot %d too large, nritems %d\n",
4441 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4443 /* first correct the data pointers */
4444 for (i
= slot
; i
< nritems
; i
++) {
4446 item
= btrfs_item_nr(i
);
4448 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4449 btrfs_set_token_item_offset(leaf
, item
,
4450 ioff
- data_size
, &token
);
4453 /* shift the data */
4454 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4455 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4456 data_end
, old_data
- data_end
);
4458 data_end
= old_data
;
4459 old_size
= btrfs_item_size_nr(leaf
, slot
);
4460 item
= btrfs_item_nr(slot
);
4461 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4462 btrfs_mark_buffer_dirty(leaf
);
4464 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4465 btrfs_print_leaf(root
, leaf
);
4471 * this is a helper for btrfs_insert_empty_items, the main goal here is
4472 * to save stack depth by doing the bulk of the work in a function
4473 * that doesn't call btrfs_search_slot
4475 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4476 struct btrfs_key
*cpu_key
, u32
*data_size
,
4477 u32 total_data
, u32 total_size
, int nr
)
4479 struct btrfs_item
*item
;
4482 unsigned int data_end
;
4483 struct btrfs_disk_key disk_key
;
4484 struct extent_buffer
*leaf
;
4486 struct btrfs_map_token token
;
4488 btrfs_init_map_token(&token
);
4490 leaf
= path
->nodes
[0];
4491 slot
= path
->slots
[0];
4493 nritems
= btrfs_header_nritems(leaf
);
4494 data_end
= leaf_data_end(root
, leaf
);
4496 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4497 btrfs_print_leaf(root
, leaf
);
4498 printk(KERN_CRIT
"not enough freespace need %u have %d\n",
4499 total_size
, btrfs_leaf_free_space(root
, leaf
));
4503 if (slot
!= nritems
) {
4504 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4506 if (old_data
< data_end
) {
4507 btrfs_print_leaf(root
, leaf
);
4508 printk(KERN_CRIT
"slot %d old_data %d data_end %d\n",
4509 slot
, old_data
, data_end
);
4513 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4515 /* first correct the data pointers */
4516 for (i
= slot
; i
< nritems
; i
++) {
4519 item
= btrfs_item_nr( i
);
4520 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4521 btrfs_set_token_item_offset(leaf
, item
,
4522 ioff
- total_data
, &token
);
4524 /* shift the items */
4525 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4526 btrfs_item_nr_offset(slot
),
4527 (nritems
- slot
) * sizeof(struct btrfs_item
));
4529 /* shift the data */
4530 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4531 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4532 data_end
, old_data
- data_end
);
4533 data_end
= old_data
;
4536 /* setup the item for the new data */
4537 for (i
= 0; i
< nr
; i
++) {
4538 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4539 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4540 item
= btrfs_item_nr(slot
+ i
);
4541 btrfs_set_token_item_offset(leaf
, item
,
4542 data_end
- data_size
[i
], &token
);
4543 data_end
-= data_size
[i
];
4544 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4547 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4550 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4551 fixup_low_keys(root
, path
, &disk_key
, 1);
4553 btrfs_unlock_up_safe(path
, 1);
4554 btrfs_mark_buffer_dirty(leaf
);
4556 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4557 btrfs_print_leaf(root
, leaf
);
4563 * Given a key and some data, insert items into the tree.
4564 * This does all the path init required, making room in the tree if needed.
4566 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4567 struct btrfs_root
*root
,
4568 struct btrfs_path
*path
,
4569 struct btrfs_key
*cpu_key
, u32
*data_size
,
4578 for (i
= 0; i
< nr
; i
++)
4579 total_data
+= data_size
[i
];
4581 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4582 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4588 slot
= path
->slots
[0];
4591 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4592 total_data
, total_size
, nr
);
4597 * Given a key and some data, insert an item into the tree.
4598 * This does all the path init required, making room in the tree if needed.
4600 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4601 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4605 struct btrfs_path
*path
;
4606 struct extent_buffer
*leaf
;
4609 path
= btrfs_alloc_path();
4612 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4614 leaf
= path
->nodes
[0];
4615 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4616 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4617 btrfs_mark_buffer_dirty(leaf
);
4619 btrfs_free_path(path
);
4624 * delete the pointer from a given node.
4626 * the tree should have been previously balanced so the deletion does not
4629 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4630 int level
, int slot
)
4632 struct extent_buffer
*parent
= path
->nodes
[level
];
4636 nritems
= btrfs_header_nritems(parent
);
4637 if (slot
!= nritems
- 1) {
4639 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4640 slot
+ 1, nritems
- slot
- 1);
4641 memmove_extent_buffer(parent
,
4642 btrfs_node_key_ptr_offset(slot
),
4643 btrfs_node_key_ptr_offset(slot
+ 1),
4644 sizeof(struct btrfs_key_ptr
) *
4645 (nritems
- slot
- 1));
4647 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4648 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4653 btrfs_set_header_nritems(parent
, nritems
);
4654 if (nritems
== 0 && parent
== root
->node
) {
4655 BUG_ON(btrfs_header_level(root
->node
) != 1);
4656 /* just turn the root into a leaf and break */
4657 btrfs_set_header_level(root
->node
, 0);
4658 } else if (slot
== 0) {
4659 struct btrfs_disk_key disk_key
;
4661 btrfs_node_key(parent
, &disk_key
, 0);
4662 fixup_low_keys(root
, path
, &disk_key
, level
+ 1);
4664 btrfs_mark_buffer_dirty(parent
);
4668 * a helper function to delete the leaf pointed to by path->slots[1] and
4671 * This deletes the pointer in path->nodes[1] and frees the leaf
4672 * block extent. zero is returned if it all worked out, < 0 otherwise.
4674 * The path must have already been setup for deleting the leaf, including
4675 * all the proper balancing. path->nodes[1] must be locked.
4677 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4678 struct btrfs_root
*root
,
4679 struct btrfs_path
*path
,
4680 struct extent_buffer
*leaf
)
4682 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4683 del_ptr(root
, path
, 1, path
->slots
[1]);
4686 * btrfs_free_extent is expensive, we want to make sure we
4687 * aren't holding any locks when we call it
4689 btrfs_unlock_up_safe(path
, 0);
4691 root_sub_used(root
, leaf
->len
);
4693 extent_buffer_get(leaf
);
4694 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4695 free_extent_buffer_stale(leaf
);
4698 * delete the item at the leaf level in path. If that empties
4699 * the leaf, remove it from the tree
4701 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4702 struct btrfs_path
*path
, int slot
, int nr
)
4704 struct extent_buffer
*leaf
;
4705 struct btrfs_item
*item
;
4712 struct btrfs_map_token token
;
4714 btrfs_init_map_token(&token
);
4716 leaf
= path
->nodes
[0];
4717 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4719 for (i
= 0; i
< nr
; i
++)
4720 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4722 nritems
= btrfs_header_nritems(leaf
);
4724 if (slot
+ nr
!= nritems
) {
4725 int data_end
= leaf_data_end(root
, leaf
);
4727 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4729 btrfs_leaf_data(leaf
) + data_end
,
4730 last_off
- data_end
);
4732 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4735 item
= btrfs_item_nr(i
);
4736 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4737 btrfs_set_token_item_offset(leaf
, item
,
4738 ioff
+ dsize
, &token
);
4741 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4742 btrfs_item_nr_offset(slot
+ nr
),
4743 sizeof(struct btrfs_item
) *
4744 (nritems
- slot
- nr
));
4746 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4749 /* delete the leaf if we've emptied it */
4751 if (leaf
== root
->node
) {
4752 btrfs_set_header_level(leaf
, 0);
4754 btrfs_set_path_blocking(path
);
4755 clean_tree_block(trans
, root
, leaf
);
4756 btrfs_del_leaf(trans
, root
, path
, leaf
);
4759 int used
= leaf_space_used(leaf
, 0, nritems
);
4761 struct btrfs_disk_key disk_key
;
4763 btrfs_item_key(leaf
, &disk_key
, 0);
4764 fixup_low_keys(root
, path
, &disk_key
, 1);
4767 /* delete the leaf if it is mostly empty */
4768 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
4769 /* push_leaf_left fixes the path.
4770 * make sure the path still points to our leaf
4771 * for possible call to del_ptr below
4773 slot
= path
->slots
[1];
4774 extent_buffer_get(leaf
);
4776 btrfs_set_path_blocking(path
);
4777 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
4779 if (wret
< 0 && wret
!= -ENOSPC
)
4782 if (path
->nodes
[0] == leaf
&&
4783 btrfs_header_nritems(leaf
)) {
4784 wret
= push_leaf_right(trans
, root
, path
, 1,
4786 if (wret
< 0 && wret
!= -ENOSPC
)
4790 if (btrfs_header_nritems(leaf
) == 0) {
4791 path
->slots
[1] = slot
;
4792 btrfs_del_leaf(trans
, root
, path
, leaf
);
4793 free_extent_buffer(leaf
);
4796 /* if we're still in the path, make sure
4797 * we're dirty. Otherwise, one of the
4798 * push_leaf functions must have already
4799 * dirtied this buffer
4801 if (path
->nodes
[0] == leaf
)
4802 btrfs_mark_buffer_dirty(leaf
);
4803 free_extent_buffer(leaf
);
4806 btrfs_mark_buffer_dirty(leaf
);
4813 * search the tree again to find a leaf with lesser keys
4814 * returns 0 if it found something or 1 if there are no lesser leaves.
4815 * returns < 0 on io errors.
4817 * This may release the path, and so you may lose any locks held at the
4820 static int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
4822 struct btrfs_key key
;
4823 struct btrfs_disk_key found_key
;
4826 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
4828 if (key
.offset
> 0) {
4830 } else if (key
.type
> 0) {
4832 key
.offset
= (u64
)-1;
4833 } else if (key
.objectid
> 0) {
4836 key
.offset
= (u64
)-1;
4841 btrfs_release_path(path
);
4842 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4845 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
4846 ret
= comp_keys(&found_key
, &key
);
4853 * A helper function to walk down the tree starting at min_key, and looking
4854 * for nodes or leaves that are have a minimum transaction id.
4855 * This is used by the btree defrag code, and tree logging
4857 * This does not cow, but it does stuff the starting key it finds back
4858 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4859 * key and get a writable path.
4861 * This does lock as it descends, and path->keep_locks should be set
4862 * to 1 by the caller.
4864 * This honors path->lowest_level to prevent descent past a given level
4867 * min_trans indicates the oldest transaction that you are interested
4868 * in walking through. Any nodes or leaves older than min_trans are
4869 * skipped over (without reading them).
4871 * returns zero if something useful was found, < 0 on error and 1 if there
4872 * was nothing in the tree that matched the search criteria.
4874 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
4875 struct btrfs_path
*path
,
4878 struct extent_buffer
*cur
;
4879 struct btrfs_key found_key
;
4886 WARN_ON(!path
->keep_locks
);
4888 cur
= btrfs_read_lock_root_node(root
);
4889 level
= btrfs_header_level(cur
);
4890 WARN_ON(path
->nodes
[level
]);
4891 path
->nodes
[level
] = cur
;
4892 path
->locks
[level
] = BTRFS_READ_LOCK
;
4894 if (btrfs_header_generation(cur
) < min_trans
) {
4899 nritems
= btrfs_header_nritems(cur
);
4900 level
= btrfs_header_level(cur
);
4901 sret
= bin_search(cur
, min_key
, level
, &slot
);
4903 /* at the lowest level, we're done, setup the path and exit */
4904 if (level
== path
->lowest_level
) {
4905 if (slot
>= nritems
)
4908 path
->slots
[level
] = slot
;
4909 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
4912 if (sret
&& slot
> 0)
4915 * check this node pointer against the min_trans parameters.
4916 * If it is too old, old, skip to the next one.
4918 while (slot
< nritems
) {
4921 gen
= btrfs_node_ptr_generation(cur
, slot
);
4922 if (gen
< min_trans
) {
4930 * we didn't find a candidate key in this node, walk forward
4931 * and find another one
4933 if (slot
>= nritems
) {
4934 path
->slots
[level
] = slot
;
4935 btrfs_set_path_blocking(path
);
4936 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
4939 btrfs_release_path(path
);
4945 /* save our key for returning back */
4946 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
4947 path
->slots
[level
] = slot
;
4948 if (level
== path
->lowest_level
) {
4950 unlock_up(path
, level
, 1, 0, NULL
);
4953 btrfs_set_path_blocking(path
);
4954 cur
= read_node_slot(root
, cur
, slot
);
4955 BUG_ON(!cur
); /* -ENOMEM */
4957 btrfs_tree_read_lock(cur
);
4959 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
4960 path
->nodes
[level
- 1] = cur
;
4961 unlock_up(path
, level
, 1, 0, NULL
);
4962 btrfs_clear_path_blocking(path
, NULL
, 0);
4966 memcpy(min_key
, &found_key
, sizeof(found_key
));
4967 btrfs_set_path_blocking(path
);
4971 static void tree_move_down(struct btrfs_root
*root
,
4972 struct btrfs_path
*path
,
4973 int *level
, int root_level
)
4975 BUG_ON(*level
== 0);
4976 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
4977 path
->slots
[*level
]);
4978 path
->slots
[*level
- 1] = 0;
4982 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
4983 struct btrfs_path
*path
,
4984 int *level
, int root_level
)
4988 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
4990 path
->slots
[*level
]++;
4992 while (path
->slots
[*level
] >= nritems
) {
4993 if (*level
== root_level
)
4997 path
->slots
[*level
] = 0;
4998 free_extent_buffer(path
->nodes
[*level
]);
4999 path
->nodes
[*level
] = NULL
;
5001 path
->slots
[*level
]++;
5003 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5010 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5013 static int tree_advance(struct btrfs_root
*root
,
5014 struct btrfs_path
*path
,
5015 int *level
, int root_level
,
5017 struct btrfs_key
*key
)
5021 if (*level
== 0 || !allow_down
) {
5022 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5024 tree_move_down(root
, path
, level
, root_level
);
5029 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5030 path
->slots
[*level
]);
5032 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5033 path
->slots
[*level
]);
5038 static int tree_compare_item(struct btrfs_root
*left_root
,
5039 struct btrfs_path
*left_path
,
5040 struct btrfs_path
*right_path
,
5045 unsigned long off1
, off2
;
5047 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5048 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5052 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5053 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5054 right_path
->slots
[0]);
5056 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5058 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5065 #define ADVANCE_ONLY_NEXT -1
5068 * This function compares two trees and calls the provided callback for
5069 * every changed/new/deleted item it finds.
5070 * If shared tree blocks are encountered, whole subtrees are skipped, making
5071 * the compare pretty fast on snapshotted subvolumes.
5073 * This currently works on commit roots only. As commit roots are read only,
5074 * we don't do any locking. The commit roots are protected with transactions.
5075 * Transactions are ended and rejoined when a commit is tried in between.
5077 * This function checks for modifications done to the trees while comparing.
5078 * If it detects a change, it aborts immediately.
5080 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5081 struct btrfs_root
*right_root
,
5082 btrfs_changed_cb_t changed_cb
, void *ctx
)
5086 struct btrfs_trans_handle
*trans
= NULL
;
5087 struct btrfs_path
*left_path
= NULL
;
5088 struct btrfs_path
*right_path
= NULL
;
5089 struct btrfs_key left_key
;
5090 struct btrfs_key right_key
;
5091 char *tmp_buf
= NULL
;
5092 int left_root_level
;
5093 int right_root_level
;
5096 int left_end_reached
;
5097 int right_end_reached
;
5102 u64 left_start_ctransid
;
5103 u64 right_start_ctransid
;
5106 left_path
= btrfs_alloc_path();
5111 right_path
= btrfs_alloc_path();
5117 tmp_buf
= kmalloc(left_root
->leafsize
, GFP_NOFS
);
5123 left_path
->search_commit_root
= 1;
5124 left_path
->skip_locking
= 1;
5125 right_path
->search_commit_root
= 1;
5126 right_path
->skip_locking
= 1;
5128 spin_lock(&left_root
->root_item_lock
);
5129 left_start_ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5130 spin_unlock(&left_root
->root_item_lock
);
5132 spin_lock(&right_root
->root_item_lock
);
5133 right_start_ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5134 spin_unlock(&right_root
->root_item_lock
);
5136 trans
= btrfs_join_transaction(left_root
);
5137 if (IS_ERR(trans
)) {
5138 ret
= PTR_ERR(trans
);
5144 * Strategy: Go to the first items of both trees. Then do
5146 * If both trees are at level 0
5147 * Compare keys of current items
5148 * If left < right treat left item as new, advance left tree
5150 * If left > right treat right item as deleted, advance right tree
5152 * If left == right do deep compare of items, treat as changed if
5153 * needed, advance both trees and repeat
5154 * If both trees are at the same level but not at level 0
5155 * Compare keys of current nodes/leafs
5156 * If left < right advance left tree and repeat
5157 * If left > right advance right tree and repeat
5158 * If left == right compare blockptrs of the next nodes/leafs
5159 * If they match advance both trees but stay at the same level
5161 * If they don't match advance both trees while allowing to go
5163 * If tree levels are different
5164 * Advance the tree that needs it and repeat
5166 * Advancing a tree means:
5167 * If we are at level 0, try to go to the next slot. If that's not
5168 * possible, go one level up and repeat. Stop when we found a level
5169 * where we could go to the next slot. We may at this point be on a
5172 * If we are not at level 0 and not on shared tree blocks, go one
5175 * If we are not at level 0 and on shared tree blocks, go one slot to
5176 * the right if possible or go up and right.
5179 left_level
= btrfs_header_level(left_root
->commit_root
);
5180 left_root_level
= left_level
;
5181 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5182 extent_buffer_get(left_path
->nodes
[left_level
]);
5184 right_level
= btrfs_header_level(right_root
->commit_root
);
5185 right_root_level
= right_level
;
5186 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5187 extent_buffer_get(right_path
->nodes
[right_level
]);
5189 if (left_level
== 0)
5190 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5191 &left_key
, left_path
->slots
[left_level
]);
5193 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5194 &left_key
, left_path
->slots
[left_level
]);
5195 if (right_level
== 0)
5196 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5197 &right_key
, right_path
->slots
[right_level
]);
5199 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5200 &right_key
, right_path
->slots
[right_level
]);
5202 left_end_reached
= right_end_reached
= 0;
5203 advance_left
= advance_right
= 0;
5207 * We need to make sure the transaction does not get committed
5208 * while we do anything on commit roots. This means, we need to
5209 * join and leave transactions for every item that we process.
5211 if (trans
&& btrfs_should_end_transaction(trans
, left_root
)) {
5212 btrfs_release_path(left_path
);
5213 btrfs_release_path(right_path
);
5215 ret
= btrfs_end_transaction(trans
, left_root
);
5220 /* now rejoin the transaction */
5222 trans
= btrfs_join_transaction(left_root
);
5223 if (IS_ERR(trans
)) {
5224 ret
= PTR_ERR(trans
);
5229 spin_lock(&left_root
->root_item_lock
);
5230 ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5231 spin_unlock(&left_root
->root_item_lock
);
5232 if (ctransid
!= left_start_ctransid
)
5233 left_start_ctransid
= 0;
5235 spin_lock(&right_root
->root_item_lock
);
5236 ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5237 spin_unlock(&right_root
->root_item_lock
);
5238 if (ctransid
!= right_start_ctransid
)
5239 right_start_ctransid
= 0;
5241 if (!left_start_ctransid
|| !right_start_ctransid
) {
5242 WARN(1, KERN_WARNING
5243 "btrfs: btrfs_compare_tree detected "
5244 "a change in one of the trees while "
5245 "iterating. This is probably a "
5252 * the commit root may have changed, so start again
5255 left_path
->lowest_level
= left_level
;
5256 right_path
->lowest_level
= right_level
;
5257 ret
= btrfs_search_slot(NULL
, left_root
,
5258 &left_key
, left_path
, 0, 0);
5261 ret
= btrfs_search_slot(NULL
, right_root
,
5262 &right_key
, right_path
, 0, 0);
5267 if (advance_left
&& !left_end_reached
) {
5268 ret
= tree_advance(left_root
, left_path
, &left_level
,
5270 advance_left
!= ADVANCE_ONLY_NEXT
,
5273 left_end_reached
= ADVANCE
;
5276 if (advance_right
&& !right_end_reached
) {
5277 ret
= tree_advance(right_root
, right_path
, &right_level
,
5279 advance_right
!= ADVANCE_ONLY_NEXT
,
5282 right_end_reached
= ADVANCE
;
5286 if (left_end_reached
&& right_end_reached
) {
5289 } else if (left_end_reached
) {
5290 if (right_level
== 0) {
5291 ret
= changed_cb(left_root
, right_root
,
5292 left_path
, right_path
,
5294 BTRFS_COMPARE_TREE_DELETED
,
5299 advance_right
= ADVANCE
;
5301 } else if (right_end_reached
) {
5302 if (left_level
== 0) {
5303 ret
= changed_cb(left_root
, right_root
,
5304 left_path
, right_path
,
5306 BTRFS_COMPARE_TREE_NEW
,
5311 advance_left
= ADVANCE
;
5315 if (left_level
== 0 && right_level
== 0) {
5316 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5318 ret
= changed_cb(left_root
, right_root
,
5319 left_path
, right_path
,
5321 BTRFS_COMPARE_TREE_NEW
,
5325 advance_left
= ADVANCE
;
5326 } else if (cmp
> 0) {
5327 ret
= changed_cb(left_root
, right_root
,
5328 left_path
, right_path
,
5330 BTRFS_COMPARE_TREE_DELETED
,
5334 advance_right
= ADVANCE
;
5336 enum btrfs_compare_tree_result cmp
;
5338 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5339 ret
= tree_compare_item(left_root
, left_path
,
5340 right_path
, tmp_buf
);
5342 cmp
= BTRFS_COMPARE_TREE_CHANGED
;
5344 cmp
= BTRFS_COMPARE_TREE_SAME
;
5345 ret
= changed_cb(left_root
, right_root
,
5346 left_path
, right_path
,
5347 &left_key
, cmp
, ctx
);
5350 advance_left
= ADVANCE
;
5351 advance_right
= ADVANCE
;
5353 } else if (left_level
== right_level
) {
5354 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5356 advance_left
= ADVANCE
;
5357 } else if (cmp
> 0) {
5358 advance_right
= ADVANCE
;
5360 left_blockptr
= btrfs_node_blockptr(
5361 left_path
->nodes
[left_level
],
5362 left_path
->slots
[left_level
]);
5363 right_blockptr
= btrfs_node_blockptr(
5364 right_path
->nodes
[right_level
],
5365 right_path
->slots
[right_level
]);
5366 if (left_blockptr
== right_blockptr
) {
5368 * As we're on a shared block, don't
5369 * allow to go deeper.
5371 advance_left
= ADVANCE_ONLY_NEXT
;
5372 advance_right
= ADVANCE_ONLY_NEXT
;
5374 advance_left
= ADVANCE
;
5375 advance_right
= ADVANCE
;
5378 } else if (left_level
< right_level
) {
5379 advance_right
= ADVANCE
;
5381 advance_left
= ADVANCE
;
5386 btrfs_free_path(left_path
);
5387 btrfs_free_path(right_path
);
5392 ret
= btrfs_end_transaction(trans
, left_root
);
5394 btrfs_end_transaction(trans
, left_root
);
5401 * this is similar to btrfs_next_leaf, but does not try to preserve
5402 * and fixup the path. It looks for and returns the next key in the
5403 * tree based on the current path and the min_trans parameters.
5405 * 0 is returned if another key is found, < 0 if there are any errors
5406 * and 1 is returned if there are no higher keys in the tree
5408 * path->keep_locks should be set to 1 on the search made before
5409 * calling this function.
5411 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5412 struct btrfs_key
*key
, int level
, u64 min_trans
)
5415 struct extent_buffer
*c
;
5417 WARN_ON(!path
->keep_locks
);
5418 while (level
< BTRFS_MAX_LEVEL
) {
5419 if (!path
->nodes
[level
])
5422 slot
= path
->slots
[level
] + 1;
5423 c
= path
->nodes
[level
];
5425 if (slot
>= btrfs_header_nritems(c
)) {
5428 struct btrfs_key cur_key
;
5429 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5430 !path
->nodes
[level
+ 1])
5433 if (path
->locks
[level
+ 1]) {
5438 slot
= btrfs_header_nritems(c
) - 1;
5440 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5442 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5444 orig_lowest
= path
->lowest_level
;
5445 btrfs_release_path(path
);
5446 path
->lowest_level
= level
;
5447 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5449 path
->lowest_level
= orig_lowest
;
5453 c
= path
->nodes
[level
];
5454 slot
= path
->slots
[level
];
5461 btrfs_item_key_to_cpu(c
, key
, slot
);
5463 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5465 if (gen
< min_trans
) {
5469 btrfs_node_key_to_cpu(c
, key
, slot
);
5477 * search the tree again to find a leaf with greater keys
5478 * returns 0 if it found something or 1 if there are no greater leaves.
5479 * returns < 0 on io errors.
5481 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5483 return btrfs_next_old_leaf(root
, path
, 0);
5486 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5491 struct extent_buffer
*c
;
5492 struct extent_buffer
*next
;
5493 struct btrfs_key key
;
5496 int old_spinning
= path
->leave_spinning
;
5497 int next_rw_lock
= 0;
5499 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5503 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5508 btrfs_release_path(path
);
5510 path
->keep_locks
= 1;
5511 path
->leave_spinning
= 1;
5514 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5516 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5517 path
->keep_locks
= 0;
5522 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5524 * by releasing the path above we dropped all our locks. A balance
5525 * could have added more items next to the key that used to be
5526 * at the very end of the block. So, check again here and
5527 * advance the path if there are now more items available.
5529 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5536 while (level
< BTRFS_MAX_LEVEL
) {
5537 if (!path
->nodes
[level
]) {
5542 slot
= path
->slots
[level
] + 1;
5543 c
= path
->nodes
[level
];
5544 if (slot
>= btrfs_header_nritems(c
)) {
5546 if (level
== BTRFS_MAX_LEVEL
) {
5554 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5555 free_extent_buffer(next
);
5559 next_rw_lock
= path
->locks
[level
];
5560 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5566 btrfs_release_path(path
);
5570 if (!path
->skip_locking
) {
5571 ret
= btrfs_try_tree_read_lock(next
);
5572 if (!ret
&& time_seq
) {
5574 * If we don't get the lock, we may be racing
5575 * with push_leaf_left, holding that lock while
5576 * itself waiting for the leaf we've currently
5577 * locked. To solve this situation, we give up
5578 * on our lock and cycle.
5580 free_extent_buffer(next
);
5581 btrfs_release_path(path
);
5586 btrfs_set_path_blocking(path
);
5587 btrfs_tree_read_lock(next
);
5588 btrfs_clear_path_blocking(path
, next
,
5591 next_rw_lock
= BTRFS_READ_LOCK
;
5595 path
->slots
[level
] = slot
;
5598 c
= path
->nodes
[level
];
5599 if (path
->locks
[level
])
5600 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5602 free_extent_buffer(c
);
5603 path
->nodes
[level
] = next
;
5604 path
->slots
[level
] = 0;
5605 if (!path
->skip_locking
)
5606 path
->locks
[level
] = next_rw_lock
;
5610 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5616 btrfs_release_path(path
);
5620 if (!path
->skip_locking
) {
5621 ret
= btrfs_try_tree_read_lock(next
);
5623 btrfs_set_path_blocking(path
);
5624 btrfs_tree_read_lock(next
);
5625 btrfs_clear_path_blocking(path
, next
,
5628 next_rw_lock
= BTRFS_READ_LOCK
;
5633 unlock_up(path
, 0, 1, 0, NULL
);
5634 path
->leave_spinning
= old_spinning
;
5636 btrfs_set_path_blocking(path
);
5642 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5643 * searching until it gets past min_objectid or finds an item of 'type'
5645 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5647 int btrfs_previous_item(struct btrfs_root
*root
,
5648 struct btrfs_path
*path
, u64 min_objectid
,
5651 struct btrfs_key found_key
;
5652 struct extent_buffer
*leaf
;
5657 if (path
->slots
[0] == 0) {
5658 btrfs_set_path_blocking(path
);
5659 ret
= btrfs_prev_leaf(root
, path
);
5665 leaf
= path
->nodes
[0];
5666 nritems
= btrfs_header_nritems(leaf
);
5669 if (path
->slots
[0] == nritems
)
5672 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5673 if (found_key
.objectid
< min_objectid
)
5675 if (found_key
.type
== type
)
5677 if (found_key
.objectid
== min_objectid
&&
5678 found_key
.type
< type
)