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
);
42 static void tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
43 struct extent_buffer
*eb
);
44 struct extent_buffer
*read_old_tree_block(struct btrfs_root
*root
, u64 bytenr
,
45 u32 blocksize
, u64 parent_transid
,
47 struct extent_buffer
*btrfs_find_old_tree_block(struct btrfs_root
*root
,
48 u64 bytenr
, u32 blocksize
,
51 struct btrfs_path
*btrfs_alloc_path(void)
53 struct btrfs_path
*path
;
54 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
59 * set all locked nodes in the path to blocking locks. This should
60 * be done before scheduling
62 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
65 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
66 if (!p
->nodes
[i
] || !p
->locks
[i
])
68 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
69 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
70 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
71 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
72 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
77 * reset all the locked nodes in the patch to spinning locks.
79 * held is used to keep lockdep happy, when lockdep is enabled
80 * we set held to a blocking lock before we go around and
81 * retake all the spinlocks in the path. You can safely use NULL
84 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
85 struct extent_buffer
*held
, int held_rw
)
89 #ifdef CONFIG_DEBUG_LOCK_ALLOC
90 /* lockdep really cares that we take all of these spinlocks
91 * in the right order. If any of the locks in the path are not
92 * currently blocking, it is going to complain. So, make really
93 * really sure by forcing the path to blocking before we clear
97 btrfs_set_lock_blocking_rw(held
, held_rw
);
98 if (held_rw
== BTRFS_WRITE_LOCK
)
99 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
100 else if (held_rw
== BTRFS_READ_LOCK
)
101 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
103 btrfs_set_path_blocking(p
);
106 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
107 if (p
->nodes
[i
] && p
->locks
[i
]) {
108 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
109 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
110 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
111 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
112 p
->locks
[i
] = BTRFS_READ_LOCK
;
116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 btrfs_clear_lock_blocking_rw(held
, held_rw
);
122 /* this also releases the path */
123 void btrfs_free_path(struct btrfs_path
*p
)
127 btrfs_release_path(p
);
128 kmem_cache_free(btrfs_path_cachep
, p
);
132 * path release drops references on the extent buffers in the path
133 * and it drops any locks held by this path
135 * It is safe to call this on paths that no locks or extent buffers held.
137 noinline
void btrfs_release_path(struct btrfs_path
*p
)
141 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
146 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
149 free_extent_buffer(p
->nodes
[i
]);
155 * safely gets a reference on the root node of a tree. A lock
156 * is not taken, so a concurrent writer may put a different node
157 * at the root of the tree. See btrfs_lock_root_node for the
160 * The extent buffer returned by this has a reference taken, so
161 * it won't disappear. It may stop being the root of the tree
162 * at any time because there are no locks held.
164 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
166 struct extent_buffer
*eb
;
170 eb
= rcu_dereference(root
->node
);
173 * RCU really hurts here, we could free up the root node because
174 * it was cow'ed but we may not get the new root node yet so do
175 * the inc_not_zero dance and if it doesn't work then
176 * synchronize_rcu and try again.
178 if (atomic_inc_not_zero(&eb
->refs
)) {
188 /* loop around taking references on and locking the root node of the
189 * tree until you end up with a lock on the root. A locked buffer
190 * is returned, with a reference held.
192 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
194 struct extent_buffer
*eb
;
197 eb
= btrfs_root_node(root
);
199 if (eb
== root
->node
)
201 btrfs_tree_unlock(eb
);
202 free_extent_buffer(eb
);
207 /* loop around taking references on and locking the root node of the
208 * tree until you end up with a lock on the root. A locked buffer
209 * is returned, with a reference held.
211 struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
213 struct extent_buffer
*eb
;
216 eb
= btrfs_root_node(root
);
217 btrfs_tree_read_lock(eb
);
218 if (eb
== root
->node
)
220 btrfs_tree_read_unlock(eb
);
221 free_extent_buffer(eb
);
226 /* cowonly root (everything not a reference counted cow subvolume), just get
227 * put onto a simple dirty list. transaction.c walks this to make sure they
228 * get properly updated on disk.
230 static void add_root_to_dirty_list(struct btrfs_root
*root
)
232 spin_lock(&root
->fs_info
->trans_lock
);
233 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
234 list_add(&root
->dirty_list
,
235 &root
->fs_info
->dirty_cowonly_roots
);
237 spin_unlock(&root
->fs_info
->trans_lock
);
241 * used by snapshot creation to make a copy of a root for a tree with
242 * a given objectid. The buffer with the new root node is returned in
243 * cow_ret, and this func returns zero on success or a negative error code.
245 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
246 struct btrfs_root
*root
,
247 struct extent_buffer
*buf
,
248 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
250 struct extent_buffer
*cow
;
253 struct btrfs_disk_key disk_key
;
255 WARN_ON(root
->ref_cows
&& trans
->transid
!=
256 root
->fs_info
->running_transaction
->transid
);
257 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
259 level
= btrfs_header_level(buf
);
261 btrfs_item_key(buf
, &disk_key
, 0);
263 btrfs_node_key(buf
, &disk_key
, 0);
265 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, 0,
266 new_root_objectid
, &disk_key
, level
,
271 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
272 btrfs_set_header_bytenr(cow
, cow
->start
);
273 btrfs_set_header_generation(cow
, trans
->transid
);
274 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
275 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
276 BTRFS_HEADER_FLAG_RELOC
);
277 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
278 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
280 btrfs_set_header_owner(cow
, new_root_objectid
);
282 write_extent_buffer(cow
, root
->fs_info
->fsid
,
283 (unsigned long)btrfs_header_fsid(cow
),
286 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
287 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
288 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
290 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
295 btrfs_mark_buffer_dirty(cow
);
304 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
305 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
307 MOD_LOG_ROOT_REPLACE
,
310 struct tree_mod_move
{
315 struct tree_mod_root
{
320 struct tree_mod_elem
{
322 u64 index
; /* shifted logical */
326 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
329 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
332 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
333 struct btrfs_disk_key key
;
336 /* this is used for op == MOD_LOG_MOVE_KEYS */
337 struct tree_mod_move move
;
339 /* this is used for op == MOD_LOG_ROOT_REPLACE */
340 struct tree_mod_root old_root
;
343 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
345 read_lock(&fs_info
->tree_mod_log_lock
);
348 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
350 read_unlock(&fs_info
->tree_mod_log_lock
);
353 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
355 write_lock(&fs_info
->tree_mod_log_lock
);
358 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
360 write_unlock(&fs_info
->tree_mod_log_lock
);
364 * This adds a new blocker to the tree mod log's blocker list if the @elem
365 * passed does not already have a sequence number set. So when a caller expects
366 * to record tree modifications, it should ensure to set elem->seq to zero
367 * before calling btrfs_get_tree_mod_seq.
368 * Returns a fresh, unused tree log modification sequence number, even if no new
371 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
372 struct seq_list
*elem
)
376 tree_mod_log_write_lock(fs_info
);
377 spin_lock(&fs_info
->tree_mod_seq_lock
);
379 elem
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
380 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
382 seq
= btrfs_inc_tree_mod_seq(fs_info
);
383 spin_unlock(&fs_info
->tree_mod_seq_lock
);
384 tree_mod_log_write_unlock(fs_info
);
389 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
390 struct seq_list
*elem
)
392 struct rb_root
*tm_root
;
393 struct rb_node
*node
;
394 struct rb_node
*next
;
395 struct seq_list
*cur_elem
;
396 struct tree_mod_elem
*tm
;
397 u64 min_seq
= (u64
)-1;
398 u64 seq_putting
= elem
->seq
;
403 spin_lock(&fs_info
->tree_mod_seq_lock
);
404 list_del(&elem
->list
);
407 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
408 if (cur_elem
->seq
< min_seq
) {
409 if (seq_putting
> cur_elem
->seq
) {
411 * blocker with lower sequence number exists, we
412 * cannot remove anything from the log
414 spin_unlock(&fs_info
->tree_mod_seq_lock
);
417 min_seq
= cur_elem
->seq
;
420 spin_unlock(&fs_info
->tree_mod_seq_lock
);
423 * anything that's lower than the lowest existing (read: blocked)
424 * sequence number can be removed from the tree.
426 tree_mod_log_write_lock(fs_info
);
427 tm_root
= &fs_info
->tree_mod_log
;
428 for (node
= rb_first(tm_root
); node
; node
= next
) {
429 next
= rb_next(node
);
430 tm
= container_of(node
, struct tree_mod_elem
, node
);
431 if (tm
->seq
> min_seq
)
433 rb_erase(node
, tm_root
);
436 tree_mod_log_write_unlock(fs_info
);
440 * key order of the log:
443 * the index is the shifted logical of the *new* root node for root replace
444 * operations, or the shifted logical of the affected block for all other
448 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
450 struct rb_root
*tm_root
;
451 struct rb_node
**new;
452 struct rb_node
*parent
= NULL
;
453 struct tree_mod_elem
*cur
;
455 BUG_ON(!tm
|| !tm
->seq
);
457 tm_root
= &fs_info
->tree_mod_log
;
458 new = &tm_root
->rb_node
;
460 cur
= container_of(*new, struct tree_mod_elem
, node
);
462 if (cur
->index
< tm
->index
)
463 new = &((*new)->rb_left
);
464 else if (cur
->index
> tm
->index
)
465 new = &((*new)->rb_right
);
466 else if (cur
->seq
< tm
->seq
)
467 new = &((*new)->rb_left
);
468 else if (cur
->seq
> tm
->seq
)
469 new = &((*new)->rb_right
);
476 rb_link_node(&tm
->node
, parent
, new);
477 rb_insert_color(&tm
->node
, tm_root
);
482 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
483 * returns zero with the tree_mod_log_lock acquired. The caller must hold
484 * this until all tree mod log insertions are recorded in the rb tree and then
485 * call tree_mod_log_write_unlock() to release.
487 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
488 struct extent_buffer
*eb
) {
490 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
492 if (eb
&& btrfs_header_level(eb
) == 0)
495 tree_mod_log_write_lock(fs_info
);
496 if (list_empty(&fs_info
->tree_mod_seq_list
)) {
498 * someone emptied the list while we were waiting for the lock.
499 * we must not add to the list when no blocker exists.
501 tree_mod_log_write_unlock(fs_info
);
509 * This allocates memory and gets a tree modification sequence number.
511 * Returns <0 on error.
512 * Returns >0 (the added sequence number) on success.
514 static inline int tree_mod_alloc(struct btrfs_fs_info
*fs_info
, gfp_t flags
,
515 struct tree_mod_elem
**tm_ret
)
517 struct tree_mod_elem
*tm
;
520 * once we switch from spin locks to something different, we should
521 * honor the flags parameter here.
523 tm
= *tm_ret
= kzalloc(sizeof(*tm
), GFP_ATOMIC
);
527 tm
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
532 __tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
533 struct extent_buffer
*eb
, int slot
,
534 enum mod_log_op op
, gfp_t flags
)
537 struct tree_mod_elem
*tm
;
539 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
543 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
544 if (op
!= MOD_LOG_KEY_ADD
) {
545 btrfs_node_key(eb
, &tm
->key
, slot
);
546 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
550 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
552 return __tree_mod_log_insert(fs_info
, tm
);
556 tree_mod_log_insert_key_mask(struct btrfs_fs_info
*fs_info
,
557 struct extent_buffer
*eb
, int slot
,
558 enum mod_log_op op
, gfp_t flags
)
562 if (tree_mod_dont_log(fs_info
, eb
))
565 ret
= __tree_mod_log_insert_key(fs_info
, eb
, slot
, op
, flags
);
567 tree_mod_log_write_unlock(fs_info
);
572 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
573 int slot
, enum mod_log_op op
)
575 return tree_mod_log_insert_key_mask(fs_info
, eb
, slot
, op
, GFP_NOFS
);
579 tree_mod_log_insert_key_locked(struct btrfs_fs_info
*fs_info
,
580 struct extent_buffer
*eb
, int slot
,
583 return __tree_mod_log_insert_key(fs_info
, eb
, slot
, op
, GFP_NOFS
);
587 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
588 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
589 int nr_items
, gfp_t flags
)
591 struct tree_mod_elem
*tm
;
595 if (tree_mod_dont_log(fs_info
, eb
))
599 * When we override something during the move, we log these removals.
600 * This can only happen when we move towards the beginning of the
601 * buffer, i.e. dst_slot < src_slot.
603 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
604 ret
= tree_mod_log_insert_key_locked(fs_info
, eb
, i
+ dst_slot
,
605 MOD_LOG_KEY_REMOVE_WHILE_MOVING
);
609 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
613 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
615 tm
->move
.dst_slot
= dst_slot
;
616 tm
->move
.nr_items
= nr_items
;
617 tm
->op
= MOD_LOG_MOVE_KEYS
;
619 ret
= __tree_mod_log_insert(fs_info
, tm
);
621 tree_mod_log_write_unlock(fs_info
);
626 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
632 if (btrfs_header_level(eb
) == 0)
635 nritems
= btrfs_header_nritems(eb
);
636 for (i
= nritems
- 1; i
>= 0; i
--) {
637 ret
= tree_mod_log_insert_key_locked(fs_info
, eb
, i
,
638 MOD_LOG_KEY_REMOVE_WHILE_FREEING
);
644 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
645 struct extent_buffer
*old_root
,
646 struct extent_buffer
*new_root
, gfp_t flags
)
648 struct tree_mod_elem
*tm
;
651 if (tree_mod_dont_log(fs_info
, NULL
))
654 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
658 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
659 tm
->old_root
.logical
= old_root
->start
;
660 tm
->old_root
.level
= btrfs_header_level(old_root
);
661 tm
->generation
= btrfs_header_generation(old_root
);
662 tm
->op
= MOD_LOG_ROOT_REPLACE
;
664 ret
= __tree_mod_log_insert(fs_info
, tm
);
666 tree_mod_log_write_unlock(fs_info
);
670 static struct tree_mod_elem
*
671 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
674 struct rb_root
*tm_root
;
675 struct rb_node
*node
;
676 struct tree_mod_elem
*cur
= NULL
;
677 struct tree_mod_elem
*found
= NULL
;
678 u64 index
= start
>> PAGE_CACHE_SHIFT
;
680 tree_mod_log_read_lock(fs_info
);
681 tm_root
= &fs_info
->tree_mod_log
;
682 node
= tm_root
->rb_node
;
684 cur
= container_of(node
, struct tree_mod_elem
, node
);
685 if (cur
->index
< index
) {
686 node
= node
->rb_left
;
687 } else if (cur
->index
> index
) {
688 node
= node
->rb_right
;
689 } else if (cur
->seq
< min_seq
) {
690 node
= node
->rb_left
;
691 } else if (!smallest
) {
692 /* we want the node with the highest seq */
694 BUG_ON(found
->seq
> cur
->seq
);
696 node
= node
->rb_left
;
697 } else if (cur
->seq
> min_seq
) {
698 /* we want the node with the smallest seq */
700 BUG_ON(found
->seq
< cur
->seq
);
702 node
= node
->rb_right
;
708 tree_mod_log_read_unlock(fs_info
);
714 * this returns the element from the log with the smallest time sequence
715 * value that's in the log (the oldest log item). any element with a time
716 * sequence lower than min_seq will be ignored.
718 static struct tree_mod_elem
*
719 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
722 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
726 * this returns the element from the log with the largest time sequence
727 * value that's in the log (the most recent log item). any element with
728 * a time sequence lower than min_seq will be ignored.
730 static struct tree_mod_elem
*
731 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
733 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
737 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
738 struct extent_buffer
*src
, unsigned long dst_offset
,
739 unsigned long src_offset
, int nr_items
)
744 if (tree_mod_dont_log(fs_info
, NULL
))
747 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0) {
748 tree_mod_log_write_unlock(fs_info
);
752 for (i
= 0; i
< nr_items
; i
++) {
753 ret
= tree_mod_log_insert_key_locked(fs_info
, src
,
757 ret
= tree_mod_log_insert_key_locked(fs_info
, dst
,
763 tree_mod_log_write_unlock(fs_info
);
767 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
768 int dst_offset
, int src_offset
, int nr_items
)
771 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
777 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
778 struct extent_buffer
*eb
, int slot
, int atomic
)
782 ret
= tree_mod_log_insert_key_mask(fs_info
, eb
, slot
,
784 atomic
? GFP_ATOMIC
: GFP_NOFS
);
789 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
791 if (tree_mod_dont_log(fs_info
, eb
))
794 __tree_mod_log_free_eb(fs_info
, eb
);
796 tree_mod_log_write_unlock(fs_info
);
800 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
801 struct extent_buffer
*new_root_node
)
804 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
805 new_root_node
, GFP_NOFS
);
810 * check if the tree block can be shared by multiple trees
812 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
813 struct extent_buffer
*buf
)
816 * Tree blocks not in refernece counted trees and tree roots
817 * are never shared. If a block was allocated after the last
818 * snapshot and the block was not allocated by tree relocation,
819 * we know the block is not shared.
821 if (root
->ref_cows
&&
822 buf
!= root
->node
&& buf
!= root
->commit_root
&&
823 (btrfs_header_generation(buf
) <=
824 btrfs_root_last_snapshot(&root
->root_item
) ||
825 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
827 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
828 if (root
->ref_cows
&&
829 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
835 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
836 struct btrfs_root
*root
,
837 struct extent_buffer
*buf
,
838 struct extent_buffer
*cow
,
848 * Backrefs update rules:
850 * Always use full backrefs for extent pointers in tree block
851 * allocated by tree relocation.
853 * If a shared tree block is no longer referenced by its owner
854 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
855 * use full backrefs for extent pointers in tree block.
857 * If a tree block is been relocating
858 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
859 * use full backrefs for extent pointers in tree block.
860 * The reason for this is some operations (such as drop tree)
861 * are only allowed for blocks use full backrefs.
864 if (btrfs_block_can_be_shared(root
, buf
)) {
865 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
866 buf
->len
, &refs
, &flags
);
871 btrfs_std_error(root
->fs_info
, ret
);
876 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
877 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
878 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
883 owner
= btrfs_header_owner(buf
);
884 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
885 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
888 if ((owner
== root
->root_key
.objectid
||
889 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
890 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
891 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
892 BUG_ON(ret
); /* -ENOMEM */
894 if (root
->root_key
.objectid
==
895 BTRFS_TREE_RELOC_OBJECTID
) {
896 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
897 BUG_ON(ret
); /* -ENOMEM */
898 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
899 BUG_ON(ret
); /* -ENOMEM */
901 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
904 if (root
->root_key
.objectid
==
905 BTRFS_TREE_RELOC_OBJECTID
)
906 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
908 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
909 BUG_ON(ret
); /* -ENOMEM */
911 if (new_flags
!= 0) {
912 ret
= btrfs_set_disk_extent_flags(trans
, root
,
920 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
921 if (root
->root_key
.objectid
==
922 BTRFS_TREE_RELOC_OBJECTID
)
923 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
925 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
926 BUG_ON(ret
); /* -ENOMEM */
927 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
928 BUG_ON(ret
); /* -ENOMEM */
930 tree_mod_log_free_eb(root
->fs_info
, buf
);
931 clean_tree_block(trans
, root
, buf
);
938 * does the dirty work in cow of a single block. The parent block (if
939 * supplied) is updated to point to the new cow copy. The new buffer is marked
940 * dirty and returned locked. If you modify the block it needs to be marked
943 * search_start -- an allocation hint for the new block
945 * empty_size -- a hint that you plan on doing more cow. This is the size in
946 * bytes the allocator should try to find free next to the block it returns.
947 * This is just a hint and may be ignored by the allocator.
949 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
950 struct btrfs_root
*root
,
951 struct extent_buffer
*buf
,
952 struct extent_buffer
*parent
, int parent_slot
,
953 struct extent_buffer
**cow_ret
,
954 u64 search_start
, u64 empty_size
)
956 struct btrfs_disk_key disk_key
;
957 struct extent_buffer
*cow
;
966 btrfs_assert_tree_locked(buf
);
968 WARN_ON(root
->ref_cows
&& trans
->transid
!=
969 root
->fs_info
->running_transaction
->transid
);
970 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
972 level
= btrfs_header_level(buf
);
975 btrfs_item_key(buf
, &disk_key
, 0);
977 btrfs_node_key(buf
, &disk_key
, 0);
979 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
981 parent_start
= parent
->start
;
987 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
988 root
->root_key
.objectid
, &disk_key
,
989 level
, search_start
, empty_size
);
993 /* cow is set to blocking by btrfs_init_new_buffer */
995 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
996 btrfs_set_header_bytenr(cow
, cow
->start
);
997 btrfs_set_header_generation(cow
, trans
->transid
);
998 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
999 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1000 BTRFS_HEADER_FLAG_RELOC
);
1001 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1002 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1004 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1006 write_extent_buffer(cow
, root
->fs_info
->fsid
,
1007 (unsigned long)btrfs_header_fsid(cow
),
1010 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1012 btrfs_abort_transaction(trans
, root
, ret
);
1017 btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1019 if (buf
== root
->node
) {
1020 WARN_ON(parent
&& parent
!= buf
);
1021 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1022 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1023 parent_start
= buf
->start
;
1027 extent_buffer_get(cow
);
1028 tree_mod_log_set_root_pointer(root
, cow
);
1029 rcu_assign_pointer(root
->node
, cow
);
1031 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1033 free_extent_buffer(buf
);
1034 add_root_to_dirty_list(root
);
1036 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1037 parent_start
= parent
->start
;
1041 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1042 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1043 MOD_LOG_KEY_REPLACE
);
1044 btrfs_set_node_blockptr(parent
, parent_slot
,
1046 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1048 btrfs_mark_buffer_dirty(parent
);
1049 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1053 btrfs_tree_unlock(buf
);
1054 free_extent_buffer_stale(buf
);
1055 btrfs_mark_buffer_dirty(cow
);
1061 * returns the logical address of the oldest predecessor of the given root.
1062 * entries older than time_seq are ignored.
1064 static struct tree_mod_elem
*
1065 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1066 struct btrfs_root
*root
, u64 time_seq
)
1068 struct tree_mod_elem
*tm
;
1069 struct tree_mod_elem
*found
= NULL
;
1070 u64 root_logical
= root
->node
->start
;
1077 * the very last operation that's logged for a root is the replacement
1078 * operation (if it is replaced at all). this has the index of the *new*
1079 * root, making it the very first operation that's logged for this root.
1082 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1087 * if there are no tree operation for the oldest root, we simply
1088 * return it. this should only happen if that (old) root is at
1095 * if there's an operation that's not a root replacement, we
1096 * found the oldest version of our root. normally, we'll find a
1097 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1099 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1103 root_logical
= tm
->old_root
.logical
;
1104 BUG_ON(root_logical
== root
->node
->start
);
1108 /* if there's no old root to return, return what we found instead */
1116 * tm is a pointer to the first operation to rewind within eb. then, all
1117 * previous operations will be rewinded (until we reach something older than
1121 __tree_mod_log_rewind(struct extent_buffer
*eb
, u64 time_seq
,
1122 struct tree_mod_elem
*first_tm
)
1125 struct rb_node
*next
;
1126 struct tree_mod_elem
*tm
= first_tm
;
1127 unsigned long o_dst
;
1128 unsigned long o_src
;
1129 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1131 n
= btrfs_header_nritems(eb
);
1132 while (tm
&& tm
->seq
>= time_seq
) {
1134 * all the operations are recorded with the operator used for
1135 * the modification. as we're going backwards, we do the
1136 * opposite of each operation here.
1139 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1140 BUG_ON(tm
->slot
< n
);
1141 case MOD_LOG_KEY_REMOVE
:
1143 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1144 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1145 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1146 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1149 case MOD_LOG_KEY_REPLACE
:
1150 BUG_ON(tm
->slot
>= n
);
1151 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1152 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1153 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1156 case MOD_LOG_KEY_ADD
:
1157 /* if a move operation is needed it's in the log */
1160 case MOD_LOG_MOVE_KEYS
:
1161 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1162 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1163 memmove_extent_buffer(eb
, o_dst
, o_src
,
1164 tm
->move
.nr_items
* p_size
);
1166 case MOD_LOG_ROOT_REPLACE
:
1168 * this operation is special. for roots, this must be
1169 * handled explicitly before rewinding.
1170 * for non-roots, this operation may exist if the node
1171 * was a root: root A -> child B; then A gets empty and
1172 * B is promoted to the new root. in the mod log, we'll
1173 * have a root-replace operation for B, a tree block
1174 * that is no root. we simply ignore that operation.
1178 next
= rb_next(&tm
->node
);
1181 tm
= container_of(next
, struct tree_mod_elem
, node
);
1182 if (tm
->index
!= first_tm
->index
)
1185 btrfs_set_header_nritems(eb
, n
);
1188 static struct extent_buffer
*
1189 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1192 struct extent_buffer
*eb_rewin
;
1193 struct tree_mod_elem
*tm
;
1198 if (btrfs_header_level(eb
) == 0)
1201 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1205 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1206 BUG_ON(tm
->slot
!= 0);
1207 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1208 fs_info
->tree_root
->nodesize
);
1210 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1211 btrfs_set_header_backref_rev(eb_rewin
,
1212 btrfs_header_backref_rev(eb
));
1213 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1214 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1216 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1220 extent_buffer_get(eb_rewin
);
1221 free_extent_buffer(eb
);
1223 __tree_mod_log_rewind(eb_rewin
, time_seq
, tm
);
1224 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1225 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->fs_root
));
1231 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1232 * value. If there are no changes, the current root->root_node is returned. If
1233 * anything changed in between, there's a fresh buffer allocated on which the
1234 * rewind operations are done. In any case, the returned buffer is read locked.
1235 * Returns NULL on error (with no locks held).
1237 static inline struct extent_buffer
*
1238 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1240 struct tree_mod_elem
*tm
;
1241 struct extent_buffer
*eb
;
1242 struct extent_buffer
*old
;
1243 struct tree_mod_root
*old_root
= NULL
;
1244 u64 old_generation
= 0;
1248 eb
= btrfs_read_lock_root_node(root
);
1249 tm
= __tree_mod_log_oldest_root(root
->fs_info
, root
, time_seq
);
1253 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1254 old_root
= &tm
->old_root
;
1255 old_generation
= tm
->generation
;
1256 logical
= old_root
->logical
;
1258 logical
= root
->node
->start
;
1261 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1262 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1263 btrfs_tree_read_unlock(root
->node
);
1264 free_extent_buffer(root
->node
);
1265 blocksize
= btrfs_level_size(root
, old_root
->level
);
1266 old
= read_tree_block(root
, logical
, blocksize
, 0);
1268 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1272 eb
= btrfs_clone_extent_buffer(old
);
1273 free_extent_buffer(old
);
1275 } else if (old_root
) {
1276 btrfs_tree_read_unlock(root
->node
);
1277 free_extent_buffer(root
->node
);
1278 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1280 eb
= btrfs_clone_extent_buffer(root
->node
);
1281 btrfs_tree_read_unlock(root
->node
);
1282 free_extent_buffer(root
->node
);
1287 extent_buffer_get(eb
);
1288 btrfs_tree_read_lock(eb
);
1290 btrfs_set_header_bytenr(eb
, eb
->start
);
1291 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1292 btrfs_set_header_owner(eb
, root
->root_key
.objectid
);
1293 btrfs_set_header_level(eb
, old_root
->level
);
1294 btrfs_set_header_generation(eb
, old_generation
);
1297 __tree_mod_log_rewind(eb
, time_seq
, tm
);
1299 WARN_ON(btrfs_header_level(eb
) != 0);
1300 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1305 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1307 struct tree_mod_elem
*tm
;
1310 tm
= __tree_mod_log_oldest_root(root
->fs_info
, root
, time_seq
);
1311 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1312 level
= tm
->old_root
.level
;
1315 level
= btrfs_header_level(root
->node
);
1322 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1323 struct btrfs_root
*root
,
1324 struct extent_buffer
*buf
)
1326 /* ensure we can see the force_cow */
1330 * We do not need to cow a block if
1331 * 1) this block is not created or changed in this transaction;
1332 * 2) this block does not belong to TREE_RELOC tree;
1333 * 3) the root is not forced COW.
1335 * What is forced COW:
1336 * when we create snapshot during commiting the transaction,
1337 * after we've finished coping src root, we must COW the shared
1338 * block to ensure the metadata consistency.
1340 if (btrfs_header_generation(buf
) == trans
->transid
&&
1341 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1342 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1343 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1350 * cows a single block, see __btrfs_cow_block for the real work.
1351 * This version of it has extra checks so that a block isn't cow'd more than
1352 * once per transaction, as long as it hasn't been written yet
1354 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1355 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1356 struct extent_buffer
*parent
, int parent_slot
,
1357 struct extent_buffer
**cow_ret
)
1362 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1363 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1364 (unsigned long long)trans
->transid
,
1365 (unsigned long long)
1366 root
->fs_info
->running_transaction
->transid
);
1368 if (trans
->transid
!= root
->fs_info
->generation
)
1369 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1370 (unsigned long long)trans
->transid
,
1371 (unsigned long long)root
->fs_info
->generation
);
1373 if (!should_cow_block(trans
, root
, buf
)) {
1378 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1381 btrfs_set_lock_blocking(parent
);
1382 btrfs_set_lock_blocking(buf
);
1384 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1385 parent_slot
, cow_ret
, search_start
, 0);
1387 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1393 * helper function for defrag to decide if two blocks pointed to by a
1394 * node are actually close by
1396 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1398 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1400 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1406 * compare two keys in a memcmp fashion
1408 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1410 struct btrfs_key k1
;
1412 btrfs_disk_key_to_cpu(&k1
, disk
);
1414 return btrfs_comp_cpu_keys(&k1
, k2
);
1418 * same as comp_keys only with two btrfs_key's
1420 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1422 if (k1
->objectid
> k2
->objectid
)
1424 if (k1
->objectid
< k2
->objectid
)
1426 if (k1
->type
> k2
->type
)
1428 if (k1
->type
< k2
->type
)
1430 if (k1
->offset
> k2
->offset
)
1432 if (k1
->offset
< k2
->offset
)
1438 * this is used by the defrag code to go through all the
1439 * leaves pointed to by a node and reallocate them so that
1440 * disk order is close to key order
1442 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1443 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1444 int start_slot
, int cache_only
, u64
*last_ret
,
1445 struct btrfs_key
*progress
)
1447 struct extent_buffer
*cur
;
1450 u64 search_start
= *last_ret
;
1460 int progress_passed
= 0;
1461 struct btrfs_disk_key disk_key
;
1463 parent_level
= btrfs_header_level(parent
);
1464 if (cache_only
&& parent_level
!= 1)
1467 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1468 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1470 parent_nritems
= btrfs_header_nritems(parent
);
1471 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1472 end_slot
= parent_nritems
;
1474 if (parent_nritems
== 1)
1477 btrfs_set_lock_blocking(parent
);
1479 for (i
= start_slot
; i
< end_slot
; i
++) {
1482 btrfs_node_key(parent
, &disk_key
, i
);
1483 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1486 progress_passed
= 1;
1487 blocknr
= btrfs_node_blockptr(parent
, i
);
1488 gen
= btrfs_node_ptr_generation(parent
, i
);
1489 if (last_block
== 0)
1490 last_block
= blocknr
;
1493 other
= btrfs_node_blockptr(parent
, i
- 1);
1494 close
= close_blocks(blocknr
, other
, blocksize
);
1496 if (!close
&& i
< end_slot
- 2) {
1497 other
= btrfs_node_blockptr(parent
, i
+ 1);
1498 close
= close_blocks(blocknr
, other
, blocksize
);
1501 last_block
= blocknr
;
1505 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1507 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1510 if (!cur
|| !uptodate
) {
1512 free_extent_buffer(cur
);
1516 cur
= read_tree_block(root
, blocknr
,
1520 } else if (!uptodate
) {
1521 err
= btrfs_read_buffer(cur
, gen
);
1523 free_extent_buffer(cur
);
1528 if (search_start
== 0)
1529 search_start
= last_block
;
1531 btrfs_tree_lock(cur
);
1532 btrfs_set_lock_blocking(cur
);
1533 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1536 (end_slot
- i
) * blocksize
));
1538 btrfs_tree_unlock(cur
);
1539 free_extent_buffer(cur
);
1542 search_start
= cur
->start
;
1543 last_block
= cur
->start
;
1544 *last_ret
= search_start
;
1545 btrfs_tree_unlock(cur
);
1546 free_extent_buffer(cur
);
1552 * The leaf data grows from end-to-front in the node.
1553 * this returns the address of the start of the last item,
1554 * which is the stop of the leaf data stack
1556 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1557 struct extent_buffer
*leaf
)
1559 u32 nr
= btrfs_header_nritems(leaf
);
1561 return BTRFS_LEAF_DATA_SIZE(root
);
1562 return btrfs_item_offset_nr(leaf
, nr
- 1);
1567 * search for key in the extent_buffer. The items start at offset p,
1568 * and they are item_size apart. There are 'max' items in p.
1570 * the slot in the array is returned via slot, and it points to
1571 * the place where you would insert key if it is not found in
1574 * slot may point to max if the key is bigger than all of the keys
1576 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1578 int item_size
, struct btrfs_key
*key
,
1585 struct btrfs_disk_key
*tmp
= NULL
;
1586 struct btrfs_disk_key unaligned
;
1587 unsigned long offset
;
1589 unsigned long map_start
= 0;
1590 unsigned long map_len
= 0;
1593 while (low
< high
) {
1594 mid
= (low
+ high
) / 2;
1595 offset
= p
+ mid
* item_size
;
1597 if (!kaddr
|| offset
< map_start
||
1598 (offset
+ sizeof(struct btrfs_disk_key
)) >
1599 map_start
+ map_len
) {
1601 err
= map_private_extent_buffer(eb
, offset
,
1602 sizeof(struct btrfs_disk_key
),
1603 &kaddr
, &map_start
, &map_len
);
1606 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1609 read_extent_buffer(eb
, &unaligned
,
1610 offset
, sizeof(unaligned
));
1615 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1618 ret
= comp_keys(tmp
, key
);
1634 * simple bin_search frontend that does the right thing for
1637 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1638 int level
, int *slot
)
1641 return generic_bin_search(eb
,
1642 offsetof(struct btrfs_leaf
, items
),
1643 sizeof(struct btrfs_item
),
1644 key
, btrfs_header_nritems(eb
),
1647 return generic_bin_search(eb
,
1648 offsetof(struct btrfs_node
, ptrs
),
1649 sizeof(struct btrfs_key_ptr
),
1650 key
, btrfs_header_nritems(eb
),
1654 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1655 int level
, int *slot
)
1657 return bin_search(eb
, key
, level
, slot
);
1660 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1662 spin_lock(&root
->accounting_lock
);
1663 btrfs_set_root_used(&root
->root_item
,
1664 btrfs_root_used(&root
->root_item
) + size
);
1665 spin_unlock(&root
->accounting_lock
);
1668 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1670 spin_lock(&root
->accounting_lock
);
1671 btrfs_set_root_used(&root
->root_item
,
1672 btrfs_root_used(&root
->root_item
) - size
);
1673 spin_unlock(&root
->accounting_lock
);
1676 /* given a node and slot number, this reads the blocks it points to. The
1677 * extent buffer is returned with a reference taken (but unlocked).
1678 * NULL is returned on error.
1680 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1681 struct extent_buffer
*parent
, int slot
)
1683 int level
= btrfs_header_level(parent
);
1686 if (slot
>= btrfs_header_nritems(parent
))
1691 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1692 btrfs_level_size(root
, level
- 1),
1693 btrfs_node_ptr_generation(parent
, slot
));
1697 * node level balancing, used to make sure nodes are in proper order for
1698 * item deletion. We balance from the top down, so we have to make sure
1699 * that a deletion won't leave an node completely empty later on.
1701 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1702 struct btrfs_root
*root
,
1703 struct btrfs_path
*path
, int level
)
1705 struct extent_buffer
*right
= NULL
;
1706 struct extent_buffer
*mid
;
1707 struct extent_buffer
*left
= NULL
;
1708 struct extent_buffer
*parent
= NULL
;
1712 int orig_slot
= path
->slots
[level
];
1718 mid
= path
->nodes
[level
];
1720 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1721 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1722 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1724 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1726 if (level
< BTRFS_MAX_LEVEL
- 1) {
1727 parent
= path
->nodes
[level
+ 1];
1728 pslot
= path
->slots
[level
+ 1];
1732 * deal with the case where there is only one pointer in the root
1733 * by promoting the node below to a root
1736 struct extent_buffer
*child
;
1738 if (btrfs_header_nritems(mid
) != 1)
1741 /* promote the child to a root */
1742 child
= read_node_slot(root
, mid
, 0);
1745 btrfs_std_error(root
->fs_info
, ret
);
1749 btrfs_tree_lock(child
);
1750 btrfs_set_lock_blocking(child
);
1751 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1753 btrfs_tree_unlock(child
);
1754 free_extent_buffer(child
);
1758 tree_mod_log_free_eb(root
->fs_info
, root
->node
);
1759 tree_mod_log_set_root_pointer(root
, child
);
1760 rcu_assign_pointer(root
->node
, child
);
1762 add_root_to_dirty_list(root
);
1763 btrfs_tree_unlock(child
);
1765 path
->locks
[level
] = 0;
1766 path
->nodes
[level
] = NULL
;
1767 clean_tree_block(trans
, root
, mid
);
1768 btrfs_tree_unlock(mid
);
1769 /* once for the path */
1770 free_extent_buffer(mid
);
1772 root_sub_used(root
, mid
->len
);
1773 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1774 /* once for the root ptr */
1775 free_extent_buffer_stale(mid
);
1778 if (btrfs_header_nritems(mid
) >
1779 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1782 left
= read_node_slot(root
, parent
, pslot
- 1);
1784 btrfs_tree_lock(left
);
1785 btrfs_set_lock_blocking(left
);
1786 wret
= btrfs_cow_block(trans
, root
, left
,
1787 parent
, pslot
- 1, &left
);
1793 right
= read_node_slot(root
, parent
, pslot
+ 1);
1795 btrfs_tree_lock(right
);
1796 btrfs_set_lock_blocking(right
);
1797 wret
= btrfs_cow_block(trans
, root
, right
,
1798 parent
, pslot
+ 1, &right
);
1805 /* first, try to make some room in the middle buffer */
1807 orig_slot
+= btrfs_header_nritems(left
);
1808 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1814 * then try to empty the right most buffer into the middle
1817 wret
= push_node_left(trans
, root
, mid
, right
, 1);
1818 if (wret
< 0 && wret
!= -ENOSPC
)
1820 if (btrfs_header_nritems(right
) == 0) {
1821 clean_tree_block(trans
, root
, right
);
1822 btrfs_tree_unlock(right
);
1823 del_ptr(trans
, root
, path
, level
+ 1, pslot
+ 1);
1824 root_sub_used(root
, right
->len
);
1825 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
1826 free_extent_buffer_stale(right
);
1829 struct btrfs_disk_key right_key
;
1830 btrfs_node_key(right
, &right_key
, 0);
1831 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1833 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1834 btrfs_mark_buffer_dirty(parent
);
1837 if (btrfs_header_nritems(mid
) == 1) {
1839 * we're not allowed to leave a node with one item in the
1840 * tree during a delete. A deletion from lower in the tree
1841 * could try to delete the only pointer in this node.
1842 * So, pull some keys from the left.
1843 * There has to be a left pointer at this point because
1844 * otherwise we would have pulled some pointers from the
1849 btrfs_std_error(root
->fs_info
, ret
);
1852 wret
= balance_node_right(trans
, root
, mid
, left
);
1858 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1864 if (btrfs_header_nritems(mid
) == 0) {
1865 clean_tree_block(trans
, root
, mid
);
1866 btrfs_tree_unlock(mid
);
1867 del_ptr(trans
, root
, path
, level
+ 1, pslot
);
1868 root_sub_used(root
, mid
->len
);
1869 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1870 free_extent_buffer_stale(mid
);
1873 /* update the parent key to reflect our changes */
1874 struct btrfs_disk_key mid_key
;
1875 btrfs_node_key(mid
, &mid_key
, 0);
1876 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1878 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1879 btrfs_mark_buffer_dirty(parent
);
1882 /* update the path */
1884 if (btrfs_header_nritems(left
) > orig_slot
) {
1885 extent_buffer_get(left
);
1886 /* left was locked after cow */
1887 path
->nodes
[level
] = left
;
1888 path
->slots
[level
+ 1] -= 1;
1889 path
->slots
[level
] = orig_slot
;
1891 btrfs_tree_unlock(mid
);
1892 free_extent_buffer(mid
);
1895 orig_slot
-= btrfs_header_nritems(left
);
1896 path
->slots
[level
] = orig_slot
;
1899 /* double check we haven't messed things up */
1901 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
1905 btrfs_tree_unlock(right
);
1906 free_extent_buffer(right
);
1909 if (path
->nodes
[level
] != left
)
1910 btrfs_tree_unlock(left
);
1911 free_extent_buffer(left
);
1916 /* Node balancing for insertion. Here we only split or push nodes around
1917 * when they are completely full. This is also done top down, so we
1918 * have to be pessimistic.
1920 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
1921 struct btrfs_root
*root
,
1922 struct btrfs_path
*path
, int level
)
1924 struct extent_buffer
*right
= NULL
;
1925 struct extent_buffer
*mid
;
1926 struct extent_buffer
*left
= NULL
;
1927 struct extent_buffer
*parent
= NULL
;
1931 int orig_slot
= path
->slots
[level
];
1936 mid
= path
->nodes
[level
];
1937 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1939 if (level
< BTRFS_MAX_LEVEL
- 1) {
1940 parent
= path
->nodes
[level
+ 1];
1941 pslot
= path
->slots
[level
+ 1];
1947 left
= read_node_slot(root
, parent
, pslot
- 1);
1949 /* first, try to make some room in the middle buffer */
1953 btrfs_tree_lock(left
);
1954 btrfs_set_lock_blocking(left
);
1956 left_nr
= btrfs_header_nritems(left
);
1957 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1960 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1965 wret
= push_node_left(trans
, root
,
1972 struct btrfs_disk_key disk_key
;
1973 orig_slot
+= left_nr
;
1974 btrfs_node_key(mid
, &disk_key
, 0);
1975 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1977 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1978 btrfs_mark_buffer_dirty(parent
);
1979 if (btrfs_header_nritems(left
) > orig_slot
) {
1980 path
->nodes
[level
] = left
;
1981 path
->slots
[level
+ 1] -= 1;
1982 path
->slots
[level
] = orig_slot
;
1983 btrfs_tree_unlock(mid
);
1984 free_extent_buffer(mid
);
1987 btrfs_header_nritems(left
);
1988 path
->slots
[level
] = orig_slot
;
1989 btrfs_tree_unlock(left
);
1990 free_extent_buffer(left
);
1994 btrfs_tree_unlock(left
);
1995 free_extent_buffer(left
);
1997 right
= read_node_slot(root
, parent
, pslot
+ 1);
2000 * then try to empty the right most buffer into the middle
2005 btrfs_tree_lock(right
);
2006 btrfs_set_lock_blocking(right
);
2008 right_nr
= btrfs_header_nritems(right
);
2009 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2012 ret
= btrfs_cow_block(trans
, root
, right
,
2018 wret
= balance_node_right(trans
, root
,
2025 struct btrfs_disk_key disk_key
;
2027 btrfs_node_key(right
, &disk_key
, 0);
2028 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2030 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2031 btrfs_mark_buffer_dirty(parent
);
2033 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2034 path
->nodes
[level
] = right
;
2035 path
->slots
[level
+ 1] += 1;
2036 path
->slots
[level
] = orig_slot
-
2037 btrfs_header_nritems(mid
);
2038 btrfs_tree_unlock(mid
);
2039 free_extent_buffer(mid
);
2041 btrfs_tree_unlock(right
);
2042 free_extent_buffer(right
);
2046 btrfs_tree_unlock(right
);
2047 free_extent_buffer(right
);
2053 * readahead one full node of leaves, finding things that are close
2054 * to the block in 'slot', and triggering ra on them.
2056 static void reada_for_search(struct btrfs_root
*root
,
2057 struct btrfs_path
*path
,
2058 int level
, int slot
, u64 objectid
)
2060 struct extent_buffer
*node
;
2061 struct btrfs_disk_key disk_key
;
2067 int direction
= path
->reada
;
2068 struct extent_buffer
*eb
;
2076 if (!path
->nodes
[level
])
2079 node
= path
->nodes
[level
];
2081 search
= btrfs_node_blockptr(node
, slot
);
2082 blocksize
= btrfs_level_size(root
, level
- 1);
2083 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
2085 free_extent_buffer(eb
);
2091 nritems
= btrfs_header_nritems(node
);
2095 if (direction
< 0) {
2099 } else if (direction
> 0) {
2104 if (path
->reada
< 0 && objectid
) {
2105 btrfs_node_key(node
, &disk_key
, nr
);
2106 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2109 search
= btrfs_node_blockptr(node
, nr
);
2110 if ((search
<= target
&& target
- search
<= 65536) ||
2111 (search
> target
&& search
- target
<= 65536)) {
2112 gen
= btrfs_node_ptr_generation(node
, nr
);
2113 readahead_tree_block(root
, search
, blocksize
, gen
);
2117 if ((nread
> 65536 || nscan
> 32))
2123 * returns -EAGAIN if it had to drop the path, or zero if everything was in
2126 static noinline
int reada_for_balance(struct btrfs_root
*root
,
2127 struct btrfs_path
*path
, int level
)
2131 struct extent_buffer
*parent
;
2132 struct extent_buffer
*eb
;
2139 parent
= path
->nodes
[level
+ 1];
2143 nritems
= btrfs_header_nritems(parent
);
2144 slot
= path
->slots
[level
+ 1];
2145 blocksize
= btrfs_level_size(root
, level
);
2148 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2149 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2150 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2152 * if we get -eagain from btrfs_buffer_uptodate, we
2153 * don't want to return eagain here. That will loop
2156 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2158 free_extent_buffer(eb
);
2160 if (slot
+ 1 < nritems
) {
2161 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2162 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2163 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2164 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2166 free_extent_buffer(eb
);
2168 if (block1
|| block2
) {
2171 /* release the whole path */
2172 btrfs_release_path(path
);
2174 /* read the blocks */
2176 readahead_tree_block(root
, block1
, blocksize
, 0);
2178 readahead_tree_block(root
, block2
, blocksize
, 0);
2181 eb
= read_tree_block(root
, block1
, blocksize
, 0);
2182 free_extent_buffer(eb
);
2185 eb
= read_tree_block(root
, block2
, blocksize
, 0);
2186 free_extent_buffer(eb
);
2194 * when we walk down the tree, it is usually safe to unlock the higher layers
2195 * in the tree. The exceptions are when our path goes through slot 0, because
2196 * operations on the tree might require changing key pointers higher up in the
2199 * callers might also have set path->keep_locks, which tells this code to keep
2200 * the lock if the path points to the last slot in the block. This is part of
2201 * walking through the tree, and selecting the next slot in the higher block.
2203 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2204 * if lowest_unlock is 1, level 0 won't be unlocked
2206 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2207 int lowest_unlock
, int min_write_lock_level
,
2208 int *write_lock_level
)
2211 int skip_level
= level
;
2213 struct extent_buffer
*t
;
2215 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2216 if (!path
->nodes
[i
])
2218 if (!path
->locks
[i
])
2220 if (!no_skips
&& path
->slots
[i
] == 0) {
2224 if (!no_skips
&& path
->keep_locks
) {
2227 nritems
= btrfs_header_nritems(t
);
2228 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2233 if (skip_level
< i
&& i
>= lowest_unlock
)
2237 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2238 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2240 if (write_lock_level
&&
2241 i
> min_write_lock_level
&&
2242 i
<= *write_lock_level
) {
2243 *write_lock_level
= i
- 1;
2250 * This releases any locks held in the path starting at level and
2251 * going all the way up to the root.
2253 * btrfs_search_slot will keep the lock held on higher nodes in a few
2254 * corner cases, such as COW of the block at slot zero in the node. This
2255 * ignores those rules, and it should only be called when there are no
2256 * more updates to be done higher up in the tree.
2258 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2262 if (path
->keep_locks
)
2265 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2266 if (!path
->nodes
[i
])
2268 if (!path
->locks
[i
])
2270 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2276 * helper function for btrfs_search_slot. The goal is to find a block
2277 * in cache without setting the path to blocking. If we find the block
2278 * we return zero and the path is unchanged.
2280 * If we can't find the block, we set the path blocking and do some
2281 * reada. -EAGAIN is returned and the search must be repeated.
2284 read_block_for_search(struct btrfs_trans_handle
*trans
,
2285 struct btrfs_root
*root
, struct btrfs_path
*p
,
2286 struct extent_buffer
**eb_ret
, int level
, int slot
,
2287 struct btrfs_key
*key
, u64 time_seq
)
2292 struct extent_buffer
*b
= *eb_ret
;
2293 struct extent_buffer
*tmp
;
2296 blocknr
= btrfs_node_blockptr(b
, slot
);
2297 gen
= btrfs_node_ptr_generation(b
, slot
);
2298 blocksize
= btrfs_level_size(root
, level
- 1);
2300 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2302 /* first we do an atomic uptodate check */
2303 if (btrfs_buffer_uptodate(tmp
, 0, 1) > 0) {
2304 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2306 * we found an up to date block without
2313 /* the pages were up to date, but we failed
2314 * the generation number check. Do a full
2315 * read for the generation number that is correct.
2316 * We must do this without dropping locks so
2317 * we can trust our generation number
2319 free_extent_buffer(tmp
);
2320 btrfs_set_path_blocking(p
);
2322 /* now we're allowed to do a blocking uptodate check */
2323 tmp
= read_tree_block(root
, blocknr
, blocksize
, gen
);
2324 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
, 0) > 0) {
2328 free_extent_buffer(tmp
);
2329 btrfs_release_path(p
);
2335 * reduce lock contention at high levels
2336 * of the btree by dropping locks before
2337 * we read. Don't release the lock on the current
2338 * level because we need to walk this node to figure
2339 * out which blocks to read.
2341 btrfs_unlock_up_safe(p
, level
+ 1);
2342 btrfs_set_path_blocking(p
);
2344 free_extent_buffer(tmp
);
2346 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2348 btrfs_release_path(p
);
2351 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2354 * If the read above didn't mark this buffer up to date,
2355 * it will never end up being up to date. Set ret to EIO now
2356 * and give up so that our caller doesn't loop forever
2359 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2361 free_extent_buffer(tmp
);
2367 * helper function for btrfs_search_slot. This does all of the checks
2368 * for node-level blocks and does any balancing required based on
2371 * If no extra work was required, zero is returned. If we had to
2372 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2376 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2377 struct btrfs_root
*root
, struct btrfs_path
*p
,
2378 struct extent_buffer
*b
, int level
, int ins_len
,
2379 int *write_lock_level
)
2382 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2383 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2386 if (*write_lock_level
< level
+ 1) {
2387 *write_lock_level
= level
+ 1;
2388 btrfs_release_path(p
);
2392 sret
= reada_for_balance(root
, p
, level
);
2396 btrfs_set_path_blocking(p
);
2397 sret
= split_node(trans
, root
, p
, level
);
2398 btrfs_clear_path_blocking(p
, NULL
, 0);
2405 b
= p
->nodes
[level
];
2406 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2407 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2410 if (*write_lock_level
< level
+ 1) {
2411 *write_lock_level
= level
+ 1;
2412 btrfs_release_path(p
);
2416 sret
= reada_for_balance(root
, p
, level
);
2420 btrfs_set_path_blocking(p
);
2421 sret
= balance_level(trans
, root
, p
, level
);
2422 btrfs_clear_path_blocking(p
, NULL
, 0);
2428 b
= p
->nodes
[level
];
2430 btrfs_release_path(p
);
2433 BUG_ON(btrfs_header_nritems(b
) == 1);
2444 * look for key in the tree. path is filled in with nodes along the way
2445 * if key is found, we return zero and you can find the item in the leaf
2446 * level of the path (level 0)
2448 * If the key isn't found, the path points to the slot where it should
2449 * be inserted, and 1 is returned. If there are other errors during the
2450 * search a negative error number is returned.
2452 * if ins_len > 0, nodes and leaves will be split as we walk down the
2453 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2456 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2457 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2460 struct extent_buffer
*b
;
2465 int lowest_unlock
= 1;
2467 /* everything at write_lock_level or lower must be write locked */
2468 int write_lock_level
= 0;
2469 u8 lowest_level
= 0;
2470 int min_write_lock_level
;
2472 lowest_level
= p
->lowest_level
;
2473 WARN_ON(lowest_level
&& ins_len
> 0);
2474 WARN_ON(p
->nodes
[0] != NULL
);
2479 /* when we are removing items, we might have to go up to level
2480 * two as we update tree pointers Make sure we keep write
2481 * for those levels as well
2483 write_lock_level
= 2;
2484 } else if (ins_len
> 0) {
2486 * for inserting items, make sure we have a write lock on
2487 * level 1 so we can update keys
2489 write_lock_level
= 1;
2493 write_lock_level
= -1;
2495 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2496 write_lock_level
= BTRFS_MAX_LEVEL
;
2498 min_write_lock_level
= write_lock_level
;
2502 * we try very hard to do read locks on the root
2504 root_lock
= BTRFS_READ_LOCK
;
2506 if (p
->search_commit_root
) {
2508 * the commit roots are read only
2509 * so we always do read locks
2511 b
= root
->commit_root
;
2512 extent_buffer_get(b
);
2513 level
= btrfs_header_level(b
);
2514 if (!p
->skip_locking
)
2515 btrfs_tree_read_lock(b
);
2517 if (p
->skip_locking
) {
2518 b
= btrfs_root_node(root
);
2519 level
= btrfs_header_level(b
);
2521 /* we don't know the level of the root node
2522 * until we actually have it read locked
2524 b
= btrfs_read_lock_root_node(root
);
2525 level
= btrfs_header_level(b
);
2526 if (level
<= write_lock_level
) {
2527 /* whoops, must trade for write lock */
2528 btrfs_tree_read_unlock(b
);
2529 free_extent_buffer(b
);
2530 b
= btrfs_lock_root_node(root
);
2531 root_lock
= BTRFS_WRITE_LOCK
;
2533 /* the level might have changed, check again */
2534 level
= btrfs_header_level(b
);
2538 p
->nodes
[level
] = b
;
2539 if (!p
->skip_locking
)
2540 p
->locks
[level
] = root_lock
;
2543 level
= btrfs_header_level(b
);
2546 * setup the path here so we can release it under lock
2547 * contention with the cow code
2551 * if we don't really need to cow this block
2552 * then we don't want to set the path blocking,
2553 * so we test it here
2555 if (!should_cow_block(trans
, root
, b
))
2558 btrfs_set_path_blocking(p
);
2561 * must have write locks on this node and the
2564 if (level
+ 1 > write_lock_level
) {
2565 write_lock_level
= level
+ 1;
2566 btrfs_release_path(p
);
2570 err
= btrfs_cow_block(trans
, root
, b
,
2571 p
->nodes
[level
+ 1],
2572 p
->slots
[level
+ 1], &b
);
2579 BUG_ON(!cow
&& ins_len
);
2581 p
->nodes
[level
] = b
;
2582 btrfs_clear_path_blocking(p
, NULL
, 0);
2585 * we have a lock on b and as long as we aren't changing
2586 * the tree, there is no way to for the items in b to change.
2587 * It is safe to drop the lock on our parent before we
2588 * go through the expensive btree search on b.
2590 * If cow is true, then we might be changing slot zero,
2591 * which may require changing the parent. So, we can't
2592 * drop the lock until after we know which slot we're
2596 btrfs_unlock_up_safe(p
, level
+ 1);
2598 ret
= bin_search(b
, key
, level
, &slot
);
2602 if (ret
&& slot
> 0) {
2606 p
->slots
[level
] = slot
;
2607 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2608 ins_len
, &write_lock_level
);
2615 b
= p
->nodes
[level
];
2616 slot
= p
->slots
[level
];
2619 * slot 0 is special, if we change the key
2620 * we have to update the parent pointer
2621 * which means we must have a write lock
2624 if (slot
== 0 && cow
&&
2625 write_lock_level
< level
+ 1) {
2626 write_lock_level
= level
+ 1;
2627 btrfs_release_path(p
);
2631 unlock_up(p
, level
, lowest_unlock
,
2632 min_write_lock_level
, &write_lock_level
);
2634 if (level
== lowest_level
) {
2640 err
= read_block_for_search(trans
, root
, p
,
2641 &b
, level
, slot
, key
, 0);
2649 if (!p
->skip_locking
) {
2650 level
= btrfs_header_level(b
);
2651 if (level
<= write_lock_level
) {
2652 err
= btrfs_try_tree_write_lock(b
);
2654 btrfs_set_path_blocking(p
);
2656 btrfs_clear_path_blocking(p
, b
,
2659 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2661 err
= btrfs_try_tree_read_lock(b
);
2663 btrfs_set_path_blocking(p
);
2664 btrfs_tree_read_lock(b
);
2665 btrfs_clear_path_blocking(p
, b
,
2668 p
->locks
[level
] = BTRFS_READ_LOCK
;
2670 p
->nodes
[level
] = b
;
2673 p
->slots
[level
] = slot
;
2675 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2676 if (write_lock_level
< 1) {
2677 write_lock_level
= 1;
2678 btrfs_release_path(p
);
2682 btrfs_set_path_blocking(p
);
2683 err
= split_leaf(trans
, root
, key
,
2684 p
, ins_len
, ret
== 0);
2685 btrfs_clear_path_blocking(p
, NULL
, 0);
2693 if (!p
->search_for_split
)
2694 unlock_up(p
, level
, lowest_unlock
,
2695 min_write_lock_level
, &write_lock_level
);
2702 * we don't really know what they plan on doing with the path
2703 * from here on, so for now just mark it as blocking
2705 if (!p
->leave_spinning
)
2706 btrfs_set_path_blocking(p
);
2708 btrfs_release_path(p
);
2713 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2714 * current state of the tree together with the operations recorded in the tree
2715 * modification log to search for the key in a previous version of this tree, as
2716 * denoted by the time_seq parameter.
2718 * Naturally, there is no support for insert, delete or cow operations.
2720 * The resulting path and return value will be set up as if we called
2721 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2723 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2724 struct btrfs_path
*p
, u64 time_seq
)
2726 struct extent_buffer
*b
;
2731 int lowest_unlock
= 1;
2732 u8 lowest_level
= 0;
2734 lowest_level
= p
->lowest_level
;
2735 WARN_ON(p
->nodes
[0] != NULL
);
2737 if (p
->search_commit_root
) {
2739 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2743 b
= get_old_root(root
, time_seq
);
2744 level
= btrfs_header_level(b
);
2745 p
->locks
[level
] = BTRFS_READ_LOCK
;
2748 level
= btrfs_header_level(b
);
2749 p
->nodes
[level
] = b
;
2750 btrfs_clear_path_blocking(p
, NULL
, 0);
2753 * we have a lock on b and as long as we aren't changing
2754 * the tree, there is no way to for the items in b to change.
2755 * It is safe to drop the lock on our parent before we
2756 * go through the expensive btree search on b.
2758 btrfs_unlock_up_safe(p
, level
+ 1);
2760 ret
= bin_search(b
, key
, level
, &slot
);
2764 if (ret
&& slot
> 0) {
2768 p
->slots
[level
] = slot
;
2769 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2771 if (level
== lowest_level
) {
2777 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
2778 slot
, key
, time_seq
);
2786 level
= btrfs_header_level(b
);
2787 err
= btrfs_try_tree_read_lock(b
);
2789 btrfs_set_path_blocking(p
);
2790 btrfs_tree_read_lock(b
);
2791 btrfs_clear_path_blocking(p
, b
,
2794 p
->locks
[level
] = BTRFS_READ_LOCK
;
2795 p
->nodes
[level
] = b
;
2796 b
= tree_mod_log_rewind(root
->fs_info
, b
, time_seq
);
2797 if (b
!= p
->nodes
[level
]) {
2798 btrfs_tree_unlock_rw(p
->nodes
[level
],
2800 p
->locks
[level
] = 0;
2801 p
->nodes
[level
] = b
;
2804 p
->slots
[level
] = slot
;
2805 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2811 if (!p
->leave_spinning
)
2812 btrfs_set_path_blocking(p
);
2814 btrfs_release_path(p
);
2820 * helper to use instead of search slot if no exact match is needed but
2821 * instead the next or previous item should be returned.
2822 * When find_higher is true, the next higher item is returned, the next lower
2824 * When return_any and find_higher are both true, and no higher item is found,
2825 * return the next lower instead.
2826 * When return_any is true and find_higher is false, and no lower item is found,
2827 * return the next higher instead.
2828 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2831 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
2832 struct btrfs_key
*key
, struct btrfs_path
*p
,
2833 int find_higher
, int return_any
)
2836 struct extent_buffer
*leaf
;
2839 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2843 * a return value of 1 means the path is at the position where the
2844 * item should be inserted. Normally this is the next bigger item,
2845 * but in case the previous item is the last in a leaf, path points
2846 * to the first free slot in the previous leaf, i.e. at an invalid
2852 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2853 ret
= btrfs_next_leaf(root
, p
);
2859 * no higher item found, return the next
2864 btrfs_release_path(p
);
2868 if (p
->slots
[0] == 0) {
2869 ret
= btrfs_prev_leaf(root
, p
);
2873 p
->slots
[0] = btrfs_header_nritems(leaf
) - 1;
2879 * no lower item found, return the next
2884 btrfs_release_path(p
);
2894 * adjust the pointers going up the tree, starting at level
2895 * making sure the right key of each node is points to 'key'.
2896 * This is used after shifting pointers to the left, so it stops
2897 * fixing up pointers when a given leaf/node is not in slot 0 of the
2901 static void fixup_low_keys(struct btrfs_trans_handle
*trans
,
2902 struct btrfs_root
*root
, struct btrfs_path
*path
,
2903 struct btrfs_disk_key
*key
, int level
)
2906 struct extent_buffer
*t
;
2908 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2909 int tslot
= path
->slots
[i
];
2910 if (!path
->nodes
[i
])
2913 tree_mod_log_set_node_key(root
->fs_info
, t
, tslot
, 1);
2914 btrfs_set_node_key(t
, key
, tslot
);
2915 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
2924 * This function isn't completely safe. It's the caller's responsibility
2925 * that the new key won't break the order
2927 void btrfs_set_item_key_safe(struct btrfs_trans_handle
*trans
,
2928 struct btrfs_root
*root
, struct btrfs_path
*path
,
2929 struct btrfs_key
*new_key
)
2931 struct btrfs_disk_key disk_key
;
2932 struct extent_buffer
*eb
;
2935 eb
= path
->nodes
[0];
2936 slot
= path
->slots
[0];
2938 btrfs_item_key(eb
, &disk_key
, slot
- 1);
2939 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
2941 if (slot
< btrfs_header_nritems(eb
) - 1) {
2942 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
2943 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
2946 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2947 btrfs_set_item_key(eb
, &disk_key
, slot
);
2948 btrfs_mark_buffer_dirty(eb
);
2950 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2954 * try to push data from one node into the next node left in the
2957 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2958 * error, and > 0 if there was no room in the left hand block.
2960 static int push_node_left(struct btrfs_trans_handle
*trans
,
2961 struct btrfs_root
*root
, struct extent_buffer
*dst
,
2962 struct extent_buffer
*src
, int empty
)
2969 src_nritems
= btrfs_header_nritems(src
);
2970 dst_nritems
= btrfs_header_nritems(dst
);
2971 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
2972 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
2973 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
2975 if (!empty
&& src_nritems
<= 8)
2978 if (push_items
<= 0)
2982 push_items
= min(src_nritems
, push_items
);
2983 if (push_items
< src_nritems
) {
2984 /* leave at least 8 pointers in the node if
2985 * we aren't going to empty it
2987 if (src_nritems
- push_items
< 8) {
2988 if (push_items
<= 8)
2994 push_items
= min(src_nritems
- 8, push_items
);
2996 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
2998 copy_extent_buffer(dst
, src
,
2999 btrfs_node_key_ptr_offset(dst_nritems
),
3000 btrfs_node_key_ptr_offset(0),
3001 push_items
* sizeof(struct btrfs_key_ptr
));
3003 if (push_items
< src_nritems
) {
3005 * don't call tree_mod_log_eb_move here, key removal was already
3006 * fully logged by tree_mod_log_eb_copy above.
3008 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3009 btrfs_node_key_ptr_offset(push_items
),
3010 (src_nritems
- push_items
) *
3011 sizeof(struct btrfs_key_ptr
));
3013 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3014 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3015 btrfs_mark_buffer_dirty(src
);
3016 btrfs_mark_buffer_dirty(dst
);
3022 * try to push data from one node into the next node right in the
3025 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3026 * error, and > 0 if there was no room in the right hand block.
3028 * this will only push up to 1/2 the contents of the left node over
3030 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3031 struct btrfs_root
*root
,
3032 struct extent_buffer
*dst
,
3033 struct extent_buffer
*src
)
3041 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3042 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3044 src_nritems
= btrfs_header_nritems(src
);
3045 dst_nritems
= btrfs_header_nritems(dst
);
3046 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3047 if (push_items
<= 0)
3050 if (src_nritems
< 4)
3053 max_push
= src_nritems
/ 2 + 1;
3054 /* don't try to empty the node */
3055 if (max_push
>= src_nritems
)
3058 if (max_push
< push_items
)
3059 push_items
= max_push
;
3061 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3062 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3063 btrfs_node_key_ptr_offset(0),
3065 sizeof(struct btrfs_key_ptr
));
3067 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3068 src_nritems
- push_items
, push_items
);
3069 copy_extent_buffer(dst
, src
,
3070 btrfs_node_key_ptr_offset(0),
3071 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3072 push_items
* sizeof(struct btrfs_key_ptr
));
3074 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3075 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3077 btrfs_mark_buffer_dirty(src
);
3078 btrfs_mark_buffer_dirty(dst
);
3084 * helper function to insert a new root level in the tree.
3085 * A new node is allocated, and a single item is inserted to
3086 * point to the existing root
3088 * returns zero on success or < 0 on failure.
3090 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3091 struct btrfs_root
*root
,
3092 struct btrfs_path
*path
, int level
)
3095 struct extent_buffer
*lower
;
3096 struct extent_buffer
*c
;
3097 struct extent_buffer
*old
;
3098 struct btrfs_disk_key lower_key
;
3100 BUG_ON(path
->nodes
[level
]);
3101 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3103 lower
= path
->nodes
[level
-1];
3105 btrfs_item_key(lower
, &lower_key
, 0);
3107 btrfs_node_key(lower
, &lower_key
, 0);
3109 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3110 root
->root_key
.objectid
, &lower_key
,
3111 level
, root
->node
->start
, 0);
3115 root_add_used(root
, root
->nodesize
);
3117 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3118 btrfs_set_header_nritems(c
, 1);
3119 btrfs_set_header_level(c
, level
);
3120 btrfs_set_header_bytenr(c
, c
->start
);
3121 btrfs_set_header_generation(c
, trans
->transid
);
3122 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3123 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3125 write_extent_buffer(c
, root
->fs_info
->fsid
,
3126 (unsigned long)btrfs_header_fsid(c
),
3129 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3130 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
3133 btrfs_set_node_key(c
, &lower_key
, 0);
3134 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3135 lower_gen
= btrfs_header_generation(lower
);
3136 WARN_ON(lower_gen
!= trans
->transid
);
3138 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3140 btrfs_mark_buffer_dirty(c
);
3143 tree_mod_log_set_root_pointer(root
, c
);
3144 rcu_assign_pointer(root
->node
, c
);
3146 /* the super has an extra ref to root->node */
3147 free_extent_buffer(old
);
3149 add_root_to_dirty_list(root
);
3150 extent_buffer_get(c
);
3151 path
->nodes
[level
] = c
;
3152 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
3153 path
->slots
[level
] = 0;
3158 * worker function to insert a single pointer in a node.
3159 * the node should have enough room for the pointer already
3161 * slot and level indicate where you want the key to go, and
3162 * blocknr is the block the key points to.
3164 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3165 struct btrfs_root
*root
, struct btrfs_path
*path
,
3166 struct btrfs_disk_key
*key
, u64 bytenr
,
3167 int slot
, int level
)
3169 struct extent_buffer
*lower
;
3173 BUG_ON(!path
->nodes
[level
]);
3174 btrfs_assert_tree_locked(path
->nodes
[level
]);
3175 lower
= path
->nodes
[level
];
3176 nritems
= btrfs_header_nritems(lower
);
3177 BUG_ON(slot
> nritems
);
3178 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3179 if (slot
!= nritems
) {
3181 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3182 slot
, nritems
- slot
);
3183 memmove_extent_buffer(lower
,
3184 btrfs_node_key_ptr_offset(slot
+ 1),
3185 btrfs_node_key_ptr_offset(slot
),
3186 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3189 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3193 btrfs_set_node_key(lower
, key
, slot
);
3194 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3195 WARN_ON(trans
->transid
== 0);
3196 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3197 btrfs_set_header_nritems(lower
, nritems
+ 1);
3198 btrfs_mark_buffer_dirty(lower
);
3202 * split the node at the specified level in path in two.
3203 * The path is corrected to point to the appropriate node after the split
3205 * Before splitting this tries to make some room in the node by pushing
3206 * left and right, if either one works, it returns right away.
3208 * returns 0 on success and < 0 on failure
3210 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3211 struct btrfs_root
*root
,
3212 struct btrfs_path
*path
, int level
)
3214 struct extent_buffer
*c
;
3215 struct extent_buffer
*split
;
3216 struct btrfs_disk_key disk_key
;
3221 c
= path
->nodes
[level
];
3222 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3223 if (c
== root
->node
) {
3224 /* trying to split the root, lets make a new one */
3225 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3229 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3230 c
= path
->nodes
[level
];
3231 if (!ret
&& btrfs_header_nritems(c
) <
3232 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3238 c_nritems
= btrfs_header_nritems(c
);
3239 mid
= (c_nritems
+ 1) / 2;
3240 btrfs_node_key(c
, &disk_key
, mid
);
3242 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3243 root
->root_key
.objectid
,
3244 &disk_key
, level
, c
->start
, 0);
3246 return PTR_ERR(split
);
3248 root_add_used(root
, root
->nodesize
);
3250 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3251 btrfs_set_header_level(split
, btrfs_header_level(c
));
3252 btrfs_set_header_bytenr(split
, split
->start
);
3253 btrfs_set_header_generation(split
, trans
->transid
);
3254 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3255 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3256 write_extent_buffer(split
, root
->fs_info
->fsid
,
3257 (unsigned long)btrfs_header_fsid(split
),
3259 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3260 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
3263 tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0, mid
, c_nritems
- mid
);
3264 copy_extent_buffer(split
, c
,
3265 btrfs_node_key_ptr_offset(0),
3266 btrfs_node_key_ptr_offset(mid
),
3267 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3268 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3269 btrfs_set_header_nritems(c
, mid
);
3272 btrfs_mark_buffer_dirty(c
);
3273 btrfs_mark_buffer_dirty(split
);
3275 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3276 path
->slots
[level
+ 1] + 1, level
+ 1);
3278 if (path
->slots
[level
] >= mid
) {
3279 path
->slots
[level
] -= mid
;
3280 btrfs_tree_unlock(c
);
3281 free_extent_buffer(c
);
3282 path
->nodes
[level
] = split
;
3283 path
->slots
[level
+ 1] += 1;
3285 btrfs_tree_unlock(split
);
3286 free_extent_buffer(split
);
3292 * how many bytes are required to store the items in a leaf. start
3293 * and nr indicate which items in the leaf to check. This totals up the
3294 * space used both by the item structs and the item data
3296 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3299 int nritems
= btrfs_header_nritems(l
);
3300 int end
= min(nritems
, start
+ nr
) - 1;
3304 data_len
= btrfs_item_end_nr(l
, start
);
3305 data_len
= data_len
- btrfs_item_offset_nr(l
, end
);
3306 data_len
+= sizeof(struct btrfs_item
) * nr
;
3307 WARN_ON(data_len
< 0);
3312 * The space between the end of the leaf items and
3313 * the start of the leaf data. IOW, how much room
3314 * the leaf has left for both items and data
3316 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3317 struct extent_buffer
*leaf
)
3319 int nritems
= btrfs_header_nritems(leaf
);
3321 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3323 printk(KERN_CRIT
"leaf free space ret %d, leaf data size %lu, "
3324 "used %d nritems %d\n",
3325 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3326 leaf_space_used(leaf
, 0, nritems
), nritems
);
3332 * min slot controls the lowest index we're willing to push to the
3333 * right. We'll push up to and including min_slot, but no lower
3335 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3336 struct btrfs_root
*root
,
3337 struct btrfs_path
*path
,
3338 int data_size
, int empty
,
3339 struct extent_buffer
*right
,
3340 int free_space
, u32 left_nritems
,
3343 struct extent_buffer
*left
= path
->nodes
[0];
3344 struct extent_buffer
*upper
= path
->nodes
[1];
3345 struct btrfs_map_token token
;
3346 struct btrfs_disk_key disk_key
;
3351 struct btrfs_item
*item
;
3357 btrfs_init_map_token(&token
);
3362 nr
= max_t(u32
, 1, min_slot
);
3364 if (path
->slots
[0] >= left_nritems
)
3365 push_space
+= data_size
;
3367 slot
= path
->slots
[1];
3368 i
= left_nritems
- 1;
3370 item
= btrfs_item_nr(left
, i
);
3372 if (!empty
&& push_items
> 0) {
3373 if (path
->slots
[0] > i
)
3375 if (path
->slots
[0] == i
) {
3376 int space
= btrfs_leaf_free_space(root
, left
);
3377 if (space
+ push_space
* 2 > free_space
)
3382 if (path
->slots
[0] == i
)
3383 push_space
+= data_size
;
3385 this_item_size
= btrfs_item_size(left
, item
);
3386 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3390 push_space
+= this_item_size
+ sizeof(*item
);
3396 if (push_items
== 0)
3399 WARN_ON(!empty
&& push_items
== left_nritems
);
3401 /* push left to right */
3402 right_nritems
= btrfs_header_nritems(right
);
3404 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3405 push_space
-= leaf_data_end(root
, left
);
3407 /* make room in the right data area */
3408 data_end
= leaf_data_end(root
, right
);
3409 memmove_extent_buffer(right
,
3410 btrfs_leaf_data(right
) + data_end
- push_space
,
3411 btrfs_leaf_data(right
) + data_end
,
3412 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3414 /* copy from the left data area */
3415 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3416 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3417 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3420 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3421 btrfs_item_nr_offset(0),
3422 right_nritems
* sizeof(struct btrfs_item
));
3424 /* copy the items from left to right */
3425 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3426 btrfs_item_nr_offset(left_nritems
- push_items
),
3427 push_items
* sizeof(struct btrfs_item
));
3429 /* update the item pointers */
3430 right_nritems
+= push_items
;
3431 btrfs_set_header_nritems(right
, right_nritems
);
3432 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3433 for (i
= 0; i
< right_nritems
; i
++) {
3434 item
= btrfs_item_nr(right
, i
);
3435 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3436 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3439 left_nritems
-= push_items
;
3440 btrfs_set_header_nritems(left
, left_nritems
);
3443 btrfs_mark_buffer_dirty(left
);
3445 clean_tree_block(trans
, root
, left
);
3447 btrfs_mark_buffer_dirty(right
);
3449 btrfs_item_key(right
, &disk_key
, 0);
3450 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3451 btrfs_mark_buffer_dirty(upper
);
3453 /* then fixup the leaf pointer in the path */
3454 if (path
->slots
[0] >= left_nritems
) {
3455 path
->slots
[0] -= left_nritems
;
3456 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3457 clean_tree_block(trans
, root
, path
->nodes
[0]);
3458 btrfs_tree_unlock(path
->nodes
[0]);
3459 free_extent_buffer(path
->nodes
[0]);
3460 path
->nodes
[0] = right
;
3461 path
->slots
[1] += 1;
3463 btrfs_tree_unlock(right
);
3464 free_extent_buffer(right
);
3469 btrfs_tree_unlock(right
);
3470 free_extent_buffer(right
);
3475 * push some data in the path leaf to the right, trying to free up at
3476 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3478 * returns 1 if the push failed because the other node didn't have enough
3479 * room, 0 if everything worked out and < 0 if there were major errors.
3481 * this will push starting from min_slot to the end of the leaf. It won't
3482 * push any slot lower than min_slot
3484 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3485 *root
, struct btrfs_path
*path
,
3486 int min_data_size
, int data_size
,
3487 int empty
, u32 min_slot
)
3489 struct extent_buffer
*left
= path
->nodes
[0];
3490 struct extent_buffer
*right
;
3491 struct extent_buffer
*upper
;
3497 if (!path
->nodes
[1])
3500 slot
= path
->slots
[1];
3501 upper
= path
->nodes
[1];
3502 if (slot
>= btrfs_header_nritems(upper
) - 1)
3505 btrfs_assert_tree_locked(path
->nodes
[1]);
3507 right
= read_node_slot(root
, upper
, slot
+ 1);
3511 btrfs_tree_lock(right
);
3512 btrfs_set_lock_blocking(right
);
3514 free_space
= btrfs_leaf_free_space(root
, right
);
3515 if (free_space
< data_size
)
3518 /* cow and double check */
3519 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3524 free_space
= btrfs_leaf_free_space(root
, right
);
3525 if (free_space
< data_size
)
3528 left_nritems
= btrfs_header_nritems(left
);
3529 if (left_nritems
== 0)
3532 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3533 right
, free_space
, left_nritems
, min_slot
);
3535 btrfs_tree_unlock(right
);
3536 free_extent_buffer(right
);
3541 * push some data in the path leaf to the left, trying to free up at
3542 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3544 * max_slot can put a limit on how far into the leaf we'll push items. The
3545 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3548 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3549 struct btrfs_root
*root
,
3550 struct btrfs_path
*path
, int data_size
,
3551 int empty
, struct extent_buffer
*left
,
3552 int free_space
, u32 right_nritems
,
3555 struct btrfs_disk_key disk_key
;
3556 struct extent_buffer
*right
= path
->nodes
[0];
3560 struct btrfs_item
*item
;
3561 u32 old_left_nritems
;
3565 u32 old_left_item_size
;
3566 struct btrfs_map_token token
;
3568 btrfs_init_map_token(&token
);
3571 nr
= min(right_nritems
, max_slot
);
3573 nr
= min(right_nritems
- 1, max_slot
);
3575 for (i
= 0; i
< nr
; i
++) {
3576 item
= btrfs_item_nr(right
, i
);
3578 if (!empty
&& push_items
> 0) {
3579 if (path
->slots
[0] < i
)
3581 if (path
->slots
[0] == i
) {
3582 int space
= btrfs_leaf_free_space(root
, right
);
3583 if (space
+ push_space
* 2 > free_space
)
3588 if (path
->slots
[0] == i
)
3589 push_space
+= data_size
;
3591 this_item_size
= btrfs_item_size(right
, item
);
3592 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3596 push_space
+= this_item_size
+ sizeof(*item
);
3599 if (push_items
== 0) {
3603 if (!empty
&& push_items
== btrfs_header_nritems(right
))
3606 /* push data from right to left */
3607 copy_extent_buffer(left
, right
,
3608 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3609 btrfs_item_nr_offset(0),
3610 push_items
* sizeof(struct btrfs_item
));
3612 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3613 btrfs_item_offset_nr(right
, push_items
- 1);
3615 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3616 leaf_data_end(root
, left
) - push_space
,
3617 btrfs_leaf_data(right
) +
3618 btrfs_item_offset_nr(right
, push_items
- 1),
3620 old_left_nritems
= btrfs_header_nritems(left
);
3621 BUG_ON(old_left_nritems
<= 0);
3623 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3624 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3627 item
= btrfs_item_nr(left
, i
);
3629 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3630 btrfs_set_token_item_offset(left
, item
,
3631 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3634 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3636 /* fixup right node */
3637 if (push_items
> right_nritems
)
3638 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3641 if (push_items
< right_nritems
) {
3642 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3643 leaf_data_end(root
, right
);
3644 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3645 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3646 btrfs_leaf_data(right
) +
3647 leaf_data_end(root
, right
), push_space
);
3649 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3650 btrfs_item_nr_offset(push_items
),
3651 (btrfs_header_nritems(right
) - push_items
) *
3652 sizeof(struct btrfs_item
));
3654 right_nritems
-= push_items
;
3655 btrfs_set_header_nritems(right
, right_nritems
);
3656 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3657 for (i
= 0; i
< right_nritems
; i
++) {
3658 item
= btrfs_item_nr(right
, i
);
3660 push_space
= push_space
- btrfs_token_item_size(right
,
3662 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3665 btrfs_mark_buffer_dirty(left
);
3667 btrfs_mark_buffer_dirty(right
);
3669 clean_tree_block(trans
, root
, right
);
3671 btrfs_item_key(right
, &disk_key
, 0);
3672 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3674 /* then fixup the leaf pointer in the path */
3675 if (path
->slots
[0] < push_items
) {
3676 path
->slots
[0] += old_left_nritems
;
3677 btrfs_tree_unlock(path
->nodes
[0]);
3678 free_extent_buffer(path
->nodes
[0]);
3679 path
->nodes
[0] = left
;
3680 path
->slots
[1] -= 1;
3682 btrfs_tree_unlock(left
);
3683 free_extent_buffer(left
);
3684 path
->slots
[0] -= push_items
;
3686 BUG_ON(path
->slots
[0] < 0);
3689 btrfs_tree_unlock(left
);
3690 free_extent_buffer(left
);
3695 * push some data in the path leaf to the left, trying to free up at
3696 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3698 * max_slot can put a limit on how far into the leaf we'll push items. The
3699 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3702 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3703 *root
, struct btrfs_path
*path
, int min_data_size
,
3704 int data_size
, int empty
, u32 max_slot
)
3706 struct extent_buffer
*right
= path
->nodes
[0];
3707 struct extent_buffer
*left
;
3713 slot
= path
->slots
[1];
3716 if (!path
->nodes
[1])
3719 right_nritems
= btrfs_header_nritems(right
);
3720 if (right_nritems
== 0)
3723 btrfs_assert_tree_locked(path
->nodes
[1]);
3725 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
3729 btrfs_tree_lock(left
);
3730 btrfs_set_lock_blocking(left
);
3732 free_space
= btrfs_leaf_free_space(root
, left
);
3733 if (free_space
< data_size
) {
3738 /* cow and double check */
3739 ret
= btrfs_cow_block(trans
, root
, left
,
3740 path
->nodes
[1], slot
- 1, &left
);
3742 /* we hit -ENOSPC, but it isn't fatal here */
3748 free_space
= btrfs_leaf_free_space(root
, left
);
3749 if (free_space
< data_size
) {
3754 return __push_leaf_left(trans
, root
, path
, min_data_size
,
3755 empty
, left
, free_space
, right_nritems
,
3758 btrfs_tree_unlock(left
);
3759 free_extent_buffer(left
);
3764 * split the path's leaf in two, making sure there is at least data_size
3765 * available for the resulting leaf level of the path.
3767 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
3768 struct btrfs_root
*root
,
3769 struct btrfs_path
*path
,
3770 struct extent_buffer
*l
,
3771 struct extent_buffer
*right
,
3772 int slot
, int mid
, int nritems
)
3777 struct btrfs_disk_key disk_key
;
3778 struct btrfs_map_token token
;
3780 btrfs_init_map_token(&token
);
3782 nritems
= nritems
- mid
;
3783 btrfs_set_header_nritems(right
, nritems
);
3784 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
3786 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
3787 btrfs_item_nr_offset(mid
),
3788 nritems
* sizeof(struct btrfs_item
));
3790 copy_extent_buffer(right
, l
,
3791 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
3792 data_copy_size
, btrfs_leaf_data(l
) +
3793 leaf_data_end(root
, l
), data_copy_size
);
3795 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
3796 btrfs_item_end_nr(l
, mid
);
3798 for (i
= 0; i
< nritems
; i
++) {
3799 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
3802 ioff
= btrfs_token_item_offset(right
, item
, &token
);
3803 btrfs_set_token_item_offset(right
, item
,
3804 ioff
+ rt_data_off
, &token
);
3807 btrfs_set_header_nritems(l
, mid
);
3808 btrfs_item_key(right
, &disk_key
, 0);
3809 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3810 path
->slots
[1] + 1, 1);
3812 btrfs_mark_buffer_dirty(right
);
3813 btrfs_mark_buffer_dirty(l
);
3814 BUG_ON(path
->slots
[0] != slot
);
3817 btrfs_tree_unlock(path
->nodes
[0]);
3818 free_extent_buffer(path
->nodes
[0]);
3819 path
->nodes
[0] = right
;
3820 path
->slots
[0] -= mid
;
3821 path
->slots
[1] += 1;
3823 btrfs_tree_unlock(right
);
3824 free_extent_buffer(right
);
3827 BUG_ON(path
->slots
[0] < 0);
3831 * double splits happen when we need to insert a big item in the middle
3832 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3833 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3836 * We avoid this by trying to push the items on either side of our target
3837 * into the adjacent leaves. If all goes well we can avoid the double split
3840 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
3841 struct btrfs_root
*root
,
3842 struct btrfs_path
*path
,
3850 slot
= path
->slots
[0];
3853 * try to push all the items after our slot into the
3856 ret
= push_leaf_right(trans
, root
, path
, 1, data_size
, 0, slot
);
3863 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3865 * our goal is to get our slot at the start or end of a leaf. If
3866 * we've done so we're done
3868 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
3871 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
3874 /* try to push all the items before our slot into the next leaf */
3875 slot
= path
->slots
[0];
3876 ret
= push_leaf_left(trans
, root
, path
, 1, data_size
, 0, slot
);
3889 * split the path's leaf in two, making sure there is at least data_size
3890 * available for the resulting leaf level of the path.
3892 * returns 0 if all went well and < 0 on failure.
3894 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
3895 struct btrfs_root
*root
,
3896 struct btrfs_key
*ins_key
,
3897 struct btrfs_path
*path
, int data_size
,
3900 struct btrfs_disk_key disk_key
;
3901 struct extent_buffer
*l
;
3905 struct extent_buffer
*right
;
3909 int num_doubles
= 0;
3910 int tried_avoid_double
= 0;
3913 slot
= path
->slots
[0];
3914 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
3915 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
3918 /* first try to make some room by pushing left and right */
3920 wret
= push_leaf_right(trans
, root
, path
, data_size
,
3925 wret
= push_leaf_left(trans
, root
, path
, data_size
,
3926 data_size
, 0, (u32
)-1);
3932 /* did the pushes work? */
3933 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
3937 if (!path
->nodes
[1]) {
3938 ret
= insert_new_root(trans
, root
, path
, 1);
3945 slot
= path
->slots
[0];
3946 nritems
= btrfs_header_nritems(l
);
3947 mid
= (nritems
+ 1) / 2;
3951 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
3952 BTRFS_LEAF_DATA_SIZE(root
)) {
3953 if (slot
>= nritems
) {
3957 if (mid
!= nritems
&&
3958 leaf_space_used(l
, mid
, nritems
- mid
) +
3959 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3960 if (data_size
&& !tried_avoid_double
)
3961 goto push_for_double
;
3967 if (leaf_space_used(l
, 0, mid
) + data_size
>
3968 BTRFS_LEAF_DATA_SIZE(root
)) {
3969 if (!extend
&& data_size
&& slot
== 0) {
3971 } else if ((extend
|| !data_size
) && slot
== 0) {
3975 if (mid
!= nritems
&&
3976 leaf_space_used(l
, mid
, nritems
- mid
) +
3977 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3978 if (data_size
&& !tried_avoid_double
)
3979 goto push_for_double
;
3987 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
3989 btrfs_item_key(l
, &disk_key
, mid
);
3991 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
3992 root
->root_key
.objectid
,
3993 &disk_key
, 0, l
->start
, 0);
3995 return PTR_ERR(right
);
3997 root_add_used(root
, root
->leafsize
);
3999 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4000 btrfs_set_header_bytenr(right
, right
->start
);
4001 btrfs_set_header_generation(right
, trans
->transid
);
4002 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4003 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4004 btrfs_set_header_level(right
, 0);
4005 write_extent_buffer(right
, root
->fs_info
->fsid
,
4006 (unsigned long)btrfs_header_fsid(right
),
4009 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
4010 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
4015 btrfs_set_header_nritems(right
, 0);
4016 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4017 path
->slots
[1] + 1, 1);
4018 btrfs_tree_unlock(path
->nodes
[0]);
4019 free_extent_buffer(path
->nodes
[0]);
4020 path
->nodes
[0] = right
;
4022 path
->slots
[1] += 1;
4024 btrfs_set_header_nritems(right
, 0);
4025 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4027 btrfs_tree_unlock(path
->nodes
[0]);
4028 free_extent_buffer(path
->nodes
[0]);
4029 path
->nodes
[0] = right
;
4031 if (path
->slots
[1] == 0)
4032 fixup_low_keys(trans
, root
, path
,
4035 btrfs_mark_buffer_dirty(right
);
4039 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4042 BUG_ON(num_doubles
!= 0);
4050 push_for_double_split(trans
, root
, path
, data_size
);
4051 tried_avoid_double
= 1;
4052 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4057 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4058 struct btrfs_root
*root
,
4059 struct btrfs_path
*path
, int ins_len
)
4061 struct btrfs_key key
;
4062 struct extent_buffer
*leaf
;
4063 struct btrfs_file_extent_item
*fi
;
4068 leaf
= path
->nodes
[0];
4069 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4071 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4072 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4074 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4077 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4078 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4079 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4080 struct btrfs_file_extent_item
);
4081 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4083 btrfs_release_path(path
);
4085 path
->keep_locks
= 1;
4086 path
->search_for_split
= 1;
4087 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4088 path
->search_for_split
= 0;
4093 leaf
= path
->nodes
[0];
4094 /* if our item isn't there or got smaller, return now */
4095 if (ret
> 0 || item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4098 /* the leaf has changed, it now has room. return now */
4099 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4102 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4103 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4104 struct btrfs_file_extent_item
);
4105 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4109 btrfs_set_path_blocking(path
);
4110 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4114 path
->keep_locks
= 0;
4115 btrfs_unlock_up_safe(path
, 1);
4118 path
->keep_locks
= 0;
4122 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4123 struct btrfs_root
*root
,
4124 struct btrfs_path
*path
,
4125 struct btrfs_key
*new_key
,
4126 unsigned long split_offset
)
4128 struct extent_buffer
*leaf
;
4129 struct btrfs_item
*item
;
4130 struct btrfs_item
*new_item
;
4136 struct btrfs_disk_key disk_key
;
4138 leaf
= path
->nodes
[0];
4139 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4141 btrfs_set_path_blocking(path
);
4143 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
4144 orig_offset
= btrfs_item_offset(leaf
, item
);
4145 item_size
= btrfs_item_size(leaf
, item
);
4147 buf
= kmalloc(item_size
, GFP_NOFS
);
4151 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4152 path
->slots
[0]), item_size
);
4154 slot
= path
->slots
[0] + 1;
4155 nritems
= btrfs_header_nritems(leaf
);
4156 if (slot
!= nritems
) {
4157 /* shift the items */
4158 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4159 btrfs_item_nr_offset(slot
),
4160 (nritems
- slot
) * sizeof(struct btrfs_item
));
4163 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4164 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4166 new_item
= btrfs_item_nr(leaf
, slot
);
4168 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4169 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4171 btrfs_set_item_offset(leaf
, item
,
4172 orig_offset
+ item_size
- split_offset
);
4173 btrfs_set_item_size(leaf
, item
, split_offset
);
4175 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4177 /* write the data for the start of the original item */
4178 write_extent_buffer(leaf
, buf
,
4179 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4182 /* write the data for the new item */
4183 write_extent_buffer(leaf
, buf
+ split_offset
,
4184 btrfs_item_ptr_offset(leaf
, slot
),
4185 item_size
- split_offset
);
4186 btrfs_mark_buffer_dirty(leaf
);
4188 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4194 * This function splits a single item into two items,
4195 * giving 'new_key' to the new item and splitting the
4196 * old one at split_offset (from the start of the item).
4198 * The path may be released by this operation. After
4199 * the split, the path is pointing to the old item. The
4200 * new item is going to be in the same node as the old one.
4202 * Note, the item being split must be smaller enough to live alone on
4203 * a tree block with room for one extra struct btrfs_item
4205 * This allows us to split the item in place, keeping a lock on the
4206 * leaf the entire time.
4208 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4209 struct btrfs_root
*root
,
4210 struct btrfs_path
*path
,
4211 struct btrfs_key
*new_key
,
4212 unsigned long split_offset
)
4215 ret
= setup_leaf_for_split(trans
, root
, path
,
4216 sizeof(struct btrfs_item
));
4220 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4225 * This function duplicate a item, giving 'new_key' to the new item.
4226 * It guarantees both items live in the same tree leaf and the new item
4227 * is contiguous with the original item.
4229 * This allows us to split file extent in place, keeping a lock on the
4230 * leaf the entire time.
4232 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4233 struct btrfs_root
*root
,
4234 struct btrfs_path
*path
,
4235 struct btrfs_key
*new_key
)
4237 struct extent_buffer
*leaf
;
4241 leaf
= path
->nodes
[0];
4242 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4243 ret
= setup_leaf_for_split(trans
, root
, path
,
4244 item_size
+ sizeof(struct btrfs_item
));
4249 setup_items_for_insert(trans
, root
, path
, new_key
, &item_size
,
4250 item_size
, item_size
+
4251 sizeof(struct btrfs_item
), 1);
4252 leaf
= path
->nodes
[0];
4253 memcpy_extent_buffer(leaf
,
4254 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4255 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4261 * make the item pointed to by the path smaller. new_size indicates
4262 * how small to make it, and from_end tells us if we just chop bytes
4263 * off the end of the item or if we shift the item to chop bytes off
4266 void btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
4267 struct btrfs_root
*root
,
4268 struct btrfs_path
*path
,
4269 u32 new_size
, int from_end
)
4272 struct extent_buffer
*leaf
;
4273 struct btrfs_item
*item
;
4275 unsigned int data_end
;
4276 unsigned int old_data_start
;
4277 unsigned int old_size
;
4278 unsigned int size_diff
;
4280 struct btrfs_map_token token
;
4282 btrfs_init_map_token(&token
);
4284 leaf
= path
->nodes
[0];
4285 slot
= path
->slots
[0];
4287 old_size
= btrfs_item_size_nr(leaf
, slot
);
4288 if (old_size
== new_size
)
4291 nritems
= btrfs_header_nritems(leaf
);
4292 data_end
= leaf_data_end(root
, leaf
);
4294 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4296 size_diff
= old_size
- new_size
;
4299 BUG_ON(slot
>= nritems
);
4302 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4304 /* first correct the data pointers */
4305 for (i
= slot
; i
< nritems
; i
++) {
4307 item
= btrfs_item_nr(leaf
, i
);
4309 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4310 btrfs_set_token_item_offset(leaf
, item
,
4311 ioff
+ size_diff
, &token
);
4314 /* shift the data */
4316 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4317 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4318 data_end
, old_data_start
+ new_size
- data_end
);
4320 struct btrfs_disk_key disk_key
;
4323 btrfs_item_key(leaf
, &disk_key
, slot
);
4325 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4327 struct btrfs_file_extent_item
*fi
;
4329 fi
= btrfs_item_ptr(leaf
, slot
,
4330 struct btrfs_file_extent_item
);
4331 fi
= (struct btrfs_file_extent_item
*)(
4332 (unsigned long)fi
- size_diff
);
4334 if (btrfs_file_extent_type(leaf
, fi
) ==
4335 BTRFS_FILE_EXTENT_INLINE
) {
4336 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4337 memmove_extent_buffer(leaf
, ptr
,
4339 offsetof(struct btrfs_file_extent_item
,
4344 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4345 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4346 data_end
, old_data_start
- data_end
);
4348 offset
= btrfs_disk_key_offset(&disk_key
);
4349 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4350 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4352 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4355 item
= btrfs_item_nr(leaf
, slot
);
4356 btrfs_set_item_size(leaf
, item
, new_size
);
4357 btrfs_mark_buffer_dirty(leaf
);
4359 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4360 btrfs_print_leaf(root
, leaf
);
4366 * make the item pointed to by the path bigger, data_size is the new size.
4368 void btrfs_extend_item(struct btrfs_trans_handle
*trans
,
4369 struct btrfs_root
*root
, struct btrfs_path
*path
,
4373 struct extent_buffer
*leaf
;
4374 struct btrfs_item
*item
;
4376 unsigned int data_end
;
4377 unsigned int old_data
;
4378 unsigned int old_size
;
4380 struct btrfs_map_token token
;
4382 btrfs_init_map_token(&token
);
4384 leaf
= path
->nodes
[0];
4386 nritems
= btrfs_header_nritems(leaf
);
4387 data_end
= leaf_data_end(root
, leaf
);
4389 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4390 btrfs_print_leaf(root
, leaf
);
4393 slot
= path
->slots
[0];
4394 old_data
= btrfs_item_end_nr(leaf
, slot
);
4397 if (slot
>= nritems
) {
4398 btrfs_print_leaf(root
, leaf
);
4399 printk(KERN_CRIT
"slot %d too large, nritems %d\n",
4405 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4407 /* first correct the data pointers */
4408 for (i
= slot
; i
< nritems
; i
++) {
4410 item
= btrfs_item_nr(leaf
, i
);
4412 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4413 btrfs_set_token_item_offset(leaf
, item
,
4414 ioff
- data_size
, &token
);
4417 /* shift the data */
4418 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4419 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4420 data_end
, old_data
- data_end
);
4422 data_end
= old_data
;
4423 old_size
= btrfs_item_size_nr(leaf
, slot
);
4424 item
= btrfs_item_nr(leaf
, slot
);
4425 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4426 btrfs_mark_buffer_dirty(leaf
);
4428 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4429 btrfs_print_leaf(root
, leaf
);
4435 * this is a helper for btrfs_insert_empty_items, the main goal here is
4436 * to save stack depth by doing the bulk of the work in a function
4437 * that doesn't call btrfs_search_slot
4439 void setup_items_for_insert(struct btrfs_trans_handle
*trans
,
4440 struct btrfs_root
*root
, struct btrfs_path
*path
,
4441 struct btrfs_key
*cpu_key
, u32
*data_size
,
4442 u32 total_data
, u32 total_size
, int nr
)
4444 struct btrfs_item
*item
;
4447 unsigned int data_end
;
4448 struct btrfs_disk_key disk_key
;
4449 struct extent_buffer
*leaf
;
4451 struct btrfs_map_token token
;
4453 btrfs_init_map_token(&token
);
4455 leaf
= path
->nodes
[0];
4456 slot
= path
->slots
[0];
4458 nritems
= btrfs_header_nritems(leaf
);
4459 data_end
= leaf_data_end(root
, leaf
);
4461 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4462 btrfs_print_leaf(root
, leaf
);
4463 printk(KERN_CRIT
"not enough freespace need %u have %d\n",
4464 total_size
, btrfs_leaf_free_space(root
, leaf
));
4468 if (slot
!= nritems
) {
4469 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4471 if (old_data
< data_end
) {
4472 btrfs_print_leaf(root
, leaf
);
4473 printk(KERN_CRIT
"slot %d old_data %d data_end %d\n",
4474 slot
, old_data
, data_end
);
4478 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4480 /* first correct the data pointers */
4481 for (i
= slot
; i
< nritems
; i
++) {
4484 item
= btrfs_item_nr(leaf
, i
);
4485 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4486 btrfs_set_token_item_offset(leaf
, item
,
4487 ioff
- total_data
, &token
);
4489 /* shift the items */
4490 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4491 btrfs_item_nr_offset(slot
),
4492 (nritems
- slot
) * sizeof(struct btrfs_item
));
4494 /* shift the data */
4495 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4496 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4497 data_end
, old_data
- data_end
);
4498 data_end
= old_data
;
4501 /* setup the item for the new data */
4502 for (i
= 0; i
< nr
; i
++) {
4503 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4504 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4505 item
= btrfs_item_nr(leaf
, slot
+ i
);
4506 btrfs_set_token_item_offset(leaf
, item
,
4507 data_end
- data_size
[i
], &token
);
4508 data_end
-= data_size
[i
];
4509 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4512 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4515 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4516 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4518 btrfs_unlock_up_safe(path
, 1);
4519 btrfs_mark_buffer_dirty(leaf
);
4521 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4522 btrfs_print_leaf(root
, leaf
);
4528 * Given a key and some data, insert items into the tree.
4529 * This does all the path init required, making room in the tree if needed.
4531 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4532 struct btrfs_root
*root
,
4533 struct btrfs_path
*path
,
4534 struct btrfs_key
*cpu_key
, u32
*data_size
,
4543 for (i
= 0; i
< nr
; i
++)
4544 total_data
+= data_size
[i
];
4546 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4547 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4553 slot
= path
->slots
[0];
4556 setup_items_for_insert(trans
, root
, path
, cpu_key
, data_size
,
4557 total_data
, total_size
, nr
);
4562 * Given a key and some data, insert an item into the tree.
4563 * This does all the path init required, making room in the tree if needed.
4565 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4566 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4570 struct btrfs_path
*path
;
4571 struct extent_buffer
*leaf
;
4574 path
= btrfs_alloc_path();
4577 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4579 leaf
= path
->nodes
[0];
4580 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4581 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4582 btrfs_mark_buffer_dirty(leaf
);
4584 btrfs_free_path(path
);
4589 * delete the pointer from a given node.
4591 * the tree should have been previously balanced so the deletion does not
4594 static void del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4595 struct btrfs_path
*path
, int level
, int slot
)
4597 struct extent_buffer
*parent
= path
->nodes
[level
];
4602 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4603 MOD_LOG_KEY_REMOVE
);
4607 nritems
= btrfs_header_nritems(parent
);
4608 if (slot
!= nritems
- 1) {
4610 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4611 slot
+ 1, nritems
- slot
- 1);
4612 memmove_extent_buffer(parent
,
4613 btrfs_node_key_ptr_offset(slot
),
4614 btrfs_node_key_ptr_offset(slot
+ 1),
4615 sizeof(struct btrfs_key_ptr
) *
4616 (nritems
- slot
- 1));
4620 btrfs_set_header_nritems(parent
, nritems
);
4621 if (nritems
== 0 && parent
== root
->node
) {
4622 BUG_ON(btrfs_header_level(root
->node
) != 1);
4623 /* just turn the root into a leaf and break */
4624 btrfs_set_header_level(root
->node
, 0);
4625 } else if (slot
== 0) {
4626 struct btrfs_disk_key disk_key
;
4628 btrfs_node_key(parent
, &disk_key
, 0);
4629 fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
4631 btrfs_mark_buffer_dirty(parent
);
4635 * a helper function to delete the leaf pointed to by path->slots[1] and
4638 * This deletes the pointer in path->nodes[1] and frees the leaf
4639 * block extent. zero is returned if it all worked out, < 0 otherwise.
4641 * The path must have already been setup for deleting the leaf, including
4642 * all the proper balancing. path->nodes[1] must be locked.
4644 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4645 struct btrfs_root
*root
,
4646 struct btrfs_path
*path
,
4647 struct extent_buffer
*leaf
)
4649 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4650 del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
4653 * btrfs_free_extent is expensive, we want to make sure we
4654 * aren't holding any locks when we call it
4656 btrfs_unlock_up_safe(path
, 0);
4658 root_sub_used(root
, leaf
->len
);
4660 extent_buffer_get(leaf
);
4661 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4662 free_extent_buffer_stale(leaf
);
4665 * delete the item at the leaf level in path. If that empties
4666 * the leaf, remove it from the tree
4668 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4669 struct btrfs_path
*path
, int slot
, int nr
)
4671 struct extent_buffer
*leaf
;
4672 struct btrfs_item
*item
;
4679 struct btrfs_map_token token
;
4681 btrfs_init_map_token(&token
);
4683 leaf
= path
->nodes
[0];
4684 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4686 for (i
= 0; i
< nr
; i
++)
4687 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4689 nritems
= btrfs_header_nritems(leaf
);
4691 if (slot
+ nr
!= nritems
) {
4692 int data_end
= leaf_data_end(root
, leaf
);
4694 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4696 btrfs_leaf_data(leaf
) + data_end
,
4697 last_off
- data_end
);
4699 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4702 item
= btrfs_item_nr(leaf
, i
);
4703 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4704 btrfs_set_token_item_offset(leaf
, item
,
4705 ioff
+ dsize
, &token
);
4708 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4709 btrfs_item_nr_offset(slot
+ nr
),
4710 sizeof(struct btrfs_item
) *
4711 (nritems
- slot
- nr
));
4713 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4716 /* delete the leaf if we've emptied it */
4718 if (leaf
== root
->node
) {
4719 btrfs_set_header_level(leaf
, 0);
4721 btrfs_set_path_blocking(path
);
4722 clean_tree_block(trans
, root
, leaf
);
4723 btrfs_del_leaf(trans
, root
, path
, leaf
);
4726 int used
= leaf_space_used(leaf
, 0, nritems
);
4728 struct btrfs_disk_key disk_key
;
4730 btrfs_item_key(leaf
, &disk_key
, 0);
4731 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4734 /* delete the leaf if it is mostly empty */
4735 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
4736 /* push_leaf_left fixes the path.
4737 * make sure the path still points to our leaf
4738 * for possible call to del_ptr below
4740 slot
= path
->slots
[1];
4741 extent_buffer_get(leaf
);
4743 btrfs_set_path_blocking(path
);
4744 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
4746 if (wret
< 0 && wret
!= -ENOSPC
)
4749 if (path
->nodes
[0] == leaf
&&
4750 btrfs_header_nritems(leaf
)) {
4751 wret
= push_leaf_right(trans
, root
, path
, 1,
4753 if (wret
< 0 && wret
!= -ENOSPC
)
4757 if (btrfs_header_nritems(leaf
) == 0) {
4758 path
->slots
[1] = slot
;
4759 btrfs_del_leaf(trans
, root
, path
, leaf
);
4760 free_extent_buffer(leaf
);
4763 /* if we're still in the path, make sure
4764 * we're dirty. Otherwise, one of the
4765 * push_leaf functions must have already
4766 * dirtied this buffer
4768 if (path
->nodes
[0] == leaf
)
4769 btrfs_mark_buffer_dirty(leaf
);
4770 free_extent_buffer(leaf
);
4773 btrfs_mark_buffer_dirty(leaf
);
4780 * search the tree again to find a leaf with lesser keys
4781 * returns 0 if it found something or 1 if there are no lesser leaves.
4782 * returns < 0 on io errors.
4784 * This may release the path, and so you may lose any locks held at the
4787 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
4789 struct btrfs_key key
;
4790 struct btrfs_disk_key found_key
;
4793 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
4797 else if (key
.type
> 0)
4799 else if (key
.objectid
> 0)
4804 btrfs_release_path(path
);
4805 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4808 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
4809 ret
= comp_keys(&found_key
, &key
);
4816 * A helper function to walk down the tree starting at min_key, and looking
4817 * for nodes or leaves that are either in cache or have a minimum
4818 * transaction id. This is used by the btree defrag code, and tree logging
4820 * This does not cow, but it does stuff the starting key it finds back
4821 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4822 * key and get a writable path.
4824 * This does lock as it descends, and path->keep_locks should be set
4825 * to 1 by the caller.
4827 * This honors path->lowest_level to prevent descent past a given level
4830 * min_trans indicates the oldest transaction that you are interested
4831 * in walking through. Any nodes or leaves older than min_trans are
4832 * skipped over (without reading them).
4834 * returns zero if something useful was found, < 0 on error and 1 if there
4835 * was nothing in the tree that matched the search criteria.
4837 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
4838 struct btrfs_key
*max_key
,
4839 struct btrfs_path
*path
, int cache_only
,
4842 struct extent_buffer
*cur
;
4843 struct btrfs_key found_key
;
4850 WARN_ON(!path
->keep_locks
);
4852 cur
= btrfs_read_lock_root_node(root
);
4853 level
= btrfs_header_level(cur
);
4854 WARN_ON(path
->nodes
[level
]);
4855 path
->nodes
[level
] = cur
;
4856 path
->locks
[level
] = BTRFS_READ_LOCK
;
4858 if (btrfs_header_generation(cur
) < min_trans
) {
4863 nritems
= btrfs_header_nritems(cur
);
4864 level
= btrfs_header_level(cur
);
4865 sret
= bin_search(cur
, min_key
, level
, &slot
);
4867 /* at the lowest level, we're done, setup the path and exit */
4868 if (level
== path
->lowest_level
) {
4869 if (slot
>= nritems
)
4872 path
->slots
[level
] = slot
;
4873 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
4876 if (sret
&& slot
> 0)
4879 * check this node pointer against the cache_only and
4880 * min_trans parameters. If it isn't in cache or is too
4881 * old, skip to the next one.
4883 while (slot
< nritems
) {
4886 struct extent_buffer
*tmp
;
4887 struct btrfs_disk_key disk_key
;
4889 blockptr
= btrfs_node_blockptr(cur
, slot
);
4890 gen
= btrfs_node_ptr_generation(cur
, slot
);
4891 if (gen
< min_trans
) {
4899 btrfs_node_key(cur
, &disk_key
, slot
);
4900 if (comp_keys(&disk_key
, max_key
) >= 0) {
4906 tmp
= btrfs_find_tree_block(root
, blockptr
,
4907 btrfs_level_size(root
, level
- 1));
4909 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
4910 free_extent_buffer(tmp
);
4914 free_extent_buffer(tmp
);
4919 * we didn't find a candidate key in this node, walk forward
4920 * and find another one
4922 if (slot
>= nritems
) {
4923 path
->slots
[level
] = slot
;
4924 btrfs_set_path_blocking(path
);
4925 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
4926 cache_only
, min_trans
);
4928 btrfs_release_path(path
);
4934 /* save our key for returning back */
4935 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
4936 path
->slots
[level
] = slot
;
4937 if (level
== path
->lowest_level
) {
4939 unlock_up(path
, level
, 1, 0, NULL
);
4942 btrfs_set_path_blocking(path
);
4943 cur
= read_node_slot(root
, cur
, slot
);
4944 BUG_ON(!cur
); /* -ENOMEM */
4946 btrfs_tree_read_lock(cur
);
4948 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
4949 path
->nodes
[level
- 1] = cur
;
4950 unlock_up(path
, level
, 1, 0, NULL
);
4951 btrfs_clear_path_blocking(path
, NULL
, 0);
4955 memcpy(min_key
, &found_key
, sizeof(found_key
));
4956 btrfs_set_path_blocking(path
);
4960 static void tree_move_down(struct btrfs_root
*root
,
4961 struct btrfs_path
*path
,
4962 int *level
, int root_level
)
4964 BUG_ON(*level
== 0);
4965 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
4966 path
->slots
[*level
]);
4967 path
->slots
[*level
- 1] = 0;
4971 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
4972 struct btrfs_path
*path
,
4973 int *level
, int root_level
)
4977 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
4979 path
->slots
[*level
]++;
4981 while (path
->slots
[*level
] >= nritems
) {
4982 if (*level
== root_level
)
4986 path
->slots
[*level
] = 0;
4987 free_extent_buffer(path
->nodes
[*level
]);
4988 path
->nodes
[*level
] = NULL
;
4990 path
->slots
[*level
]++;
4992 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
4999 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5002 static int tree_advance(struct btrfs_root
*root
,
5003 struct btrfs_path
*path
,
5004 int *level
, int root_level
,
5006 struct btrfs_key
*key
)
5010 if (*level
== 0 || !allow_down
) {
5011 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5013 tree_move_down(root
, path
, level
, root_level
);
5018 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5019 path
->slots
[*level
]);
5021 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5022 path
->slots
[*level
]);
5027 static int tree_compare_item(struct btrfs_root
*left_root
,
5028 struct btrfs_path
*left_path
,
5029 struct btrfs_path
*right_path
,
5034 unsigned long off1
, off2
;
5036 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5037 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5041 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5042 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5043 right_path
->slots
[0]);
5045 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5047 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5054 #define ADVANCE_ONLY_NEXT -1
5057 * This function compares two trees and calls the provided callback for
5058 * every changed/new/deleted item it finds.
5059 * If shared tree blocks are encountered, whole subtrees are skipped, making
5060 * the compare pretty fast on snapshotted subvolumes.
5062 * This currently works on commit roots only. As commit roots are read only,
5063 * we don't do any locking. The commit roots are protected with transactions.
5064 * Transactions are ended and rejoined when a commit is tried in between.
5066 * This function checks for modifications done to the trees while comparing.
5067 * If it detects a change, it aborts immediately.
5069 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5070 struct btrfs_root
*right_root
,
5071 btrfs_changed_cb_t changed_cb
, void *ctx
)
5075 struct btrfs_trans_handle
*trans
= NULL
;
5076 struct btrfs_path
*left_path
= NULL
;
5077 struct btrfs_path
*right_path
= NULL
;
5078 struct btrfs_key left_key
;
5079 struct btrfs_key right_key
;
5080 char *tmp_buf
= NULL
;
5081 int left_root_level
;
5082 int right_root_level
;
5085 int left_end_reached
;
5086 int right_end_reached
;
5091 u64 left_start_ctransid
;
5092 u64 right_start_ctransid
;
5095 left_path
= btrfs_alloc_path();
5100 right_path
= btrfs_alloc_path();
5106 tmp_buf
= kmalloc(left_root
->leafsize
, GFP_NOFS
);
5112 left_path
->search_commit_root
= 1;
5113 left_path
->skip_locking
= 1;
5114 right_path
->search_commit_root
= 1;
5115 right_path
->skip_locking
= 1;
5117 spin_lock(&left_root
->root_times_lock
);
5118 left_start_ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5119 spin_unlock(&left_root
->root_times_lock
);
5121 spin_lock(&right_root
->root_times_lock
);
5122 right_start_ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5123 spin_unlock(&right_root
->root_times_lock
);
5125 trans
= btrfs_join_transaction(left_root
);
5126 if (IS_ERR(trans
)) {
5127 ret
= PTR_ERR(trans
);
5133 * Strategy: Go to the first items of both trees. Then do
5135 * If both trees are at level 0
5136 * Compare keys of current items
5137 * If left < right treat left item as new, advance left tree
5139 * If left > right treat right item as deleted, advance right tree
5141 * If left == right do deep compare of items, treat as changed if
5142 * needed, advance both trees and repeat
5143 * If both trees are at the same level but not at level 0
5144 * Compare keys of current nodes/leafs
5145 * If left < right advance left tree and repeat
5146 * If left > right advance right tree and repeat
5147 * If left == right compare blockptrs of the next nodes/leafs
5148 * If they match advance both trees but stay at the same level
5150 * If they don't match advance both trees while allowing to go
5152 * If tree levels are different
5153 * Advance the tree that needs it and repeat
5155 * Advancing a tree means:
5156 * If we are at level 0, try to go to the next slot. If that's not
5157 * possible, go one level up and repeat. Stop when we found a level
5158 * where we could go to the next slot. We may at this point be on a
5161 * If we are not at level 0 and not on shared tree blocks, go one
5164 * If we are not at level 0 and on shared tree blocks, go one slot to
5165 * the right if possible or go up and right.
5168 left_level
= btrfs_header_level(left_root
->commit_root
);
5169 left_root_level
= left_level
;
5170 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5171 extent_buffer_get(left_path
->nodes
[left_level
]);
5173 right_level
= btrfs_header_level(right_root
->commit_root
);
5174 right_root_level
= right_level
;
5175 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5176 extent_buffer_get(right_path
->nodes
[right_level
]);
5178 if (left_level
== 0)
5179 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5180 &left_key
, left_path
->slots
[left_level
]);
5182 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5183 &left_key
, left_path
->slots
[left_level
]);
5184 if (right_level
== 0)
5185 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5186 &right_key
, right_path
->slots
[right_level
]);
5188 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5189 &right_key
, right_path
->slots
[right_level
]);
5191 left_end_reached
= right_end_reached
= 0;
5192 advance_left
= advance_right
= 0;
5196 * We need to make sure the transaction does not get committed
5197 * while we do anything on commit roots. This means, we need to
5198 * join and leave transactions for every item that we process.
5200 if (trans
&& btrfs_should_end_transaction(trans
, left_root
)) {
5201 btrfs_release_path(left_path
);
5202 btrfs_release_path(right_path
);
5204 ret
= btrfs_end_transaction(trans
, left_root
);
5209 /* now rejoin the transaction */
5211 trans
= btrfs_join_transaction(left_root
);
5212 if (IS_ERR(trans
)) {
5213 ret
= PTR_ERR(trans
);
5218 spin_lock(&left_root
->root_times_lock
);
5219 ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5220 spin_unlock(&left_root
->root_times_lock
);
5221 if (ctransid
!= left_start_ctransid
)
5222 left_start_ctransid
= 0;
5224 spin_lock(&right_root
->root_times_lock
);
5225 ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5226 spin_unlock(&right_root
->root_times_lock
);
5227 if (ctransid
!= right_start_ctransid
)
5228 right_start_ctransid
= 0;
5230 if (!left_start_ctransid
|| !right_start_ctransid
) {
5231 WARN(1, KERN_WARNING
5232 "btrfs: btrfs_compare_tree detected "
5233 "a change in one of the trees while "
5234 "iterating. This is probably a "
5241 * the commit root may have changed, so start again
5244 left_path
->lowest_level
= left_level
;
5245 right_path
->lowest_level
= right_level
;
5246 ret
= btrfs_search_slot(NULL
, left_root
,
5247 &left_key
, left_path
, 0, 0);
5250 ret
= btrfs_search_slot(NULL
, right_root
,
5251 &right_key
, right_path
, 0, 0);
5256 if (advance_left
&& !left_end_reached
) {
5257 ret
= tree_advance(left_root
, left_path
, &left_level
,
5259 advance_left
!= ADVANCE_ONLY_NEXT
,
5262 left_end_reached
= ADVANCE
;
5265 if (advance_right
&& !right_end_reached
) {
5266 ret
= tree_advance(right_root
, right_path
, &right_level
,
5268 advance_right
!= ADVANCE_ONLY_NEXT
,
5271 right_end_reached
= ADVANCE
;
5275 if (left_end_reached
&& right_end_reached
) {
5278 } else if (left_end_reached
) {
5279 if (right_level
== 0) {
5280 ret
= changed_cb(left_root
, right_root
,
5281 left_path
, right_path
,
5283 BTRFS_COMPARE_TREE_DELETED
,
5288 advance_right
= ADVANCE
;
5290 } else if (right_end_reached
) {
5291 if (left_level
== 0) {
5292 ret
= changed_cb(left_root
, right_root
,
5293 left_path
, right_path
,
5295 BTRFS_COMPARE_TREE_NEW
,
5300 advance_left
= ADVANCE
;
5304 if (left_level
== 0 && right_level
== 0) {
5305 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5307 ret
= changed_cb(left_root
, right_root
,
5308 left_path
, right_path
,
5310 BTRFS_COMPARE_TREE_NEW
,
5314 advance_left
= ADVANCE
;
5315 } else if (cmp
> 0) {
5316 ret
= changed_cb(left_root
, right_root
,
5317 left_path
, right_path
,
5319 BTRFS_COMPARE_TREE_DELETED
,
5323 advance_right
= ADVANCE
;
5325 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5326 ret
= tree_compare_item(left_root
, left_path
,
5327 right_path
, tmp_buf
);
5329 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5330 ret
= changed_cb(left_root
, right_root
,
5331 left_path
, right_path
,
5333 BTRFS_COMPARE_TREE_CHANGED
,
5338 advance_left
= ADVANCE
;
5339 advance_right
= ADVANCE
;
5341 } else if (left_level
== right_level
) {
5342 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5344 advance_left
= ADVANCE
;
5345 } else if (cmp
> 0) {
5346 advance_right
= ADVANCE
;
5348 left_blockptr
= btrfs_node_blockptr(
5349 left_path
->nodes
[left_level
],
5350 left_path
->slots
[left_level
]);
5351 right_blockptr
= btrfs_node_blockptr(
5352 right_path
->nodes
[right_level
],
5353 right_path
->slots
[right_level
]);
5354 if (left_blockptr
== right_blockptr
) {
5356 * As we're on a shared block, don't
5357 * allow to go deeper.
5359 advance_left
= ADVANCE_ONLY_NEXT
;
5360 advance_right
= ADVANCE_ONLY_NEXT
;
5362 advance_left
= ADVANCE
;
5363 advance_right
= ADVANCE
;
5366 } else if (left_level
< right_level
) {
5367 advance_right
= ADVANCE
;
5369 advance_left
= ADVANCE
;
5374 btrfs_free_path(left_path
);
5375 btrfs_free_path(right_path
);
5380 ret
= btrfs_end_transaction(trans
, left_root
);
5382 btrfs_end_transaction(trans
, left_root
);
5389 * this is similar to btrfs_next_leaf, but does not try to preserve
5390 * and fixup the path. It looks for and returns the next key in the
5391 * tree based on the current path and the cache_only and min_trans
5394 * 0 is returned if another key is found, < 0 if there are any errors
5395 * and 1 is returned if there are no higher keys in the tree
5397 * path->keep_locks should be set to 1 on the search made before
5398 * calling this function.
5400 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5401 struct btrfs_key
*key
, int level
,
5402 int cache_only
, u64 min_trans
)
5405 struct extent_buffer
*c
;
5407 WARN_ON(!path
->keep_locks
);
5408 while (level
< BTRFS_MAX_LEVEL
) {
5409 if (!path
->nodes
[level
])
5412 slot
= path
->slots
[level
] + 1;
5413 c
= path
->nodes
[level
];
5415 if (slot
>= btrfs_header_nritems(c
)) {
5418 struct btrfs_key cur_key
;
5419 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5420 !path
->nodes
[level
+ 1])
5423 if (path
->locks
[level
+ 1]) {
5428 slot
= btrfs_header_nritems(c
) - 1;
5430 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5432 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5434 orig_lowest
= path
->lowest_level
;
5435 btrfs_release_path(path
);
5436 path
->lowest_level
= level
;
5437 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5439 path
->lowest_level
= orig_lowest
;
5443 c
= path
->nodes
[level
];
5444 slot
= path
->slots
[level
];
5451 btrfs_item_key_to_cpu(c
, key
, slot
);
5453 u64 blockptr
= btrfs_node_blockptr(c
, slot
);
5454 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5457 struct extent_buffer
*cur
;
5458 cur
= btrfs_find_tree_block(root
, blockptr
,
5459 btrfs_level_size(root
, level
- 1));
5461 btrfs_buffer_uptodate(cur
, gen
, 1) <= 0) {
5464 free_extent_buffer(cur
);
5467 free_extent_buffer(cur
);
5469 if (gen
< min_trans
) {
5473 btrfs_node_key_to_cpu(c
, key
, slot
);
5481 * search the tree again to find a leaf with greater keys
5482 * returns 0 if it found something or 1 if there are no greater leaves.
5483 * returns < 0 on io errors.
5485 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5487 return btrfs_next_old_leaf(root
, path
, 0);
5490 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5495 struct extent_buffer
*c
;
5496 struct extent_buffer
*next
;
5497 struct btrfs_key key
;
5500 int old_spinning
= path
->leave_spinning
;
5501 int next_rw_lock
= 0;
5503 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5507 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5512 btrfs_release_path(path
);
5514 path
->keep_locks
= 1;
5515 path
->leave_spinning
= 1;
5518 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5520 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5521 path
->keep_locks
= 0;
5526 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5528 * by releasing the path above we dropped all our locks. A balance
5529 * could have added more items next to the key that used to be
5530 * at the very end of the block. So, check again here and
5531 * advance the path if there are now more items available.
5533 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5540 while (level
< BTRFS_MAX_LEVEL
) {
5541 if (!path
->nodes
[level
]) {
5546 slot
= path
->slots
[level
] + 1;
5547 c
= path
->nodes
[level
];
5548 if (slot
>= btrfs_header_nritems(c
)) {
5550 if (level
== BTRFS_MAX_LEVEL
) {
5558 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5559 free_extent_buffer(next
);
5563 next_rw_lock
= path
->locks
[level
];
5564 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5570 btrfs_release_path(path
);
5574 if (!path
->skip_locking
) {
5575 ret
= btrfs_try_tree_read_lock(next
);
5576 if (!ret
&& time_seq
) {
5578 * If we don't get the lock, we may be racing
5579 * with push_leaf_left, holding that lock while
5580 * itself waiting for the leaf we've currently
5581 * locked. To solve this situation, we give up
5582 * on our lock and cycle.
5584 free_extent_buffer(next
);
5585 btrfs_release_path(path
);
5590 btrfs_set_path_blocking(path
);
5591 btrfs_tree_read_lock(next
);
5592 btrfs_clear_path_blocking(path
, next
,
5595 next_rw_lock
= BTRFS_READ_LOCK
;
5599 path
->slots
[level
] = slot
;
5602 c
= path
->nodes
[level
];
5603 if (path
->locks
[level
])
5604 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5606 free_extent_buffer(c
);
5607 path
->nodes
[level
] = next
;
5608 path
->slots
[level
] = 0;
5609 if (!path
->skip_locking
)
5610 path
->locks
[level
] = next_rw_lock
;
5614 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5620 btrfs_release_path(path
);
5624 if (!path
->skip_locking
) {
5625 ret
= btrfs_try_tree_read_lock(next
);
5627 btrfs_set_path_blocking(path
);
5628 btrfs_tree_read_lock(next
);
5629 btrfs_clear_path_blocking(path
, next
,
5632 next_rw_lock
= BTRFS_READ_LOCK
;
5637 unlock_up(path
, 0, 1, 0, NULL
);
5638 path
->leave_spinning
= old_spinning
;
5640 btrfs_set_path_blocking(path
);
5646 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5647 * searching until it gets past min_objectid or finds an item of 'type'
5649 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5651 int btrfs_previous_item(struct btrfs_root
*root
,
5652 struct btrfs_path
*path
, u64 min_objectid
,
5655 struct btrfs_key found_key
;
5656 struct extent_buffer
*leaf
;
5661 if (path
->slots
[0] == 0) {
5662 btrfs_set_path_blocking(path
);
5663 ret
= btrfs_prev_leaf(root
, path
);
5669 leaf
= path
->nodes
[0];
5670 nritems
= btrfs_header_nritems(leaf
);
5673 if (path
->slots
[0] == nritems
)
5676 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5677 if (found_key
.objectid
< min_objectid
)
5679 if (found_key
.type
== type
)
5681 if (found_key
.objectid
== min_objectid
&&
5682 found_key
.type
< type
)