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 int tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
43 struct extent_buffer
*eb
);
45 struct btrfs_path
*btrfs_alloc_path(void)
47 struct btrfs_path
*path
;
48 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
59 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
60 if (!p
->nodes
[i
] || !p
->locks
[i
])
62 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
63 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
64 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
65 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
66 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
79 struct extent_buffer
*held
, int held_rw
)
84 btrfs_set_lock_blocking_rw(held
, held_rw
);
85 if (held_rw
== BTRFS_WRITE_LOCK
)
86 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
87 else if (held_rw
== BTRFS_READ_LOCK
)
88 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
90 btrfs_set_path_blocking(p
);
92 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
93 if (p
->nodes
[i
] && p
->locks
[i
]) {
94 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
95 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
96 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
97 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
98 p
->locks
[i
] = BTRFS_READ_LOCK
;
103 btrfs_clear_lock_blocking_rw(held
, held_rw
);
106 /* this also releases the path */
107 void btrfs_free_path(struct btrfs_path
*p
)
111 btrfs_release_path(p
);
112 kmem_cache_free(btrfs_path_cachep
, p
);
116 * path release drops references on the extent buffers in the path
117 * and it drops any locks held by this path
119 * It is safe to call this on paths that no locks or extent buffers held.
121 noinline
void btrfs_release_path(struct btrfs_path
*p
)
125 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
130 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
133 free_extent_buffer(p
->nodes
[i
]);
139 * safely gets a reference on the root node of a tree. A lock
140 * is not taken, so a concurrent writer may put a different node
141 * at the root of the tree. See btrfs_lock_root_node for the
144 * The extent buffer returned by this has a reference taken, so
145 * it won't disappear. It may stop being the root of the tree
146 * at any time because there are no locks held.
148 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
150 struct extent_buffer
*eb
;
154 eb
= rcu_dereference(root
->node
);
157 * RCU really hurts here, we could free up the root node because
158 * it was cow'ed but we may not get the new root node yet so do
159 * the inc_not_zero dance and if it doesn't work then
160 * synchronize_rcu and try again.
162 if (atomic_inc_not_zero(&eb
->refs
)) {
172 /* loop around taking references on and locking the root node of the
173 * tree until you end up with a lock on the root. A locked buffer
174 * is returned, with a reference held.
176 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
178 struct extent_buffer
*eb
;
181 eb
= btrfs_root_node(root
);
183 if (eb
== root
->node
)
185 btrfs_tree_unlock(eb
);
186 free_extent_buffer(eb
);
191 /* loop around taking references on and locking the root node of the
192 * tree until you end up with a lock on the root. A locked buffer
193 * is returned, with a reference held.
195 static struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
197 struct extent_buffer
*eb
;
200 eb
= btrfs_root_node(root
);
201 btrfs_tree_read_lock(eb
);
202 if (eb
== root
->node
)
204 btrfs_tree_read_unlock(eb
);
205 free_extent_buffer(eb
);
210 /* cowonly root (everything not a reference counted cow subvolume), just get
211 * put onto a simple dirty list. transaction.c walks this to make sure they
212 * get properly updated on disk.
214 static void add_root_to_dirty_list(struct btrfs_root
*root
)
216 if (test_bit(BTRFS_ROOT_DIRTY
, &root
->state
) ||
217 !test_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
))
220 spin_lock(&root
->fs_info
->trans_lock
);
221 if (!test_and_set_bit(BTRFS_ROOT_DIRTY
, &root
->state
)) {
222 /* Want the extent tree to be the last on the list */
223 if (root
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
224 list_move_tail(&root
->dirty_list
,
225 &root
->fs_info
->dirty_cowonly_roots
);
227 list_move(&root
->dirty_list
,
228 &root
->fs_info
->dirty_cowonly_roots
);
230 spin_unlock(&root
->fs_info
->trans_lock
);
234 * used by snapshot creation to make a copy of a root for a tree with
235 * a given objectid. The buffer with the new root node is returned in
236 * cow_ret, and this func returns zero on success or a negative error code.
238 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
239 struct btrfs_root
*root
,
240 struct extent_buffer
*buf
,
241 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
243 struct extent_buffer
*cow
;
246 struct btrfs_disk_key disk_key
;
248 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
249 trans
->transid
!= root
->fs_info
->running_transaction
->transid
);
250 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
251 trans
->transid
!= root
->last_trans
);
253 level
= btrfs_header_level(buf
);
255 btrfs_item_key(buf
, &disk_key
, 0);
257 btrfs_node_key(buf
, &disk_key
, 0);
259 cow
= btrfs_alloc_tree_block(trans
, root
, 0, new_root_objectid
,
260 &disk_key
, level
, buf
->start
, 0);
264 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
265 btrfs_set_header_bytenr(cow
, cow
->start
);
266 btrfs_set_header_generation(cow
, trans
->transid
);
267 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
268 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
269 BTRFS_HEADER_FLAG_RELOC
);
270 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
271 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
273 btrfs_set_header_owner(cow
, new_root_objectid
);
275 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
278 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
279 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
280 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
282 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
287 btrfs_mark_buffer_dirty(cow
);
296 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
297 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
299 MOD_LOG_ROOT_REPLACE
,
302 struct tree_mod_move
{
307 struct tree_mod_root
{
312 struct tree_mod_elem
{
314 u64 index
; /* shifted logical */
318 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
321 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
324 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
325 struct btrfs_disk_key key
;
328 /* this is used for op == MOD_LOG_MOVE_KEYS */
329 struct tree_mod_move move
;
331 /* this is used for op == MOD_LOG_ROOT_REPLACE */
332 struct tree_mod_root old_root
;
335 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
337 read_lock(&fs_info
->tree_mod_log_lock
);
340 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
342 read_unlock(&fs_info
->tree_mod_log_lock
);
345 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
347 write_lock(&fs_info
->tree_mod_log_lock
);
350 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
352 write_unlock(&fs_info
->tree_mod_log_lock
);
356 * Pull a new tree mod seq number for our operation.
358 static inline u64
btrfs_inc_tree_mod_seq(struct btrfs_fs_info
*fs_info
)
360 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
364 * This adds a new blocker to the tree mod log's blocker list if the @elem
365 * passed does not already have a sequence number set. So when a caller expects
366 * to record tree modifications, it should ensure to set elem->seq to zero
367 * before calling btrfs_get_tree_mod_seq.
368 * Returns a fresh, unused tree log modification sequence number, even if no new
371 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
372 struct seq_list
*elem
)
374 tree_mod_log_write_lock(fs_info
);
375 spin_lock(&fs_info
->tree_mod_seq_lock
);
377 elem
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
378 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
380 spin_unlock(&fs_info
->tree_mod_seq_lock
);
381 tree_mod_log_write_unlock(fs_info
);
386 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
387 struct seq_list
*elem
)
389 struct rb_root
*tm_root
;
390 struct rb_node
*node
;
391 struct rb_node
*next
;
392 struct seq_list
*cur_elem
;
393 struct tree_mod_elem
*tm
;
394 u64 min_seq
= (u64
)-1;
395 u64 seq_putting
= elem
->seq
;
400 spin_lock(&fs_info
->tree_mod_seq_lock
);
401 list_del(&elem
->list
);
404 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
405 if (cur_elem
->seq
< min_seq
) {
406 if (seq_putting
> cur_elem
->seq
) {
408 * blocker with lower sequence number exists, we
409 * cannot remove anything from the log
411 spin_unlock(&fs_info
->tree_mod_seq_lock
);
414 min_seq
= cur_elem
->seq
;
417 spin_unlock(&fs_info
->tree_mod_seq_lock
);
420 * anything that's lower than the lowest existing (read: blocked)
421 * sequence number can be removed from the tree.
423 tree_mod_log_write_lock(fs_info
);
424 tm_root
= &fs_info
->tree_mod_log
;
425 for (node
= rb_first(tm_root
); node
; node
= next
) {
426 next
= rb_next(node
);
427 tm
= container_of(node
, struct tree_mod_elem
, node
);
428 if (tm
->seq
> min_seq
)
430 rb_erase(node
, tm_root
);
433 tree_mod_log_write_unlock(fs_info
);
437 * key order of the log:
440 * the index is the shifted logical of the *new* root node for root replace
441 * operations, or the shifted logical of the affected block for all other
444 * Note: must be called with write lock (tree_mod_log_write_lock).
447 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
449 struct rb_root
*tm_root
;
450 struct rb_node
**new;
451 struct rb_node
*parent
= NULL
;
452 struct tree_mod_elem
*cur
;
456 tm
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
458 tm_root
= &fs_info
->tree_mod_log
;
459 new = &tm_root
->rb_node
;
461 cur
= container_of(*new, struct tree_mod_elem
, node
);
463 if (cur
->index
< tm
->index
)
464 new = &((*new)->rb_left
);
465 else if (cur
->index
> tm
->index
)
466 new = &((*new)->rb_right
);
467 else if (cur
->seq
< tm
->seq
)
468 new = &((*new)->rb_left
);
469 else if (cur
->seq
> tm
->seq
)
470 new = &((*new)->rb_right
);
475 rb_link_node(&tm
->node
, parent
, new);
476 rb_insert_color(&tm
->node
, tm_root
);
481 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
482 * returns zero with the tree_mod_log_lock acquired. The caller must hold
483 * this until all tree mod log insertions are recorded in the rb tree and then
484 * call tree_mod_log_write_unlock() to release.
486 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
487 struct extent_buffer
*eb
) {
489 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
491 if (eb
&& btrfs_header_level(eb
) == 0)
494 tree_mod_log_write_lock(fs_info
);
495 if (list_empty(&(fs_info
)->tree_mod_seq_list
)) {
496 tree_mod_log_write_unlock(fs_info
);
503 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
504 static inline int tree_mod_need_log(const struct btrfs_fs_info
*fs_info
,
505 struct extent_buffer
*eb
)
508 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
510 if (eb
&& btrfs_header_level(eb
) == 0)
516 static struct tree_mod_elem
*
517 alloc_tree_mod_elem(struct extent_buffer
*eb
, int slot
,
518 enum mod_log_op op
, gfp_t flags
)
520 struct tree_mod_elem
*tm
;
522 tm
= kzalloc(sizeof(*tm
), flags
);
526 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
527 if (op
!= MOD_LOG_KEY_ADD
) {
528 btrfs_node_key(eb
, &tm
->key
, slot
);
529 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
533 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
534 RB_CLEAR_NODE(&tm
->node
);
540 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
541 struct extent_buffer
*eb
, int slot
,
542 enum mod_log_op op
, gfp_t flags
)
544 struct tree_mod_elem
*tm
;
547 if (!tree_mod_need_log(fs_info
, eb
))
550 tm
= alloc_tree_mod_elem(eb
, slot
, op
, flags
);
554 if (tree_mod_dont_log(fs_info
, eb
)) {
559 ret
= __tree_mod_log_insert(fs_info
, tm
);
560 tree_mod_log_write_unlock(fs_info
);
568 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
569 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
570 int nr_items
, gfp_t flags
)
572 struct tree_mod_elem
*tm
= NULL
;
573 struct tree_mod_elem
**tm_list
= NULL
;
578 if (!tree_mod_need_log(fs_info
, eb
))
581 tm_list
= kcalloc(nr_items
, sizeof(struct tree_mod_elem
*), flags
);
585 tm
= kzalloc(sizeof(*tm
), flags
);
591 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
593 tm
->move
.dst_slot
= dst_slot
;
594 tm
->move
.nr_items
= nr_items
;
595 tm
->op
= MOD_LOG_MOVE_KEYS
;
597 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
598 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
+ dst_slot
,
599 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, flags
);
606 if (tree_mod_dont_log(fs_info
, eb
))
611 * When we override something during the move, we log these removals.
612 * This can only happen when we move towards the beginning of the
613 * buffer, i.e. dst_slot < src_slot.
615 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
616 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
621 ret
= __tree_mod_log_insert(fs_info
, tm
);
624 tree_mod_log_write_unlock(fs_info
);
629 for (i
= 0; i
< nr_items
; i
++) {
630 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
631 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
635 tree_mod_log_write_unlock(fs_info
);
643 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
644 struct tree_mod_elem
**tm_list
,
650 for (i
= nritems
- 1; i
>= 0; i
--) {
651 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
653 for (j
= nritems
- 1; j
> i
; j
--)
654 rb_erase(&tm_list
[j
]->node
,
655 &fs_info
->tree_mod_log
);
664 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
665 struct extent_buffer
*old_root
,
666 struct extent_buffer
*new_root
, gfp_t flags
,
669 struct tree_mod_elem
*tm
= NULL
;
670 struct tree_mod_elem
**tm_list
= NULL
;
675 if (!tree_mod_need_log(fs_info
, NULL
))
678 if (log_removal
&& btrfs_header_level(old_root
) > 0) {
679 nritems
= btrfs_header_nritems(old_root
);
680 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*),
686 for (i
= 0; i
< nritems
; i
++) {
687 tm_list
[i
] = alloc_tree_mod_elem(old_root
, i
,
688 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, flags
);
696 tm
= kzalloc(sizeof(*tm
), flags
);
702 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
703 tm
->old_root
.logical
= old_root
->start
;
704 tm
->old_root
.level
= btrfs_header_level(old_root
);
705 tm
->generation
= btrfs_header_generation(old_root
);
706 tm
->op
= MOD_LOG_ROOT_REPLACE
;
708 if (tree_mod_dont_log(fs_info
, NULL
))
712 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
714 ret
= __tree_mod_log_insert(fs_info
, tm
);
716 tree_mod_log_write_unlock(fs_info
);
725 for (i
= 0; i
< nritems
; i
++)
734 static struct tree_mod_elem
*
735 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
738 struct rb_root
*tm_root
;
739 struct rb_node
*node
;
740 struct tree_mod_elem
*cur
= NULL
;
741 struct tree_mod_elem
*found
= NULL
;
742 u64 index
= start
>> PAGE_CACHE_SHIFT
;
744 tree_mod_log_read_lock(fs_info
);
745 tm_root
= &fs_info
->tree_mod_log
;
746 node
= tm_root
->rb_node
;
748 cur
= container_of(node
, struct tree_mod_elem
, node
);
749 if (cur
->index
< index
) {
750 node
= node
->rb_left
;
751 } else if (cur
->index
> index
) {
752 node
= node
->rb_right
;
753 } else if (cur
->seq
< min_seq
) {
754 node
= node
->rb_left
;
755 } else if (!smallest
) {
756 /* we want the node with the highest seq */
758 BUG_ON(found
->seq
> cur
->seq
);
760 node
= node
->rb_left
;
761 } else if (cur
->seq
> min_seq
) {
762 /* we want the node with the smallest seq */
764 BUG_ON(found
->seq
< cur
->seq
);
766 node
= node
->rb_right
;
772 tree_mod_log_read_unlock(fs_info
);
778 * this returns the element from the log with the smallest time sequence
779 * value that's in the log (the oldest log item). any element with a time
780 * sequence lower than min_seq will be ignored.
782 static struct tree_mod_elem
*
783 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
786 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
790 * this returns the element from the log with the largest time sequence
791 * value that's in the log (the most recent log item). any element with
792 * a time sequence lower than min_seq will be ignored.
794 static struct tree_mod_elem
*
795 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
797 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
801 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
802 struct extent_buffer
*src
, unsigned long dst_offset
,
803 unsigned long src_offset
, int nr_items
)
806 struct tree_mod_elem
**tm_list
= NULL
;
807 struct tree_mod_elem
**tm_list_add
, **tm_list_rem
;
811 if (!tree_mod_need_log(fs_info
, NULL
))
814 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
817 tm_list
= kcalloc(nr_items
* 2, sizeof(struct tree_mod_elem
*),
822 tm_list_add
= tm_list
;
823 tm_list_rem
= tm_list
+ nr_items
;
824 for (i
= 0; i
< nr_items
; i
++) {
825 tm_list_rem
[i
] = alloc_tree_mod_elem(src
, i
+ src_offset
,
826 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
827 if (!tm_list_rem
[i
]) {
832 tm_list_add
[i
] = alloc_tree_mod_elem(dst
, i
+ dst_offset
,
833 MOD_LOG_KEY_ADD
, GFP_NOFS
);
834 if (!tm_list_add
[i
]) {
840 if (tree_mod_dont_log(fs_info
, NULL
))
844 for (i
= 0; i
< nr_items
; i
++) {
845 ret
= __tree_mod_log_insert(fs_info
, tm_list_rem
[i
]);
848 ret
= __tree_mod_log_insert(fs_info
, tm_list_add
[i
]);
853 tree_mod_log_write_unlock(fs_info
);
859 for (i
= 0; i
< nr_items
* 2; i
++) {
860 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
861 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
865 tree_mod_log_write_unlock(fs_info
);
872 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
873 int dst_offset
, int src_offset
, int nr_items
)
876 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
882 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
883 struct extent_buffer
*eb
, int slot
, int atomic
)
887 ret
= tree_mod_log_insert_key(fs_info
, eb
, slot
,
889 atomic
? GFP_ATOMIC
: GFP_NOFS
);
894 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
896 struct tree_mod_elem
**tm_list
= NULL
;
901 if (btrfs_header_level(eb
) == 0)
904 if (!tree_mod_need_log(fs_info
, NULL
))
907 nritems
= btrfs_header_nritems(eb
);
908 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*), GFP_NOFS
);
912 for (i
= 0; i
< nritems
; i
++) {
913 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
,
914 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
921 if (tree_mod_dont_log(fs_info
, eb
))
924 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
925 tree_mod_log_write_unlock(fs_info
);
933 for (i
= 0; i
< nritems
; i
++)
941 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
942 struct extent_buffer
*new_root_node
,
946 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
947 new_root_node
, GFP_NOFS
, log_removal
);
952 * check if the tree block can be shared by multiple trees
954 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
955 struct extent_buffer
*buf
)
958 * Tree blocks not in refernece counted trees and tree roots
959 * are never shared. If a block was allocated after the last
960 * snapshot and the block was not allocated by tree relocation,
961 * we know the block is not shared.
963 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
964 buf
!= root
->node
&& buf
!= root
->commit_root
&&
965 (btrfs_header_generation(buf
) <=
966 btrfs_root_last_snapshot(&root
->root_item
) ||
967 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
969 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
970 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
971 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
977 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
978 struct btrfs_root
*root
,
979 struct extent_buffer
*buf
,
980 struct extent_buffer
*cow
,
990 * Backrefs update rules:
992 * Always use full backrefs for extent pointers in tree block
993 * allocated by tree relocation.
995 * If a shared tree block is no longer referenced by its owner
996 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 * use full backrefs for extent pointers in tree block.
999 * If a tree block is been relocating
1000 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 * use full backrefs for extent pointers in tree block.
1002 * The reason for this is some operations (such as drop tree)
1003 * are only allowed for blocks use full backrefs.
1006 if (btrfs_block_can_be_shared(root
, buf
)) {
1007 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
1008 btrfs_header_level(buf
), 1,
1014 btrfs_std_error(root
->fs_info
, ret
, NULL
);
1019 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1020 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1021 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1026 owner
= btrfs_header_owner(buf
);
1027 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
1028 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
1031 if ((owner
== root
->root_key
.objectid
||
1032 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
1033 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
1034 ret
= btrfs_inc_ref(trans
, root
, buf
, 1);
1035 BUG_ON(ret
); /* -ENOMEM */
1037 if (root
->root_key
.objectid
==
1038 BTRFS_TREE_RELOC_OBJECTID
) {
1039 ret
= btrfs_dec_ref(trans
, root
, buf
, 0);
1040 BUG_ON(ret
); /* -ENOMEM */
1041 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1042 BUG_ON(ret
); /* -ENOMEM */
1044 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1047 if (root
->root_key
.objectid
==
1048 BTRFS_TREE_RELOC_OBJECTID
)
1049 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1051 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
1052 BUG_ON(ret
); /* -ENOMEM */
1054 if (new_flags
!= 0) {
1055 int level
= btrfs_header_level(buf
);
1057 ret
= btrfs_set_disk_extent_flags(trans
, root
,
1060 new_flags
, level
, 0);
1065 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
1066 if (root
->root_key
.objectid
==
1067 BTRFS_TREE_RELOC_OBJECTID
)
1068 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1070 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
1071 BUG_ON(ret
); /* -ENOMEM */
1072 ret
= btrfs_dec_ref(trans
, root
, buf
, 1);
1073 BUG_ON(ret
); /* -ENOMEM */
1075 clean_tree_block(trans
, root
->fs_info
, buf
);
1082 * does the dirty work in cow of a single block. The parent block (if
1083 * supplied) is updated to point to the new cow copy. The new buffer is marked
1084 * dirty and returned locked. If you modify the block it needs to be marked
1087 * search_start -- an allocation hint for the new block
1089 * empty_size -- a hint that you plan on doing more cow. This is the size in
1090 * bytes the allocator should try to find free next to the block it returns.
1091 * This is just a hint and may be ignored by the allocator.
1093 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1094 struct btrfs_root
*root
,
1095 struct extent_buffer
*buf
,
1096 struct extent_buffer
*parent
, int parent_slot
,
1097 struct extent_buffer
**cow_ret
,
1098 u64 search_start
, u64 empty_size
)
1100 struct btrfs_disk_key disk_key
;
1101 struct extent_buffer
*cow
;
1104 int unlock_orig
= 0;
1107 if (*cow_ret
== buf
)
1110 btrfs_assert_tree_locked(buf
);
1112 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
1113 trans
->transid
!= root
->fs_info
->running_transaction
->transid
);
1114 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
1115 trans
->transid
!= root
->last_trans
);
1117 level
= btrfs_header_level(buf
);
1120 btrfs_item_key(buf
, &disk_key
, 0);
1122 btrfs_node_key(buf
, &disk_key
, 0);
1124 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
1126 parent_start
= parent
->start
;
1132 cow
= btrfs_alloc_tree_block(trans
, root
, parent_start
,
1133 root
->root_key
.objectid
, &disk_key
, level
,
1134 search_start
, empty_size
);
1136 return PTR_ERR(cow
);
1138 /* cow is set to blocking by btrfs_init_new_buffer */
1140 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
1141 btrfs_set_header_bytenr(cow
, cow
->start
);
1142 btrfs_set_header_generation(cow
, trans
->transid
);
1143 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1144 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1145 BTRFS_HEADER_FLAG_RELOC
);
1146 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1147 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1149 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1151 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
1154 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1156 btrfs_abort_transaction(trans
, root
, ret
);
1160 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
)) {
1161 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1163 btrfs_abort_transaction(trans
, root
, ret
);
1168 if (buf
== root
->node
) {
1169 WARN_ON(parent
&& parent
!= buf
);
1170 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1171 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1172 parent_start
= buf
->start
;
1176 extent_buffer_get(cow
);
1177 tree_mod_log_set_root_pointer(root
, cow
, 1);
1178 rcu_assign_pointer(root
->node
, cow
);
1180 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1182 free_extent_buffer(buf
);
1183 add_root_to_dirty_list(root
);
1185 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1186 parent_start
= parent
->start
;
1190 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1191 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1192 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1193 btrfs_set_node_blockptr(parent
, parent_slot
,
1195 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1197 btrfs_mark_buffer_dirty(parent
);
1199 ret
= tree_mod_log_free_eb(root
->fs_info
, buf
);
1201 btrfs_abort_transaction(trans
, root
, ret
);
1205 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1209 btrfs_tree_unlock(buf
);
1210 free_extent_buffer_stale(buf
);
1211 btrfs_mark_buffer_dirty(cow
);
1217 * returns the logical address of the oldest predecessor of the given root.
1218 * entries older than time_seq are ignored.
1220 static struct tree_mod_elem
*
1221 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1222 struct extent_buffer
*eb_root
, u64 time_seq
)
1224 struct tree_mod_elem
*tm
;
1225 struct tree_mod_elem
*found
= NULL
;
1226 u64 root_logical
= eb_root
->start
;
1233 * the very last operation that's logged for a root is the replacement
1234 * operation (if it is replaced at all). this has the index of the *new*
1235 * root, making it the very first operation that's logged for this root.
1238 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1243 * if there are no tree operation for the oldest root, we simply
1244 * return it. this should only happen if that (old) root is at
1251 * if there's an operation that's not a root replacement, we
1252 * found the oldest version of our root. normally, we'll find a
1253 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1255 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1259 root_logical
= tm
->old_root
.logical
;
1263 /* if there's no old root to return, return what we found instead */
1271 * tm is a pointer to the first operation to rewind within eb. then, all
1272 * previous operations will be rewinded (until we reach something older than
1276 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1277 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1280 struct rb_node
*next
;
1281 struct tree_mod_elem
*tm
= first_tm
;
1282 unsigned long o_dst
;
1283 unsigned long o_src
;
1284 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1286 n
= btrfs_header_nritems(eb
);
1287 tree_mod_log_read_lock(fs_info
);
1288 while (tm
&& tm
->seq
>= time_seq
) {
1290 * all the operations are recorded with the operator used for
1291 * the modification. as we're going backwards, we do the
1292 * opposite of each operation here.
1295 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1296 BUG_ON(tm
->slot
< n
);
1298 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1299 case MOD_LOG_KEY_REMOVE
:
1300 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1301 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1302 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1306 case MOD_LOG_KEY_REPLACE
:
1307 BUG_ON(tm
->slot
>= n
);
1308 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1309 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1310 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1313 case MOD_LOG_KEY_ADD
:
1314 /* if a move operation is needed it's in the log */
1317 case MOD_LOG_MOVE_KEYS
:
1318 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1319 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1320 memmove_extent_buffer(eb
, o_dst
, o_src
,
1321 tm
->move
.nr_items
* p_size
);
1323 case MOD_LOG_ROOT_REPLACE
:
1325 * this operation is special. for roots, this must be
1326 * handled explicitly before rewinding.
1327 * for non-roots, this operation may exist if the node
1328 * was a root: root A -> child B; then A gets empty and
1329 * B is promoted to the new root. in the mod log, we'll
1330 * have a root-replace operation for B, a tree block
1331 * that is no root. we simply ignore that operation.
1335 next
= rb_next(&tm
->node
);
1338 tm
= container_of(next
, struct tree_mod_elem
, node
);
1339 if (tm
->index
!= first_tm
->index
)
1342 tree_mod_log_read_unlock(fs_info
);
1343 btrfs_set_header_nritems(eb
, n
);
1347 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1348 * is returned. If rewind operations happen, a fresh buffer is returned. The
1349 * returned buffer is always read-locked. If the returned buffer is not the
1350 * input buffer, the lock on the input buffer is released and the input buffer
1351 * is freed (its refcount is decremented).
1353 static struct extent_buffer
*
1354 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1355 struct extent_buffer
*eb
, u64 time_seq
)
1357 struct extent_buffer
*eb_rewin
;
1358 struct tree_mod_elem
*tm
;
1363 if (btrfs_header_level(eb
) == 0)
1366 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1370 btrfs_set_path_blocking(path
);
1371 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1373 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1374 BUG_ON(tm
->slot
!= 0);
1375 eb_rewin
= alloc_dummy_extent_buffer(fs_info
, eb
->start
);
1377 btrfs_tree_read_unlock_blocking(eb
);
1378 free_extent_buffer(eb
);
1381 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1382 btrfs_set_header_backref_rev(eb_rewin
,
1383 btrfs_header_backref_rev(eb
));
1384 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1385 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1387 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1389 btrfs_tree_read_unlock_blocking(eb
);
1390 free_extent_buffer(eb
);
1395 btrfs_clear_path_blocking(path
, NULL
, BTRFS_READ_LOCK
);
1396 btrfs_tree_read_unlock_blocking(eb
);
1397 free_extent_buffer(eb
);
1399 extent_buffer_get(eb_rewin
);
1400 btrfs_tree_read_lock(eb_rewin
);
1401 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1402 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1403 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1409 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1410 * value. If there are no changes, the current root->root_node is returned. If
1411 * anything changed in between, there's a fresh buffer allocated on which the
1412 * rewind operations are done. In any case, the returned buffer is read locked.
1413 * Returns NULL on error (with no locks held).
1415 static inline struct extent_buffer
*
1416 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1418 struct tree_mod_elem
*tm
;
1419 struct extent_buffer
*eb
= NULL
;
1420 struct extent_buffer
*eb_root
;
1421 struct extent_buffer
*old
;
1422 struct tree_mod_root
*old_root
= NULL
;
1423 u64 old_generation
= 0;
1426 eb_root
= btrfs_read_lock_root_node(root
);
1427 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1431 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1432 old_root
= &tm
->old_root
;
1433 old_generation
= tm
->generation
;
1434 logical
= old_root
->logical
;
1436 logical
= eb_root
->start
;
1439 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1440 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1441 btrfs_tree_read_unlock(eb_root
);
1442 free_extent_buffer(eb_root
);
1443 old
= read_tree_block(root
, logical
, 0);
1444 if (WARN_ON(IS_ERR(old
) || !extent_buffer_uptodate(old
))) {
1446 free_extent_buffer(old
);
1447 btrfs_warn(root
->fs_info
,
1448 "failed to read tree block %llu from get_old_root", logical
);
1450 eb
= btrfs_clone_extent_buffer(old
);
1451 free_extent_buffer(old
);
1453 } else if (old_root
) {
1454 btrfs_tree_read_unlock(eb_root
);
1455 free_extent_buffer(eb_root
);
1456 eb
= alloc_dummy_extent_buffer(root
->fs_info
, logical
);
1458 btrfs_set_lock_blocking_rw(eb_root
, BTRFS_READ_LOCK
);
1459 eb
= btrfs_clone_extent_buffer(eb_root
);
1460 btrfs_tree_read_unlock_blocking(eb_root
);
1461 free_extent_buffer(eb_root
);
1466 extent_buffer_get(eb
);
1467 btrfs_tree_read_lock(eb
);
1469 btrfs_set_header_bytenr(eb
, eb
->start
);
1470 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1471 btrfs_set_header_owner(eb
, btrfs_header_owner(eb_root
));
1472 btrfs_set_header_level(eb
, old_root
->level
);
1473 btrfs_set_header_generation(eb
, old_generation
);
1476 __tree_mod_log_rewind(root
->fs_info
, eb
, time_seq
, tm
);
1478 WARN_ON(btrfs_header_level(eb
) != 0);
1479 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1484 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1486 struct tree_mod_elem
*tm
;
1488 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1490 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1491 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1492 level
= tm
->old_root
.level
;
1494 level
= btrfs_header_level(eb_root
);
1496 free_extent_buffer(eb_root
);
1501 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1502 struct btrfs_root
*root
,
1503 struct extent_buffer
*buf
)
1505 if (btrfs_test_is_dummy_root(root
))
1508 /* ensure we can see the force_cow */
1512 * We do not need to cow a block if
1513 * 1) this block is not created or changed in this transaction;
1514 * 2) this block does not belong to TREE_RELOC tree;
1515 * 3) the root is not forced COW.
1517 * What is forced COW:
1518 * when we create snapshot during commiting the transaction,
1519 * after we've finished coping src root, we must COW the shared
1520 * block to ensure the metadata consistency.
1522 if (btrfs_header_generation(buf
) == trans
->transid
&&
1523 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1524 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1525 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1526 !test_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
))
1532 * cows a single block, see __btrfs_cow_block for the real work.
1533 * This version of it has extra checks so that a block isn't cow'd more than
1534 * once per transaction, as long as it hasn't been written yet
1536 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1537 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1538 struct extent_buffer
*parent
, int parent_slot
,
1539 struct extent_buffer
**cow_ret
)
1544 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1545 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1547 root
->fs_info
->running_transaction
->transid
);
1549 if (trans
->transid
!= root
->fs_info
->generation
)
1550 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1551 trans
->transid
, root
->fs_info
->generation
);
1553 if (!should_cow_block(trans
, root
, buf
)) {
1558 search_start
= buf
->start
& ~((u64
)SZ_1G
- 1);
1561 btrfs_set_lock_blocking(parent
);
1562 btrfs_set_lock_blocking(buf
);
1564 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1565 parent_slot
, cow_ret
, search_start
, 0);
1567 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1573 * helper function for defrag to decide if two blocks pointed to by a
1574 * node are actually close by
1576 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1578 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1580 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1586 * compare two keys in a memcmp fashion
1588 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1590 struct btrfs_key k1
;
1592 btrfs_disk_key_to_cpu(&k1
, disk
);
1594 return btrfs_comp_cpu_keys(&k1
, k2
);
1598 * same as comp_keys only with two btrfs_key's
1600 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1602 if (k1
->objectid
> k2
->objectid
)
1604 if (k1
->objectid
< k2
->objectid
)
1606 if (k1
->type
> k2
->type
)
1608 if (k1
->type
< k2
->type
)
1610 if (k1
->offset
> k2
->offset
)
1612 if (k1
->offset
< k2
->offset
)
1618 * this is used by the defrag code to go through all the
1619 * leaves pointed to by a node and reallocate them so that
1620 * disk order is close to key order
1622 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1623 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1624 int start_slot
, u64
*last_ret
,
1625 struct btrfs_key
*progress
)
1627 struct extent_buffer
*cur
;
1630 u64 search_start
= *last_ret
;
1640 int progress_passed
= 0;
1641 struct btrfs_disk_key disk_key
;
1643 parent_level
= btrfs_header_level(parent
);
1645 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1646 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1648 parent_nritems
= btrfs_header_nritems(parent
);
1649 blocksize
= root
->nodesize
;
1650 end_slot
= parent_nritems
- 1;
1652 if (parent_nritems
<= 1)
1655 btrfs_set_lock_blocking(parent
);
1657 for (i
= start_slot
; i
<= end_slot
; i
++) {
1660 btrfs_node_key(parent
, &disk_key
, i
);
1661 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1664 progress_passed
= 1;
1665 blocknr
= btrfs_node_blockptr(parent
, i
);
1666 gen
= btrfs_node_ptr_generation(parent
, i
);
1667 if (last_block
== 0)
1668 last_block
= blocknr
;
1671 other
= btrfs_node_blockptr(parent
, i
- 1);
1672 close
= close_blocks(blocknr
, other
, blocksize
);
1674 if (!close
&& i
< end_slot
) {
1675 other
= btrfs_node_blockptr(parent
, i
+ 1);
1676 close
= close_blocks(blocknr
, other
, blocksize
);
1679 last_block
= blocknr
;
1683 cur
= btrfs_find_tree_block(root
->fs_info
, blocknr
);
1685 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1688 if (!cur
|| !uptodate
) {
1690 cur
= read_tree_block(root
, blocknr
, gen
);
1692 return PTR_ERR(cur
);
1693 } else if (!extent_buffer_uptodate(cur
)) {
1694 free_extent_buffer(cur
);
1697 } else if (!uptodate
) {
1698 err
= btrfs_read_buffer(cur
, gen
);
1700 free_extent_buffer(cur
);
1705 if (search_start
== 0)
1706 search_start
= last_block
;
1708 btrfs_tree_lock(cur
);
1709 btrfs_set_lock_blocking(cur
);
1710 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1713 (end_slot
- i
) * blocksize
));
1715 btrfs_tree_unlock(cur
);
1716 free_extent_buffer(cur
);
1719 search_start
= cur
->start
;
1720 last_block
= cur
->start
;
1721 *last_ret
= search_start
;
1722 btrfs_tree_unlock(cur
);
1723 free_extent_buffer(cur
);
1729 * The leaf data grows from end-to-front in the node.
1730 * this returns the address of the start of the last item,
1731 * which is the stop of the leaf data stack
1733 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1734 struct extent_buffer
*leaf
)
1736 u32 nr
= btrfs_header_nritems(leaf
);
1738 return BTRFS_LEAF_DATA_SIZE(root
);
1739 return btrfs_item_offset_nr(leaf
, nr
- 1);
1744 * search for key in the extent_buffer. The items start at offset p,
1745 * and they are item_size apart. There are 'max' items in p.
1747 * the slot in the array is returned via slot, and it points to
1748 * the place where you would insert key if it is not found in
1751 * slot may point to max if the key is bigger than all of the keys
1753 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1755 int item_size
, struct btrfs_key
*key
,
1762 struct btrfs_disk_key
*tmp
= NULL
;
1763 struct btrfs_disk_key unaligned
;
1764 unsigned long offset
;
1766 unsigned long map_start
= 0;
1767 unsigned long map_len
= 0;
1770 while (low
< high
) {
1771 mid
= (low
+ high
) / 2;
1772 offset
= p
+ mid
* item_size
;
1774 if (!kaddr
|| offset
< map_start
||
1775 (offset
+ sizeof(struct btrfs_disk_key
)) >
1776 map_start
+ map_len
) {
1778 err
= map_private_extent_buffer(eb
, offset
,
1779 sizeof(struct btrfs_disk_key
),
1780 &kaddr
, &map_start
, &map_len
);
1783 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1786 read_extent_buffer(eb
, &unaligned
,
1787 offset
, sizeof(unaligned
));
1792 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1795 ret
= comp_keys(tmp
, key
);
1811 * simple bin_search frontend that does the right thing for
1814 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1815 int level
, int *slot
)
1818 return generic_bin_search(eb
,
1819 offsetof(struct btrfs_leaf
, items
),
1820 sizeof(struct btrfs_item
),
1821 key
, btrfs_header_nritems(eb
),
1824 return generic_bin_search(eb
,
1825 offsetof(struct btrfs_node
, ptrs
),
1826 sizeof(struct btrfs_key_ptr
),
1827 key
, btrfs_header_nritems(eb
),
1831 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1832 int level
, int *slot
)
1834 return bin_search(eb
, key
, level
, slot
);
1837 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1839 spin_lock(&root
->accounting_lock
);
1840 btrfs_set_root_used(&root
->root_item
,
1841 btrfs_root_used(&root
->root_item
) + size
);
1842 spin_unlock(&root
->accounting_lock
);
1845 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1847 spin_lock(&root
->accounting_lock
);
1848 btrfs_set_root_used(&root
->root_item
,
1849 btrfs_root_used(&root
->root_item
) - size
);
1850 spin_unlock(&root
->accounting_lock
);
1853 /* given a node and slot number, this reads the blocks it points to. The
1854 * extent buffer is returned with a reference taken (but unlocked).
1855 * NULL is returned on error.
1857 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1858 struct extent_buffer
*parent
, int slot
)
1860 int level
= btrfs_header_level(parent
);
1861 struct extent_buffer
*eb
;
1865 if (slot
>= btrfs_header_nritems(parent
))
1870 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1871 btrfs_node_ptr_generation(parent
, slot
));
1872 if (IS_ERR(eb
) || !extent_buffer_uptodate(eb
)) {
1874 free_extent_buffer(eb
);
1882 * node level balancing, used to make sure nodes are in proper order for
1883 * item deletion. We balance from the top down, so we have to make sure
1884 * that a deletion won't leave an node completely empty later on.
1886 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1887 struct btrfs_root
*root
,
1888 struct btrfs_path
*path
, int level
)
1890 struct extent_buffer
*right
= NULL
;
1891 struct extent_buffer
*mid
;
1892 struct extent_buffer
*left
= NULL
;
1893 struct extent_buffer
*parent
= NULL
;
1897 int orig_slot
= path
->slots
[level
];
1903 mid
= path
->nodes
[level
];
1905 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1906 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1907 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1909 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1911 if (level
< BTRFS_MAX_LEVEL
- 1) {
1912 parent
= path
->nodes
[level
+ 1];
1913 pslot
= path
->slots
[level
+ 1];
1917 * deal with the case where there is only one pointer in the root
1918 * by promoting the node below to a root
1921 struct extent_buffer
*child
;
1923 if (btrfs_header_nritems(mid
) != 1)
1926 /* promote the child to a root */
1927 child
= read_node_slot(root
, mid
, 0);
1930 btrfs_std_error(root
->fs_info
, ret
, NULL
);
1934 btrfs_tree_lock(child
);
1935 btrfs_set_lock_blocking(child
);
1936 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1938 btrfs_tree_unlock(child
);
1939 free_extent_buffer(child
);
1943 tree_mod_log_set_root_pointer(root
, child
, 1);
1944 rcu_assign_pointer(root
->node
, child
);
1946 add_root_to_dirty_list(root
);
1947 btrfs_tree_unlock(child
);
1949 path
->locks
[level
] = 0;
1950 path
->nodes
[level
] = NULL
;
1951 clean_tree_block(trans
, root
->fs_info
, mid
);
1952 btrfs_tree_unlock(mid
);
1953 /* once for the path */
1954 free_extent_buffer(mid
);
1956 root_sub_used(root
, mid
->len
);
1957 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1958 /* once for the root ptr */
1959 free_extent_buffer_stale(mid
);
1962 if (btrfs_header_nritems(mid
) >
1963 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1966 left
= read_node_slot(root
, parent
, pslot
- 1);
1968 btrfs_tree_lock(left
);
1969 btrfs_set_lock_blocking(left
);
1970 wret
= btrfs_cow_block(trans
, root
, left
,
1971 parent
, pslot
- 1, &left
);
1977 right
= read_node_slot(root
, parent
, pslot
+ 1);
1979 btrfs_tree_lock(right
);
1980 btrfs_set_lock_blocking(right
);
1981 wret
= btrfs_cow_block(trans
, root
, right
,
1982 parent
, pslot
+ 1, &right
);
1989 /* first, try to make some room in the middle buffer */
1991 orig_slot
+= btrfs_header_nritems(left
);
1992 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1998 * then try to empty the right most buffer into the middle
2001 wret
= push_node_left(trans
, root
, mid
, right
, 1);
2002 if (wret
< 0 && wret
!= -ENOSPC
)
2004 if (btrfs_header_nritems(right
) == 0) {
2005 clean_tree_block(trans
, root
->fs_info
, right
);
2006 btrfs_tree_unlock(right
);
2007 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
2008 root_sub_used(root
, right
->len
);
2009 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
2010 free_extent_buffer_stale(right
);
2013 struct btrfs_disk_key right_key
;
2014 btrfs_node_key(right
, &right_key
, 0);
2015 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2017 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
2018 btrfs_mark_buffer_dirty(parent
);
2021 if (btrfs_header_nritems(mid
) == 1) {
2023 * we're not allowed to leave a node with one item in the
2024 * tree during a delete. A deletion from lower in the tree
2025 * could try to delete the only pointer in this node.
2026 * So, pull some keys from the left.
2027 * There has to be a left pointer at this point because
2028 * otherwise we would have pulled some pointers from the
2033 btrfs_std_error(root
->fs_info
, ret
, NULL
);
2036 wret
= balance_node_right(trans
, root
, mid
, left
);
2042 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2048 if (btrfs_header_nritems(mid
) == 0) {
2049 clean_tree_block(trans
, root
->fs_info
, mid
);
2050 btrfs_tree_unlock(mid
);
2051 del_ptr(root
, path
, level
+ 1, pslot
);
2052 root_sub_used(root
, mid
->len
);
2053 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
2054 free_extent_buffer_stale(mid
);
2057 /* update the parent key to reflect our changes */
2058 struct btrfs_disk_key mid_key
;
2059 btrfs_node_key(mid
, &mid_key
, 0);
2060 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2062 btrfs_set_node_key(parent
, &mid_key
, pslot
);
2063 btrfs_mark_buffer_dirty(parent
);
2066 /* update the path */
2068 if (btrfs_header_nritems(left
) > orig_slot
) {
2069 extent_buffer_get(left
);
2070 /* left was locked after cow */
2071 path
->nodes
[level
] = left
;
2072 path
->slots
[level
+ 1] -= 1;
2073 path
->slots
[level
] = orig_slot
;
2075 btrfs_tree_unlock(mid
);
2076 free_extent_buffer(mid
);
2079 orig_slot
-= btrfs_header_nritems(left
);
2080 path
->slots
[level
] = orig_slot
;
2083 /* double check we haven't messed things up */
2085 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2089 btrfs_tree_unlock(right
);
2090 free_extent_buffer(right
);
2093 if (path
->nodes
[level
] != left
)
2094 btrfs_tree_unlock(left
);
2095 free_extent_buffer(left
);
2100 /* Node balancing for insertion. Here we only split or push nodes around
2101 * when they are completely full. This is also done top down, so we
2102 * have to be pessimistic.
2104 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2105 struct btrfs_root
*root
,
2106 struct btrfs_path
*path
, int level
)
2108 struct extent_buffer
*right
= NULL
;
2109 struct extent_buffer
*mid
;
2110 struct extent_buffer
*left
= NULL
;
2111 struct extent_buffer
*parent
= NULL
;
2115 int orig_slot
= path
->slots
[level
];
2120 mid
= path
->nodes
[level
];
2121 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2123 if (level
< BTRFS_MAX_LEVEL
- 1) {
2124 parent
= path
->nodes
[level
+ 1];
2125 pslot
= path
->slots
[level
+ 1];
2131 left
= read_node_slot(root
, parent
, pslot
- 1);
2133 /* first, try to make some room in the middle buffer */
2137 btrfs_tree_lock(left
);
2138 btrfs_set_lock_blocking(left
);
2140 left_nr
= btrfs_header_nritems(left
);
2141 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2144 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2149 wret
= push_node_left(trans
, root
,
2156 struct btrfs_disk_key disk_key
;
2157 orig_slot
+= left_nr
;
2158 btrfs_node_key(mid
, &disk_key
, 0);
2159 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2161 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2162 btrfs_mark_buffer_dirty(parent
);
2163 if (btrfs_header_nritems(left
) > orig_slot
) {
2164 path
->nodes
[level
] = left
;
2165 path
->slots
[level
+ 1] -= 1;
2166 path
->slots
[level
] = orig_slot
;
2167 btrfs_tree_unlock(mid
);
2168 free_extent_buffer(mid
);
2171 btrfs_header_nritems(left
);
2172 path
->slots
[level
] = orig_slot
;
2173 btrfs_tree_unlock(left
);
2174 free_extent_buffer(left
);
2178 btrfs_tree_unlock(left
);
2179 free_extent_buffer(left
);
2181 right
= read_node_slot(root
, parent
, pslot
+ 1);
2184 * then try to empty the right most buffer into the middle
2189 btrfs_tree_lock(right
);
2190 btrfs_set_lock_blocking(right
);
2192 right_nr
= btrfs_header_nritems(right
);
2193 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2196 ret
= btrfs_cow_block(trans
, root
, right
,
2202 wret
= balance_node_right(trans
, root
,
2209 struct btrfs_disk_key disk_key
;
2211 btrfs_node_key(right
, &disk_key
, 0);
2212 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2214 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2215 btrfs_mark_buffer_dirty(parent
);
2217 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2218 path
->nodes
[level
] = right
;
2219 path
->slots
[level
+ 1] += 1;
2220 path
->slots
[level
] = orig_slot
-
2221 btrfs_header_nritems(mid
);
2222 btrfs_tree_unlock(mid
);
2223 free_extent_buffer(mid
);
2225 btrfs_tree_unlock(right
);
2226 free_extent_buffer(right
);
2230 btrfs_tree_unlock(right
);
2231 free_extent_buffer(right
);
2237 * readahead one full node of leaves, finding things that are close
2238 * to the block in 'slot', and triggering ra on them.
2240 static void reada_for_search(struct btrfs_root
*root
,
2241 struct btrfs_path
*path
,
2242 int level
, int slot
, u64 objectid
)
2244 struct extent_buffer
*node
;
2245 struct btrfs_disk_key disk_key
;
2251 struct extent_buffer
*eb
;
2259 if (!path
->nodes
[level
])
2262 node
= path
->nodes
[level
];
2264 search
= btrfs_node_blockptr(node
, slot
);
2265 blocksize
= root
->nodesize
;
2266 eb
= btrfs_find_tree_block(root
->fs_info
, search
);
2268 free_extent_buffer(eb
);
2274 nritems
= btrfs_header_nritems(node
);
2278 if (path
->reada
== READA_BACK
) {
2282 } else if (path
->reada
== READA_FORWARD
) {
2287 if (path
->reada
== READA_BACK
&& objectid
) {
2288 btrfs_node_key(node
, &disk_key
, nr
);
2289 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2292 search
= btrfs_node_blockptr(node
, nr
);
2293 if ((search
<= target
&& target
- search
<= 65536) ||
2294 (search
> target
&& search
- target
<= 65536)) {
2295 gen
= btrfs_node_ptr_generation(node
, nr
);
2296 readahead_tree_block(root
, search
);
2300 if ((nread
> 65536 || nscan
> 32))
2305 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2306 struct btrfs_path
*path
, int level
)
2310 struct extent_buffer
*parent
;
2311 struct extent_buffer
*eb
;
2316 parent
= path
->nodes
[level
+ 1];
2320 nritems
= btrfs_header_nritems(parent
);
2321 slot
= path
->slots
[level
+ 1];
2324 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2325 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2326 eb
= btrfs_find_tree_block(root
->fs_info
, block1
);
2328 * if we get -eagain from btrfs_buffer_uptodate, we
2329 * don't want to return eagain here. That will loop
2332 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2334 free_extent_buffer(eb
);
2336 if (slot
+ 1 < nritems
) {
2337 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2338 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2339 eb
= btrfs_find_tree_block(root
->fs_info
, block2
);
2340 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2342 free_extent_buffer(eb
);
2346 readahead_tree_block(root
, block1
);
2348 readahead_tree_block(root
, block2
);
2353 * when we walk down the tree, it is usually safe to unlock the higher layers
2354 * in the tree. The exceptions are when our path goes through slot 0, because
2355 * operations on the tree might require changing key pointers higher up in the
2358 * callers might also have set path->keep_locks, which tells this code to keep
2359 * the lock if the path points to the last slot in the block. This is part of
2360 * walking through the tree, and selecting the next slot in the higher block.
2362 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2363 * if lowest_unlock is 1, level 0 won't be unlocked
2365 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2366 int lowest_unlock
, int min_write_lock_level
,
2367 int *write_lock_level
)
2370 int skip_level
= level
;
2372 struct extent_buffer
*t
;
2374 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2375 if (!path
->nodes
[i
])
2377 if (!path
->locks
[i
])
2379 if (!no_skips
&& path
->slots
[i
] == 0) {
2383 if (!no_skips
&& path
->keep_locks
) {
2386 nritems
= btrfs_header_nritems(t
);
2387 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2392 if (skip_level
< i
&& i
>= lowest_unlock
)
2396 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2397 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2399 if (write_lock_level
&&
2400 i
> min_write_lock_level
&&
2401 i
<= *write_lock_level
) {
2402 *write_lock_level
= i
- 1;
2409 * This releases any locks held in the path starting at level and
2410 * going all the way up to the root.
2412 * btrfs_search_slot will keep the lock held on higher nodes in a few
2413 * corner cases, such as COW of the block at slot zero in the node. This
2414 * ignores those rules, and it should only be called when there are no
2415 * more updates to be done higher up in the tree.
2417 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2421 if (path
->keep_locks
)
2424 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2425 if (!path
->nodes
[i
])
2427 if (!path
->locks
[i
])
2429 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2435 * helper function for btrfs_search_slot. The goal is to find a block
2436 * in cache without setting the path to blocking. If we find the block
2437 * we return zero and the path is unchanged.
2439 * If we can't find the block, we set the path blocking and do some
2440 * reada. -EAGAIN is returned and the search must be repeated.
2443 read_block_for_search(struct btrfs_trans_handle
*trans
,
2444 struct btrfs_root
*root
, struct btrfs_path
*p
,
2445 struct extent_buffer
**eb_ret
, int level
, int slot
,
2446 struct btrfs_key
*key
, u64 time_seq
)
2450 struct extent_buffer
*b
= *eb_ret
;
2451 struct extent_buffer
*tmp
;
2454 blocknr
= btrfs_node_blockptr(b
, slot
);
2455 gen
= btrfs_node_ptr_generation(b
, slot
);
2457 tmp
= btrfs_find_tree_block(root
->fs_info
, blocknr
);
2459 /* first we do an atomic uptodate check */
2460 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2465 /* the pages were up to date, but we failed
2466 * the generation number check. Do a full
2467 * read for the generation number that is correct.
2468 * We must do this without dropping locks so
2469 * we can trust our generation number
2471 btrfs_set_path_blocking(p
);
2473 /* now we're allowed to do a blocking uptodate check */
2474 ret
= btrfs_read_buffer(tmp
, gen
);
2479 free_extent_buffer(tmp
);
2480 btrfs_release_path(p
);
2485 * reduce lock contention at high levels
2486 * of the btree by dropping locks before
2487 * we read. Don't release the lock on the current
2488 * level because we need to walk this node to figure
2489 * out which blocks to read.
2491 btrfs_unlock_up_safe(p
, level
+ 1);
2492 btrfs_set_path_blocking(p
);
2494 free_extent_buffer(tmp
);
2495 if (p
->reada
!= READA_NONE
)
2496 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2498 btrfs_release_path(p
);
2501 tmp
= read_tree_block(root
, blocknr
, 0);
2504 * If the read above didn't mark this buffer up to date,
2505 * it will never end up being up to date. Set ret to EIO now
2506 * and give up so that our caller doesn't loop forever
2509 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2511 free_extent_buffer(tmp
);
2517 * helper function for btrfs_search_slot. This does all of the checks
2518 * for node-level blocks and does any balancing required based on
2521 * If no extra work was required, zero is returned. If we had to
2522 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2526 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2527 struct btrfs_root
*root
, struct btrfs_path
*p
,
2528 struct extent_buffer
*b
, int level
, int ins_len
,
2529 int *write_lock_level
)
2532 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2533 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2536 if (*write_lock_level
< level
+ 1) {
2537 *write_lock_level
= level
+ 1;
2538 btrfs_release_path(p
);
2542 btrfs_set_path_blocking(p
);
2543 reada_for_balance(root
, p
, level
);
2544 sret
= split_node(trans
, root
, p
, level
);
2545 btrfs_clear_path_blocking(p
, NULL
, 0);
2552 b
= p
->nodes
[level
];
2553 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2554 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2557 if (*write_lock_level
< level
+ 1) {
2558 *write_lock_level
= level
+ 1;
2559 btrfs_release_path(p
);
2563 btrfs_set_path_blocking(p
);
2564 reada_for_balance(root
, p
, level
);
2565 sret
= balance_level(trans
, root
, p
, level
);
2566 btrfs_clear_path_blocking(p
, NULL
, 0);
2572 b
= p
->nodes
[level
];
2574 btrfs_release_path(p
);
2577 BUG_ON(btrfs_header_nritems(b
) == 1);
2587 static void key_search_validate(struct extent_buffer
*b
,
2588 struct btrfs_key
*key
,
2591 #ifdef CONFIG_BTRFS_ASSERT
2592 struct btrfs_disk_key disk_key
;
2594 btrfs_cpu_key_to_disk(&disk_key
, key
);
2597 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2598 offsetof(struct btrfs_leaf
, items
[0].key
),
2601 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2602 offsetof(struct btrfs_node
, ptrs
[0].key
),
2607 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2608 int level
, int *prev_cmp
, int *slot
)
2610 if (*prev_cmp
!= 0) {
2611 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2615 key_search_validate(b
, key
, level
);
2621 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*path
,
2622 u64 iobjectid
, u64 ioff
, u8 key_type
,
2623 struct btrfs_key
*found_key
)
2626 struct btrfs_key key
;
2627 struct extent_buffer
*eb
;
2632 key
.type
= key_type
;
2633 key
.objectid
= iobjectid
;
2636 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2640 eb
= path
->nodes
[0];
2641 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2642 ret
= btrfs_next_leaf(fs_root
, path
);
2645 eb
= path
->nodes
[0];
2648 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2649 if (found_key
->type
!= key
.type
||
2650 found_key
->objectid
!= key
.objectid
)
2657 * look for key in the tree. path is filled in with nodes along the way
2658 * if key is found, we return zero and you can find the item in the leaf
2659 * level of the path (level 0)
2661 * If the key isn't found, the path points to the slot where it should
2662 * be inserted, and 1 is returned. If there are other errors during the
2663 * search a negative error number is returned.
2665 * if ins_len > 0, nodes and leaves will be split as we walk down the
2666 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2669 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2670 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2673 struct extent_buffer
*b
;
2678 int lowest_unlock
= 1;
2680 /* everything at write_lock_level or lower must be write locked */
2681 int write_lock_level
= 0;
2682 u8 lowest_level
= 0;
2683 int min_write_lock_level
;
2686 lowest_level
= p
->lowest_level
;
2687 WARN_ON(lowest_level
&& ins_len
> 0);
2688 WARN_ON(p
->nodes
[0] != NULL
);
2689 BUG_ON(!cow
&& ins_len
);
2694 /* when we are removing items, we might have to go up to level
2695 * two as we update tree pointers Make sure we keep write
2696 * for those levels as well
2698 write_lock_level
= 2;
2699 } else if (ins_len
> 0) {
2701 * for inserting items, make sure we have a write lock on
2702 * level 1 so we can update keys
2704 write_lock_level
= 1;
2708 write_lock_level
= -1;
2710 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2711 write_lock_level
= BTRFS_MAX_LEVEL
;
2713 min_write_lock_level
= write_lock_level
;
2718 * we try very hard to do read locks on the root
2720 root_lock
= BTRFS_READ_LOCK
;
2722 if (p
->search_commit_root
) {
2724 * the commit roots are read only
2725 * so we always do read locks
2727 if (p
->need_commit_sem
)
2728 down_read(&root
->fs_info
->commit_root_sem
);
2729 b
= root
->commit_root
;
2730 extent_buffer_get(b
);
2731 level
= btrfs_header_level(b
);
2732 if (p
->need_commit_sem
)
2733 up_read(&root
->fs_info
->commit_root_sem
);
2734 if (!p
->skip_locking
)
2735 btrfs_tree_read_lock(b
);
2737 if (p
->skip_locking
) {
2738 b
= btrfs_root_node(root
);
2739 level
= btrfs_header_level(b
);
2741 /* we don't know the level of the root node
2742 * until we actually have it read locked
2744 b
= btrfs_read_lock_root_node(root
);
2745 level
= btrfs_header_level(b
);
2746 if (level
<= write_lock_level
) {
2747 /* whoops, must trade for write lock */
2748 btrfs_tree_read_unlock(b
);
2749 free_extent_buffer(b
);
2750 b
= btrfs_lock_root_node(root
);
2751 root_lock
= BTRFS_WRITE_LOCK
;
2753 /* the level might have changed, check again */
2754 level
= btrfs_header_level(b
);
2758 p
->nodes
[level
] = b
;
2759 if (!p
->skip_locking
)
2760 p
->locks
[level
] = root_lock
;
2763 level
= btrfs_header_level(b
);
2766 * setup the path here so we can release it under lock
2767 * contention with the cow code
2771 * if we don't really need to cow this block
2772 * then we don't want to set the path blocking,
2773 * so we test it here
2775 if (!should_cow_block(trans
, root
, b
))
2779 * must have write locks on this node and the
2782 if (level
> write_lock_level
||
2783 (level
+ 1 > write_lock_level
&&
2784 level
+ 1 < BTRFS_MAX_LEVEL
&&
2785 p
->nodes
[level
+ 1])) {
2786 write_lock_level
= level
+ 1;
2787 btrfs_release_path(p
);
2791 btrfs_set_path_blocking(p
);
2792 err
= btrfs_cow_block(trans
, root
, b
,
2793 p
->nodes
[level
+ 1],
2794 p
->slots
[level
+ 1], &b
);
2801 p
->nodes
[level
] = b
;
2802 btrfs_clear_path_blocking(p
, NULL
, 0);
2805 * we have a lock on b and as long as we aren't changing
2806 * the tree, there is no way to for the items in b to change.
2807 * It is safe to drop the lock on our parent before we
2808 * go through the expensive btree search on b.
2810 * If we're inserting or deleting (ins_len != 0), then we might
2811 * be changing slot zero, which may require changing the parent.
2812 * So, we can't drop the lock until after we know which slot
2813 * we're operating on.
2815 if (!ins_len
&& !p
->keep_locks
) {
2818 if (u
< BTRFS_MAX_LEVEL
&& p
->locks
[u
]) {
2819 btrfs_tree_unlock_rw(p
->nodes
[u
], p
->locks
[u
]);
2824 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2828 if (ret
&& slot
> 0) {
2832 p
->slots
[level
] = slot
;
2833 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2834 ins_len
, &write_lock_level
);
2841 b
= p
->nodes
[level
];
2842 slot
= p
->slots
[level
];
2845 * slot 0 is special, if we change the key
2846 * we have to update the parent pointer
2847 * which means we must have a write lock
2850 if (slot
== 0 && ins_len
&&
2851 write_lock_level
< level
+ 1) {
2852 write_lock_level
= level
+ 1;
2853 btrfs_release_path(p
);
2857 unlock_up(p
, level
, lowest_unlock
,
2858 min_write_lock_level
, &write_lock_level
);
2860 if (level
== lowest_level
) {
2866 err
= read_block_for_search(trans
, root
, p
,
2867 &b
, level
, slot
, key
, 0);
2875 if (!p
->skip_locking
) {
2876 level
= btrfs_header_level(b
);
2877 if (level
<= write_lock_level
) {
2878 err
= btrfs_try_tree_write_lock(b
);
2880 btrfs_set_path_blocking(p
);
2882 btrfs_clear_path_blocking(p
, b
,
2885 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2887 err
= btrfs_tree_read_lock_atomic(b
);
2889 btrfs_set_path_blocking(p
);
2890 btrfs_tree_read_lock(b
);
2891 btrfs_clear_path_blocking(p
, b
,
2894 p
->locks
[level
] = BTRFS_READ_LOCK
;
2896 p
->nodes
[level
] = b
;
2899 p
->slots
[level
] = slot
;
2901 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2902 if (write_lock_level
< 1) {
2903 write_lock_level
= 1;
2904 btrfs_release_path(p
);
2908 btrfs_set_path_blocking(p
);
2909 err
= split_leaf(trans
, root
, key
,
2910 p
, ins_len
, ret
== 0);
2911 btrfs_clear_path_blocking(p
, NULL
, 0);
2919 if (!p
->search_for_split
)
2920 unlock_up(p
, level
, lowest_unlock
,
2921 min_write_lock_level
, &write_lock_level
);
2928 * we don't really know what they plan on doing with the path
2929 * from here on, so for now just mark it as blocking
2931 if (!p
->leave_spinning
)
2932 btrfs_set_path_blocking(p
);
2933 if (ret
< 0 && !p
->skip_release_on_error
)
2934 btrfs_release_path(p
);
2939 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2940 * current state of the tree together with the operations recorded in the tree
2941 * modification log to search for the key in a previous version of this tree, as
2942 * denoted by the time_seq parameter.
2944 * Naturally, there is no support for insert, delete or cow operations.
2946 * The resulting path and return value will be set up as if we called
2947 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2949 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2950 struct btrfs_path
*p
, u64 time_seq
)
2952 struct extent_buffer
*b
;
2957 int lowest_unlock
= 1;
2958 u8 lowest_level
= 0;
2961 lowest_level
= p
->lowest_level
;
2962 WARN_ON(p
->nodes
[0] != NULL
);
2964 if (p
->search_commit_root
) {
2966 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2970 b
= get_old_root(root
, time_seq
);
2971 level
= btrfs_header_level(b
);
2972 p
->locks
[level
] = BTRFS_READ_LOCK
;
2975 level
= btrfs_header_level(b
);
2976 p
->nodes
[level
] = b
;
2977 btrfs_clear_path_blocking(p
, NULL
, 0);
2980 * we have a lock on b and as long as we aren't changing
2981 * the tree, there is no way to for the items in b to change.
2982 * It is safe to drop the lock on our parent before we
2983 * go through the expensive btree search on b.
2985 btrfs_unlock_up_safe(p
, level
+ 1);
2988 * Since we can unwind eb's we want to do a real search every
2992 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2996 if (ret
&& slot
> 0) {
3000 p
->slots
[level
] = slot
;
3001 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3003 if (level
== lowest_level
) {
3009 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
3010 slot
, key
, time_seq
);
3018 level
= btrfs_header_level(b
);
3019 err
= btrfs_tree_read_lock_atomic(b
);
3021 btrfs_set_path_blocking(p
);
3022 btrfs_tree_read_lock(b
);
3023 btrfs_clear_path_blocking(p
, b
,
3026 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
3031 p
->locks
[level
] = BTRFS_READ_LOCK
;
3032 p
->nodes
[level
] = b
;
3034 p
->slots
[level
] = slot
;
3035 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3041 if (!p
->leave_spinning
)
3042 btrfs_set_path_blocking(p
);
3044 btrfs_release_path(p
);
3050 * helper to use instead of search slot if no exact match is needed but
3051 * instead the next or previous item should be returned.
3052 * When find_higher is true, the next higher item is returned, the next lower
3054 * When return_any and find_higher are both true, and no higher item is found,
3055 * return the next lower instead.
3056 * When return_any is true and find_higher is false, and no lower item is found,
3057 * return the next higher instead.
3058 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3061 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3062 struct btrfs_key
*key
, struct btrfs_path
*p
,
3063 int find_higher
, int return_any
)
3066 struct extent_buffer
*leaf
;
3069 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3073 * a return value of 1 means the path is at the position where the
3074 * item should be inserted. Normally this is the next bigger item,
3075 * but in case the previous item is the last in a leaf, path points
3076 * to the first free slot in the previous leaf, i.e. at an invalid
3082 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3083 ret
= btrfs_next_leaf(root
, p
);
3089 * no higher item found, return the next
3094 btrfs_release_path(p
);
3098 if (p
->slots
[0] == 0) {
3099 ret
= btrfs_prev_leaf(root
, p
);
3104 if (p
->slots
[0] == btrfs_header_nritems(leaf
))
3111 * no lower item found, return the next
3116 btrfs_release_path(p
);
3126 * adjust the pointers going up the tree, starting at level
3127 * making sure the right key of each node is points to 'key'.
3128 * This is used after shifting pointers to the left, so it stops
3129 * fixing up pointers when a given leaf/node is not in slot 0 of the
3133 static void fixup_low_keys(struct btrfs_fs_info
*fs_info
,
3134 struct btrfs_path
*path
,
3135 struct btrfs_disk_key
*key
, int level
)
3138 struct extent_buffer
*t
;
3140 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3141 int tslot
= path
->slots
[i
];
3142 if (!path
->nodes
[i
])
3145 tree_mod_log_set_node_key(fs_info
, t
, tslot
, 1);
3146 btrfs_set_node_key(t
, key
, tslot
);
3147 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3156 * This function isn't completely safe. It's the caller's responsibility
3157 * that the new key won't break the order
3159 void btrfs_set_item_key_safe(struct btrfs_fs_info
*fs_info
,
3160 struct btrfs_path
*path
,
3161 struct btrfs_key
*new_key
)
3163 struct btrfs_disk_key disk_key
;
3164 struct extent_buffer
*eb
;
3167 eb
= path
->nodes
[0];
3168 slot
= path
->slots
[0];
3170 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3171 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
3173 if (slot
< btrfs_header_nritems(eb
) - 1) {
3174 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3175 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
3178 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3179 btrfs_set_item_key(eb
, &disk_key
, slot
);
3180 btrfs_mark_buffer_dirty(eb
);
3182 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
3186 * try to push data from one node into the next node left in the
3189 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3190 * error, and > 0 if there was no room in the left hand block.
3192 static int push_node_left(struct btrfs_trans_handle
*trans
,
3193 struct btrfs_root
*root
, struct extent_buffer
*dst
,
3194 struct extent_buffer
*src
, int empty
)
3201 src_nritems
= btrfs_header_nritems(src
);
3202 dst_nritems
= btrfs_header_nritems(dst
);
3203 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3204 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3205 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3207 if (!empty
&& src_nritems
<= 8)
3210 if (push_items
<= 0)
3214 push_items
= min(src_nritems
, push_items
);
3215 if (push_items
< src_nritems
) {
3216 /* leave at least 8 pointers in the node if
3217 * we aren't going to empty it
3219 if (src_nritems
- push_items
< 8) {
3220 if (push_items
<= 8)
3226 push_items
= min(src_nritems
- 8, push_items
);
3228 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3231 btrfs_abort_transaction(trans
, root
, ret
);
3234 copy_extent_buffer(dst
, src
,
3235 btrfs_node_key_ptr_offset(dst_nritems
),
3236 btrfs_node_key_ptr_offset(0),
3237 push_items
* sizeof(struct btrfs_key_ptr
));
3239 if (push_items
< src_nritems
) {
3241 * don't call tree_mod_log_eb_move here, key removal was already
3242 * fully logged by tree_mod_log_eb_copy above.
3244 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3245 btrfs_node_key_ptr_offset(push_items
),
3246 (src_nritems
- push_items
) *
3247 sizeof(struct btrfs_key_ptr
));
3249 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3250 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3251 btrfs_mark_buffer_dirty(src
);
3252 btrfs_mark_buffer_dirty(dst
);
3258 * try to push data from one node into the next node right in the
3261 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3262 * error, and > 0 if there was no room in the right hand block.
3264 * this will only push up to 1/2 the contents of the left node over
3266 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3267 struct btrfs_root
*root
,
3268 struct extent_buffer
*dst
,
3269 struct extent_buffer
*src
)
3277 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3278 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3280 src_nritems
= btrfs_header_nritems(src
);
3281 dst_nritems
= btrfs_header_nritems(dst
);
3282 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3283 if (push_items
<= 0)
3286 if (src_nritems
< 4)
3289 max_push
= src_nritems
/ 2 + 1;
3290 /* don't try to empty the node */
3291 if (max_push
>= src_nritems
)
3294 if (max_push
< push_items
)
3295 push_items
= max_push
;
3297 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3298 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3299 btrfs_node_key_ptr_offset(0),
3301 sizeof(struct btrfs_key_ptr
));
3303 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3304 src_nritems
- push_items
, push_items
);
3306 btrfs_abort_transaction(trans
, root
, ret
);
3309 copy_extent_buffer(dst
, src
,
3310 btrfs_node_key_ptr_offset(0),
3311 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3312 push_items
* sizeof(struct btrfs_key_ptr
));
3314 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3315 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3317 btrfs_mark_buffer_dirty(src
);
3318 btrfs_mark_buffer_dirty(dst
);
3324 * helper function to insert a new root level in the tree.
3325 * A new node is allocated, and a single item is inserted to
3326 * point to the existing root
3328 * returns zero on success or < 0 on failure.
3330 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3331 struct btrfs_root
*root
,
3332 struct btrfs_path
*path
, int level
)
3335 struct extent_buffer
*lower
;
3336 struct extent_buffer
*c
;
3337 struct extent_buffer
*old
;
3338 struct btrfs_disk_key lower_key
;
3340 BUG_ON(path
->nodes
[level
]);
3341 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3343 lower
= path
->nodes
[level
-1];
3345 btrfs_item_key(lower
, &lower_key
, 0);
3347 btrfs_node_key(lower
, &lower_key
, 0);
3349 c
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3350 &lower_key
, level
, root
->node
->start
, 0);
3354 root_add_used(root
, root
->nodesize
);
3356 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3357 btrfs_set_header_nritems(c
, 1);
3358 btrfs_set_header_level(c
, level
);
3359 btrfs_set_header_bytenr(c
, c
->start
);
3360 btrfs_set_header_generation(c
, trans
->transid
);
3361 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3362 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3364 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3367 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3368 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3370 btrfs_set_node_key(c
, &lower_key
, 0);
3371 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3372 lower_gen
= btrfs_header_generation(lower
);
3373 WARN_ON(lower_gen
!= trans
->transid
);
3375 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3377 btrfs_mark_buffer_dirty(c
);
3380 tree_mod_log_set_root_pointer(root
, c
, 0);
3381 rcu_assign_pointer(root
->node
, c
);
3383 /* the super has an extra ref to root->node */
3384 free_extent_buffer(old
);
3386 add_root_to_dirty_list(root
);
3387 extent_buffer_get(c
);
3388 path
->nodes
[level
] = c
;
3389 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
3390 path
->slots
[level
] = 0;
3395 * worker function to insert a single pointer in a node.
3396 * the node should have enough room for the pointer already
3398 * slot and level indicate where you want the key to go, and
3399 * blocknr is the block the key points to.
3401 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3402 struct btrfs_root
*root
, struct btrfs_path
*path
,
3403 struct btrfs_disk_key
*key
, u64 bytenr
,
3404 int slot
, int level
)
3406 struct extent_buffer
*lower
;
3410 BUG_ON(!path
->nodes
[level
]);
3411 btrfs_assert_tree_locked(path
->nodes
[level
]);
3412 lower
= path
->nodes
[level
];
3413 nritems
= btrfs_header_nritems(lower
);
3414 BUG_ON(slot
> nritems
);
3415 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3416 if (slot
!= nritems
) {
3418 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3419 slot
, nritems
- slot
);
3420 memmove_extent_buffer(lower
,
3421 btrfs_node_key_ptr_offset(slot
+ 1),
3422 btrfs_node_key_ptr_offset(slot
),
3423 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3426 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3427 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3430 btrfs_set_node_key(lower
, key
, slot
);
3431 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3432 WARN_ON(trans
->transid
== 0);
3433 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3434 btrfs_set_header_nritems(lower
, nritems
+ 1);
3435 btrfs_mark_buffer_dirty(lower
);
3439 * split the node at the specified level in path in two.
3440 * The path is corrected to point to the appropriate node after the split
3442 * Before splitting this tries to make some room in the node by pushing
3443 * left and right, if either one works, it returns right away.
3445 * returns 0 on success and < 0 on failure
3447 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3448 struct btrfs_root
*root
,
3449 struct btrfs_path
*path
, int level
)
3451 struct extent_buffer
*c
;
3452 struct extent_buffer
*split
;
3453 struct btrfs_disk_key disk_key
;
3458 c
= path
->nodes
[level
];
3459 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3460 if (c
== root
->node
) {
3462 * trying to split the root, lets make a new one
3464 * tree mod log: We don't log_removal old root in
3465 * insert_new_root, because that root buffer will be kept as a
3466 * normal node. We are going to log removal of half of the
3467 * elements below with tree_mod_log_eb_copy. We're holding a
3468 * tree lock on the buffer, which is why we cannot race with
3469 * other tree_mod_log users.
3471 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3475 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3476 c
= path
->nodes
[level
];
3477 if (!ret
&& btrfs_header_nritems(c
) <
3478 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3484 c_nritems
= btrfs_header_nritems(c
);
3485 mid
= (c_nritems
+ 1) / 2;
3486 btrfs_node_key(c
, &disk_key
, mid
);
3488 split
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3489 &disk_key
, level
, c
->start
, 0);
3491 return PTR_ERR(split
);
3493 root_add_used(root
, root
->nodesize
);
3495 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3496 btrfs_set_header_level(split
, btrfs_header_level(c
));
3497 btrfs_set_header_bytenr(split
, split
->start
);
3498 btrfs_set_header_generation(split
, trans
->transid
);
3499 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3500 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3501 write_extent_buffer(split
, root
->fs_info
->fsid
,
3502 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3503 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3504 btrfs_header_chunk_tree_uuid(split
),
3507 ret
= tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0,
3508 mid
, c_nritems
- mid
);
3510 btrfs_abort_transaction(trans
, root
, ret
);
3513 copy_extent_buffer(split
, c
,
3514 btrfs_node_key_ptr_offset(0),
3515 btrfs_node_key_ptr_offset(mid
),
3516 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3517 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3518 btrfs_set_header_nritems(c
, mid
);
3521 btrfs_mark_buffer_dirty(c
);
3522 btrfs_mark_buffer_dirty(split
);
3524 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3525 path
->slots
[level
+ 1] + 1, level
+ 1);
3527 if (path
->slots
[level
] >= mid
) {
3528 path
->slots
[level
] -= mid
;
3529 btrfs_tree_unlock(c
);
3530 free_extent_buffer(c
);
3531 path
->nodes
[level
] = split
;
3532 path
->slots
[level
+ 1] += 1;
3534 btrfs_tree_unlock(split
);
3535 free_extent_buffer(split
);
3541 * how many bytes are required to store the items in a leaf. start
3542 * and nr indicate which items in the leaf to check. This totals up the
3543 * space used both by the item structs and the item data
3545 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3547 struct btrfs_item
*start_item
;
3548 struct btrfs_item
*end_item
;
3549 struct btrfs_map_token token
;
3551 int nritems
= btrfs_header_nritems(l
);
3552 int end
= min(nritems
, start
+ nr
) - 1;
3556 btrfs_init_map_token(&token
);
3557 start_item
= btrfs_item_nr(start
);
3558 end_item
= btrfs_item_nr(end
);
3559 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3560 btrfs_token_item_size(l
, start_item
, &token
);
3561 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3562 data_len
+= sizeof(struct btrfs_item
) * nr
;
3563 WARN_ON(data_len
< 0);
3568 * The space between the end of the leaf items and
3569 * the start of the leaf data. IOW, how much room
3570 * the leaf has left for both items and data
3572 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3573 struct extent_buffer
*leaf
)
3575 int nritems
= btrfs_header_nritems(leaf
);
3577 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3579 btrfs_crit(root
->fs_info
,
3580 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3581 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3582 leaf_space_used(leaf
, 0, nritems
), nritems
);
3588 * min slot controls the lowest index we're willing to push to the
3589 * right. We'll push up to and including min_slot, but no lower
3591 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3592 struct btrfs_root
*root
,
3593 struct btrfs_path
*path
,
3594 int data_size
, int empty
,
3595 struct extent_buffer
*right
,
3596 int free_space
, u32 left_nritems
,
3599 struct extent_buffer
*left
= path
->nodes
[0];
3600 struct extent_buffer
*upper
= path
->nodes
[1];
3601 struct btrfs_map_token token
;
3602 struct btrfs_disk_key disk_key
;
3607 struct btrfs_item
*item
;
3613 btrfs_init_map_token(&token
);
3618 nr
= max_t(u32
, 1, min_slot
);
3620 if (path
->slots
[0] >= left_nritems
)
3621 push_space
+= data_size
;
3623 slot
= path
->slots
[1];
3624 i
= left_nritems
- 1;
3626 item
= btrfs_item_nr(i
);
3628 if (!empty
&& push_items
> 0) {
3629 if (path
->slots
[0] > i
)
3631 if (path
->slots
[0] == i
) {
3632 int space
= btrfs_leaf_free_space(root
, left
);
3633 if (space
+ push_space
* 2 > free_space
)
3638 if (path
->slots
[0] == i
)
3639 push_space
+= data_size
;
3641 this_item_size
= btrfs_item_size(left
, item
);
3642 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3646 push_space
+= this_item_size
+ sizeof(*item
);
3652 if (push_items
== 0)
3655 WARN_ON(!empty
&& push_items
== left_nritems
);
3657 /* push left to right */
3658 right_nritems
= btrfs_header_nritems(right
);
3660 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3661 push_space
-= leaf_data_end(root
, left
);
3663 /* make room in the right data area */
3664 data_end
= leaf_data_end(root
, right
);
3665 memmove_extent_buffer(right
,
3666 btrfs_leaf_data(right
) + data_end
- push_space
,
3667 btrfs_leaf_data(right
) + data_end
,
3668 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3670 /* copy from the left data area */
3671 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3672 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3673 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3676 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3677 btrfs_item_nr_offset(0),
3678 right_nritems
* sizeof(struct btrfs_item
));
3680 /* copy the items from left to right */
3681 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3682 btrfs_item_nr_offset(left_nritems
- push_items
),
3683 push_items
* sizeof(struct btrfs_item
));
3685 /* update the item pointers */
3686 right_nritems
+= push_items
;
3687 btrfs_set_header_nritems(right
, right_nritems
);
3688 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3689 for (i
= 0; i
< right_nritems
; i
++) {
3690 item
= btrfs_item_nr(i
);
3691 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3692 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3695 left_nritems
-= push_items
;
3696 btrfs_set_header_nritems(left
, left_nritems
);
3699 btrfs_mark_buffer_dirty(left
);
3701 clean_tree_block(trans
, root
->fs_info
, left
);
3703 btrfs_mark_buffer_dirty(right
);
3705 btrfs_item_key(right
, &disk_key
, 0);
3706 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3707 btrfs_mark_buffer_dirty(upper
);
3709 /* then fixup the leaf pointer in the path */
3710 if (path
->slots
[0] >= left_nritems
) {
3711 path
->slots
[0] -= left_nritems
;
3712 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3713 clean_tree_block(trans
, root
->fs_info
, path
->nodes
[0]);
3714 btrfs_tree_unlock(path
->nodes
[0]);
3715 free_extent_buffer(path
->nodes
[0]);
3716 path
->nodes
[0] = right
;
3717 path
->slots
[1] += 1;
3719 btrfs_tree_unlock(right
);
3720 free_extent_buffer(right
);
3725 btrfs_tree_unlock(right
);
3726 free_extent_buffer(right
);
3731 * push some data in the path leaf to the right, trying to free up at
3732 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3734 * returns 1 if the push failed because the other node didn't have enough
3735 * room, 0 if everything worked out and < 0 if there were major errors.
3737 * this will push starting from min_slot to the end of the leaf. It won't
3738 * push any slot lower than min_slot
3740 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3741 *root
, struct btrfs_path
*path
,
3742 int min_data_size
, int data_size
,
3743 int empty
, u32 min_slot
)
3745 struct extent_buffer
*left
= path
->nodes
[0];
3746 struct extent_buffer
*right
;
3747 struct extent_buffer
*upper
;
3753 if (!path
->nodes
[1])
3756 slot
= path
->slots
[1];
3757 upper
= path
->nodes
[1];
3758 if (slot
>= btrfs_header_nritems(upper
) - 1)
3761 btrfs_assert_tree_locked(path
->nodes
[1]);
3763 right
= read_node_slot(root
, upper
, slot
+ 1);
3767 btrfs_tree_lock(right
);
3768 btrfs_set_lock_blocking(right
);
3770 free_space
= btrfs_leaf_free_space(root
, right
);
3771 if (free_space
< data_size
)
3774 /* cow and double check */
3775 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3780 free_space
= btrfs_leaf_free_space(root
, right
);
3781 if (free_space
< data_size
)
3784 left_nritems
= btrfs_header_nritems(left
);
3785 if (left_nritems
== 0)
3788 if (path
->slots
[0] == left_nritems
&& !empty
) {
3789 /* Key greater than all keys in the leaf, right neighbor has
3790 * enough room for it and we're not emptying our leaf to delete
3791 * it, therefore use right neighbor to insert the new item and
3792 * no need to touch/dirty our left leaft. */
3793 btrfs_tree_unlock(left
);
3794 free_extent_buffer(left
);
3795 path
->nodes
[0] = right
;
3801 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3802 right
, free_space
, left_nritems
, min_slot
);
3804 btrfs_tree_unlock(right
);
3805 free_extent_buffer(right
);
3810 * push some data in the path leaf to the left, trying to free up at
3811 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3813 * max_slot can put a limit on how far into the leaf we'll push items. The
3814 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3817 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3818 struct btrfs_root
*root
,
3819 struct btrfs_path
*path
, int data_size
,
3820 int empty
, struct extent_buffer
*left
,
3821 int free_space
, u32 right_nritems
,
3824 struct btrfs_disk_key disk_key
;
3825 struct extent_buffer
*right
= path
->nodes
[0];
3829 struct btrfs_item
*item
;
3830 u32 old_left_nritems
;
3834 u32 old_left_item_size
;
3835 struct btrfs_map_token token
;
3837 btrfs_init_map_token(&token
);
3840 nr
= min(right_nritems
, max_slot
);
3842 nr
= min(right_nritems
- 1, max_slot
);
3844 for (i
= 0; i
< nr
; i
++) {
3845 item
= btrfs_item_nr(i
);
3847 if (!empty
&& push_items
> 0) {
3848 if (path
->slots
[0] < i
)
3850 if (path
->slots
[0] == i
) {
3851 int space
= btrfs_leaf_free_space(root
, right
);
3852 if (space
+ push_space
* 2 > free_space
)
3857 if (path
->slots
[0] == i
)
3858 push_space
+= data_size
;
3860 this_item_size
= btrfs_item_size(right
, item
);
3861 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3865 push_space
+= this_item_size
+ sizeof(*item
);
3868 if (push_items
== 0) {
3872 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3874 /* push data from right to left */
3875 copy_extent_buffer(left
, right
,
3876 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3877 btrfs_item_nr_offset(0),
3878 push_items
* sizeof(struct btrfs_item
));
3880 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3881 btrfs_item_offset_nr(right
, push_items
- 1);
3883 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3884 leaf_data_end(root
, left
) - push_space
,
3885 btrfs_leaf_data(right
) +
3886 btrfs_item_offset_nr(right
, push_items
- 1),
3888 old_left_nritems
= btrfs_header_nritems(left
);
3889 BUG_ON(old_left_nritems
<= 0);
3891 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3892 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3895 item
= btrfs_item_nr(i
);
3897 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3898 btrfs_set_token_item_offset(left
, item
,
3899 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3902 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3904 /* fixup right node */
3905 if (push_items
> right_nritems
)
3906 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3909 if (push_items
< right_nritems
) {
3910 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3911 leaf_data_end(root
, right
);
3912 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3913 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3914 btrfs_leaf_data(right
) +
3915 leaf_data_end(root
, right
), push_space
);
3917 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3918 btrfs_item_nr_offset(push_items
),
3919 (btrfs_header_nritems(right
) - push_items
) *
3920 sizeof(struct btrfs_item
));
3922 right_nritems
-= push_items
;
3923 btrfs_set_header_nritems(right
, right_nritems
);
3924 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3925 for (i
= 0; i
< right_nritems
; i
++) {
3926 item
= btrfs_item_nr(i
);
3928 push_space
= push_space
- btrfs_token_item_size(right
,
3930 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3933 btrfs_mark_buffer_dirty(left
);
3935 btrfs_mark_buffer_dirty(right
);
3937 clean_tree_block(trans
, root
->fs_info
, right
);
3939 btrfs_item_key(right
, &disk_key
, 0);
3940 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
3942 /* then fixup the leaf pointer in the path */
3943 if (path
->slots
[0] < push_items
) {
3944 path
->slots
[0] += old_left_nritems
;
3945 btrfs_tree_unlock(path
->nodes
[0]);
3946 free_extent_buffer(path
->nodes
[0]);
3947 path
->nodes
[0] = left
;
3948 path
->slots
[1] -= 1;
3950 btrfs_tree_unlock(left
);
3951 free_extent_buffer(left
);
3952 path
->slots
[0] -= push_items
;
3954 BUG_ON(path
->slots
[0] < 0);
3957 btrfs_tree_unlock(left
);
3958 free_extent_buffer(left
);
3963 * push some data in the path leaf to the left, trying to free up at
3964 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3966 * max_slot can put a limit on how far into the leaf we'll push items. The
3967 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3970 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3971 *root
, struct btrfs_path
*path
, int min_data_size
,
3972 int data_size
, int empty
, u32 max_slot
)
3974 struct extent_buffer
*right
= path
->nodes
[0];
3975 struct extent_buffer
*left
;
3981 slot
= path
->slots
[1];
3984 if (!path
->nodes
[1])
3987 right_nritems
= btrfs_header_nritems(right
);
3988 if (right_nritems
== 0)
3991 btrfs_assert_tree_locked(path
->nodes
[1]);
3993 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
3997 btrfs_tree_lock(left
);
3998 btrfs_set_lock_blocking(left
);
4000 free_space
= btrfs_leaf_free_space(root
, left
);
4001 if (free_space
< data_size
) {
4006 /* cow and double check */
4007 ret
= btrfs_cow_block(trans
, root
, left
,
4008 path
->nodes
[1], slot
- 1, &left
);
4010 /* we hit -ENOSPC, but it isn't fatal here */
4016 free_space
= btrfs_leaf_free_space(root
, left
);
4017 if (free_space
< data_size
) {
4022 return __push_leaf_left(trans
, root
, path
, min_data_size
,
4023 empty
, left
, free_space
, right_nritems
,
4026 btrfs_tree_unlock(left
);
4027 free_extent_buffer(left
);
4032 * split the path's leaf in two, making sure there is at least data_size
4033 * available for the resulting leaf level of the path.
4035 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4036 struct btrfs_root
*root
,
4037 struct btrfs_path
*path
,
4038 struct extent_buffer
*l
,
4039 struct extent_buffer
*right
,
4040 int slot
, int mid
, int nritems
)
4045 struct btrfs_disk_key disk_key
;
4046 struct btrfs_map_token token
;
4048 btrfs_init_map_token(&token
);
4050 nritems
= nritems
- mid
;
4051 btrfs_set_header_nritems(right
, nritems
);
4052 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
4054 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4055 btrfs_item_nr_offset(mid
),
4056 nritems
* sizeof(struct btrfs_item
));
4058 copy_extent_buffer(right
, l
,
4059 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
4060 data_copy_size
, btrfs_leaf_data(l
) +
4061 leaf_data_end(root
, l
), data_copy_size
);
4063 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
4064 btrfs_item_end_nr(l
, mid
);
4066 for (i
= 0; i
< nritems
; i
++) {
4067 struct btrfs_item
*item
= btrfs_item_nr(i
);
4070 ioff
= btrfs_token_item_offset(right
, item
, &token
);
4071 btrfs_set_token_item_offset(right
, item
,
4072 ioff
+ rt_data_off
, &token
);
4075 btrfs_set_header_nritems(l
, mid
);
4076 btrfs_item_key(right
, &disk_key
, 0);
4077 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4078 path
->slots
[1] + 1, 1);
4080 btrfs_mark_buffer_dirty(right
);
4081 btrfs_mark_buffer_dirty(l
);
4082 BUG_ON(path
->slots
[0] != slot
);
4085 btrfs_tree_unlock(path
->nodes
[0]);
4086 free_extent_buffer(path
->nodes
[0]);
4087 path
->nodes
[0] = right
;
4088 path
->slots
[0] -= mid
;
4089 path
->slots
[1] += 1;
4091 btrfs_tree_unlock(right
);
4092 free_extent_buffer(right
);
4095 BUG_ON(path
->slots
[0] < 0);
4099 * double splits happen when we need to insert a big item in the middle
4100 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4101 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4104 * We avoid this by trying to push the items on either side of our target
4105 * into the adjacent leaves. If all goes well we can avoid the double split
4108 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4109 struct btrfs_root
*root
,
4110 struct btrfs_path
*path
,
4117 int space_needed
= data_size
;
4119 slot
= path
->slots
[0];
4120 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4121 space_needed
-= btrfs_leaf_free_space(root
, path
->nodes
[0]);
4124 * try to push all the items after our slot into the
4127 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4134 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4136 * our goal is to get our slot at the start or end of a leaf. If
4137 * we've done so we're done
4139 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4142 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4145 /* try to push all the items before our slot into the next leaf */
4146 slot
= path
->slots
[0];
4147 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4160 * split the path's leaf in two, making sure there is at least data_size
4161 * available for the resulting leaf level of the path.
4163 * returns 0 if all went well and < 0 on failure.
4165 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4166 struct btrfs_root
*root
,
4167 struct btrfs_key
*ins_key
,
4168 struct btrfs_path
*path
, int data_size
,
4171 struct btrfs_disk_key disk_key
;
4172 struct extent_buffer
*l
;
4176 struct extent_buffer
*right
;
4177 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4181 int num_doubles
= 0;
4182 int tried_avoid_double
= 0;
4185 slot
= path
->slots
[0];
4186 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4187 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
4190 /* first try to make some room by pushing left and right */
4191 if (data_size
&& path
->nodes
[1]) {
4192 int space_needed
= data_size
;
4194 if (slot
< btrfs_header_nritems(l
))
4195 space_needed
-= btrfs_leaf_free_space(root
, l
);
4197 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4198 space_needed
, 0, 0);
4202 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4203 space_needed
, 0, (u32
)-1);
4209 /* did the pushes work? */
4210 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
4214 if (!path
->nodes
[1]) {
4215 ret
= insert_new_root(trans
, root
, path
, 1);
4222 slot
= path
->slots
[0];
4223 nritems
= btrfs_header_nritems(l
);
4224 mid
= (nritems
+ 1) / 2;
4228 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4229 BTRFS_LEAF_DATA_SIZE(root
)) {
4230 if (slot
>= nritems
) {
4234 if (mid
!= nritems
&&
4235 leaf_space_used(l
, mid
, nritems
- mid
) +
4236 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4237 if (data_size
&& !tried_avoid_double
)
4238 goto push_for_double
;
4244 if (leaf_space_used(l
, 0, mid
) + data_size
>
4245 BTRFS_LEAF_DATA_SIZE(root
)) {
4246 if (!extend
&& data_size
&& slot
== 0) {
4248 } else if ((extend
|| !data_size
) && slot
== 0) {
4252 if (mid
!= nritems
&&
4253 leaf_space_used(l
, mid
, nritems
- mid
) +
4254 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4255 if (data_size
&& !tried_avoid_double
)
4256 goto push_for_double
;
4264 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4266 btrfs_item_key(l
, &disk_key
, mid
);
4268 right
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
4269 &disk_key
, 0, l
->start
, 0);
4271 return PTR_ERR(right
);
4273 root_add_used(root
, root
->nodesize
);
4275 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4276 btrfs_set_header_bytenr(right
, right
->start
);
4277 btrfs_set_header_generation(right
, trans
->transid
);
4278 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4279 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4280 btrfs_set_header_level(right
, 0);
4281 write_extent_buffer(right
, fs_info
->fsid
,
4282 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4284 write_extent_buffer(right
, fs_info
->chunk_tree_uuid
,
4285 btrfs_header_chunk_tree_uuid(right
),
4290 btrfs_set_header_nritems(right
, 0);
4291 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4292 path
->slots
[1] + 1, 1);
4293 btrfs_tree_unlock(path
->nodes
[0]);
4294 free_extent_buffer(path
->nodes
[0]);
4295 path
->nodes
[0] = right
;
4297 path
->slots
[1] += 1;
4299 btrfs_set_header_nritems(right
, 0);
4300 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4302 btrfs_tree_unlock(path
->nodes
[0]);
4303 free_extent_buffer(path
->nodes
[0]);
4304 path
->nodes
[0] = right
;
4306 if (path
->slots
[1] == 0)
4307 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
4309 btrfs_mark_buffer_dirty(right
);
4313 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4316 BUG_ON(num_doubles
!= 0);
4324 push_for_double_split(trans
, root
, path
, data_size
);
4325 tried_avoid_double
= 1;
4326 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4331 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4332 struct btrfs_root
*root
,
4333 struct btrfs_path
*path
, int ins_len
)
4335 struct btrfs_key key
;
4336 struct extent_buffer
*leaf
;
4337 struct btrfs_file_extent_item
*fi
;
4342 leaf
= path
->nodes
[0];
4343 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4345 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4346 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4348 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4351 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4352 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4353 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4354 struct btrfs_file_extent_item
);
4355 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4357 btrfs_release_path(path
);
4359 path
->keep_locks
= 1;
4360 path
->search_for_split
= 1;
4361 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4362 path
->search_for_split
= 0;
4369 leaf
= path
->nodes
[0];
4370 /* if our item isn't there, return now */
4371 if (item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4374 /* the leaf has changed, it now has room. return now */
4375 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4378 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4379 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4380 struct btrfs_file_extent_item
);
4381 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4385 btrfs_set_path_blocking(path
);
4386 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4390 path
->keep_locks
= 0;
4391 btrfs_unlock_up_safe(path
, 1);
4394 path
->keep_locks
= 0;
4398 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4399 struct btrfs_root
*root
,
4400 struct btrfs_path
*path
,
4401 struct btrfs_key
*new_key
,
4402 unsigned long split_offset
)
4404 struct extent_buffer
*leaf
;
4405 struct btrfs_item
*item
;
4406 struct btrfs_item
*new_item
;
4412 struct btrfs_disk_key disk_key
;
4414 leaf
= path
->nodes
[0];
4415 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4417 btrfs_set_path_blocking(path
);
4419 item
= btrfs_item_nr(path
->slots
[0]);
4420 orig_offset
= btrfs_item_offset(leaf
, item
);
4421 item_size
= btrfs_item_size(leaf
, item
);
4423 buf
= kmalloc(item_size
, GFP_NOFS
);
4427 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4428 path
->slots
[0]), item_size
);
4430 slot
= path
->slots
[0] + 1;
4431 nritems
= btrfs_header_nritems(leaf
);
4432 if (slot
!= nritems
) {
4433 /* shift the items */
4434 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4435 btrfs_item_nr_offset(slot
),
4436 (nritems
- slot
) * sizeof(struct btrfs_item
));
4439 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4440 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4442 new_item
= btrfs_item_nr(slot
);
4444 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4445 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4447 btrfs_set_item_offset(leaf
, item
,
4448 orig_offset
+ item_size
- split_offset
);
4449 btrfs_set_item_size(leaf
, item
, split_offset
);
4451 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4453 /* write the data for the start of the original item */
4454 write_extent_buffer(leaf
, buf
,
4455 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4458 /* write the data for the new item */
4459 write_extent_buffer(leaf
, buf
+ split_offset
,
4460 btrfs_item_ptr_offset(leaf
, slot
),
4461 item_size
- split_offset
);
4462 btrfs_mark_buffer_dirty(leaf
);
4464 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4470 * This function splits a single item into two items,
4471 * giving 'new_key' to the new item and splitting the
4472 * old one at split_offset (from the start of the item).
4474 * The path may be released by this operation. After
4475 * the split, the path is pointing to the old item. The
4476 * new item is going to be in the same node as the old one.
4478 * Note, the item being split must be smaller enough to live alone on
4479 * a tree block with room for one extra struct btrfs_item
4481 * This allows us to split the item in place, keeping a lock on the
4482 * leaf the entire time.
4484 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4485 struct btrfs_root
*root
,
4486 struct btrfs_path
*path
,
4487 struct btrfs_key
*new_key
,
4488 unsigned long split_offset
)
4491 ret
= setup_leaf_for_split(trans
, root
, path
,
4492 sizeof(struct btrfs_item
));
4496 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4501 * This function duplicate a item, giving 'new_key' to the new item.
4502 * It guarantees both items live in the same tree leaf and the new item
4503 * is contiguous with the original item.
4505 * This allows us to split file extent in place, keeping a lock on the
4506 * leaf the entire time.
4508 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4509 struct btrfs_root
*root
,
4510 struct btrfs_path
*path
,
4511 struct btrfs_key
*new_key
)
4513 struct extent_buffer
*leaf
;
4517 leaf
= path
->nodes
[0];
4518 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4519 ret
= setup_leaf_for_split(trans
, root
, path
,
4520 item_size
+ sizeof(struct btrfs_item
));
4525 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4526 item_size
, item_size
+
4527 sizeof(struct btrfs_item
), 1);
4528 leaf
= path
->nodes
[0];
4529 memcpy_extent_buffer(leaf
,
4530 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4531 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4537 * make the item pointed to by the path smaller. new_size indicates
4538 * how small to make it, and from_end tells us if we just chop bytes
4539 * off the end of the item or if we shift the item to chop bytes off
4542 void btrfs_truncate_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4543 u32 new_size
, int from_end
)
4546 struct extent_buffer
*leaf
;
4547 struct btrfs_item
*item
;
4549 unsigned int data_end
;
4550 unsigned int old_data_start
;
4551 unsigned int old_size
;
4552 unsigned int size_diff
;
4554 struct btrfs_map_token token
;
4556 btrfs_init_map_token(&token
);
4558 leaf
= path
->nodes
[0];
4559 slot
= path
->slots
[0];
4561 old_size
= btrfs_item_size_nr(leaf
, slot
);
4562 if (old_size
== new_size
)
4565 nritems
= btrfs_header_nritems(leaf
);
4566 data_end
= leaf_data_end(root
, leaf
);
4568 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4570 size_diff
= old_size
- new_size
;
4573 BUG_ON(slot
>= nritems
);
4576 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4578 /* first correct the data pointers */
4579 for (i
= slot
; i
< nritems
; i
++) {
4581 item
= btrfs_item_nr(i
);
4583 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4584 btrfs_set_token_item_offset(leaf
, item
,
4585 ioff
+ size_diff
, &token
);
4588 /* shift the data */
4590 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4591 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4592 data_end
, old_data_start
+ new_size
- data_end
);
4594 struct btrfs_disk_key disk_key
;
4597 btrfs_item_key(leaf
, &disk_key
, slot
);
4599 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4601 struct btrfs_file_extent_item
*fi
;
4603 fi
= btrfs_item_ptr(leaf
, slot
,
4604 struct btrfs_file_extent_item
);
4605 fi
= (struct btrfs_file_extent_item
*)(
4606 (unsigned long)fi
- size_diff
);
4608 if (btrfs_file_extent_type(leaf
, fi
) ==
4609 BTRFS_FILE_EXTENT_INLINE
) {
4610 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4611 memmove_extent_buffer(leaf
, ptr
,
4613 BTRFS_FILE_EXTENT_INLINE_DATA_START
);
4617 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4618 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4619 data_end
, old_data_start
- data_end
);
4621 offset
= btrfs_disk_key_offset(&disk_key
);
4622 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4623 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4625 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4628 item
= btrfs_item_nr(slot
);
4629 btrfs_set_item_size(leaf
, item
, new_size
);
4630 btrfs_mark_buffer_dirty(leaf
);
4632 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4633 btrfs_print_leaf(root
, leaf
);
4639 * make the item pointed to by the path bigger, data_size is the added size.
4641 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4645 struct extent_buffer
*leaf
;
4646 struct btrfs_item
*item
;
4648 unsigned int data_end
;
4649 unsigned int old_data
;
4650 unsigned int old_size
;
4652 struct btrfs_map_token token
;
4654 btrfs_init_map_token(&token
);
4656 leaf
= path
->nodes
[0];
4658 nritems
= btrfs_header_nritems(leaf
);
4659 data_end
= leaf_data_end(root
, leaf
);
4661 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4662 btrfs_print_leaf(root
, leaf
);
4665 slot
= path
->slots
[0];
4666 old_data
= btrfs_item_end_nr(leaf
, slot
);
4669 if (slot
>= nritems
) {
4670 btrfs_print_leaf(root
, leaf
);
4671 btrfs_crit(root
->fs_info
, "slot %d too large, nritems %d",
4677 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4679 /* first correct the data pointers */
4680 for (i
= slot
; i
< nritems
; i
++) {
4682 item
= btrfs_item_nr(i
);
4684 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4685 btrfs_set_token_item_offset(leaf
, item
,
4686 ioff
- data_size
, &token
);
4689 /* shift the data */
4690 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4691 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4692 data_end
, old_data
- data_end
);
4694 data_end
= old_data
;
4695 old_size
= btrfs_item_size_nr(leaf
, slot
);
4696 item
= btrfs_item_nr(slot
);
4697 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4698 btrfs_mark_buffer_dirty(leaf
);
4700 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4701 btrfs_print_leaf(root
, leaf
);
4707 * this is a helper for btrfs_insert_empty_items, the main goal here is
4708 * to save stack depth by doing the bulk of the work in a function
4709 * that doesn't call btrfs_search_slot
4711 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4712 struct btrfs_key
*cpu_key
, u32
*data_size
,
4713 u32 total_data
, u32 total_size
, int nr
)
4715 struct btrfs_item
*item
;
4718 unsigned int data_end
;
4719 struct btrfs_disk_key disk_key
;
4720 struct extent_buffer
*leaf
;
4722 struct btrfs_map_token token
;
4724 if (path
->slots
[0] == 0) {
4725 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4726 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4728 btrfs_unlock_up_safe(path
, 1);
4730 btrfs_init_map_token(&token
);
4732 leaf
= path
->nodes
[0];
4733 slot
= path
->slots
[0];
4735 nritems
= btrfs_header_nritems(leaf
);
4736 data_end
= leaf_data_end(root
, leaf
);
4738 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4739 btrfs_print_leaf(root
, leaf
);
4740 btrfs_crit(root
->fs_info
, "not enough freespace need %u have %d",
4741 total_size
, btrfs_leaf_free_space(root
, leaf
));
4745 if (slot
!= nritems
) {
4746 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4748 if (old_data
< data_end
) {
4749 btrfs_print_leaf(root
, leaf
);
4750 btrfs_crit(root
->fs_info
, "slot %d old_data %d data_end %d",
4751 slot
, old_data
, data_end
);
4755 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4757 /* first correct the data pointers */
4758 for (i
= slot
; i
< nritems
; i
++) {
4761 item
= btrfs_item_nr( i
);
4762 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4763 btrfs_set_token_item_offset(leaf
, item
,
4764 ioff
- total_data
, &token
);
4766 /* shift the items */
4767 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4768 btrfs_item_nr_offset(slot
),
4769 (nritems
- slot
) * sizeof(struct btrfs_item
));
4771 /* shift the data */
4772 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4773 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4774 data_end
, old_data
- data_end
);
4775 data_end
= old_data
;
4778 /* setup the item for the new data */
4779 for (i
= 0; i
< nr
; i
++) {
4780 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4781 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4782 item
= btrfs_item_nr(slot
+ i
);
4783 btrfs_set_token_item_offset(leaf
, item
,
4784 data_end
- data_size
[i
], &token
);
4785 data_end
-= data_size
[i
];
4786 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4789 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4790 btrfs_mark_buffer_dirty(leaf
);
4792 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4793 btrfs_print_leaf(root
, leaf
);
4799 * Given a key and some data, insert items into the tree.
4800 * This does all the path init required, making room in the tree if needed.
4802 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4803 struct btrfs_root
*root
,
4804 struct btrfs_path
*path
,
4805 struct btrfs_key
*cpu_key
, u32
*data_size
,
4814 for (i
= 0; i
< nr
; i
++)
4815 total_data
+= data_size
[i
];
4817 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4818 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4824 slot
= path
->slots
[0];
4827 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4828 total_data
, total_size
, nr
);
4833 * Given a key and some data, insert an item into the tree.
4834 * This does all the path init required, making room in the tree if needed.
4836 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4837 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4841 struct btrfs_path
*path
;
4842 struct extent_buffer
*leaf
;
4845 path
= btrfs_alloc_path();
4848 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4850 leaf
= path
->nodes
[0];
4851 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4852 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4853 btrfs_mark_buffer_dirty(leaf
);
4855 btrfs_free_path(path
);
4860 * delete the pointer from a given node.
4862 * the tree should have been previously balanced so the deletion does not
4865 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4866 int level
, int slot
)
4868 struct extent_buffer
*parent
= path
->nodes
[level
];
4872 nritems
= btrfs_header_nritems(parent
);
4873 if (slot
!= nritems
- 1) {
4875 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4876 slot
+ 1, nritems
- slot
- 1);
4877 memmove_extent_buffer(parent
,
4878 btrfs_node_key_ptr_offset(slot
),
4879 btrfs_node_key_ptr_offset(slot
+ 1),
4880 sizeof(struct btrfs_key_ptr
) *
4881 (nritems
- slot
- 1));
4883 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4884 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4889 btrfs_set_header_nritems(parent
, nritems
);
4890 if (nritems
== 0 && parent
== root
->node
) {
4891 BUG_ON(btrfs_header_level(root
->node
) != 1);
4892 /* just turn the root into a leaf and break */
4893 btrfs_set_header_level(root
->node
, 0);
4894 } else if (slot
== 0) {
4895 struct btrfs_disk_key disk_key
;
4897 btrfs_node_key(parent
, &disk_key
, 0);
4898 fixup_low_keys(root
->fs_info
, path
, &disk_key
, level
+ 1);
4900 btrfs_mark_buffer_dirty(parent
);
4904 * a helper function to delete the leaf pointed to by path->slots[1] and
4907 * This deletes the pointer in path->nodes[1] and frees the leaf
4908 * block extent. zero is returned if it all worked out, < 0 otherwise.
4910 * The path must have already been setup for deleting the leaf, including
4911 * all the proper balancing. path->nodes[1] must be locked.
4913 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4914 struct btrfs_root
*root
,
4915 struct btrfs_path
*path
,
4916 struct extent_buffer
*leaf
)
4918 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4919 del_ptr(root
, path
, 1, path
->slots
[1]);
4922 * btrfs_free_extent is expensive, we want to make sure we
4923 * aren't holding any locks when we call it
4925 btrfs_unlock_up_safe(path
, 0);
4927 root_sub_used(root
, leaf
->len
);
4929 extent_buffer_get(leaf
);
4930 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4931 free_extent_buffer_stale(leaf
);
4934 * delete the item at the leaf level in path. If that empties
4935 * the leaf, remove it from the tree
4937 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4938 struct btrfs_path
*path
, int slot
, int nr
)
4940 struct extent_buffer
*leaf
;
4941 struct btrfs_item
*item
;
4948 struct btrfs_map_token token
;
4950 btrfs_init_map_token(&token
);
4952 leaf
= path
->nodes
[0];
4953 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4955 for (i
= 0; i
< nr
; i
++)
4956 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4958 nritems
= btrfs_header_nritems(leaf
);
4960 if (slot
+ nr
!= nritems
) {
4961 int data_end
= leaf_data_end(root
, leaf
);
4963 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4965 btrfs_leaf_data(leaf
) + data_end
,
4966 last_off
- data_end
);
4968 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4971 item
= btrfs_item_nr(i
);
4972 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4973 btrfs_set_token_item_offset(leaf
, item
,
4974 ioff
+ dsize
, &token
);
4977 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4978 btrfs_item_nr_offset(slot
+ nr
),
4979 sizeof(struct btrfs_item
) *
4980 (nritems
- slot
- nr
));
4982 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4985 /* delete the leaf if we've emptied it */
4987 if (leaf
== root
->node
) {
4988 btrfs_set_header_level(leaf
, 0);
4990 btrfs_set_path_blocking(path
);
4991 clean_tree_block(trans
, root
->fs_info
, leaf
);
4992 btrfs_del_leaf(trans
, root
, path
, leaf
);
4995 int used
= leaf_space_used(leaf
, 0, nritems
);
4997 struct btrfs_disk_key disk_key
;
4999 btrfs_item_key(leaf
, &disk_key
, 0);
5000 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
5003 /* delete the leaf if it is mostly empty */
5004 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
5005 /* push_leaf_left fixes the path.
5006 * make sure the path still points to our leaf
5007 * for possible call to del_ptr below
5009 slot
= path
->slots
[1];
5010 extent_buffer_get(leaf
);
5012 btrfs_set_path_blocking(path
);
5013 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
5015 if (wret
< 0 && wret
!= -ENOSPC
)
5018 if (path
->nodes
[0] == leaf
&&
5019 btrfs_header_nritems(leaf
)) {
5020 wret
= push_leaf_right(trans
, root
, path
, 1,
5022 if (wret
< 0 && wret
!= -ENOSPC
)
5026 if (btrfs_header_nritems(leaf
) == 0) {
5027 path
->slots
[1] = slot
;
5028 btrfs_del_leaf(trans
, root
, path
, leaf
);
5029 free_extent_buffer(leaf
);
5032 /* if we're still in the path, make sure
5033 * we're dirty. Otherwise, one of the
5034 * push_leaf functions must have already
5035 * dirtied this buffer
5037 if (path
->nodes
[0] == leaf
)
5038 btrfs_mark_buffer_dirty(leaf
);
5039 free_extent_buffer(leaf
);
5042 btrfs_mark_buffer_dirty(leaf
);
5049 * search the tree again to find a leaf with lesser keys
5050 * returns 0 if it found something or 1 if there are no lesser leaves.
5051 * returns < 0 on io errors.
5053 * This may release the path, and so you may lose any locks held at the
5056 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5058 struct btrfs_key key
;
5059 struct btrfs_disk_key found_key
;
5062 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5064 if (key
.offset
> 0) {
5066 } else if (key
.type
> 0) {
5068 key
.offset
= (u64
)-1;
5069 } else if (key
.objectid
> 0) {
5072 key
.offset
= (u64
)-1;
5077 btrfs_release_path(path
);
5078 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5081 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5082 ret
= comp_keys(&found_key
, &key
);
5084 * We might have had an item with the previous key in the tree right
5085 * before we released our path. And after we released our path, that
5086 * item might have been pushed to the first slot (0) of the leaf we
5087 * were holding due to a tree balance. Alternatively, an item with the
5088 * previous key can exist as the only element of a leaf (big fat item).
5089 * Therefore account for these 2 cases, so that our callers (like
5090 * btrfs_previous_item) don't miss an existing item with a key matching
5091 * the previous key we computed above.
5099 * A helper function to walk down the tree starting at min_key, and looking
5100 * for nodes or leaves that are have a minimum transaction id.
5101 * This is used by the btree defrag code, and tree logging
5103 * This does not cow, but it does stuff the starting key it finds back
5104 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5105 * key and get a writable path.
5107 * This does lock as it descends, and path->keep_locks should be set
5108 * to 1 by the caller.
5110 * This honors path->lowest_level to prevent descent past a given level
5113 * min_trans indicates the oldest transaction that you are interested
5114 * in walking through. Any nodes or leaves older than min_trans are
5115 * skipped over (without reading them).
5117 * returns zero if something useful was found, < 0 on error and 1 if there
5118 * was nothing in the tree that matched the search criteria.
5120 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5121 struct btrfs_path
*path
,
5124 struct extent_buffer
*cur
;
5125 struct btrfs_key found_key
;
5131 int keep_locks
= path
->keep_locks
;
5133 path
->keep_locks
= 1;
5135 cur
= btrfs_read_lock_root_node(root
);
5136 level
= btrfs_header_level(cur
);
5137 WARN_ON(path
->nodes
[level
]);
5138 path
->nodes
[level
] = cur
;
5139 path
->locks
[level
] = BTRFS_READ_LOCK
;
5141 if (btrfs_header_generation(cur
) < min_trans
) {
5146 nritems
= btrfs_header_nritems(cur
);
5147 level
= btrfs_header_level(cur
);
5148 sret
= bin_search(cur
, min_key
, level
, &slot
);
5150 /* at the lowest level, we're done, setup the path and exit */
5151 if (level
== path
->lowest_level
) {
5152 if (slot
>= nritems
)
5155 path
->slots
[level
] = slot
;
5156 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5159 if (sret
&& slot
> 0)
5162 * check this node pointer against the min_trans parameters.
5163 * If it is too old, old, skip to the next one.
5165 while (slot
< nritems
) {
5168 gen
= btrfs_node_ptr_generation(cur
, slot
);
5169 if (gen
< min_trans
) {
5177 * we didn't find a candidate key in this node, walk forward
5178 * and find another one
5180 if (slot
>= nritems
) {
5181 path
->slots
[level
] = slot
;
5182 btrfs_set_path_blocking(path
);
5183 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5186 btrfs_release_path(path
);
5192 /* save our key for returning back */
5193 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5194 path
->slots
[level
] = slot
;
5195 if (level
== path
->lowest_level
) {
5199 btrfs_set_path_blocking(path
);
5200 cur
= read_node_slot(root
, cur
, slot
);
5201 BUG_ON(!cur
); /* -ENOMEM */
5203 btrfs_tree_read_lock(cur
);
5205 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5206 path
->nodes
[level
- 1] = cur
;
5207 unlock_up(path
, level
, 1, 0, NULL
);
5208 btrfs_clear_path_blocking(path
, NULL
, 0);
5211 path
->keep_locks
= keep_locks
;
5213 btrfs_unlock_up_safe(path
, path
->lowest_level
+ 1);
5214 btrfs_set_path_blocking(path
);
5215 memcpy(min_key
, &found_key
, sizeof(found_key
));
5220 static void tree_move_down(struct btrfs_root
*root
,
5221 struct btrfs_path
*path
,
5222 int *level
, int root_level
)
5224 BUG_ON(*level
== 0);
5225 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
5226 path
->slots
[*level
]);
5227 path
->slots
[*level
- 1] = 0;
5231 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
5232 struct btrfs_path
*path
,
5233 int *level
, int root_level
)
5237 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5239 path
->slots
[*level
]++;
5241 while (path
->slots
[*level
] >= nritems
) {
5242 if (*level
== root_level
)
5246 path
->slots
[*level
] = 0;
5247 free_extent_buffer(path
->nodes
[*level
]);
5248 path
->nodes
[*level
] = NULL
;
5250 path
->slots
[*level
]++;
5252 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5259 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5262 static int tree_advance(struct btrfs_root
*root
,
5263 struct btrfs_path
*path
,
5264 int *level
, int root_level
,
5266 struct btrfs_key
*key
)
5270 if (*level
== 0 || !allow_down
) {
5271 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5273 tree_move_down(root
, path
, level
, root_level
);
5278 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5279 path
->slots
[*level
]);
5281 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5282 path
->slots
[*level
]);
5287 static int tree_compare_item(struct btrfs_root
*left_root
,
5288 struct btrfs_path
*left_path
,
5289 struct btrfs_path
*right_path
,
5294 unsigned long off1
, off2
;
5296 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5297 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5301 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5302 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5303 right_path
->slots
[0]);
5305 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5307 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5314 #define ADVANCE_ONLY_NEXT -1
5317 * This function compares two trees and calls the provided callback for
5318 * every changed/new/deleted item it finds.
5319 * If shared tree blocks are encountered, whole subtrees are skipped, making
5320 * the compare pretty fast on snapshotted subvolumes.
5322 * This currently works on commit roots only. As commit roots are read only,
5323 * we don't do any locking. The commit roots are protected with transactions.
5324 * Transactions are ended and rejoined when a commit is tried in between.
5326 * This function checks for modifications done to the trees while comparing.
5327 * If it detects a change, it aborts immediately.
5329 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5330 struct btrfs_root
*right_root
,
5331 btrfs_changed_cb_t changed_cb
, void *ctx
)
5335 struct btrfs_path
*left_path
= NULL
;
5336 struct btrfs_path
*right_path
= NULL
;
5337 struct btrfs_key left_key
;
5338 struct btrfs_key right_key
;
5339 char *tmp_buf
= NULL
;
5340 int left_root_level
;
5341 int right_root_level
;
5344 int left_end_reached
;
5345 int right_end_reached
;
5353 left_path
= btrfs_alloc_path();
5358 right_path
= btrfs_alloc_path();
5364 tmp_buf
= kmalloc(left_root
->nodesize
, GFP_NOFS
);
5370 left_path
->search_commit_root
= 1;
5371 left_path
->skip_locking
= 1;
5372 right_path
->search_commit_root
= 1;
5373 right_path
->skip_locking
= 1;
5376 * Strategy: Go to the first items of both trees. Then do
5378 * If both trees are at level 0
5379 * Compare keys of current items
5380 * If left < right treat left item as new, advance left tree
5382 * If left > right treat right item as deleted, advance right tree
5384 * If left == right do deep compare of items, treat as changed if
5385 * needed, advance both trees and repeat
5386 * If both trees are at the same level but not at level 0
5387 * Compare keys of current nodes/leafs
5388 * If left < right advance left tree and repeat
5389 * If left > right advance right tree and repeat
5390 * If left == right compare blockptrs of the next nodes/leafs
5391 * If they match advance both trees but stay at the same level
5393 * If they don't match advance both trees while allowing to go
5395 * If tree levels are different
5396 * Advance the tree that needs it and repeat
5398 * Advancing a tree means:
5399 * If we are at level 0, try to go to the next slot. If that's not
5400 * possible, go one level up and repeat. Stop when we found a level
5401 * where we could go to the next slot. We may at this point be on a
5404 * If we are not at level 0 and not on shared tree blocks, go one
5407 * If we are not at level 0 and on shared tree blocks, go one slot to
5408 * the right if possible or go up and right.
5411 down_read(&left_root
->fs_info
->commit_root_sem
);
5412 left_level
= btrfs_header_level(left_root
->commit_root
);
5413 left_root_level
= left_level
;
5414 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5415 extent_buffer_get(left_path
->nodes
[left_level
]);
5417 right_level
= btrfs_header_level(right_root
->commit_root
);
5418 right_root_level
= right_level
;
5419 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5420 extent_buffer_get(right_path
->nodes
[right_level
]);
5421 up_read(&left_root
->fs_info
->commit_root_sem
);
5423 if (left_level
== 0)
5424 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5425 &left_key
, left_path
->slots
[left_level
]);
5427 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5428 &left_key
, left_path
->slots
[left_level
]);
5429 if (right_level
== 0)
5430 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5431 &right_key
, right_path
->slots
[right_level
]);
5433 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5434 &right_key
, right_path
->slots
[right_level
]);
5436 left_end_reached
= right_end_reached
= 0;
5437 advance_left
= advance_right
= 0;
5440 if (advance_left
&& !left_end_reached
) {
5441 ret
= tree_advance(left_root
, left_path
, &left_level
,
5443 advance_left
!= ADVANCE_ONLY_NEXT
,
5446 left_end_reached
= ADVANCE
;
5449 if (advance_right
&& !right_end_reached
) {
5450 ret
= tree_advance(right_root
, right_path
, &right_level
,
5452 advance_right
!= ADVANCE_ONLY_NEXT
,
5455 right_end_reached
= ADVANCE
;
5459 if (left_end_reached
&& right_end_reached
) {
5462 } else if (left_end_reached
) {
5463 if (right_level
== 0) {
5464 ret
= changed_cb(left_root
, right_root
,
5465 left_path
, right_path
,
5467 BTRFS_COMPARE_TREE_DELETED
,
5472 advance_right
= ADVANCE
;
5474 } else if (right_end_reached
) {
5475 if (left_level
== 0) {
5476 ret
= changed_cb(left_root
, right_root
,
5477 left_path
, right_path
,
5479 BTRFS_COMPARE_TREE_NEW
,
5484 advance_left
= ADVANCE
;
5488 if (left_level
== 0 && right_level
== 0) {
5489 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5491 ret
= changed_cb(left_root
, right_root
,
5492 left_path
, right_path
,
5494 BTRFS_COMPARE_TREE_NEW
,
5498 advance_left
= ADVANCE
;
5499 } else if (cmp
> 0) {
5500 ret
= changed_cb(left_root
, right_root
,
5501 left_path
, right_path
,
5503 BTRFS_COMPARE_TREE_DELETED
,
5507 advance_right
= ADVANCE
;
5509 enum btrfs_compare_tree_result result
;
5511 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5512 ret
= tree_compare_item(left_root
, left_path
,
5513 right_path
, tmp_buf
);
5515 result
= BTRFS_COMPARE_TREE_CHANGED
;
5517 result
= BTRFS_COMPARE_TREE_SAME
;
5518 ret
= changed_cb(left_root
, right_root
,
5519 left_path
, right_path
,
5520 &left_key
, result
, ctx
);
5523 advance_left
= ADVANCE
;
5524 advance_right
= ADVANCE
;
5526 } else if (left_level
== right_level
) {
5527 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5529 advance_left
= ADVANCE
;
5530 } else if (cmp
> 0) {
5531 advance_right
= ADVANCE
;
5533 left_blockptr
= btrfs_node_blockptr(
5534 left_path
->nodes
[left_level
],
5535 left_path
->slots
[left_level
]);
5536 right_blockptr
= btrfs_node_blockptr(
5537 right_path
->nodes
[right_level
],
5538 right_path
->slots
[right_level
]);
5539 left_gen
= btrfs_node_ptr_generation(
5540 left_path
->nodes
[left_level
],
5541 left_path
->slots
[left_level
]);
5542 right_gen
= btrfs_node_ptr_generation(
5543 right_path
->nodes
[right_level
],
5544 right_path
->slots
[right_level
]);
5545 if (left_blockptr
== right_blockptr
&&
5546 left_gen
== right_gen
) {
5548 * As we're on a shared block, don't
5549 * allow to go deeper.
5551 advance_left
= ADVANCE_ONLY_NEXT
;
5552 advance_right
= ADVANCE_ONLY_NEXT
;
5554 advance_left
= ADVANCE
;
5555 advance_right
= ADVANCE
;
5558 } else if (left_level
< right_level
) {
5559 advance_right
= ADVANCE
;
5561 advance_left
= ADVANCE
;
5566 btrfs_free_path(left_path
);
5567 btrfs_free_path(right_path
);
5573 * this is similar to btrfs_next_leaf, but does not try to preserve
5574 * and fixup the path. It looks for and returns the next key in the
5575 * tree based on the current path and the min_trans parameters.
5577 * 0 is returned if another key is found, < 0 if there are any errors
5578 * and 1 is returned if there are no higher keys in the tree
5580 * path->keep_locks should be set to 1 on the search made before
5581 * calling this function.
5583 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5584 struct btrfs_key
*key
, int level
, u64 min_trans
)
5587 struct extent_buffer
*c
;
5589 WARN_ON(!path
->keep_locks
);
5590 while (level
< BTRFS_MAX_LEVEL
) {
5591 if (!path
->nodes
[level
])
5594 slot
= path
->slots
[level
] + 1;
5595 c
= path
->nodes
[level
];
5597 if (slot
>= btrfs_header_nritems(c
)) {
5600 struct btrfs_key cur_key
;
5601 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5602 !path
->nodes
[level
+ 1])
5605 if (path
->locks
[level
+ 1]) {
5610 slot
= btrfs_header_nritems(c
) - 1;
5612 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5614 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5616 orig_lowest
= path
->lowest_level
;
5617 btrfs_release_path(path
);
5618 path
->lowest_level
= level
;
5619 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5621 path
->lowest_level
= orig_lowest
;
5625 c
= path
->nodes
[level
];
5626 slot
= path
->slots
[level
];
5633 btrfs_item_key_to_cpu(c
, key
, slot
);
5635 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5637 if (gen
< min_trans
) {
5641 btrfs_node_key_to_cpu(c
, key
, slot
);
5649 * search the tree again to find a leaf with greater keys
5650 * returns 0 if it found something or 1 if there are no greater leaves.
5651 * returns < 0 on io errors.
5653 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5655 return btrfs_next_old_leaf(root
, path
, 0);
5658 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5663 struct extent_buffer
*c
;
5664 struct extent_buffer
*next
;
5665 struct btrfs_key key
;
5668 int old_spinning
= path
->leave_spinning
;
5669 int next_rw_lock
= 0;
5671 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5675 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5680 btrfs_release_path(path
);
5682 path
->keep_locks
= 1;
5683 path
->leave_spinning
= 1;
5686 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5688 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5689 path
->keep_locks
= 0;
5694 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5696 * by releasing the path above we dropped all our locks. A balance
5697 * could have added more items next to the key that used to be
5698 * at the very end of the block. So, check again here and
5699 * advance the path if there are now more items available.
5701 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5708 * So the above check misses one case:
5709 * - after releasing the path above, someone has removed the item that
5710 * used to be at the very end of the block, and balance between leafs
5711 * gets another one with bigger key.offset to replace it.
5713 * This one should be returned as well, or we can get leaf corruption
5714 * later(esp. in __btrfs_drop_extents()).
5716 * And a bit more explanation about this check,
5717 * with ret > 0, the key isn't found, the path points to the slot
5718 * where it should be inserted, so the path->slots[0] item must be the
5721 if (nritems
> 0 && ret
> 0 && path
->slots
[0] == nritems
- 1) {
5726 while (level
< BTRFS_MAX_LEVEL
) {
5727 if (!path
->nodes
[level
]) {
5732 slot
= path
->slots
[level
] + 1;
5733 c
= path
->nodes
[level
];
5734 if (slot
>= btrfs_header_nritems(c
)) {
5736 if (level
== BTRFS_MAX_LEVEL
) {
5744 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5745 free_extent_buffer(next
);
5749 next_rw_lock
= path
->locks
[level
];
5750 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5756 btrfs_release_path(path
);
5760 if (!path
->skip_locking
) {
5761 ret
= btrfs_try_tree_read_lock(next
);
5762 if (!ret
&& time_seq
) {
5764 * If we don't get the lock, we may be racing
5765 * with push_leaf_left, holding that lock while
5766 * itself waiting for the leaf we've currently
5767 * locked. To solve this situation, we give up
5768 * on our lock and cycle.
5770 free_extent_buffer(next
);
5771 btrfs_release_path(path
);
5776 btrfs_set_path_blocking(path
);
5777 btrfs_tree_read_lock(next
);
5778 btrfs_clear_path_blocking(path
, next
,
5781 next_rw_lock
= BTRFS_READ_LOCK
;
5785 path
->slots
[level
] = slot
;
5788 c
= path
->nodes
[level
];
5789 if (path
->locks
[level
])
5790 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5792 free_extent_buffer(c
);
5793 path
->nodes
[level
] = next
;
5794 path
->slots
[level
] = 0;
5795 if (!path
->skip_locking
)
5796 path
->locks
[level
] = next_rw_lock
;
5800 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5806 btrfs_release_path(path
);
5810 if (!path
->skip_locking
) {
5811 ret
= btrfs_try_tree_read_lock(next
);
5813 btrfs_set_path_blocking(path
);
5814 btrfs_tree_read_lock(next
);
5815 btrfs_clear_path_blocking(path
, next
,
5818 next_rw_lock
= BTRFS_READ_LOCK
;
5823 unlock_up(path
, 0, 1, 0, NULL
);
5824 path
->leave_spinning
= old_spinning
;
5826 btrfs_set_path_blocking(path
);
5832 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5833 * searching until it gets past min_objectid or finds an item of 'type'
5835 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5837 int btrfs_previous_item(struct btrfs_root
*root
,
5838 struct btrfs_path
*path
, u64 min_objectid
,
5841 struct btrfs_key found_key
;
5842 struct extent_buffer
*leaf
;
5847 if (path
->slots
[0] == 0) {
5848 btrfs_set_path_blocking(path
);
5849 ret
= btrfs_prev_leaf(root
, path
);
5855 leaf
= path
->nodes
[0];
5856 nritems
= btrfs_header_nritems(leaf
);
5859 if (path
->slots
[0] == nritems
)
5862 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5863 if (found_key
.objectid
< min_objectid
)
5865 if (found_key
.type
== type
)
5867 if (found_key
.objectid
== min_objectid
&&
5868 found_key
.type
< type
)
5875 * search in extent tree to find a previous Metadata/Data extent item with
5878 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5880 int btrfs_previous_extent_item(struct btrfs_root
*root
,
5881 struct btrfs_path
*path
, u64 min_objectid
)
5883 struct btrfs_key found_key
;
5884 struct extent_buffer
*leaf
;
5889 if (path
->slots
[0] == 0) {
5890 btrfs_set_path_blocking(path
);
5891 ret
= btrfs_prev_leaf(root
, path
);
5897 leaf
= path
->nodes
[0];
5898 nritems
= btrfs_header_nritems(leaf
);
5901 if (path
->slots
[0] == nritems
)
5904 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5905 if (found_key
.objectid
< min_objectid
)
5907 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
||
5908 found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
5910 if (found_key
.objectid
== min_objectid
&&
5911 found_key
.type
< BTRFS_EXTENT_ITEM_KEY
)