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_trans_handle
*trans
, struct btrfs_root
*root
,
41 struct btrfs_path
*path
, int level
, int slot
,
43 static void tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
44 struct extent_buffer
*eb
);
45 struct extent_buffer
*read_old_tree_block(struct btrfs_root
*root
, u64 bytenr
,
46 u32 blocksize
, u64 parent_transid
,
48 struct extent_buffer
*btrfs_find_old_tree_block(struct btrfs_root
*root
,
49 u64 bytenr
, u32 blocksize
,
52 struct btrfs_path
*btrfs_alloc_path(void)
54 struct btrfs_path
*path
;
55 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
60 * set all locked nodes in the path to blocking locks. This should
61 * be done before scheduling
63 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
66 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
67 if (!p
->nodes
[i
] || !p
->locks
[i
])
69 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
70 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
71 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
72 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
73 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
78 * reset all the locked nodes in the patch to spinning locks.
80 * held is used to keep lockdep happy, when lockdep is enabled
81 * we set held to a blocking lock before we go around and
82 * retake all the spinlocks in the path. You can safely use NULL
85 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
86 struct extent_buffer
*held
, int held_rw
)
90 #ifdef CONFIG_DEBUG_LOCK_ALLOC
91 /* lockdep really cares that we take all of these spinlocks
92 * in the right order. If any of the locks in the path are not
93 * currently blocking, it is going to complain. So, make really
94 * really sure by forcing the path to blocking before we clear
98 btrfs_set_lock_blocking_rw(held
, held_rw
);
99 if (held_rw
== BTRFS_WRITE_LOCK
)
100 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
101 else if (held_rw
== BTRFS_READ_LOCK
)
102 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
104 btrfs_set_path_blocking(p
);
107 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
108 if (p
->nodes
[i
] && p
->locks
[i
]) {
109 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
110 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
111 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
112 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
113 p
->locks
[i
] = BTRFS_READ_LOCK
;
117 #ifdef CONFIG_DEBUG_LOCK_ALLOC
119 btrfs_clear_lock_blocking_rw(held
, held_rw
);
123 /* this also releases the path */
124 void btrfs_free_path(struct btrfs_path
*p
)
128 btrfs_release_path(p
);
129 kmem_cache_free(btrfs_path_cachep
, p
);
133 * path release drops references on the extent buffers in the path
134 * and it drops any locks held by this path
136 * It is safe to call this on paths that no locks or extent buffers held.
138 noinline
void btrfs_release_path(struct btrfs_path
*p
)
142 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
147 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
150 free_extent_buffer(p
->nodes
[i
]);
156 * safely gets a reference on the root node of a tree. A lock
157 * is not taken, so a concurrent writer may put a different node
158 * at the root of the tree. See btrfs_lock_root_node for the
161 * The extent buffer returned by this has a reference taken, so
162 * it won't disappear. It may stop being the root of the tree
163 * at any time because there are no locks held.
165 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
167 struct extent_buffer
*eb
;
171 eb
= rcu_dereference(root
->node
);
174 * RCU really hurts here, we could free up the root node because
175 * it was cow'ed but we may not get the new root node yet so do
176 * the inc_not_zero dance and if it doesn't work then
177 * synchronize_rcu and try again.
179 if (atomic_inc_not_zero(&eb
->refs
)) {
189 /* loop around taking references on and locking the root node of the
190 * tree until you end up with a lock on the root. A locked buffer
191 * is returned, with a reference held.
193 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
195 struct extent_buffer
*eb
;
198 eb
= btrfs_root_node(root
);
200 if (eb
== root
->node
)
202 btrfs_tree_unlock(eb
);
203 free_extent_buffer(eb
);
208 /* loop around taking references on and locking the root node of the
209 * tree until you end up with a lock on the root. A locked buffer
210 * is returned, with a reference held.
212 struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
214 struct extent_buffer
*eb
;
217 eb
= btrfs_root_node(root
);
218 btrfs_tree_read_lock(eb
);
219 if (eb
== root
->node
)
221 btrfs_tree_read_unlock(eb
);
222 free_extent_buffer(eb
);
227 /* cowonly root (everything not a reference counted cow subvolume), just get
228 * put onto a simple dirty list. transaction.c walks this to make sure they
229 * get properly updated on disk.
231 static void add_root_to_dirty_list(struct btrfs_root
*root
)
233 spin_lock(&root
->fs_info
->trans_lock
);
234 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
235 list_add(&root
->dirty_list
,
236 &root
->fs_info
->dirty_cowonly_roots
);
238 spin_unlock(&root
->fs_info
->trans_lock
);
242 * used by snapshot creation to make a copy of a root for a tree with
243 * a given objectid. The buffer with the new root node is returned in
244 * cow_ret, and this func returns zero on success or a negative error code.
246 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
247 struct btrfs_root
*root
,
248 struct extent_buffer
*buf
,
249 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
251 struct extent_buffer
*cow
;
254 struct btrfs_disk_key disk_key
;
256 WARN_ON(root
->ref_cows
&& trans
->transid
!=
257 root
->fs_info
->running_transaction
->transid
);
258 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
260 level
= btrfs_header_level(buf
);
262 btrfs_item_key(buf
, &disk_key
, 0);
264 btrfs_node_key(buf
, &disk_key
, 0);
266 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, 0,
267 new_root_objectid
, &disk_key
, level
,
272 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
273 btrfs_set_header_bytenr(cow
, cow
->start
);
274 btrfs_set_header_generation(cow
, trans
->transid
);
275 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
276 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
277 BTRFS_HEADER_FLAG_RELOC
);
278 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
279 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
281 btrfs_set_header_owner(cow
, new_root_objectid
);
283 write_extent_buffer(cow
, root
->fs_info
->fsid
,
284 (unsigned long)btrfs_header_fsid(cow
),
287 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
288 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
289 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
291 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
296 btrfs_mark_buffer_dirty(cow
);
305 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
306 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
308 MOD_LOG_ROOT_REPLACE
,
311 struct tree_mod_move
{
316 struct tree_mod_root
{
321 struct tree_mod_elem
{
323 u64 index
; /* shifted logical */
324 struct seq_list elem
;
327 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
330 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
333 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
334 struct btrfs_disk_key key
;
337 /* this is used for op == MOD_LOG_MOVE_KEYS */
338 struct tree_mod_move move
;
340 /* this is used for op == MOD_LOG_ROOT_REPLACE */
341 struct tree_mod_root old_root
;
345 __get_tree_mod_seq(struct btrfs_fs_info
*fs_info
, struct seq_list
*elem
)
347 elem
->seq
= atomic_inc_return(&fs_info
->tree_mod_seq
);
348 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
351 void btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
352 struct seq_list
*elem
)
355 spin_lock(&fs_info
->tree_mod_seq_lock
);
356 __get_tree_mod_seq(fs_info
, elem
);
357 spin_unlock(&fs_info
->tree_mod_seq_lock
);
360 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
361 struct seq_list
*elem
)
363 struct rb_root
*tm_root
;
364 struct rb_node
*node
;
365 struct rb_node
*next
;
366 struct seq_list
*cur_elem
;
367 struct tree_mod_elem
*tm
;
368 u64 min_seq
= (u64
)-1;
369 u64 seq_putting
= elem
->seq
;
374 BUG_ON(!(elem
->flags
& 1));
375 spin_lock(&fs_info
->tree_mod_seq_lock
);
376 list_del(&elem
->list
);
378 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
379 if ((cur_elem
->flags
& 1) && cur_elem
->seq
< min_seq
) {
380 if (seq_putting
> cur_elem
->seq
) {
382 * blocker with lower sequence number exists, we
383 * cannot remove anything from the log
387 min_seq
= cur_elem
->seq
;
392 * anything that's lower than the lowest existing (read: blocked)
393 * sequence number can be removed from the tree.
395 write_lock(&fs_info
->tree_mod_log_lock
);
396 tm_root
= &fs_info
->tree_mod_log
;
397 for (node
= rb_first(tm_root
); node
; node
= next
) {
398 next
= rb_next(node
);
399 tm
= container_of(node
, struct tree_mod_elem
, node
);
400 if (tm
->elem
.seq
> min_seq
)
402 rb_erase(node
, tm_root
);
403 list_del(&tm
->elem
.list
);
406 write_unlock(&fs_info
->tree_mod_log_lock
);
408 spin_unlock(&fs_info
->tree_mod_seq_lock
);
412 * key order of the log:
415 * the index is the shifted logical of the *new* root node for root replace
416 * operations, or the shifted logical of the affected block for all other
420 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
422 struct rb_root
*tm_root
;
423 struct rb_node
**new;
424 struct rb_node
*parent
= NULL
;
425 struct tree_mod_elem
*cur
;
428 BUG_ON(!tm
|| !tm
->elem
.seq
);
430 write_lock(&fs_info
->tree_mod_log_lock
);
431 tm_root
= &fs_info
->tree_mod_log
;
432 new = &tm_root
->rb_node
;
434 cur
= container_of(*new, struct tree_mod_elem
, node
);
436 if (cur
->index
< tm
->index
)
437 new = &((*new)->rb_left
);
438 else if (cur
->index
> tm
->index
)
439 new = &((*new)->rb_right
);
440 else if (cur
->elem
.seq
< tm
->elem
.seq
)
441 new = &((*new)->rb_left
);
442 else if (cur
->elem
.seq
> tm
->elem
.seq
)
443 new = &((*new)->rb_right
);
451 rb_link_node(&tm
->node
, parent
, new);
452 rb_insert_color(&tm
->node
, tm_root
);
454 write_unlock(&fs_info
->tree_mod_log_lock
);
458 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
459 struct extent_buffer
*eb
) {
461 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
465 if (btrfs_header_level(eb
) == 0)
471 * This allocates memory and gets a tree modification sequence number when
474 * Returns 0 when no sequence number is needed, < 0 on error.
475 * Returns 1 when a sequence number was added. In this case,
476 * fs_info->tree_mod_seq_lock was acquired and must be released by the caller
477 * after inserting into the rb tree.
479 static inline int tree_mod_alloc(struct btrfs_fs_info
*fs_info
, gfp_t flags
,
480 struct tree_mod_elem
**tm_ret
)
482 struct tree_mod_elem
*tm
;
485 if (tree_mod_dont_log(fs_info
, NULL
))
488 tm
= *tm_ret
= kzalloc(sizeof(*tm
), flags
);
493 spin_lock(&fs_info
->tree_mod_seq_lock
);
494 if (list_empty(&fs_info
->tree_mod_seq_list
)) {
496 * someone emptied the list while we were waiting for the lock.
497 * we must not add to the list, because no blocker exists. items
498 * are removed from the list only when the existing blocker is
499 * removed from the list.
503 spin_unlock(&fs_info
->tree_mod_seq_lock
);
505 __get_tree_mod_seq(fs_info
, &tm
->elem
);
513 tree_mod_log_insert_key_mask(struct btrfs_fs_info
*fs_info
,
514 struct extent_buffer
*eb
, int slot
,
515 enum mod_log_op op
, gfp_t flags
)
517 struct tree_mod_elem
*tm
;
520 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
524 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
525 if (op
!= MOD_LOG_KEY_ADD
) {
526 btrfs_node_key(eb
, &tm
->key
, slot
);
527 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
531 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
533 ret
= __tree_mod_log_insert(fs_info
, tm
);
534 spin_unlock(&fs_info
->tree_mod_seq_lock
);
539 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
540 int slot
, enum mod_log_op op
)
542 return tree_mod_log_insert_key_mask(fs_info
, eb
, slot
, op
, GFP_NOFS
);
546 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
547 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
548 int nr_items
, gfp_t flags
)
550 struct tree_mod_elem
*tm
;
554 if (tree_mod_dont_log(fs_info
, eb
))
557 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
558 ret
= tree_mod_log_insert_key(fs_info
, eb
, i
+ dst_slot
,
559 MOD_LOG_KEY_REMOVE_WHILE_MOVING
);
563 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
567 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
569 tm
->move
.dst_slot
= dst_slot
;
570 tm
->move
.nr_items
= nr_items
;
571 tm
->op
= MOD_LOG_MOVE_KEYS
;
573 ret
= __tree_mod_log_insert(fs_info
, tm
);
574 spin_unlock(&fs_info
->tree_mod_seq_lock
);
579 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
580 struct extent_buffer
*old_root
,
581 struct extent_buffer
*new_root
, gfp_t flags
)
583 struct tree_mod_elem
*tm
;
586 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
590 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
591 tm
->old_root
.logical
= old_root
->start
;
592 tm
->old_root
.level
= btrfs_header_level(old_root
);
593 tm
->generation
= btrfs_header_generation(old_root
);
594 tm
->op
= MOD_LOG_ROOT_REPLACE
;
596 ret
= __tree_mod_log_insert(fs_info
, tm
);
597 spin_unlock(&fs_info
->tree_mod_seq_lock
);
601 static struct tree_mod_elem
*
602 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
605 struct rb_root
*tm_root
;
606 struct rb_node
*node
;
607 struct tree_mod_elem
*cur
= NULL
;
608 struct tree_mod_elem
*found
= NULL
;
609 u64 index
= start
>> PAGE_CACHE_SHIFT
;
611 read_lock(&fs_info
->tree_mod_log_lock
);
612 tm_root
= &fs_info
->tree_mod_log
;
613 node
= tm_root
->rb_node
;
615 cur
= container_of(node
, struct tree_mod_elem
, node
);
616 if (cur
->index
< index
) {
617 node
= node
->rb_left
;
618 } else if (cur
->index
> index
) {
619 node
= node
->rb_right
;
620 } else if (cur
->elem
.seq
< min_seq
) {
621 node
= node
->rb_left
;
622 } else if (!smallest
) {
623 /* we want the node with the highest seq */
625 BUG_ON(found
->elem
.seq
> cur
->elem
.seq
);
627 node
= node
->rb_left
;
628 } else if (cur
->elem
.seq
> min_seq
) {
629 /* we want the node with the smallest seq */
631 BUG_ON(found
->elem
.seq
< cur
->elem
.seq
);
633 node
= node
->rb_right
;
639 read_unlock(&fs_info
->tree_mod_log_lock
);
645 * this returns the element from the log with the smallest time sequence
646 * value that's in the log (the oldest log item). any element with a time
647 * sequence lower than min_seq will be ignored.
649 static struct tree_mod_elem
*
650 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
653 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
657 * this returns the element from the log with the largest time sequence
658 * value that's in the log (the most recent log item). any element with
659 * a time sequence lower than min_seq will be ignored.
661 static struct tree_mod_elem
*
662 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
664 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
668 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
669 struct extent_buffer
*src
, unsigned long dst_offset
,
670 unsigned long src_offset
, int nr_items
)
675 if (tree_mod_dont_log(fs_info
, NULL
))
678 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
681 /* speed this up by single seq for all operations? */
682 for (i
= 0; i
< nr_items
; i
++) {
683 ret
= tree_mod_log_insert_key(fs_info
, src
, i
+ src_offset
,
686 ret
= tree_mod_log_insert_key(fs_info
, dst
, i
+ dst_offset
,
693 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
694 int dst_offset
, int src_offset
, int nr_items
)
697 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
703 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
704 struct extent_buffer
*eb
,
705 struct btrfs_disk_key
*disk_key
, int slot
, int atomic
)
709 ret
= tree_mod_log_insert_key_mask(fs_info
, eb
, slot
,
711 atomic
? GFP_ATOMIC
: GFP_NOFS
);
715 static void tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
716 struct extent_buffer
*eb
)
722 if (tree_mod_dont_log(fs_info
, eb
))
725 nritems
= btrfs_header_nritems(eb
);
726 for (i
= nritems
- 1; i
>= 0; i
--) {
727 ret
= tree_mod_log_insert_key(fs_info
, eb
, i
,
728 MOD_LOG_KEY_REMOVE_WHILE_FREEING
);
734 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
735 struct extent_buffer
*new_root_node
)
738 tree_mod_log_free_eb(root
->fs_info
, root
->node
);
739 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
740 new_root_node
, GFP_NOFS
);
745 * check if the tree block can be shared by multiple trees
747 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
748 struct extent_buffer
*buf
)
751 * Tree blocks not in refernece counted trees and tree roots
752 * are never shared. If a block was allocated after the last
753 * snapshot and the block was not allocated by tree relocation,
754 * we know the block is not shared.
756 if (root
->ref_cows
&&
757 buf
!= root
->node
&& buf
!= root
->commit_root
&&
758 (btrfs_header_generation(buf
) <=
759 btrfs_root_last_snapshot(&root
->root_item
) ||
760 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
762 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
763 if (root
->ref_cows
&&
764 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
770 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
771 struct btrfs_root
*root
,
772 struct extent_buffer
*buf
,
773 struct extent_buffer
*cow
,
783 * Backrefs update rules:
785 * Always use full backrefs for extent pointers in tree block
786 * allocated by tree relocation.
788 * If a shared tree block is no longer referenced by its owner
789 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
790 * use full backrefs for extent pointers in tree block.
792 * If a tree block is been relocating
793 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
794 * use full backrefs for extent pointers in tree block.
795 * The reason for this is some operations (such as drop tree)
796 * are only allowed for blocks use full backrefs.
799 if (btrfs_block_can_be_shared(root
, buf
)) {
800 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
801 buf
->len
, &refs
, &flags
);
806 btrfs_std_error(root
->fs_info
, ret
);
811 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
812 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
813 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
818 owner
= btrfs_header_owner(buf
);
819 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
820 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
823 if ((owner
== root
->root_key
.objectid
||
824 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
825 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
826 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
827 BUG_ON(ret
); /* -ENOMEM */
829 if (root
->root_key
.objectid
==
830 BTRFS_TREE_RELOC_OBJECTID
) {
831 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
832 BUG_ON(ret
); /* -ENOMEM */
833 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
834 BUG_ON(ret
); /* -ENOMEM */
836 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
839 if (root
->root_key
.objectid
==
840 BTRFS_TREE_RELOC_OBJECTID
)
841 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
843 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
844 BUG_ON(ret
); /* -ENOMEM */
846 if (new_flags
!= 0) {
847 ret
= btrfs_set_disk_extent_flags(trans
, root
,
855 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
856 if (root
->root_key
.objectid
==
857 BTRFS_TREE_RELOC_OBJECTID
)
858 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
860 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
861 BUG_ON(ret
); /* -ENOMEM */
862 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
863 BUG_ON(ret
); /* -ENOMEM */
866 * don't log freeing in case we're freeing the root node, this
867 * is done by tree_mod_log_set_root_pointer later
869 if (buf
!= root
->node
&& btrfs_header_level(buf
) != 0)
870 tree_mod_log_free_eb(root
->fs_info
, buf
);
871 clean_tree_block(trans
, root
, buf
);
878 * does the dirty work in cow of a single block. The parent block (if
879 * supplied) is updated to point to the new cow copy. The new buffer is marked
880 * dirty and returned locked. If you modify the block it needs to be marked
883 * search_start -- an allocation hint for the new block
885 * empty_size -- a hint that you plan on doing more cow. This is the size in
886 * bytes the allocator should try to find free next to the block it returns.
887 * This is just a hint and may be ignored by the allocator.
889 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
890 struct btrfs_root
*root
,
891 struct extent_buffer
*buf
,
892 struct extent_buffer
*parent
, int parent_slot
,
893 struct extent_buffer
**cow_ret
,
894 u64 search_start
, u64 empty_size
)
896 struct btrfs_disk_key disk_key
;
897 struct extent_buffer
*cow
;
906 btrfs_assert_tree_locked(buf
);
908 WARN_ON(root
->ref_cows
&& trans
->transid
!=
909 root
->fs_info
->running_transaction
->transid
);
910 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
912 level
= btrfs_header_level(buf
);
915 btrfs_item_key(buf
, &disk_key
, 0);
917 btrfs_node_key(buf
, &disk_key
, 0);
919 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
921 parent_start
= parent
->start
;
927 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
928 root
->root_key
.objectid
, &disk_key
,
929 level
, search_start
, empty_size
);
933 /* cow is set to blocking by btrfs_init_new_buffer */
935 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
936 btrfs_set_header_bytenr(cow
, cow
->start
);
937 btrfs_set_header_generation(cow
, trans
->transid
);
938 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
939 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
940 BTRFS_HEADER_FLAG_RELOC
);
941 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
942 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
944 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
946 write_extent_buffer(cow
, root
->fs_info
->fsid
,
947 (unsigned long)btrfs_header_fsid(cow
),
950 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
952 btrfs_abort_transaction(trans
, root
, ret
);
957 btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
959 if (buf
== root
->node
) {
960 WARN_ON(parent
&& parent
!= buf
);
961 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
962 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
963 parent_start
= buf
->start
;
967 extent_buffer_get(cow
);
968 tree_mod_log_set_root_pointer(root
, cow
);
969 rcu_assign_pointer(root
->node
, cow
);
971 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
973 free_extent_buffer(buf
);
974 add_root_to_dirty_list(root
);
976 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
977 parent_start
= parent
->start
;
981 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
982 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
983 MOD_LOG_KEY_REPLACE
);
984 btrfs_set_node_blockptr(parent
, parent_slot
,
986 btrfs_set_node_ptr_generation(parent
, parent_slot
,
988 btrfs_mark_buffer_dirty(parent
);
989 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
993 btrfs_tree_unlock(buf
);
994 free_extent_buffer_stale(buf
);
995 btrfs_mark_buffer_dirty(cow
);
1001 * returns the logical address of the oldest predecessor of the given root.
1002 * entries older than time_seq are ignored.
1004 static struct tree_mod_elem
*
1005 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1006 struct btrfs_root
*root
, u64 time_seq
)
1008 struct tree_mod_elem
*tm
;
1009 struct tree_mod_elem
*found
= NULL
;
1010 u64 root_logical
= root
->node
->start
;
1017 * the very last operation that's logged for a root is the replacement
1018 * operation (if it is replaced at all). this has the index of the *new*
1019 * root, making it the very first operation that's logged for this root.
1022 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1027 * if there are no tree operation for the oldest root, we simply
1028 * return it. this should only happen if that (old) root is at
1035 * if there's an operation that's not a root replacement, we
1036 * found the oldest version of our root. normally, we'll find a
1037 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1039 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1043 root_logical
= tm
->old_root
.logical
;
1044 BUG_ON(root_logical
== root
->node
->start
);
1048 /* if there's no old root to return, return what we found instead */
1056 * tm is a pointer to the first operation to rewind within eb. then, all
1057 * previous operations will be rewinded (until we reach something older than
1061 __tree_mod_log_rewind(struct extent_buffer
*eb
, u64 time_seq
,
1062 struct tree_mod_elem
*first_tm
)
1065 struct rb_node
*next
;
1066 struct tree_mod_elem
*tm
= first_tm
;
1067 unsigned long o_dst
;
1068 unsigned long o_src
;
1069 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1071 n
= btrfs_header_nritems(eb
);
1072 while (tm
&& tm
->elem
.seq
>= time_seq
) {
1074 * all the operations are recorded with the operator used for
1075 * the modification. as we're going backwards, we do the
1076 * opposite of each operation here.
1079 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1080 BUG_ON(tm
->slot
< n
);
1081 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1082 case MOD_LOG_KEY_REMOVE
:
1083 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1084 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1085 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1089 case MOD_LOG_KEY_REPLACE
:
1090 BUG_ON(tm
->slot
>= n
);
1091 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1092 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1093 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1096 case MOD_LOG_KEY_ADD
:
1097 if (tm
->slot
!= n
- 1) {
1098 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1099 o_src
= btrfs_node_key_ptr_offset(tm
->slot
+ 1);
1100 memmove_extent_buffer(eb
, o_dst
, o_src
, p_size
);
1104 case MOD_LOG_MOVE_KEYS
:
1105 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1106 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1107 memmove_extent_buffer(eb
, o_dst
, o_src
,
1108 tm
->move
.nr_items
* p_size
);
1110 case MOD_LOG_ROOT_REPLACE
:
1112 * this operation is special. for roots, this must be
1113 * handled explicitly before rewinding.
1114 * for non-roots, this operation may exist if the node
1115 * was a root: root A -> child B; then A gets empty and
1116 * B is promoted to the new root. in the mod log, we'll
1117 * have a root-replace operation for B, a tree block
1118 * that is no root. we simply ignore that operation.
1122 next
= rb_next(&tm
->node
);
1125 tm
= container_of(next
, struct tree_mod_elem
, node
);
1126 if (tm
->index
!= first_tm
->index
)
1129 btrfs_set_header_nritems(eb
, n
);
1132 static struct extent_buffer
*
1133 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1136 struct extent_buffer
*eb_rewin
;
1137 struct tree_mod_elem
*tm
;
1142 if (btrfs_header_level(eb
) == 0)
1145 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1149 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1150 BUG_ON(tm
->slot
!= 0);
1151 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1152 fs_info
->tree_root
->nodesize
);
1154 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1155 btrfs_set_header_backref_rev(eb_rewin
,
1156 btrfs_header_backref_rev(eb
));
1157 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1158 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1160 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1164 extent_buffer_get(eb_rewin
);
1165 free_extent_buffer(eb
);
1167 __tree_mod_log_rewind(eb_rewin
, time_seq
, tm
);
1173 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1174 * value. If there are no changes, the current root->root_node is returned. If
1175 * anything changed in between, there's a fresh buffer allocated on which the
1176 * rewind operations are done. In any case, the returned buffer is read locked.
1177 * Returns NULL on error (with no locks held).
1179 static inline struct extent_buffer
*
1180 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1182 struct tree_mod_elem
*tm
;
1183 struct extent_buffer
*eb
;
1184 struct tree_mod_root
*old_root
= NULL
;
1185 u64 old_generation
= 0;
1188 eb
= btrfs_read_lock_root_node(root
);
1189 tm
= __tree_mod_log_oldest_root(root
->fs_info
, root
, time_seq
);
1193 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1194 old_root
= &tm
->old_root
;
1195 old_generation
= tm
->generation
;
1196 logical
= old_root
->logical
;
1198 logical
= root
->node
->start
;
1201 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1203 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1205 eb
= btrfs_clone_extent_buffer(root
->node
);
1206 btrfs_tree_read_unlock(root
->node
);
1207 free_extent_buffer(root
->node
);
1210 btrfs_tree_read_lock(eb
);
1212 btrfs_set_header_bytenr(eb
, eb
->start
);
1213 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1214 btrfs_set_header_owner(eb
, root
->root_key
.objectid
);
1215 btrfs_set_header_level(eb
, old_root
->level
);
1216 btrfs_set_header_generation(eb
, old_generation
);
1219 __tree_mod_log_rewind(eb
, time_seq
, tm
);
1221 WARN_ON(btrfs_header_level(eb
) != 0);
1222 extent_buffer_get(eb
);
1227 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1228 struct btrfs_root
*root
,
1229 struct extent_buffer
*buf
)
1231 /* ensure we can see the force_cow */
1235 * We do not need to cow a block if
1236 * 1) this block is not created or changed in this transaction;
1237 * 2) this block does not belong to TREE_RELOC tree;
1238 * 3) the root is not forced COW.
1240 * What is forced COW:
1241 * when we create snapshot during commiting the transaction,
1242 * after we've finished coping src root, we must COW the shared
1243 * block to ensure the metadata consistency.
1245 if (btrfs_header_generation(buf
) == trans
->transid
&&
1246 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1247 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1248 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1255 * cows a single block, see __btrfs_cow_block for the real work.
1256 * This version of it has extra checks so that a block isn't cow'd more than
1257 * once per transaction, as long as it hasn't been written yet
1259 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1260 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1261 struct extent_buffer
*parent
, int parent_slot
,
1262 struct extent_buffer
**cow_ret
)
1267 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
1268 printk(KERN_CRIT
"trans %llu running %llu\n",
1269 (unsigned long long)trans
->transid
,
1270 (unsigned long long)
1271 root
->fs_info
->running_transaction
->transid
);
1274 if (trans
->transid
!= root
->fs_info
->generation
) {
1275 printk(KERN_CRIT
"trans %llu running %llu\n",
1276 (unsigned long long)trans
->transid
,
1277 (unsigned long long)root
->fs_info
->generation
);
1281 if (!should_cow_block(trans
, root
, buf
)) {
1286 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1289 btrfs_set_lock_blocking(parent
);
1290 btrfs_set_lock_blocking(buf
);
1292 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1293 parent_slot
, cow_ret
, search_start
, 0);
1295 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1301 * helper function for defrag to decide if two blocks pointed to by a
1302 * node are actually close by
1304 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1306 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1308 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1314 * compare two keys in a memcmp fashion
1316 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1318 struct btrfs_key k1
;
1320 btrfs_disk_key_to_cpu(&k1
, disk
);
1322 return btrfs_comp_cpu_keys(&k1
, k2
);
1326 * same as comp_keys only with two btrfs_key's
1328 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1330 if (k1
->objectid
> k2
->objectid
)
1332 if (k1
->objectid
< k2
->objectid
)
1334 if (k1
->type
> k2
->type
)
1336 if (k1
->type
< k2
->type
)
1338 if (k1
->offset
> k2
->offset
)
1340 if (k1
->offset
< k2
->offset
)
1346 * this is used by the defrag code to go through all the
1347 * leaves pointed to by a node and reallocate them so that
1348 * disk order is close to key order
1350 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1351 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1352 int start_slot
, int cache_only
, u64
*last_ret
,
1353 struct btrfs_key
*progress
)
1355 struct extent_buffer
*cur
;
1358 u64 search_start
= *last_ret
;
1368 int progress_passed
= 0;
1369 struct btrfs_disk_key disk_key
;
1371 parent_level
= btrfs_header_level(parent
);
1372 if (cache_only
&& parent_level
!= 1)
1375 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1377 if (trans
->transid
!= root
->fs_info
->generation
)
1380 parent_nritems
= btrfs_header_nritems(parent
);
1381 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1382 end_slot
= parent_nritems
;
1384 if (parent_nritems
== 1)
1387 btrfs_set_lock_blocking(parent
);
1389 for (i
= start_slot
; i
< end_slot
; i
++) {
1392 btrfs_node_key(parent
, &disk_key
, i
);
1393 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1396 progress_passed
= 1;
1397 blocknr
= btrfs_node_blockptr(parent
, i
);
1398 gen
= btrfs_node_ptr_generation(parent
, i
);
1399 if (last_block
== 0)
1400 last_block
= blocknr
;
1403 other
= btrfs_node_blockptr(parent
, i
- 1);
1404 close
= close_blocks(blocknr
, other
, blocksize
);
1406 if (!close
&& i
< end_slot
- 2) {
1407 other
= btrfs_node_blockptr(parent
, i
+ 1);
1408 close
= close_blocks(blocknr
, other
, blocksize
);
1411 last_block
= blocknr
;
1415 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1417 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1420 if (!cur
|| !uptodate
) {
1422 free_extent_buffer(cur
);
1426 cur
= read_tree_block(root
, blocknr
,
1430 } else if (!uptodate
) {
1431 err
= btrfs_read_buffer(cur
, gen
);
1433 free_extent_buffer(cur
);
1438 if (search_start
== 0)
1439 search_start
= last_block
;
1441 btrfs_tree_lock(cur
);
1442 btrfs_set_lock_blocking(cur
);
1443 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1446 (end_slot
- i
) * blocksize
));
1448 btrfs_tree_unlock(cur
);
1449 free_extent_buffer(cur
);
1452 search_start
= cur
->start
;
1453 last_block
= cur
->start
;
1454 *last_ret
= search_start
;
1455 btrfs_tree_unlock(cur
);
1456 free_extent_buffer(cur
);
1462 * The leaf data grows from end-to-front in the node.
1463 * this returns the address of the start of the last item,
1464 * which is the stop of the leaf data stack
1466 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1467 struct extent_buffer
*leaf
)
1469 u32 nr
= btrfs_header_nritems(leaf
);
1471 return BTRFS_LEAF_DATA_SIZE(root
);
1472 return btrfs_item_offset_nr(leaf
, nr
- 1);
1477 * search for key in the extent_buffer. The items start at offset p,
1478 * and they are item_size apart. There are 'max' items in p.
1480 * the slot in the array is returned via slot, and it points to
1481 * the place where you would insert key if it is not found in
1484 * slot may point to max if the key is bigger than all of the keys
1486 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1488 int item_size
, struct btrfs_key
*key
,
1495 struct btrfs_disk_key
*tmp
= NULL
;
1496 struct btrfs_disk_key unaligned
;
1497 unsigned long offset
;
1499 unsigned long map_start
= 0;
1500 unsigned long map_len
= 0;
1503 while (low
< high
) {
1504 mid
= (low
+ high
) / 2;
1505 offset
= p
+ mid
* item_size
;
1507 if (!kaddr
|| offset
< map_start
||
1508 (offset
+ sizeof(struct btrfs_disk_key
)) >
1509 map_start
+ map_len
) {
1511 err
= map_private_extent_buffer(eb
, offset
,
1512 sizeof(struct btrfs_disk_key
),
1513 &kaddr
, &map_start
, &map_len
);
1516 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1519 read_extent_buffer(eb
, &unaligned
,
1520 offset
, sizeof(unaligned
));
1525 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1528 ret
= comp_keys(tmp
, key
);
1544 * simple bin_search frontend that does the right thing for
1547 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1548 int level
, int *slot
)
1551 return generic_bin_search(eb
,
1552 offsetof(struct btrfs_leaf
, items
),
1553 sizeof(struct btrfs_item
),
1554 key
, btrfs_header_nritems(eb
),
1557 return generic_bin_search(eb
,
1558 offsetof(struct btrfs_node
, ptrs
),
1559 sizeof(struct btrfs_key_ptr
),
1560 key
, btrfs_header_nritems(eb
),
1564 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1565 int level
, int *slot
)
1567 return bin_search(eb
, key
, level
, slot
);
1570 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1572 spin_lock(&root
->accounting_lock
);
1573 btrfs_set_root_used(&root
->root_item
,
1574 btrfs_root_used(&root
->root_item
) + size
);
1575 spin_unlock(&root
->accounting_lock
);
1578 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1580 spin_lock(&root
->accounting_lock
);
1581 btrfs_set_root_used(&root
->root_item
,
1582 btrfs_root_used(&root
->root_item
) - size
);
1583 spin_unlock(&root
->accounting_lock
);
1586 /* given a node and slot number, this reads the blocks it points to. The
1587 * extent buffer is returned with a reference taken (but unlocked).
1588 * NULL is returned on error.
1590 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1591 struct extent_buffer
*parent
, int slot
)
1593 int level
= btrfs_header_level(parent
);
1596 if (slot
>= btrfs_header_nritems(parent
))
1601 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1602 btrfs_level_size(root
, level
- 1),
1603 btrfs_node_ptr_generation(parent
, slot
));
1607 * node level balancing, used to make sure nodes are in proper order for
1608 * item deletion. We balance from the top down, so we have to make sure
1609 * that a deletion won't leave an node completely empty later on.
1611 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1612 struct btrfs_root
*root
,
1613 struct btrfs_path
*path
, int level
)
1615 struct extent_buffer
*right
= NULL
;
1616 struct extent_buffer
*mid
;
1617 struct extent_buffer
*left
= NULL
;
1618 struct extent_buffer
*parent
= NULL
;
1622 int orig_slot
= path
->slots
[level
];
1628 mid
= path
->nodes
[level
];
1630 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1631 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1632 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1634 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1636 if (level
< BTRFS_MAX_LEVEL
- 1) {
1637 parent
= path
->nodes
[level
+ 1];
1638 pslot
= path
->slots
[level
+ 1];
1642 * deal with the case where there is only one pointer in the root
1643 * by promoting the node below to a root
1646 struct extent_buffer
*child
;
1648 if (btrfs_header_nritems(mid
) != 1)
1651 /* promote the child to a root */
1652 child
= read_node_slot(root
, mid
, 0);
1655 btrfs_std_error(root
->fs_info
, ret
);
1659 btrfs_tree_lock(child
);
1660 btrfs_set_lock_blocking(child
);
1661 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1663 btrfs_tree_unlock(child
);
1664 free_extent_buffer(child
);
1668 tree_mod_log_set_root_pointer(root
, child
);
1669 rcu_assign_pointer(root
->node
, child
);
1671 add_root_to_dirty_list(root
);
1672 btrfs_tree_unlock(child
);
1674 path
->locks
[level
] = 0;
1675 path
->nodes
[level
] = NULL
;
1676 clean_tree_block(trans
, root
, mid
);
1677 btrfs_tree_unlock(mid
);
1678 /* once for the path */
1679 free_extent_buffer(mid
);
1681 root_sub_used(root
, mid
->len
);
1682 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1683 /* once for the root ptr */
1684 free_extent_buffer_stale(mid
);
1687 if (btrfs_header_nritems(mid
) >
1688 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1691 left
= read_node_slot(root
, parent
, pslot
- 1);
1693 btrfs_tree_lock(left
);
1694 btrfs_set_lock_blocking(left
);
1695 wret
= btrfs_cow_block(trans
, root
, left
,
1696 parent
, pslot
- 1, &left
);
1702 right
= read_node_slot(root
, parent
, pslot
+ 1);
1704 btrfs_tree_lock(right
);
1705 btrfs_set_lock_blocking(right
);
1706 wret
= btrfs_cow_block(trans
, root
, right
,
1707 parent
, pslot
+ 1, &right
);
1714 /* first, try to make some room in the middle buffer */
1716 orig_slot
+= btrfs_header_nritems(left
);
1717 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1723 * then try to empty the right most buffer into the middle
1726 wret
= push_node_left(trans
, root
, mid
, right
, 1);
1727 if (wret
< 0 && wret
!= -ENOSPC
)
1729 if (btrfs_header_nritems(right
) == 0) {
1730 clean_tree_block(trans
, root
, right
);
1731 btrfs_tree_unlock(right
);
1732 del_ptr(trans
, root
, path
, level
+ 1, pslot
+ 1, 1);
1733 root_sub_used(root
, right
->len
);
1734 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
1735 free_extent_buffer_stale(right
);
1738 struct btrfs_disk_key right_key
;
1739 btrfs_node_key(right
, &right_key
, 0);
1740 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1741 &right_key
, pslot
+ 1, 0);
1742 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1743 btrfs_mark_buffer_dirty(parent
);
1746 if (btrfs_header_nritems(mid
) == 1) {
1748 * we're not allowed to leave a node with one item in the
1749 * tree during a delete. A deletion from lower in the tree
1750 * could try to delete the only pointer in this node.
1751 * So, pull some keys from the left.
1752 * There has to be a left pointer at this point because
1753 * otherwise we would have pulled some pointers from the
1758 btrfs_std_error(root
->fs_info
, ret
);
1761 wret
= balance_node_right(trans
, root
, mid
, left
);
1767 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1773 if (btrfs_header_nritems(mid
) == 0) {
1774 clean_tree_block(trans
, root
, mid
);
1775 btrfs_tree_unlock(mid
);
1776 del_ptr(trans
, root
, path
, level
+ 1, pslot
, 1);
1777 root_sub_used(root
, mid
->len
);
1778 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1779 free_extent_buffer_stale(mid
);
1782 /* update the parent key to reflect our changes */
1783 struct btrfs_disk_key mid_key
;
1784 btrfs_node_key(mid
, &mid_key
, 0);
1785 tree_mod_log_set_node_key(root
->fs_info
, parent
, &mid_key
,
1787 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1788 btrfs_mark_buffer_dirty(parent
);
1791 /* update the path */
1793 if (btrfs_header_nritems(left
) > orig_slot
) {
1794 extent_buffer_get(left
);
1795 /* left was locked after cow */
1796 path
->nodes
[level
] = left
;
1797 path
->slots
[level
+ 1] -= 1;
1798 path
->slots
[level
] = orig_slot
;
1800 btrfs_tree_unlock(mid
);
1801 free_extent_buffer(mid
);
1804 orig_slot
-= btrfs_header_nritems(left
);
1805 path
->slots
[level
] = orig_slot
;
1808 /* double check we haven't messed things up */
1810 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
1814 btrfs_tree_unlock(right
);
1815 free_extent_buffer(right
);
1818 if (path
->nodes
[level
] != left
)
1819 btrfs_tree_unlock(left
);
1820 free_extent_buffer(left
);
1825 /* Node balancing for insertion. Here we only split or push nodes around
1826 * when they are completely full. This is also done top down, so we
1827 * have to be pessimistic.
1829 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
1830 struct btrfs_root
*root
,
1831 struct btrfs_path
*path
, int level
)
1833 struct extent_buffer
*right
= NULL
;
1834 struct extent_buffer
*mid
;
1835 struct extent_buffer
*left
= NULL
;
1836 struct extent_buffer
*parent
= NULL
;
1840 int orig_slot
= path
->slots
[level
];
1845 mid
= path
->nodes
[level
];
1846 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1848 if (level
< BTRFS_MAX_LEVEL
- 1) {
1849 parent
= path
->nodes
[level
+ 1];
1850 pslot
= path
->slots
[level
+ 1];
1856 left
= read_node_slot(root
, parent
, pslot
- 1);
1858 /* first, try to make some room in the middle buffer */
1862 btrfs_tree_lock(left
);
1863 btrfs_set_lock_blocking(left
);
1865 left_nr
= btrfs_header_nritems(left
);
1866 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1869 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1874 wret
= push_node_left(trans
, root
,
1881 struct btrfs_disk_key disk_key
;
1882 orig_slot
+= left_nr
;
1883 btrfs_node_key(mid
, &disk_key
, 0);
1884 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1885 &disk_key
, pslot
, 0);
1886 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1887 btrfs_mark_buffer_dirty(parent
);
1888 if (btrfs_header_nritems(left
) > orig_slot
) {
1889 path
->nodes
[level
] = left
;
1890 path
->slots
[level
+ 1] -= 1;
1891 path
->slots
[level
] = orig_slot
;
1892 btrfs_tree_unlock(mid
);
1893 free_extent_buffer(mid
);
1896 btrfs_header_nritems(left
);
1897 path
->slots
[level
] = orig_slot
;
1898 btrfs_tree_unlock(left
);
1899 free_extent_buffer(left
);
1903 btrfs_tree_unlock(left
);
1904 free_extent_buffer(left
);
1906 right
= read_node_slot(root
, parent
, pslot
+ 1);
1909 * then try to empty the right most buffer into the middle
1914 btrfs_tree_lock(right
);
1915 btrfs_set_lock_blocking(right
);
1917 right_nr
= btrfs_header_nritems(right
);
1918 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1921 ret
= btrfs_cow_block(trans
, root
, right
,
1927 wret
= balance_node_right(trans
, root
,
1934 struct btrfs_disk_key disk_key
;
1936 btrfs_node_key(right
, &disk_key
, 0);
1937 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1938 &disk_key
, pslot
+ 1, 0);
1939 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
1940 btrfs_mark_buffer_dirty(parent
);
1942 if (btrfs_header_nritems(mid
) <= orig_slot
) {
1943 path
->nodes
[level
] = right
;
1944 path
->slots
[level
+ 1] += 1;
1945 path
->slots
[level
] = orig_slot
-
1946 btrfs_header_nritems(mid
);
1947 btrfs_tree_unlock(mid
);
1948 free_extent_buffer(mid
);
1950 btrfs_tree_unlock(right
);
1951 free_extent_buffer(right
);
1955 btrfs_tree_unlock(right
);
1956 free_extent_buffer(right
);
1962 * readahead one full node of leaves, finding things that are close
1963 * to the block in 'slot', and triggering ra on them.
1965 static void reada_for_search(struct btrfs_root
*root
,
1966 struct btrfs_path
*path
,
1967 int level
, int slot
, u64 objectid
)
1969 struct extent_buffer
*node
;
1970 struct btrfs_disk_key disk_key
;
1976 int direction
= path
->reada
;
1977 struct extent_buffer
*eb
;
1985 if (!path
->nodes
[level
])
1988 node
= path
->nodes
[level
];
1990 search
= btrfs_node_blockptr(node
, slot
);
1991 blocksize
= btrfs_level_size(root
, level
- 1);
1992 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
1994 free_extent_buffer(eb
);
2000 nritems
= btrfs_header_nritems(node
);
2004 if (direction
< 0) {
2008 } else if (direction
> 0) {
2013 if (path
->reada
< 0 && objectid
) {
2014 btrfs_node_key(node
, &disk_key
, nr
);
2015 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2018 search
= btrfs_node_blockptr(node
, nr
);
2019 if ((search
<= target
&& target
- search
<= 65536) ||
2020 (search
> target
&& search
- target
<= 65536)) {
2021 gen
= btrfs_node_ptr_generation(node
, nr
);
2022 readahead_tree_block(root
, search
, blocksize
, gen
);
2026 if ((nread
> 65536 || nscan
> 32))
2032 * returns -EAGAIN if it had to drop the path, or zero if everything was in
2035 static noinline
int reada_for_balance(struct btrfs_root
*root
,
2036 struct btrfs_path
*path
, int level
)
2040 struct extent_buffer
*parent
;
2041 struct extent_buffer
*eb
;
2048 parent
= path
->nodes
[level
+ 1];
2052 nritems
= btrfs_header_nritems(parent
);
2053 slot
= path
->slots
[level
+ 1];
2054 blocksize
= btrfs_level_size(root
, level
);
2057 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2058 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2059 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2061 * if we get -eagain from btrfs_buffer_uptodate, we
2062 * don't want to return eagain here. That will loop
2065 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2067 free_extent_buffer(eb
);
2069 if (slot
+ 1 < nritems
) {
2070 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2071 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2072 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2073 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2075 free_extent_buffer(eb
);
2077 if (block1
|| block2
) {
2080 /* release the whole path */
2081 btrfs_release_path(path
);
2083 /* read the blocks */
2085 readahead_tree_block(root
, block1
, blocksize
, 0);
2087 readahead_tree_block(root
, block2
, blocksize
, 0);
2090 eb
= read_tree_block(root
, block1
, blocksize
, 0);
2091 free_extent_buffer(eb
);
2094 eb
= read_tree_block(root
, block2
, blocksize
, 0);
2095 free_extent_buffer(eb
);
2103 * when we walk down the tree, it is usually safe to unlock the higher layers
2104 * in the tree. The exceptions are when our path goes through slot 0, because
2105 * operations on the tree might require changing key pointers higher up in the
2108 * callers might also have set path->keep_locks, which tells this code to keep
2109 * the lock if the path points to the last slot in the block. This is part of
2110 * walking through the tree, and selecting the next slot in the higher block.
2112 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2113 * if lowest_unlock is 1, level 0 won't be unlocked
2115 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2116 int lowest_unlock
, int min_write_lock_level
,
2117 int *write_lock_level
)
2120 int skip_level
= level
;
2122 struct extent_buffer
*t
;
2124 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2125 if (!path
->nodes
[i
])
2127 if (!path
->locks
[i
])
2129 if (!no_skips
&& path
->slots
[i
] == 0) {
2133 if (!no_skips
&& path
->keep_locks
) {
2136 nritems
= btrfs_header_nritems(t
);
2137 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2142 if (skip_level
< i
&& i
>= lowest_unlock
)
2146 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2147 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2149 if (write_lock_level
&&
2150 i
> min_write_lock_level
&&
2151 i
<= *write_lock_level
) {
2152 *write_lock_level
= i
- 1;
2159 * This releases any locks held in the path starting at level and
2160 * going all the way up to the root.
2162 * btrfs_search_slot will keep the lock held on higher nodes in a few
2163 * corner cases, such as COW of the block at slot zero in the node. This
2164 * ignores those rules, and it should only be called when there are no
2165 * more updates to be done higher up in the tree.
2167 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2171 if (path
->keep_locks
)
2174 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2175 if (!path
->nodes
[i
])
2177 if (!path
->locks
[i
])
2179 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2185 * helper function for btrfs_search_slot. The goal is to find a block
2186 * in cache without setting the path to blocking. If we find the block
2187 * we return zero and the path is unchanged.
2189 * If we can't find the block, we set the path blocking and do some
2190 * reada. -EAGAIN is returned and the search must be repeated.
2193 read_block_for_search(struct btrfs_trans_handle
*trans
,
2194 struct btrfs_root
*root
, struct btrfs_path
*p
,
2195 struct extent_buffer
**eb_ret
, int level
, int slot
,
2196 struct btrfs_key
*key
, u64 time_seq
)
2201 struct extent_buffer
*b
= *eb_ret
;
2202 struct extent_buffer
*tmp
;
2205 blocknr
= btrfs_node_blockptr(b
, slot
);
2206 gen
= btrfs_node_ptr_generation(b
, slot
);
2207 blocksize
= btrfs_level_size(root
, level
- 1);
2209 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2211 /* first we do an atomic uptodate check */
2212 if (btrfs_buffer_uptodate(tmp
, 0, 1) > 0) {
2213 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2215 * we found an up to date block without
2222 /* the pages were up to date, but we failed
2223 * the generation number check. Do a full
2224 * read for the generation number that is correct.
2225 * We must do this without dropping locks so
2226 * we can trust our generation number
2228 free_extent_buffer(tmp
);
2229 btrfs_set_path_blocking(p
);
2231 /* now we're allowed to do a blocking uptodate check */
2232 tmp
= read_tree_block(root
, blocknr
, blocksize
, gen
);
2233 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
, 0) > 0) {
2237 free_extent_buffer(tmp
);
2238 btrfs_release_path(p
);
2244 * reduce lock contention at high levels
2245 * of the btree by dropping locks before
2246 * we read. Don't release the lock on the current
2247 * level because we need to walk this node to figure
2248 * out which blocks to read.
2250 btrfs_unlock_up_safe(p
, level
+ 1);
2251 btrfs_set_path_blocking(p
);
2253 free_extent_buffer(tmp
);
2255 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2257 btrfs_release_path(p
);
2260 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2263 * If the read above didn't mark this buffer up to date,
2264 * it will never end up being up to date. Set ret to EIO now
2265 * and give up so that our caller doesn't loop forever
2268 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2270 free_extent_buffer(tmp
);
2276 * helper function for btrfs_search_slot. This does all of the checks
2277 * for node-level blocks and does any balancing required based on
2280 * If no extra work was required, zero is returned. If we had to
2281 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2285 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2286 struct btrfs_root
*root
, struct btrfs_path
*p
,
2287 struct extent_buffer
*b
, int level
, int ins_len
,
2288 int *write_lock_level
)
2291 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2292 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2295 if (*write_lock_level
< level
+ 1) {
2296 *write_lock_level
= level
+ 1;
2297 btrfs_release_path(p
);
2301 sret
= reada_for_balance(root
, p
, level
);
2305 btrfs_set_path_blocking(p
);
2306 sret
= split_node(trans
, root
, p
, level
);
2307 btrfs_clear_path_blocking(p
, NULL
, 0);
2314 b
= p
->nodes
[level
];
2315 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2316 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2319 if (*write_lock_level
< level
+ 1) {
2320 *write_lock_level
= level
+ 1;
2321 btrfs_release_path(p
);
2325 sret
= reada_for_balance(root
, p
, level
);
2329 btrfs_set_path_blocking(p
);
2330 sret
= balance_level(trans
, root
, p
, level
);
2331 btrfs_clear_path_blocking(p
, NULL
, 0);
2337 b
= p
->nodes
[level
];
2339 btrfs_release_path(p
);
2342 BUG_ON(btrfs_header_nritems(b
) == 1);
2353 * look for key in the tree. path is filled in with nodes along the way
2354 * if key is found, we return zero and you can find the item in the leaf
2355 * level of the path (level 0)
2357 * If the key isn't found, the path points to the slot where it should
2358 * be inserted, and 1 is returned. If there are other errors during the
2359 * search a negative error number is returned.
2361 * if ins_len > 0, nodes and leaves will be split as we walk down the
2362 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2365 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2366 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2369 struct extent_buffer
*b
;
2374 int lowest_unlock
= 1;
2376 /* everything at write_lock_level or lower must be write locked */
2377 int write_lock_level
= 0;
2378 u8 lowest_level
= 0;
2379 int min_write_lock_level
;
2381 lowest_level
= p
->lowest_level
;
2382 WARN_ON(lowest_level
&& ins_len
> 0);
2383 WARN_ON(p
->nodes
[0] != NULL
);
2388 /* when we are removing items, we might have to go up to level
2389 * two as we update tree pointers Make sure we keep write
2390 * for those levels as well
2392 write_lock_level
= 2;
2393 } else if (ins_len
> 0) {
2395 * for inserting items, make sure we have a write lock on
2396 * level 1 so we can update keys
2398 write_lock_level
= 1;
2402 write_lock_level
= -1;
2404 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2405 write_lock_level
= BTRFS_MAX_LEVEL
;
2407 min_write_lock_level
= write_lock_level
;
2411 * we try very hard to do read locks on the root
2413 root_lock
= BTRFS_READ_LOCK
;
2415 if (p
->search_commit_root
) {
2417 * the commit roots are read only
2418 * so we always do read locks
2420 b
= root
->commit_root
;
2421 extent_buffer_get(b
);
2422 level
= btrfs_header_level(b
);
2423 if (!p
->skip_locking
)
2424 btrfs_tree_read_lock(b
);
2426 if (p
->skip_locking
) {
2427 b
= btrfs_root_node(root
);
2428 level
= btrfs_header_level(b
);
2430 /* we don't know the level of the root node
2431 * until we actually have it read locked
2433 b
= btrfs_read_lock_root_node(root
);
2434 level
= btrfs_header_level(b
);
2435 if (level
<= write_lock_level
) {
2436 /* whoops, must trade for write lock */
2437 btrfs_tree_read_unlock(b
);
2438 free_extent_buffer(b
);
2439 b
= btrfs_lock_root_node(root
);
2440 root_lock
= BTRFS_WRITE_LOCK
;
2442 /* the level might have changed, check again */
2443 level
= btrfs_header_level(b
);
2447 p
->nodes
[level
] = b
;
2448 if (!p
->skip_locking
)
2449 p
->locks
[level
] = root_lock
;
2452 level
= btrfs_header_level(b
);
2455 * setup the path here so we can release it under lock
2456 * contention with the cow code
2460 * if we don't really need to cow this block
2461 * then we don't want to set the path blocking,
2462 * so we test it here
2464 if (!should_cow_block(trans
, root
, b
))
2467 btrfs_set_path_blocking(p
);
2470 * must have write locks on this node and the
2473 if (level
+ 1 > write_lock_level
) {
2474 write_lock_level
= level
+ 1;
2475 btrfs_release_path(p
);
2479 err
= btrfs_cow_block(trans
, root
, b
,
2480 p
->nodes
[level
+ 1],
2481 p
->slots
[level
+ 1], &b
);
2488 BUG_ON(!cow
&& ins_len
);
2490 p
->nodes
[level
] = b
;
2491 btrfs_clear_path_blocking(p
, NULL
, 0);
2494 * we have a lock on b and as long as we aren't changing
2495 * the tree, there is no way to for the items in b to change.
2496 * It is safe to drop the lock on our parent before we
2497 * go through the expensive btree search on b.
2499 * If cow is true, then we might be changing slot zero,
2500 * which may require changing the parent. So, we can't
2501 * drop the lock until after we know which slot we're
2505 btrfs_unlock_up_safe(p
, level
+ 1);
2507 ret
= bin_search(b
, key
, level
, &slot
);
2511 if (ret
&& slot
> 0) {
2515 p
->slots
[level
] = slot
;
2516 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2517 ins_len
, &write_lock_level
);
2524 b
= p
->nodes
[level
];
2525 slot
= p
->slots
[level
];
2528 * slot 0 is special, if we change the key
2529 * we have to update the parent pointer
2530 * which means we must have a write lock
2533 if (slot
== 0 && cow
&&
2534 write_lock_level
< level
+ 1) {
2535 write_lock_level
= level
+ 1;
2536 btrfs_release_path(p
);
2540 unlock_up(p
, level
, lowest_unlock
,
2541 min_write_lock_level
, &write_lock_level
);
2543 if (level
== lowest_level
) {
2549 err
= read_block_for_search(trans
, root
, p
,
2550 &b
, level
, slot
, key
, 0);
2558 if (!p
->skip_locking
) {
2559 level
= btrfs_header_level(b
);
2560 if (level
<= write_lock_level
) {
2561 err
= btrfs_try_tree_write_lock(b
);
2563 btrfs_set_path_blocking(p
);
2565 btrfs_clear_path_blocking(p
, b
,
2568 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2570 err
= btrfs_try_tree_read_lock(b
);
2572 btrfs_set_path_blocking(p
);
2573 btrfs_tree_read_lock(b
);
2574 btrfs_clear_path_blocking(p
, b
,
2577 p
->locks
[level
] = BTRFS_READ_LOCK
;
2579 p
->nodes
[level
] = b
;
2582 p
->slots
[level
] = slot
;
2584 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2585 if (write_lock_level
< 1) {
2586 write_lock_level
= 1;
2587 btrfs_release_path(p
);
2591 btrfs_set_path_blocking(p
);
2592 err
= split_leaf(trans
, root
, key
,
2593 p
, ins_len
, ret
== 0);
2594 btrfs_clear_path_blocking(p
, NULL
, 0);
2602 if (!p
->search_for_split
)
2603 unlock_up(p
, level
, lowest_unlock
,
2604 min_write_lock_level
, &write_lock_level
);
2611 * we don't really know what they plan on doing with the path
2612 * from here on, so for now just mark it as blocking
2614 if (!p
->leave_spinning
)
2615 btrfs_set_path_blocking(p
);
2617 btrfs_release_path(p
);
2622 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2623 * current state of the tree together with the operations recorded in the tree
2624 * modification log to search for the key in a previous version of this tree, as
2625 * denoted by the time_seq parameter.
2627 * Naturally, there is no support for insert, delete or cow operations.
2629 * The resulting path and return value will be set up as if we called
2630 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2632 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2633 struct btrfs_path
*p
, u64 time_seq
)
2635 struct extent_buffer
*b
;
2640 int lowest_unlock
= 1;
2641 u8 lowest_level
= 0;
2643 lowest_level
= p
->lowest_level
;
2644 WARN_ON(p
->nodes
[0] != NULL
);
2646 if (p
->search_commit_root
) {
2648 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2652 b
= get_old_root(root
, time_seq
);
2653 level
= btrfs_header_level(b
);
2654 p
->locks
[level
] = BTRFS_READ_LOCK
;
2657 level
= btrfs_header_level(b
);
2658 p
->nodes
[level
] = b
;
2659 btrfs_clear_path_blocking(p
, NULL
, 0);
2662 * we have a lock on b and as long as we aren't changing
2663 * the tree, there is no way to for the items in b to change.
2664 * It is safe to drop the lock on our parent before we
2665 * go through the expensive btree search on b.
2667 btrfs_unlock_up_safe(p
, level
+ 1);
2669 ret
= bin_search(b
, key
, level
, &slot
);
2673 if (ret
&& slot
> 0) {
2677 p
->slots
[level
] = slot
;
2678 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2680 if (level
== lowest_level
) {
2686 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
2687 slot
, key
, time_seq
);
2695 level
= btrfs_header_level(b
);
2696 err
= btrfs_try_tree_read_lock(b
);
2698 btrfs_set_path_blocking(p
);
2699 btrfs_tree_read_lock(b
);
2700 btrfs_clear_path_blocking(p
, b
,
2703 p
->locks
[level
] = BTRFS_READ_LOCK
;
2704 p
->nodes
[level
] = b
;
2705 b
= tree_mod_log_rewind(root
->fs_info
, b
, time_seq
);
2706 if (b
!= p
->nodes
[level
]) {
2707 btrfs_tree_unlock_rw(p
->nodes
[level
],
2709 p
->locks
[level
] = 0;
2710 p
->nodes
[level
] = b
;
2713 p
->slots
[level
] = slot
;
2714 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2720 if (!p
->leave_spinning
)
2721 btrfs_set_path_blocking(p
);
2723 btrfs_release_path(p
);
2729 * adjust the pointers going up the tree, starting at level
2730 * making sure the right key of each node is points to 'key'.
2731 * This is used after shifting pointers to the left, so it stops
2732 * fixing up pointers when a given leaf/node is not in slot 0 of the
2736 static void fixup_low_keys(struct btrfs_trans_handle
*trans
,
2737 struct btrfs_root
*root
, struct btrfs_path
*path
,
2738 struct btrfs_disk_key
*key
, int level
)
2741 struct extent_buffer
*t
;
2743 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2744 int tslot
= path
->slots
[i
];
2745 if (!path
->nodes
[i
])
2748 tree_mod_log_set_node_key(root
->fs_info
, t
, key
, tslot
, 1);
2749 btrfs_set_node_key(t
, key
, tslot
);
2750 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
2759 * This function isn't completely safe. It's the caller's responsibility
2760 * that the new key won't break the order
2762 void btrfs_set_item_key_safe(struct btrfs_trans_handle
*trans
,
2763 struct btrfs_root
*root
, struct btrfs_path
*path
,
2764 struct btrfs_key
*new_key
)
2766 struct btrfs_disk_key disk_key
;
2767 struct extent_buffer
*eb
;
2770 eb
= path
->nodes
[0];
2771 slot
= path
->slots
[0];
2773 btrfs_item_key(eb
, &disk_key
, slot
- 1);
2774 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
2776 if (slot
< btrfs_header_nritems(eb
) - 1) {
2777 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
2778 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
2781 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2782 btrfs_set_item_key(eb
, &disk_key
, slot
);
2783 btrfs_mark_buffer_dirty(eb
);
2785 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2789 * try to push data from one node into the next node left in the
2792 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2793 * error, and > 0 if there was no room in the left hand block.
2795 static int push_node_left(struct btrfs_trans_handle
*trans
,
2796 struct btrfs_root
*root
, struct extent_buffer
*dst
,
2797 struct extent_buffer
*src
, int empty
)
2804 src_nritems
= btrfs_header_nritems(src
);
2805 dst_nritems
= btrfs_header_nritems(dst
);
2806 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
2807 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
2808 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
2810 if (!empty
&& src_nritems
<= 8)
2813 if (push_items
<= 0)
2817 push_items
= min(src_nritems
, push_items
);
2818 if (push_items
< src_nritems
) {
2819 /* leave at least 8 pointers in the node if
2820 * we aren't going to empty it
2822 if (src_nritems
- push_items
< 8) {
2823 if (push_items
<= 8)
2829 push_items
= min(src_nritems
- 8, push_items
);
2831 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
2833 copy_extent_buffer(dst
, src
,
2834 btrfs_node_key_ptr_offset(dst_nritems
),
2835 btrfs_node_key_ptr_offset(0),
2836 push_items
* sizeof(struct btrfs_key_ptr
));
2838 if (push_items
< src_nritems
) {
2839 tree_mod_log_eb_move(root
->fs_info
, src
, 0, push_items
,
2840 src_nritems
- push_items
);
2841 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
2842 btrfs_node_key_ptr_offset(push_items
),
2843 (src_nritems
- push_items
) *
2844 sizeof(struct btrfs_key_ptr
));
2846 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
2847 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
2848 btrfs_mark_buffer_dirty(src
);
2849 btrfs_mark_buffer_dirty(dst
);
2855 * try to push data from one node into the next node right in the
2858 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
2859 * error, and > 0 if there was no room in the right hand block.
2861 * this will only push up to 1/2 the contents of the left node over
2863 static int balance_node_right(struct btrfs_trans_handle
*trans
,
2864 struct btrfs_root
*root
,
2865 struct extent_buffer
*dst
,
2866 struct extent_buffer
*src
)
2874 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
2875 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
2877 src_nritems
= btrfs_header_nritems(src
);
2878 dst_nritems
= btrfs_header_nritems(dst
);
2879 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
2880 if (push_items
<= 0)
2883 if (src_nritems
< 4)
2886 max_push
= src_nritems
/ 2 + 1;
2887 /* don't try to empty the node */
2888 if (max_push
>= src_nritems
)
2891 if (max_push
< push_items
)
2892 push_items
= max_push
;
2894 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
2895 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
2896 btrfs_node_key_ptr_offset(0),
2898 sizeof(struct btrfs_key_ptr
));
2900 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
2901 src_nritems
- push_items
, push_items
);
2902 copy_extent_buffer(dst
, src
,
2903 btrfs_node_key_ptr_offset(0),
2904 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
2905 push_items
* sizeof(struct btrfs_key_ptr
));
2907 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
2908 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
2910 btrfs_mark_buffer_dirty(src
);
2911 btrfs_mark_buffer_dirty(dst
);
2917 * helper function to insert a new root level in the tree.
2918 * A new node is allocated, and a single item is inserted to
2919 * point to the existing root
2921 * returns zero on success or < 0 on failure.
2923 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
2924 struct btrfs_root
*root
,
2925 struct btrfs_path
*path
, int level
)
2928 struct extent_buffer
*lower
;
2929 struct extent_buffer
*c
;
2930 struct extent_buffer
*old
;
2931 struct btrfs_disk_key lower_key
;
2933 BUG_ON(path
->nodes
[level
]);
2934 BUG_ON(path
->nodes
[level
-1] != root
->node
);
2936 lower
= path
->nodes
[level
-1];
2938 btrfs_item_key(lower
, &lower_key
, 0);
2940 btrfs_node_key(lower
, &lower_key
, 0);
2942 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
2943 root
->root_key
.objectid
, &lower_key
,
2944 level
, root
->node
->start
, 0);
2948 root_add_used(root
, root
->nodesize
);
2950 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
2951 btrfs_set_header_nritems(c
, 1);
2952 btrfs_set_header_level(c
, level
);
2953 btrfs_set_header_bytenr(c
, c
->start
);
2954 btrfs_set_header_generation(c
, trans
->transid
);
2955 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
2956 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
2958 write_extent_buffer(c
, root
->fs_info
->fsid
,
2959 (unsigned long)btrfs_header_fsid(c
),
2962 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
2963 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
2966 btrfs_set_node_key(c
, &lower_key
, 0);
2967 btrfs_set_node_blockptr(c
, 0, lower
->start
);
2968 lower_gen
= btrfs_header_generation(lower
);
2969 WARN_ON(lower_gen
!= trans
->transid
);
2971 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
2973 btrfs_mark_buffer_dirty(c
);
2976 tree_mod_log_set_root_pointer(root
, c
);
2977 rcu_assign_pointer(root
->node
, c
);
2979 /* the super has an extra ref to root->node */
2980 free_extent_buffer(old
);
2982 add_root_to_dirty_list(root
);
2983 extent_buffer_get(c
);
2984 path
->nodes
[level
] = c
;
2985 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
2986 path
->slots
[level
] = 0;
2991 * worker function to insert a single pointer in a node.
2992 * the node should have enough room for the pointer already
2994 * slot and level indicate where you want the key to go, and
2995 * blocknr is the block the key points to.
2997 static void insert_ptr(struct btrfs_trans_handle
*trans
,
2998 struct btrfs_root
*root
, struct btrfs_path
*path
,
2999 struct btrfs_disk_key
*key
, u64 bytenr
,
3000 int slot
, int level
)
3002 struct extent_buffer
*lower
;
3006 BUG_ON(!path
->nodes
[level
]);
3007 btrfs_assert_tree_locked(path
->nodes
[level
]);
3008 lower
= path
->nodes
[level
];
3009 nritems
= btrfs_header_nritems(lower
);
3010 BUG_ON(slot
> nritems
);
3011 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3012 if (slot
!= nritems
) {
3014 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3015 slot
, nritems
- slot
);
3016 memmove_extent_buffer(lower
,
3017 btrfs_node_key_ptr_offset(slot
+ 1),
3018 btrfs_node_key_ptr_offset(slot
),
3019 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3022 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3026 btrfs_set_node_key(lower
, key
, slot
);
3027 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3028 WARN_ON(trans
->transid
== 0);
3029 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3030 btrfs_set_header_nritems(lower
, nritems
+ 1);
3031 btrfs_mark_buffer_dirty(lower
);
3035 * split the node at the specified level in path in two.
3036 * The path is corrected to point to the appropriate node after the split
3038 * Before splitting this tries to make some room in the node by pushing
3039 * left and right, if either one works, it returns right away.
3041 * returns 0 on success and < 0 on failure
3043 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3044 struct btrfs_root
*root
,
3045 struct btrfs_path
*path
, int level
)
3047 struct extent_buffer
*c
;
3048 struct extent_buffer
*split
;
3049 struct btrfs_disk_key disk_key
;
3054 c
= path
->nodes
[level
];
3055 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3056 if (c
== root
->node
) {
3057 /* trying to split the root, lets make a new one */
3058 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3062 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3063 c
= path
->nodes
[level
];
3064 if (!ret
&& btrfs_header_nritems(c
) <
3065 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3071 c_nritems
= btrfs_header_nritems(c
);
3072 mid
= (c_nritems
+ 1) / 2;
3073 btrfs_node_key(c
, &disk_key
, mid
);
3075 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3076 root
->root_key
.objectid
,
3077 &disk_key
, level
, c
->start
, 0);
3079 return PTR_ERR(split
);
3081 root_add_used(root
, root
->nodesize
);
3083 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3084 btrfs_set_header_level(split
, btrfs_header_level(c
));
3085 btrfs_set_header_bytenr(split
, split
->start
);
3086 btrfs_set_header_generation(split
, trans
->transid
);
3087 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3088 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3089 write_extent_buffer(split
, root
->fs_info
->fsid
,
3090 (unsigned long)btrfs_header_fsid(split
),
3092 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3093 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
3096 tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0, mid
, c_nritems
- mid
);
3097 copy_extent_buffer(split
, c
,
3098 btrfs_node_key_ptr_offset(0),
3099 btrfs_node_key_ptr_offset(mid
),
3100 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3101 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3102 btrfs_set_header_nritems(c
, mid
);
3105 btrfs_mark_buffer_dirty(c
);
3106 btrfs_mark_buffer_dirty(split
);
3108 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3109 path
->slots
[level
+ 1] + 1, level
+ 1);
3111 if (path
->slots
[level
] >= mid
) {
3112 path
->slots
[level
] -= mid
;
3113 btrfs_tree_unlock(c
);
3114 free_extent_buffer(c
);
3115 path
->nodes
[level
] = split
;
3116 path
->slots
[level
+ 1] += 1;
3118 btrfs_tree_unlock(split
);
3119 free_extent_buffer(split
);
3125 * how many bytes are required to store the items in a leaf. start
3126 * and nr indicate which items in the leaf to check. This totals up the
3127 * space used both by the item structs and the item data
3129 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3132 int nritems
= btrfs_header_nritems(l
);
3133 int end
= min(nritems
, start
+ nr
) - 1;
3137 data_len
= btrfs_item_end_nr(l
, start
);
3138 data_len
= data_len
- btrfs_item_offset_nr(l
, end
);
3139 data_len
+= sizeof(struct btrfs_item
) * nr
;
3140 WARN_ON(data_len
< 0);
3145 * The space between the end of the leaf items and
3146 * the start of the leaf data. IOW, how much room
3147 * the leaf has left for both items and data
3149 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3150 struct extent_buffer
*leaf
)
3152 int nritems
= btrfs_header_nritems(leaf
);
3154 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3156 printk(KERN_CRIT
"leaf free space ret %d, leaf data size %lu, "
3157 "used %d nritems %d\n",
3158 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3159 leaf_space_used(leaf
, 0, nritems
), nritems
);
3165 * min slot controls the lowest index we're willing to push to the
3166 * right. We'll push up to and including min_slot, but no lower
3168 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3169 struct btrfs_root
*root
,
3170 struct btrfs_path
*path
,
3171 int data_size
, int empty
,
3172 struct extent_buffer
*right
,
3173 int free_space
, u32 left_nritems
,
3176 struct extent_buffer
*left
= path
->nodes
[0];
3177 struct extent_buffer
*upper
= path
->nodes
[1];
3178 struct btrfs_map_token token
;
3179 struct btrfs_disk_key disk_key
;
3184 struct btrfs_item
*item
;
3190 btrfs_init_map_token(&token
);
3195 nr
= max_t(u32
, 1, min_slot
);
3197 if (path
->slots
[0] >= left_nritems
)
3198 push_space
+= data_size
;
3200 slot
= path
->slots
[1];
3201 i
= left_nritems
- 1;
3203 item
= btrfs_item_nr(left
, i
);
3205 if (!empty
&& push_items
> 0) {
3206 if (path
->slots
[0] > i
)
3208 if (path
->slots
[0] == i
) {
3209 int space
= btrfs_leaf_free_space(root
, left
);
3210 if (space
+ push_space
* 2 > free_space
)
3215 if (path
->slots
[0] == i
)
3216 push_space
+= data_size
;
3218 this_item_size
= btrfs_item_size(left
, item
);
3219 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3223 push_space
+= this_item_size
+ sizeof(*item
);
3229 if (push_items
== 0)
3232 if (!empty
&& push_items
== left_nritems
)
3235 /* push left to right */
3236 right_nritems
= btrfs_header_nritems(right
);
3238 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3239 push_space
-= leaf_data_end(root
, left
);
3241 /* make room in the right data area */
3242 data_end
= leaf_data_end(root
, right
);
3243 memmove_extent_buffer(right
,
3244 btrfs_leaf_data(right
) + data_end
- push_space
,
3245 btrfs_leaf_data(right
) + data_end
,
3246 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3248 /* copy from the left data area */
3249 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3250 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3251 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3254 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3255 btrfs_item_nr_offset(0),
3256 right_nritems
* sizeof(struct btrfs_item
));
3258 /* copy the items from left to right */
3259 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3260 btrfs_item_nr_offset(left_nritems
- push_items
),
3261 push_items
* sizeof(struct btrfs_item
));
3263 /* update the item pointers */
3264 right_nritems
+= push_items
;
3265 btrfs_set_header_nritems(right
, right_nritems
);
3266 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3267 for (i
= 0; i
< right_nritems
; i
++) {
3268 item
= btrfs_item_nr(right
, i
);
3269 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3270 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3273 left_nritems
-= push_items
;
3274 btrfs_set_header_nritems(left
, left_nritems
);
3277 btrfs_mark_buffer_dirty(left
);
3279 clean_tree_block(trans
, root
, left
);
3281 btrfs_mark_buffer_dirty(right
);
3283 btrfs_item_key(right
, &disk_key
, 0);
3284 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3285 btrfs_mark_buffer_dirty(upper
);
3287 /* then fixup the leaf pointer in the path */
3288 if (path
->slots
[0] >= left_nritems
) {
3289 path
->slots
[0] -= left_nritems
;
3290 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3291 clean_tree_block(trans
, root
, path
->nodes
[0]);
3292 btrfs_tree_unlock(path
->nodes
[0]);
3293 free_extent_buffer(path
->nodes
[0]);
3294 path
->nodes
[0] = right
;
3295 path
->slots
[1] += 1;
3297 btrfs_tree_unlock(right
);
3298 free_extent_buffer(right
);
3303 btrfs_tree_unlock(right
);
3304 free_extent_buffer(right
);
3309 * push some data in the path leaf to the right, trying to free up at
3310 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3312 * returns 1 if the push failed because the other node didn't have enough
3313 * room, 0 if everything worked out and < 0 if there were major errors.
3315 * this will push starting from min_slot to the end of the leaf. It won't
3316 * push any slot lower than min_slot
3318 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3319 *root
, struct btrfs_path
*path
,
3320 int min_data_size
, int data_size
,
3321 int empty
, u32 min_slot
)
3323 struct extent_buffer
*left
= path
->nodes
[0];
3324 struct extent_buffer
*right
;
3325 struct extent_buffer
*upper
;
3331 if (!path
->nodes
[1])
3334 slot
= path
->slots
[1];
3335 upper
= path
->nodes
[1];
3336 if (slot
>= btrfs_header_nritems(upper
) - 1)
3339 btrfs_assert_tree_locked(path
->nodes
[1]);
3341 right
= read_node_slot(root
, upper
, slot
+ 1);
3345 btrfs_tree_lock(right
);
3346 btrfs_set_lock_blocking(right
);
3348 free_space
= btrfs_leaf_free_space(root
, right
);
3349 if (free_space
< data_size
)
3352 /* cow and double check */
3353 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3358 free_space
= btrfs_leaf_free_space(root
, right
);
3359 if (free_space
< data_size
)
3362 left_nritems
= btrfs_header_nritems(left
);
3363 if (left_nritems
== 0)
3366 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3367 right
, free_space
, left_nritems
, min_slot
);
3369 btrfs_tree_unlock(right
);
3370 free_extent_buffer(right
);
3375 * push some data in the path leaf to the left, trying to free up at
3376 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3378 * max_slot can put a limit on how far into the leaf we'll push items. The
3379 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3382 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3383 struct btrfs_root
*root
,
3384 struct btrfs_path
*path
, int data_size
,
3385 int empty
, struct extent_buffer
*left
,
3386 int free_space
, u32 right_nritems
,
3389 struct btrfs_disk_key disk_key
;
3390 struct extent_buffer
*right
= path
->nodes
[0];
3394 struct btrfs_item
*item
;
3395 u32 old_left_nritems
;
3399 u32 old_left_item_size
;
3400 struct btrfs_map_token token
;
3402 btrfs_init_map_token(&token
);
3405 nr
= min(right_nritems
, max_slot
);
3407 nr
= min(right_nritems
- 1, max_slot
);
3409 for (i
= 0; i
< nr
; i
++) {
3410 item
= btrfs_item_nr(right
, i
);
3412 if (!empty
&& push_items
> 0) {
3413 if (path
->slots
[0] < i
)
3415 if (path
->slots
[0] == i
) {
3416 int space
= btrfs_leaf_free_space(root
, right
);
3417 if (space
+ push_space
* 2 > free_space
)
3422 if (path
->slots
[0] == i
)
3423 push_space
+= data_size
;
3425 this_item_size
= btrfs_item_size(right
, item
);
3426 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3430 push_space
+= this_item_size
+ sizeof(*item
);
3433 if (push_items
== 0) {
3437 if (!empty
&& push_items
== btrfs_header_nritems(right
))
3440 /* push data from right to left */
3441 copy_extent_buffer(left
, right
,
3442 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3443 btrfs_item_nr_offset(0),
3444 push_items
* sizeof(struct btrfs_item
));
3446 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3447 btrfs_item_offset_nr(right
, push_items
- 1);
3449 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3450 leaf_data_end(root
, left
) - push_space
,
3451 btrfs_leaf_data(right
) +
3452 btrfs_item_offset_nr(right
, push_items
- 1),
3454 old_left_nritems
= btrfs_header_nritems(left
);
3455 BUG_ON(old_left_nritems
<= 0);
3457 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3458 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3461 item
= btrfs_item_nr(left
, i
);
3463 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3464 btrfs_set_token_item_offset(left
, item
,
3465 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3468 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3470 /* fixup right node */
3471 if (push_items
> right_nritems
) {
3472 printk(KERN_CRIT
"push items %d nr %u\n", push_items
,
3477 if (push_items
< right_nritems
) {
3478 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3479 leaf_data_end(root
, right
);
3480 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3481 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3482 btrfs_leaf_data(right
) +
3483 leaf_data_end(root
, right
), push_space
);
3485 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3486 btrfs_item_nr_offset(push_items
),
3487 (btrfs_header_nritems(right
) - push_items
) *
3488 sizeof(struct btrfs_item
));
3490 right_nritems
-= push_items
;
3491 btrfs_set_header_nritems(right
, right_nritems
);
3492 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3493 for (i
= 0; i
< right_nritems
; i
++) {
3494 item
= btrfs_item_nr(right
, i
);
3496 push_space
= push_space
- btrfs_token_item_size(right
,
3498 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3501 btrfs_mark_buffer_dirty(left
);
3503 btrfs_mark_buffer_dirty(right
);
3505 clean_tree_block(trans
, root
, right
);
3507 btrfs_item_key(right
, &disk_key
, 0);
3508 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3510 /* then fixup the leaf pointer in the path */
3511 if (path
->slots
[0] < push_items
) {
3512 path
->slots
[0] += old_left_nritems
;
3513 btrfs_tree_unlock(path
->nodes
[0]);
3514 free_extent_buffer(path
->nodes
[0]);
3515 path
->nodes
[0] = left
;
3516 path
->slots
[1] -= 1;
3518 btrfs_tree_unlock(left
);
3519 free_extent_buffer(left
);
3520 path
->slots
[0] -= push_items
;
3522 BUG_ON(path
->slots
[0] < 0);
3525 btrfs_tree_unlock(left
);
3526 free_extent_buffer(left
);
3531 * push some data in the path leaf to the left, trying to free up at
3532 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3534 * max_slot can put a limit on how far into the leaf we'll push items. The
3535 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3538 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3539 *root
, struct btrfs_path
*path
, int min_data_size
,
3540 int data_size
, int empty
, u32 max_slot
)
3542 struct extent_buffer
*right
= path
->nodes
[0];
3543 struct extent_buffer
*left
;
3549 slot
= path
->slots
[1];
3552 if (!path
->nodes
[1])
3555 right_nritems
= btrfs_header_nritems(right
);
3556 if (right_nritems
== 0)
3559 btrfs_assert_tree_locked(path
->nodes
[1]);
3561 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
3565 btrfs_tree_lock(left
);
3566 btrfs_set_lock_blocking(left
);
3568 free_space
= btrfs_leaf_free_space(root
, left
);
3569 if (free_space
< data_size
) {
3574 /* cow and double check */
3575 ret
= btrfs_cow_block(trans
, root
, left
,
3576 path
->nodes
[1], slot
- 1, &left
);
3578 /* we hit -ENOSPC, but it isn't fatal here */
3584 free_space
= btrfs_leaf_free_space(root
, left
);
3585 if (free_space
< data_size
) {
3590 return __push_leaf_left(trans
, root
, path
, min_data_size
,
3591 empty
, left
, free_space
, right_nritems
,
3594 btrfs_tree_unlock(left
);
3595 free_extent_buffer(left
);
3600 * split the path's leaf in two, making sure there is at least data_size
3601 * available for the resulting leaf level of the path.
3603 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
3604 struct btrfs_root
*root
,
3605 struct btrfs_path
*path
,
3606 struct extent_buffer
*l
,
3607 struct extent_buffer
*right
,
3608 int slot
, int mid
, int nritems
)
3613 struct btrfs_disk_key disk_key
;
3614 struct btrfs_map_token token
;
3616 btrfs_init_map_token(&token
);
3618 nritems
= nritems
- mid
;
3619 btrfs_set_header_nritems(right
, nritems
);
3620 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
3622 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
3623 btrfs_item_nr_offset(mid
),
3624 nritems
* sizeof(struct btrfs_item
));
3626 copy_extent_buffer(right
, l
,
3627 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
3628 data_copy_size
, btrfs_leaf_data(l
) +
3629 leaf_data_end(root
, l
), data_copy_size
);
3631 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
3632 btrfs_item_end_nr(l
, mid
);
3634 for (i
= 0; i
< nritems
; i
++) {
3635 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
3638 ioff
= btrfs_token_item_offset(right
, item
, &token
);
3639 btrfs_set_token_item_offset(right
, item
,
3640 ioff
+ rt_data_off
, &token
);
3643 btrfs_set_header_nritems(l
, mid
);
3644 btrfs_item_key(right
, &disk_key
, 0);
3645 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3646 path
->slots
[1] + 1, 1);
3648 btrfs_mark_buffer_dirty(right
);
3649 btrfs_mark_buffer_dirty(l
);
3650 BUG_ON(path
->slots
[0] != slot
);
3653 btrfs_tree_unlock(path
->nodes
[0]);
3654 free_extent_buffer(path
->nodes
[0]);
3655 path
->nodes
[0] = right
;
3656 path
->slots
[0] -= mid
;
3657 path
->slots
[1] += 1;
3659 btrfs_tree_unlock(right
);
3660 free_extent_buffer(right
);
3663 BUG_ON(path
->slots
[0] < 0);
3667 * double splits happen when we need to insert a big item in the middle
3668 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3669 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3672 * We avoid this by trying to push the items on either side of our target
3673 * into the adjacent leaves. If all goes well we can avoid the double split
3676 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
3677 struct btrfs_root
*root
,
3678 struct btrfs_path
*path
,
3686 slot
= path
->slots
[0];
3689 * try to push all the items after our slot into the
3692 ret
= push_leaf_right(trans
, root
, path
, 1, data_size
, 0, slot
);
3699 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3701 * our goal is to get our slot at the start or end of a leaf. If
3702 * we've done so we're done
3704 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
3707 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
3710 /* try to push all the items before our slot into the next leaf */
3711 slot
= path
->slots
[0];
3712 ret
= push_leaf_left(trans
, root
, path
, 1, data_size
, 0, slot
);
3725 * split the path's leaf in two, making sure there is at least data_size
3726 * available for the resulting leaf level of the path.
3728 * returns 0 if all went well and < 0 on failure.
3730 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
3731 struct btrfs_root
*root
,
3732 struct btrfs_key
*ins_key
,
3733 struct btrfs_path
*path
, int data_size
,
3736 struct btrfs_disk_key disk_key
;
3737 struct extent_buffer
*l
;
3741 struct extent_buffer
*right
;
3745 int num_doubles
= 0;
3746 int tried_avoid_double
= 0;
3749 slot
= path
->slots
[0];
3750 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
3751 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
3754 /* first try to make some room by pushing left and right */
3756 wret
= push_leaf_right(trans
, root
, path
, data_size
,
3761 wret
= push_leaf_left(trans
, root
, path
, data_size
,
3762 data_size
, 0, (u32
)-1);
3768 /* did the pushes work? */
3769 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
3773 if (!path
->nodes
[1]) {
3774 ret
= insert_new_root(trans
, root
, path
, 1);
3781 slot
= path
->slots
[0];
3782 nritems
= btrfs_header_nritems(l
);
3783 mid
= (nritems
+ 1) / 2;
3787 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
3788 BTRFS_LEAF_DATA_SIZE(root
)) {
3789 if (slot
>= nritems
) {
3793 if (mid
!= nritems
&&
3794 leaf_space_used(l
, mid
, nritems
- mid
) +
3795 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3796 if (data_size
&& !tried_avoid_double
)
3797 goto push_for_double
;
3803 if (leaf_space_used(l
, 0, mid
) + data_size
>
3804 BTRFS_LEAF_DATA_SIZE(root
)) {
3805 if (!extend
&& data_size
&& slot
== 0) {
3807 } else if ((extend
|| !data_size
) && slot
== 0) {
3811 if (mid
!= nritems
&&
3812 leaf_space_used(l
, mid
, nritems
- mid
) +
3813 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3814 if (data_size
&& !tried_avoid_double
)
3815 goto push_for_double
;
3823 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
3825 btrfs_item_key(l
, &disk_key
, mid
);
3827 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
3828 root
->root_key
.objectid
,
3829 &disk_key
, 0, l
->start
, 0);
3831 return PTR_ERR(right
);
3833 root_add_used(root
, root
->leafsize
);
3835 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
3836 btrfs_set_header_bytenr(right
, right
->start
);
3837 btrfs_set_header_generation(right
, trans
->transid
);
3838 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
3839 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
3840 btrfs_set_header_level(right
, 0);
3841 write_extent_buffer(right
, root
->fs_info
->fsid
,
3842 (unsigned long)btrfs_header_fsid(right
),
3845 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
3846 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
3851 btrfs_set_header_nritems(right
, 0);
3852 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3853 path
->slots
[1] + 1, 1);
3854 btrfs_tree_unlock(path
->nodes
[0]);
3855 free_extent_buffer(path
->nodes
[0]);
3856 path
->nodes
[0] = right
;
3858 path
->slots
[1] += 1;
3860 btrfs_set_header_nritems(right
, 0);
3861 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3863 btrfs_tree_unlock(path
->nodes
[0]);
3864 free_extent_buffer(path
->nodes
[0]);
3865 path
->nodes
[0] = right
;
3867 if (path
->slots
[1] == 0)
3868 fixup_low_keys(trans
, root
, path
,
3871 btrfs_mark_buffer_dirty(right
);
3875 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
3878 BUG_ON(num_doubles
!= 0);
3886 push_for_double_split(trans
, root
, path
, data_size
);
3887 tried_avoid_double
= 1;
3888 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
3893 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
3894 struct btrfs_root
*root
,
3895 struct btrfs_path
*path
, int ins_len
)
3897 struct btrfs_key key
;
3898 struct extent_buffer
*leaf
;
3899 struct btrfs_file_extent_item
*fi
;
3904 leaf
= path
->nodes
[0];
3905 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
3907 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
3908 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
3910 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
3913 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
3914 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
3915 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3916 struct btrfs_file_extent_item
);
3917 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
3919 btrfs_release_path(path
);
3921 path
->keep_locks
= 1;
3922 path
->search_for_split
= 1;
3923 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
3924 path
->search_for_split
= 0;
3929 leaf
= path
->nodes
[0];
3930 /* if our item isn't there or got smaller, return now */
3931 if (ret
> 0 || item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
3934 /* the leaf has changed, it now has room. return now */
3935 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
3938 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
3939 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
3940 struct btrfs_file_extent_item
);
3941 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
3945 btrfs_set_path_blocking(path
);
3946 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
3950 path
->keep_locks
= 0;
3951 btrfs_unlock_up_safe(path
, 1);
3954 path
->keep_locks
= 0;
3958 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
3959 struct btrfs_root
*root
,
3960 struct btrfs_path
*path
,
3961 struct btrfs_key
*new_key
,
3962 unsigned long split_offset
)
3964 struct extent_buffer
*leaf
;
3965 struct btrfs_item
*item
;
3966 struct btrfs_item
*new_item
;
3972 struct btrfs_disk_key disk_key
;
3974 leaf
= path
->nodes
[0];
3975 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
3977 btrfs_set_path_blocking(path
);
3979 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
3980 orig_offset
= btrfs_item_offset(leaf
, item
);
3981 item_size
= btrfs_item_size(leaf
, item
);
3983 buf
= kmalloc(item_size
, GFP_NOFS
);
3987 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
3988 path
->slots
[0]), item_size
);
3990 slot
= path
->slots
[0] + 1;
3991 nritems
= btrfs_header_nritems(leaf
);
3992 if (slot
!= nritems
) {
3993 /* shift the items */
3994 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
3995 btrfs_item_nr_offset(slot
),
3996 (nritems
- slot
) * sizeof(struct btrfs_item
));
3999 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4000 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4002 new_item
= btrfs_item_nr(leaf
, slot
);
4004 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4005 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4007 btrfs_set_item_offset(leaf
, item
,
4008 orig_offset
+ item_size
- split_offset
);
4009 btrfs_set_item_size(leaf
, item
, split_offset
);
4011 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4013 /* write the data for the start of the original item */
4014 write_extent_buffer(leaf
, buf
,
4015 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4018 /* write the data for the new item */
4019 write_extent_buffer(leaf
, buf
+ split_offset
,
4020 btrfs_item_ptr_offset(leaf
, slot
),
4021 item_size
- split_offset
);
4022 btrfs_mark_buffer_dirty(leaf
);
4024 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4030 * This function splits a single item into two items,
4031 * giving 'new_key' to the new item and splitting the
4032 * old one at split_offset (from the start of the item).
4034 * The path may be released by this operation. After
4035 * the split, the path is pointing to the old item. The
4036 * new item is going to be in the same node as the old one.
4038 * Note, the item being split must be smaller enough to live alone on
4039 * a tree block with room for one extra struct btrfs_item
4041 * This allows us to split the item in place, keeping a lock on the
4042 * leaf the entire time.
4044 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4045 struct btrfs_root
*root
,
4046 struct btrfs_path
*path
,
4047 struct btrfs_key
*new_key
,
4048 unsigned long split_offset
)
4051 ret
= setup_leaf_for_split(trans
, root
, path
,
4052 sizeof(struct btrfs_item
));
4056 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4061 * This function duplicate a item, giving 'new_key' to the new item.
4062 * It guarantees both items live in the same tree leaf and the new item
4063 * is contiguous with the original item.
4065 * This allows us to split file extent in place, keeping a lock on the
4066 * leaf the entire time.
4068 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4069 struct btrfs_root
*root
,
4070 struct btrfs_path
*path
,
4071 struct btrfs_key
*new_key
)
4073 struct extent_buffer
*leaf
;
4077 leaf
= path
->nodes
[0];
4078 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4079 ret
= setup_leaf_for_split(trans
, root
, path
,
4080 item_size
+ sizeof(struct btrfs_item
));
4085 setup_items_for_insert(trans
, root
, path
, new_key
, &item_size
,
4086 item_size
, item_size
+
4087 sizeof(struct btrfs_item
), 1);
4088 leaf
= path
->nodes
[0];
4089 memcpy_extent_buffer(leaf
,
4090 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4091 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4097 * make the item pointed to by the path smaller. new_size indicates
4098 * how small to make it, and from_end tells us if we just chop bytes
4099 * off the end of the item or if we shift the item to chop bytes off
4102 void btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
4103 struct btrfs_root
*root
,
4104 struct btrfs_path
*path
,
4105 u32 new_size
, int from_end
)
4108 struct extent_buffer
*leaf
;
4109 struct btrfs_item
*item
;
4111 unsigned int data_end
;
4112 unsigned int old_data_start
;
4113 unsigned int old_size
;
4114 unsigned int size_diff
;
4116 struct btrfs_map_token token
;
4118 btrfs_init_map_token(&token
);
4120 leaf
= path
->nodes
[0];
4121 slot
= path
->slots
[0];
4123 old_size
= btrfs_item_size_nr(leaf
, slot
);
4124 if (old_size
== new_size
)
4127 nritems
= btrfs_header_nritems(leaf
);
4128 data_end
= leaf_data_end(root
, leaf
);
4130 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4132 size_diff
= old_size
- new_size
;
4135 BUG_ON(slot
>= nritems
);
4138 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4140 /* first correct the data pointers */
4141 for (i
= slot
; i
< nritems
; i
++) {
4143 item
= btrfs_item_nr(leaf
, i
);
4145 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4146 btrfs_set_token_item_offset(leaf
, item
,
4147 ioff
+ size_diff
, &token
);
4150 /* shift the data */
4152 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4153 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4154 data_end
, old_data_start
+ new_size
- data_end
);
4156 struct btrfs_disk_key disk_key
;
4159 btrfs_item_key(leaf
, &disk_key
, slot
);
4161 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4163 struct btrfs_file_extent_item
*fi
;
4165 fi
= btrfs_item_ptr(leaf
, slot
,
4166 struct btrfs_file_extent_item
);
4167 fi
= (struct btrfs_file_extent_item
*)(
4168 (unsigned long)fi
- size_diff
);
4170 if (btrfs_file_extent_type(leaf
, fi
) ==
4171 BTRFS_FILE_EXTENT_INLINE
) {
4172 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4173 memmove_extent_buffer(leaf
, ptr
,
4175 offsetof(struct btrfs_file_extent_item
,
4180 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4181 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4182 data_end
, old_data_start
- data_end
);
4184 offset
= btrfs_disk_key_offset(&disk_key
);
4185 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4186 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4188 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4191 item
= btrfs_item_nr(leaf
, slot
);
4192 btrfs_set_item_size(leaf
, item
, new_size
);
4193 btrfs_mark_buffer_dirty(leaf
);
4195 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4196 btrfs_print_leaf(root
, leaf
);
4202 * make the item pointed to by the path bigger, data_size is the new size.
4204 void btrfs_extend_item(struct btrfs_trans_handle
*trans
,
4205 struct btrfs_root
*root
, struct btrfs_path
*path
,
4209 struct extent_buffer
*leaf
;
4210 struct btrfs_item
*item
;
4212 unsigned int data_end
;
4213 unsigned int old_data
;
4214 unsigned int old_size
;
4216 struct btrfs_map_token token
;
4218 btrfs_init_map_token(&token
);
4220 leaf
= path
->nodes
[0];
4222 nritems
= btrfs_header_nritems(leaf
);
4223 data_end
= leaf_data_end(root
, leaf
);
4225 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4226 btrfs_print_leaf(root
, leaf
);
4229 slot
= path
->slots
[0];
4230 old_data
= btrfs_item_end_nr(leaf
, slot
);
4233 if (slot
>= nritems
) {
4234 btrfs_print_leaf(root
, leaf
);
4235 printk(KERN_CRIT
"slot %d too large, nritems %d\n",
4241 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4243 /* first correct the data pointers */
4244 for (i
= slot
; i
< nritems
; i
++) {
4246 item
= btrfs_item_nr(leaf
, i
);
4248 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4249 btrfs_set_token_item_offset(leaf
, item
,
4250 ioff
- data_size
, &token
);
4253 /* shift the data */
4254 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4255 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4256 data_end
, old_data
- data_end
);
4258 data_end
= old_data
;
4259 old_size
= btrfs_item_size_nr(leaf
, slot
);
4260 item
= btrfs_item_nr(leaf
, slot
);
4261 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4262 btrfs_mark_buffer_dirty(leaf
);
4264 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4265 btrfs_print_leaf(root
, leaf
);
4271 * Given a key and some data, insert items into the tree.
4272 * This does all the path init required, making room in the tree if needed.
4273 * Returns the number of keys that were inserted.
4275 int btrfs_insert_some_items(struct btrfs_trans_handle
*trans
,
4276 struct btrfs_root
*root
,
4277 struct btrfs_path
*path
,
4278 struct btrfs_key
*cpu_key
, u32
*data_size
,
4281 struct extent_buffer
*leaf
;
4282 struct btrfs_item
*item
;
4289 unsigned int data_end
;
4290 struct btrfs_disk_key disk_key
;
4291 struct btrfs_key found_key
;
4292 struct btrfs_map_token token
;
4294 btrfs_init_map_token(&token
);
4296 for (i
= 0; i
< nr
; i
++) {
4297 if (total_size
+ data_size
[i
] + sizeof(struct btrfs_item
) >
4298 BTRFS_LEAF_DATA_SIZE(root
)) {
4302 total_data
+= data_size
[i
];
4303 total_size
+= data_size
[i
] + sizeof(struct btrfs_item
);
4307 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4313 leaf
= path
->nodes
[0];
4315 nritems
= btrfs_header_nritems(leaf
);
4316 data_end
= leaf_data_end(root
, leaf
);
4318 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4319 for (i
= nr
; i
>= 0; i
--) {
4320 total_data
-= data_size
[i
];
4321 total_size
-= data_size
[i
] + sizeof(struct btrfs_item
);
4322 if (total_size
< btrfs_leaf_free_space(root
, leaf
))
4328 slot
= path
->slots
[0];
4331 if (slot
!= nritems
) {
4332 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4334 item
= btrfs_item_nr(leaf
, slot
);
4335 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
4337 /* figure out how many keys we can insert in here */
4338 total_data
= data_size
[0];
4339 for (i
= 1; i
< nr
; i
++) {
4340 if (btrfs_comp_cpu_keys(&found_key
, cpu_key
+ i
) <= 0)
4342 total_data
+= data_size
[i
];
4346 if (old_data
< data_end
) {
4347 btrfs_print_leaf(root
, leaf
);
4348 printk(KERN_CRIT
"slot %d old_data %d data_end %d\n",
4349 slot
, old_data
, data_end
);
4353 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4355 /* first correct the data pointers */
4356 for (i
= slot
; i
< nritems
; i
++) {
4359 item
= btrfs_item_nr(leaf
, i
);
4360 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4361 btrfs_set_token_item_offset(leaf
, item
,
4362 ioff
- total_data
, &token
);
4364 /* shift the items */
4365 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4366 btrfs_item_nr_offset(slot
),
4367 (nritems
- slot
) * sizeof(struct btrfs_item
));
4369 /* shift the data */
4370 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4371 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4372 data_end
, old_data
- data_end
);
4373 data_end
= old_data
;
4376 * this sucks but it has to be done, if we are inserting at
4377 * the end of the leaf only insert 1 of the items, since we
4378 * have no way of knowing whats on the next leaf and we'd have
4379 * to drop our current locks to figure it out
4384 /* setup the item for the new data */
4385 for (i
= 0; i
< nr
; i
++) {
4386 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4387 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4388 item
= btrfs_item_nr(leaf
, slot
+ i
);
4389 btrfs_set_token_item_offset(leaf
, item
,
4390 data_end
- data_size
[i
], &token
);
4391 data_end
-= data_size
[i
];
4392 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4394 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4395 btrfs_mark_buffer_dirty(leaf
);
4399 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4400 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4403 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4404 btrfs_print_leaf(root
, leaf
);
4414 * this is a helper for btrfs_insert_empty_items, the main goal here is
4415 * to save stack depth by doing the bulk of the work in a function
4416 * that doesn't call btrfs_search_slot
4418 void setup_items_for_insert(struct btrfs_trans_handle
*trans
,
4419 struct btrfs_root
*root
, struct btrfs_path
*path
,
4420 struct btrfs_key
*cpu_key
, u32
*data_size
,
4421 u32 total_data
, u32 total_size
, int nr
)
4423 struct btrfs_item
*item
;
4426 unsigned int data_end
;
4427 struct btrfs_disk_key disk_key
;
4428 struct extent_buffer
*leaf
;
4430 struct btrfs_map_token token
;
4432 btrfs_init_map_token(&token
);
4434 leaf
= path
->nodes
[0];
4435 slot
= path
->slots
[0];
4437 nritems
= btrfs_header_nritems(leaf
);
4438 data_end
= leaf_data_end(root
, leaf
);
4440 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4441 btrfs_print_leaf(root
, leaf
);
4442 printk(KERN_CRIT
"not enough freespace need %u have %d\n",
4443 total_size
, btrfs_leaf_free_space(root
, leaf
));
4447 if (slot
!= nritems
) {
4448 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4450 if (old_data
< data_end
) {
4451 btrfs_print_leaf(root
, leaf
);
4452 printk(KERN_CRIT
"slot %d old_data %d data_end %d\n",
4453 slot
, old_data
, data_end
);
4457 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4459 /* first correct the data pointers */
4460 for (i
= slot
; i
< nritems
; i
++) {
4463 item
= btrfs_item_nr(leaf
, i
);
4464 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4465 btrfs_set_token_item_offset(leaf
, item
,
4466 ioff
- total_data
, &token
);
4468 /* shift the items */
4469 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4470 btrfs_item_nr_offset(slot
),
4471 (nritems
- slot
) * sizeof(struct btrfs_item
));
4473 /* shift the data */
4474 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4475 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4476 data_end
, old_data
- data_end
);
4477 data_end
= old_data
;
4480 /* setup the item for the new data */
4481 for (i
= 0; i
< nr
; i
++) {
4482 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4483 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4484 item
= btrfs_item_nr(leaf
, slot
+ i
);
4485 btrfs_set_token_item_offset(leaf
, item
,
4486 data_end
- data_size
[i
], &token
);
4487 data_end
-= data_size
[i
];
4488 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4491 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4494 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4495 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4497 btrfs_unlock_up_safe(path
, 1);
4498 btrfs_mark_buffer_dirty(leaf
);
4500 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4501 btrfs_print_leaf(root
, leaf
);
4507 * Given a key and some data, insert items into the tree.
4508 * This does all the path init required, making room in the tree if needed.
4510 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4511 struct btrfs_root
*root
,
4512 struct btrfs_path
*path
,
4513 struct btrfs_key
*cpu_key
, u32
*data_size
,
4522 for (i
= 0; i
< nr
; i
++)
4523 total_data
+= data_size
[i
];
4525 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4526 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4532 slot
= path
->slots
[0];
4535 setup_items_for_insert(trans
, root
, path
, cpu_key
, data_size
,
4536 total_data
, total_size
, nr
);
4541 * Given a key and some data, insert an item into the tree.
4542 * This does all the path init required, making room in the tree if needed.
4544 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4545 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4549 struct btrfs_path
*path
;
4550 struct extent_buffer
*leaf
;
4553 path
= btrfs_alloc_path();
4556 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4558 leaf
= path
->nodes
[0];
4559 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4560 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4561 btrfs_mark_buffer_dirty(leaf
);
4563 btrfs_free_path(path
);
4568 * delete the pointer from a given node.
4570 * the tree should have been previously balanced so the deletion does not
4573 static void del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4574 struct btrfs_path
*path
, int level
, int slot
,
4577 struct extent_buffer
*parent
= path
->nodes
[level
];
4581 nritems
= btrfs_header_nritems(parent
);
4582 if (slot
!= nritems
- 1) {
4583 if (tree_mod_log
&& level
)
4584 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4585 slot
+ 1, nritems
- slot
- 1);
4586 memmove_extent_buffer(parent
,
4587 btrfs_node_key_ptr_offset(slot
),
4588 btrfs_node_key_ptr_offset(slot
+ 1),
4589 sizeof(struct btrfs_key_ptr
) *
4590 (nritems
- slot
- 1));
4591 } else if (tree_mod_log
&& level
) {
4592 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4593 MOD_LOG_KEY_REMOVE
);
4598 btrfs_set_header_nritems(parent
, nritems
);
4599 if (nritems
== 0 && parent
== root
->node
) {
4600 BUG_ON(btrfs_header_level(root
->node
) != 1);
4601 /* just turn the root into a leaf and break */
4602 btrfs_set_header_level(root
->node
, 0);
4603 } else if (slot
== 0) {
4604 struct btrfs_disk_key disk_key
;
4606 btrfs_node_key(parent
, &disk_key
, 0);
4607 fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
4609 btrfs_mark_buffer_dirty(parent
);
4613 * a helper function to delete the leaf pointed to by path->slots[1] and
4616 * This deletes the pointer in path->nodes[1] and frees the leaf
4617 * block extent. zero is returned if it all worked out, < 0 otherwise.
4619 * The path must have already been setup for deleting the leaf, including
4620 * all the proper balancing. path->nodes[1] must be locked.
4622 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4623 struct btrfs_root
*root
,
4624 struct btrfs_path
*path
,
4625 struct extent_buffer
*leaf
)
4627 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4628 del_ptr(trans
, root
, path
, 1, path
->slots
[1], 1);
4631 * btrfs_free_extent is expensive, we want to make sure we
4632 * aren't holding any locks when we call it
4634 btrfs_unlock_up_safe(path
, 0);
4636 root_sub_used(root
, leaf
->len
);
4638 extent_buffer_get(leaf
);
4639 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4640 free_extent_buffer_stale(leaf
);
4643 * delete the item at the leaf level in path. If that empties
4644 * the leaf, remove it from the tree
4646 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4647 struct btrfs_path
*path
, int slot
, int nr
)
4649 struct extent_buffer
*leaf
;
4650 struct btrfs_item
*item
;
4657 struct btrfs_map_token token
;
4659 btrfs_init_map_token(&token
);
4661 leaf
= path
->nodes
[0];
4662 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4664 for (i
= 0; i
< nr
; i
++)
4665 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4667 nritems
= btrfs_header_nritems(leaf
);
4669 if (slot
+ nr
!= nritems
) {
4670 int data_end
= leaf_data_end(root
, leaf
);
4672 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4674 btrfs_leaf_data(leaf
) + data_end
,
4675 last_off
- data_end
);
4677 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4680 item
= btrfs_item_nr(leaf
, i
);
4681 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4682 btrfs_set_token_item_offset(leaf
, item
,
4683 ioff
+ dsize
, &token
);
4686 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4687 btrfs_item_nr_offset(slot
+ nr
),
4688 sizeof(struct btrfs_item
) *
4689 (nritems
- slot
- nr
));
4691 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4694 /* delete the leaf if we've emptied it */
4696 if (leaf
== root
->node
) {
4697 btrfs_set_header_level(leaf
, 0);
4699 btrfs_set_path_blocking(path
);
4700 clean_tree_block(trans
, root
, leaf
);
4701 btrfs_del_leaf(trans
, root
, path
, leaf
);
4704 int used
= leaf_space_used(leaf
, 0, nritems
);
4706 struct btrfs_disk_key disk_key
;
4708 btrfs_item_key(leaf
, &disk_key
, 0);
4709 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4712 /* delete the leaf if it is mostly empty */
4713 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
4714 /* push_leaf_left fixes the path.
4715 * make sure the path still points to our leaf
4716 * for possible call to del_ptr below
4718 slot
= path
->slots
[1];
4719 extent_buffer_get(leaf
);
4721 btrfs_set_path_blocking(path
);
4722 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
4724 if (wret
< 0 && wret
!= -ENOSPC
)
4727 if (path
->nodes
[0] == leaf
&&
4728 btrfs_header_nritems(leaf
)) {
4729 wret
= push_leaf_right(trans
, root
, path
, 1,
4731 if (wret
< 0 && wret
!= -ENOSPC
)
4735 if (btrfs_header_nritems(leaf
) == 0) {
4736 path
->slots
[1] = slot
;
4737 btrfs_del_leaf(trans
, root
, path
, leaf
);
4738 free_extent_buffer(leaf
);
4741 /* if we're still in the path, make sure
4742 * we're dirty. Otherwise, one of the
4743 * push_leaf functions must have already
4744 * dirtied this buffer
4746 if (path
->nodes
[0] == leaf
)
4747 btrfs_mark_buffer_dirty(leaf
);
4748 free_extent_buffer(leaf
);
4751 btrfs_mark_buffer_dirty(leaf
);
4758 * search the tree again to find a leaf with lesser keys
4759 * returns 0 if it found something or 1 if there are no lesser leaves.
4760 * returns < 0 on io errors.
4762 * This may release the path, and so you may lose any locks held at the
4765 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
4767 struct btrfs_key key
;
4768 struct btrfs_disk_key found_key
;
4771 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
4775 else if (key
.type
> 0)
4777 else if (key
.objectid
> 0)
4782 btrfs_release_path(path
);
4783 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4786 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
4787 ret
= comp_keys(&found_key
, &key
);
4794 * A helper function to walk down the tree starting at min_key, and looking
4795 * for nodes or leaves that are either in cache or have a minimum
4796 * transaction id. This is used by the btree defrag code, and tree logging
4798 * This does not cow, but it does stuff the starting key it finds back
4799 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4800 * key and get a writable path.
4802 * This does lock as it descends, and path->keep_locks should be set
4803 * to 1 by the caller.
4805 * This honors path->lowest_level to prevent descent past a given level
4808 * min_trans indicates the oldest transaction that you are interested
4809 * in walking through. Any nodes or leaves older than min_trans are
4810 * skipped over (without reading them).
4812 * returns zero if something useful was found, < 0 on error and 1 if there
4813 * was nothing in the tree that matched the search criteria.
4815 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
4816 struct btrfs_key
*max_key
,
4817 struct btrfs_path
*path
, int cache_only
,
4820 struct extent_buffer
*cur
;
4821 struct btrfs_key found_key
;
4828 WARN_ON(!path
->keep_locks
);
4830 cur
= btrfs_read_lock_root_node(root
);
4831 level
= btrfs_header_level(cur
);
4832 WARN_ON(path
->nodes
[level
]);
4833 path
->nodes
[level
] = cur
;
4834 path
->locks
[level
] = BTRFS_READ_LOCK
;
4836 if (btrfs_header_generation(cur
) < min_trans
) {
4841 nritems
= btrfs_header_nritems(cur
);
4842 level
= btrfs_header_level(cur
);
4843 sret
= bin_search(cur
, min_key
, level
, &slot
);
4845 /* at the lowest level, we're done, setup the path and exit */
4846 if (level
== path
->lowest_level
) {
4847 if (slot
>= nritems
)
4850 path
->slots
[level
] = slot
;
4851 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
4854 if (sret
&& slot
> 0)
4857 * check this node pointer against the cache_only and
4858 * min_trans parameters. If it isn't in cache or is too
4859 * old, skip to the next one.
4861 while (slot
< nritems
) {
4864 struct extent_buffer
*tmp
;
4865 struct btrfs_disk_key disk_key
;
4867 blockptr
= btrfs_node_blockptr(cur
, slot
);
4868 gen
= btrfs_node_ptr_generation(cur
, slot
);
4869 if (gen
< min_trans
) {
4877 btrfs_node_key(cur
, &disk_key
, slot
);
4878 if (comp_keys(&disk_key
, max_key
) >= 0) {
4884 tmp
= btrfs_find_tree_block(root
, blockptr
,
4885 btrfs_level_size(root
, level
- 1));
4887 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
4888 free_extent_buffer(tmp
);
4892 free_extent_buffer(tmp
);
4897 * we didn't find a candidate key in this node, walk forward
4898 * and find another one
4900 if (slot
>= nritems
) {
4901 path
->slots
[level
] = slot
;
4902 btrfs_set_path_blocking(path
);
4903 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
4904 cache_only
, min_trans
);
4906 btrfs_release_path(path
);
4912 /* save our key for returning back */
4913 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
4914 path
->slots
[level
] = slot
;
4915 if (level
== path
->lowest_level
) {
4917 unlock_up(path
, level
, 1, 0, NULL
);
4920 btrfs_set_path_blocking(path
);
4921 cur
= read_node_slot(root
, cur
, slot
);
4922 BUG_ON(!cur
); /* -ENOMEM */
4924 btrfs_tree_read_lock(cur
);
4926 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
4927 path
->nodes
[level
- 1] = cur
;
4928 unlock_up(path
, level
, 1, 0, NULL
);
4929 btrfs_clear_path_blocking(path
, NULL
, 0);
4933 memcpy(min_key
, &found_key
, sizeof(found_key
));
4934 btrfs_set_path_blocking(path
);
4939 * this is similar to btrfs_next_leaf, but does not try to preserve
4940 * and fixup the path. It looks for and returns the next key in the
4941 * tree based on the current path and the cache_only and min_trans
4944 * 0 is returned if another key is found, < 0 if there are any errors
4945 * and 1 is returned if there are no higher keys in the tree
4947 * path->keep_locks should be set to 1 on the search made before
4948 * calling this function.
4950 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
4951 struct btrfs_key
*key
, int level
,
4952 int cache_only
, u64 min_trans
)
4955 struct extent_buffer
*c
;
4957 WARN_ON(!path
->keep_locks
);
4958 while (level
< BTRFS_MAX_LEVEL
) {
4959 if (!path
->nodes
[level
])
4962 slot
= path
->slots
[level
] + 1;
4963 c
= path
->nodes
[level
];
4965 if (slot
>= btrfs_header_nritems(c
)) {
4968 struct btrfs_key cur_key
;
4969 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
4970 !path
->nodes
[level
+ 1])
4973 if (path
->locks
[level
+ 1]) {
4978 slot
= btrfs_header_nritems(c
) - 1;
4980 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
4982 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
4984 orig_lowest
= path
->lowest_level
;
4985 btrfs_release_path(path
);
4986 path
->lowest_level
= level
;
4987 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
4989 path
->lowest_level
= orig_lowest
;
4993 c
= path
->nodes
[level
];
4994 slot
= path
->slots
[level
];
5001 btrfs_item_key_to_cpu(c
, key
, slot
);
5003 u64 blockptr
= btrfs_node_blockptr(c
, slot
);
5004 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5007 struct extent_buffer
*cur
;
5008 cur
= btrfs_find_tree_block(root
, blockptr
,
5009 btrfs_level_size(root
, level
- 1));
5011 btrfs_buffer_uptodate(cur
, gen
, 1) <= 0) {
5014 free_extent_buffer(cur
);
5017 free_extent_buffer(cur
);
5019 if (gen
< min_trans
) {
5023 btrfs_node_key_to_cpu(c
, key
, slot
);
5031 * search the tree again to find a leaf with greater keys
5032 * returns 0 if it found something or 1 if there are no greater leaves.
5033 * returns < 0 on io errors.
5035 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5037 return btrfs_next_old_leaf(root
, path
, 0);
5040 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5045 struct extent_buffer
*c
;
5046 struct extent_buffer
*next
;
5047 struct btrfs_key key
;
5050 int old_spinning
= path
->leave_spinning
;
5051 int next_rw_lock
= 0;
5053 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5057 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5062 btrfs_release_path(path
);
5064 path
->keep_locks
= 1;
5065 path
->leave_spinning
= 1;
5068 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5070 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5071 path
->keep_locks
= 0;
5076 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5078 * by releasing the path above we dropped all our locks. A balance
5079 * could have added more items next to the key that used to be
5080 * at the very end of the block. So, check again here and
5081 * advance the path if there are now more items available.
5083 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5090 while (level
< BTRFS_MAX_LEVEL
) {
5091 if (!path
->nodes
[level
]) {
5096 slot
= path
->slots
[level
] + 1;
5097 c
= path
->nodes
[level
];
5098 if (slot
>= btrfs_header_nritems(c
)) {
5100 if (level
== BTRFS_MAX_LEVEL
) {
5108 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5109 free_extent_buffer(next
);
5113 next_rw_lock
= path
->locks
[level
];
5114 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5120 btrfs_release_path(path
);
5124 if (!path
->skip_locking
) {
5125 ret
= btrfs_try_tree_read_lock(next
);
5127 btrfs_set_path_blocking(path
);
5128 btrfs_tree_read_lock(next
);
5129 btrfs_clear_path_blocking(path
, next
,
5132 next_rw_lock
= BTRFS_READ_LOCK
;
5136 path
->slots
[level
] = slot
;
5139 c
= path
->nodes
[level
];
5140 if (path
->locks
[level
])
5141 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5143 free_extent_buffer(c
);
5144 path
->nodes
[level
] = next
;
5145 path
->slots
[level
] = 0;
5146 if (!path
->skip_locking
)
5147 path
->locks
[level
] = next_rw_lock
;
5151 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5157 btrfs_release_path(path
);
5161 if (!path
->skip_locking
) {
5162 ret
= btrfs_try_tree_read_lock(next
);
5164 btrfs_set_path_blocking(path
);
5165 btrfs_tree_read_lock(next
);
5166 btrfs_clear_path_blocking(path
, next
,
5169 next_rw_lock
= BTRFS_READ_LOCK
;
5174 unlock_up(path
, 0, 1, 0, NULL
);
5175 path
->leave_spinning
= old_spinning
;
5177 btrfs_set_path_blocking(path
);
5183 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5184 * searching until it gets past min_objectid or finds an item of 'type'
5186 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5188 int btrfs_previous_item(struct btrfs_root
*root
,
5189 struct btrfs_path
*path
, u64 min_objectid
,
5192 struct btrfs_key found_key
;
5193 struct extent_buffer
*leaf
;
5198 if (path
->slots
[0] == 0) {
5199 btrfs_set_path_blocking(path
);
5200 ret
= btrfs_prev_leaf(root
, path
);
5206 leaf
= path
->nodes
[0];
5207 nritems
= btrfs_header_nritems(leaf
);
5210 if (path
->slots
[0] == nritems
)
5213 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5214 if (found_key
.objectid
< min_objectid
)
5216 if (found_key
.type
== type
)
5218 if (found_key
.objectid
== min_objectid
&&
5219 found_key
.type
< type
)