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>
22 #include <linux/vmalloc.h>
25 #include "transaction.h"
26 #include "print-tree.h"
29 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
30 *root
, struct btrfs_path
*path
, int level
);
31 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
32 *root
, struct btrfs_key
*ins_key
,
33 struct btrfs_path
*path
, int data_size
, int extend
);
34 static int push_node_left(struct btrfs_trans_handle
*trans
,
35 struct btrfs_root
*root
, struct extent_buffer
*dst
,
36 struct extent_buffer
*src
, int empty
);
37 static int balance_node_right(struct btrfs_trans_handle
*trans
,
38 struct btrfs_root
*root
,
39 struct extent_buffer
*dst_buf
,
40 struct extent_buffer
*src_buf
);
41 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
43 static int tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
44 struct extent_buffer
*eb
);
46 struct btrfs_path
*btrfs_alloc_path(void)
48 struct btrfs_path
*path
;
49 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
54 * set all locked nodes in the path to blocking locks. This should
55 * be done before scheduling
57 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
60 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
61 if (!p
->nodes
[i
] || !p
->locks
[i
])
63 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
64 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
65 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
66 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
67 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
72 * reset all the locked nodes in the patch to spinning locks.
74 * held is used to keep lockdep happy, when lockdep is enabled
75 * we set held to a blocking lock before we go around and
76 * retake all the spinlocks in the path. You can safely use NULL
79 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
80 struct extent_buffer
*held
, int held_rw
)
85 btrfs_set_lock_blocking_rw(held
, held_rw
);
86 if (held_rw
== BTRFS_WRITE_LOCK
)
87 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
88 else if (held_rw
== BTRFS_READ_LOCK
)
89 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
91 btrfs_set_path_blocking(p
);
93 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
94 if (p
->nodes
[i
] && p
->locks
[i
]) {
95 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
96 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
97 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
98 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
99 p
->locks
[i
] = BTRFS_READ_LOCK
;
104 btrfs_clear_lock_blocking_rw(held
, held_rw
);
107 /* this also releases the path */
108 void btrfs_free_path(struct btrfs_path
*p
)
112 btrfs_release_path(p
);
113 kmem_cache_free(btrfs_path_cachep
, p
);
117 * path release drops references on the extent buffers in the path
118 * and it drops any locks held by this path
120 * It is safe to call this on paths that no locks or extent buffers held.
122 noinline
void btrfs_release_path(struct btrfs_path
*p
)
126 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
131 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
134 free_extent_buffer(p
->nodes
[i
]);
140 * safely gets a reference on the root node of a tree. A lock
141 * is not taken, so a concurrent writer may put a different node
142 * at the root of the tree. See btrfs_lock_root_node for the
145 * The extent buffer returned by this has a reference taken, so
146 * it won't disappear. It may stop being the root of the tree
147 * at any time because there are no locks held.
149 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
151 struct extent_buffer
*eb
;
155 eb
= rcu_dereference(root
->node
);
158 * RCU really hurts here, we could free up the root node because
159 * it was COWed but we may not get the new root node yet so do
160 * the inc_not_zero dance and if it doesn't work then
161 * synchronize_rcu and try again.
163 if (atomic_inc_not_zero(&eb
->refs
)) {
173 /* loop around taking references on and locking the root node of the
174 * tree until you end up with a lock on the root. A locked buffer
175 * is returned, with a reference held.
177 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
179 struct extent_buffer
*eb
;
182 eb
= btrfs_root_node(root
);
184 if (eb
== root
->node
)
186 btrfs_tree_unlock(eb
);
187 free_extent_buffer(eb
);
192 /* loop around taking references on and locking the root node of the
193 * tree until you end up with a lock on the root. A locked buffer
194 * is returned, with a reference held.
196 static struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
198 struct extent_buffer
*eb
;
201 eb
= btrfs_root_node(root
);
202 btrfs_tree_read_lock(eb
);
203 if (eb
== root
->node
)
205 btrfs_tree_read_unlock(eb
);
206 free_extent_buffer(eb
);
211 /* cowonly root (everything not a reference counted cow subvolume), just get
212 * put onto a simple dirty list. transaction.c walks this to make sure they
213 * get properly updated on disk.
215 static void add_root_to_dirty_list(struct btrfs_root
*root
)
217 if (test_bit(BTRFS_ROOT_DIRTY
, &root
->state
) ||
218 !test_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
))
221 spin_lock(&root
->fs_info
->trans_lock
);
222 if (!test_and_set_bit(BTRFS_ROOT_DIRTY
, &root
->state
)) {
223 /* Want the extent tree to be the last on the list */
224 if (root
->objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
225 list_move_tail(&root
->dirty_list
,
226 &root
->fs_info
->dirty_cowonly_roots
);
228 list_move(&root
->dirty_list
,
229 &root
->fs_info
->dirty_cowonly_roots
);
231 spin_unlock(&root
->fs_info
->trans_lock
);
235 * used by snapshot creation to make a copy of a root for a tree with
236 * a given objectid. The buffer with the new root node is returned in
237 * cow_ret, and this func returns zero on success or a negative error code.
239 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
240 struct btrfs_root
*root
,
241 struct extent_buffer
*buf
,
242 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
244 struct extent_buffer
*cow
;
247 struct btrfs_disk_key disk_key
;
249 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
250 trans
->transid
!= root
->fs_info
->running_transaction
->transid
);
251 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
252 trans
->transid
!= root
->last_trans
);
254 level
= btrfs_header_level(buf
);
256 btrfs_item_key(buf
, &disk_key
, 0);
258 btrfs_node_key(buf
, &disk_key
, 0);
260 cow
= btrfs_alloc_tree_block(trans
, root
, 0, new_root_objectid
,
261 &disk_key
, level
, buf
->start
, 0);
265 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
266 btrfs_set_header_bytenr(cow
, cow
->start
);
267 btrfs_set_header_generation(cow
, trans
->transid
);
268 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
269 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
270 BTRFS_HEADER_FLAG_RELOC
);
271 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
272 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
274 btrfs_set_header_owner(cow
, new_root_objectid
);
276 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
279 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
280 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
281 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
283 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
288 btrfs_mark_buffer_dirty(cow
);
297 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
298 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
300 MOD_LOG_ROOT_REPLACE
,
303 struct tree_mod_move
{
308 struct tree_mod_root
{
313 struct tree_mod_elem
{
319 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
322 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
325 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
326 struct btrfs_disk_key key
;
329 /* this is used for op == MOD_LOG_MOVE_KEYS */
330 struct tree_mod_move move
;
332 /* this is used for op == MOD_LOG_ROOT_REPLACE */
333 struct tree_mod_root old_root
;
336 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
338 read_lock(&fs_info
->tree_mod_log_lock
);
341 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
343 read_unlock(&fs_info
->tree_mod_log_lock
);
346 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
348 write_lock(&fs_info
->tree_mod_log_lock
);
351 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
353 write_unlock(&fs_info
->tree_mod_log_lock
);
357 * Pull a new tree mod seq number for our operation.
359 static inline u64
btrfs_inc_tree_mod_seq(struct btrfs_fs_info
*fs_info
)
361 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
365 * This adds a new blocker to the tree mod log's blocker list if the @elem
366 * passed does not already have a sequence number set. So when a caller expects
367 * to record tree modifications, it should ensure to set elem->seq to zero
368 * before calling btrfs_get_tree_mod_seq.
369 * Returns a fresh, unused tree log modification sequence number, even if no new
372 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
373 struct seq_list
*elem
)
375 tree_mod_log_write_lock(fs_info
);
376 spin_lock(&fs_info
->tree_mod_seq_lock
);
378 elem
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
379 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
381 spin_unlock(&fs_info
->tree_mod_seq_lock
);
382 tree_mod_log_write_unlock(fs_info
);
387 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
388 struct seq_list
*elem
)
390 struct rb_root
*tm_root
;
391 struct rb_node
*node
;
392 struct rb_node
*next
;
393 struct seq_list
*cur_elem
;
394 struct tree_mod_elem
*tm
;
395 u64 min_seq
= (u64
)-1;
396 u64 seq_putting
= elem
->seq
;
401 spin_lock(&fs_info
->tree_mod_seq_lock
);
402 list_del(&elem
->list
);
405 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
406 if (cur_elem
->seq
< min_seq
) {
407 if (seq_putting
> cur_elem
->seq
) {
409 * blocker with lower sequence number exists, we
410 * cannot remove anything from the log
412 spin_unlock(&fs_info
->tree_mod_seq_lock
);
415 min_seq
= cur_elem
->seq
;
418 spin_unlock(&fs_info
->tree_mod_seq_lock
);
421 * anything that's lower than the lowest existing (read: blocked)
422 * sequence number can be removed from the tree.
424 tree_mod_log_write_lock(fs_info
);
425 tm_root
= &fs_info
->tree_mod_log
;
426 for (node
= rb_first(tm_root
); node
; node
= next
) {
427 next
= rb_next(node
);
428 tm
= container_of(node
, struct tree_mod_elem
, node
);
429 if (tm
->seq
> min_seq
)
431 rb_erase(node
, tm_root
);
434 tree_mod_log_write_unlock(fs_info
);
438 * key order of the log:
439 * node/leaf start address -> sequence
441 * The 'start address' is the logical address of the *new* root node
442 * for root replace operations, or the logical address of the affected
443 * block for all other operations.
445 * Note: must be called with write lock (tree_mod_log_write_lock).
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
;
457 tm
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
459 tm_root
= &fs_info
->tree_mod_log
;
460 new = &tm_root
->rb_node
;
462 cur
= container_of(*new, struct tree_mod_elem
, node
);
464 if (cur
->logical
< tm
->logical
)
465 new = &((*new)->rb_left
);
466 else if (cur
->logical
> tm
->logical
)
467 new = &((*new)->rb_right
);
468 else if (cur
->seq
< tm
->seq
)
469 new = &((*new)->rb_left
);
470 else if (cur
->seq
> tm
->seq
)
471 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
)) {
497 tree_mod_log_write_unlock(fs_info
);
504 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
505 static inline int tree_mod_need_log(const struct btrfs_fs_info
*fs_info
,
506 struct extent_buffer
*eb
)
509 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
511 if (eb
&& btrfs_header_level(eb
) == 0)
517 static struct tree_mod_elem
*
518 alloc_tree_mod_elem(struct extent_buffer
*eb
, int slot
,
519 enum mod_log_op op
, gfp_t flags
)
521 struct tree_mod_elem
*tm
;
523 tm
= kzalloc(sizeof(*tm
), flags
);
527 tm
->logical
= eb
->start
;
528 if (op
!= MOD_LOG_KEY_ADD
) {
529 btrfs_node_key(eb
, &tm
->key
, slot
);
530 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
534 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
535 RB_CLEAR_NODE(&tm
->node
);
541 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
542 struct extent_buffer
*eb
, int slot
,
543 enum mod_log_op op
, gfp_t flags
)
545 struct tree_mod_elem
*tm
;
548 if (!tree_mod_need_log(fs_info
, eb
))
551 tm
= alloc_tree_mod_elem(eb
, slot
, op
, flags
);
555 if (tree_mod_dont_log(fs_info
, eb
)) {
560 ret
= __tree_mod_log_insert(fs_info
, tm
);
561 tree_mod_log_write_unlock(fs_info
);
569 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
570 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
571 int nr_items
, gfp_t flags
)
573 struct tree_mod_elem
*tm
= NULL
;
574 struct tree_mod_elem
**tm_list
= NULL
;
579 if (!tree_mod_need_log(fs_info
, eb
))
582 tm_list
= kcalloc(nr_items
, sizeof(struct tree_mod_elem
*), flags
);
586 tm
= kzalloc(sizeof(*tm
), flags
);
592 tm
->logical
= eb
->start
;
594 tm
->move
.dst_slot
= dst_slot
;
595 tm
->move
.nr_items
= nr_items
;
596 tm
->op
= MOD_LOG_MOVE_KEYS
;
598 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
599 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
+ dst_slot
,
600 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, flags
);
607 if (tree_mod_dont_log(fs_info
, eb
))
612 * When we override something during the move, we log these removals.
613 * This can only happen when we move towards the beginning of the
614 * buffer, i.e. dst_slot < src_slot.
616 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
617 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
622 ret
= __tree_mod_log_insert(fs_info
, tm
);
625 tree_mod_log_write_unlock(fs_info
);
630 for (i
= 0; i
< nr_items
; i
++) {
631 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
632 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
636 tree_mod_log_write_unlock(fs_info
);
644 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
645 struct tree_mod_elem
**tm_list
,
651 for (i
= nritems
- 1; i
>= 0; i
--) {
652 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
654 for (j
= nritems
- 1; j
> i
; j
--)
655 rb_erase(&tm_list
[j
]->node
,
656 &fs_info
->tree_mod_log
);
665 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
666 struct extent_buffer
*old_root
,
667 struct extent_buffer
*new_root
, gfp_t flags
,
670 struct tree_mod_elem
*tm
= NULL
;
671 struct tree_mod_elem
**tm_list
= NULL
;
676 if (!tree_mod_need_log(fs_info
, NULL
))
679 if (log_removal
&& btrfs_header_level(old_root
) > 0) {
680 nritems
= btrfs_header_nritems(old_root
);
681 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*),
687 for (i
= 0; i
< nritems
; i
++) {
688 tm_list
[i
] = alloc_tree_mod_elem(old_root
, i
,
689 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, flags
);
697 tm
= kzalloc(sizeof(*tm
), flags
);
703 tm
->logical
= new_root
->start
;
704 tm
->old_root
.logical
= old_root
->start
;
705 tm
->old_root
.level
= btrfs_header_level(old_root
);
706 tm
->generation
= btrfs_header_generation(old_root
);
707 tm
->op
= MOD_LOG_ROOT_REPLACE
;
709 if (tree_mod_dont_log(fs_info
, NULL
))
713 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
715 ret
= __tree_mod_log_insert(fs_info
, tm
);
717 tree_mod_log_write_unlock(fs_info
);
726 for (i
= 0; i
< nritems
; i
++)
735 static struct tree_mod_elem
*
736 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
739 struct rb_root
*tm_root
;
740 struct rb_node
*node
;
741 struct tree_mod_elem
*cur
= NULL
;
742 struct tree_mod_elem
*found
= NULL
;
744 tree_mod_log_read_lock(fs_info
);
745 tm_root
= &fs_info
->tree_mod_log
;
746 node
= tm_root
->rb_node
;
748 cur
= container_of(node
, struct tree_mod_elem
, node
);
749 if (cur
->logical
< start
) {
750 node
= node
->rb_left
;
751 } else if (cur
->logical
> start
) {
752 node
= node
->rb_right
;
753 } else if (cur
->seq
< min_seq
) {
754 node
= node
->rb_left
;
755 } else if (!smallest
) {
756 /* we want the node with the highest seq */
758 BUG_ON(found
->seq
> cur
->seq
);
760 node
= node
->rb_left
;
761 } else if (cur
->seq
> min_seq
) {
762 /* we want the node with the smallest seq */
764 BUG_ON(found
->seq
< cur
->seq
);
766 node
= node
->rb_right
;
772 tree_mod_log_read_unlock(fs_info
);
778 * this returns the element from the log with the smallest time sequence
779 * value that's in the log (the oldest log item). any element with a time
780 * sequence lower than min_seq will be ignored.
782 static struct tree_mod_elem
*
783 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
786 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
790 * this returns the element from the log with the largest time sequence
791 * value that's in the log (the most recent log item). any element with
792 * a time sequence lower than min_seq will be ignored.
794 static struct tree_mod_elem
*
795 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
797 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
801 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
802 struct extent_buffer
*src
, unsigned long dst_offset
,
803 unsigned long src_offset
, int nr_items
)
806 struct tree_mod_elem
**tm_list
= NULL
;
807 struct tree_mod_elem
**tm_list_add
, **tm_list_rem
;
811 if (!tree_mod_need_log(fs_info
, NULL
))
814 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
817 tm_list
= kcalloc(nr_items
* 2, sizeof(struct tree_mod_elem
*),
822 tm_list_add
= tm_list
;
823 tm_list_rem
= tm_list
+ nr_items
;
824 for (i
= 0; i
< nr_items
; i
++) {
825 tm_list_rem
[i
] = alloc_tree_mod_elem(src
, i
+ src_offset
,
826 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
827 if (!tm_list_rem
[i
]) {
832 tm_list_add
[i
] = alloc_tree_mod_elem(dst
, i
+ dst_offset
,
833 MOD_LOG_KEY_ADD
, GFP_NOFS
);
834 if (!tm_list_add
[i
]) {
840 if (tree_mod_dont_log(fs_info
, NULL
))
844 for (i
= 0; i
< nr_items
; i
++) {
845 ret
= __tree_mod_log_insert(fs_info
, tm_list_rem
[i
]);
848 ret
= __tree_mod_log_insert(fs_info
, tm_list_add
[i
]);
853 tree_mod_log_write_unlock(fs_info
);
859 for (i
= 0; i
< nr_items
* 2; i
++) {
860 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
861 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
865 tree_mod_log_write_unlock(fs_info
);
872 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
873 int dst_offset
, int src_offset
, int nr_items
)
876 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
882 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
883 struct extent_buffer
*eb
, int slot
, int atomic
)
887 ret
= tree_mod_log_insert_key(fs_info
, eb
, slot
,
889 atomic
? GFP_ATOMIC
: GFP_NOFS
);
894 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
896 struct tree_mod_elem
**tm_list
= NULL
;
901 if (btrfs_header_level(eb
) == 0)
904 if (!tree_mod_need_log(fs_info
, NULL
))
907 nritems
= btrfs_header_nritems(eb
);
908 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*), GFP_NOFS
);
912 for (i
= 0; i
< nritems
; i
++) {
913 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
,
914 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
921 if (tree_mod_dont_log(fs_info
, eb
))
924 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
925 tree_mod_log_write_unlock(fs_info
);
933 for (i
= 0; i
< nritems
; i
++)
941 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
942 struct extent_buffer
*new_root_node
,
946 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
947 new_root_node
, GFP_NOFS
, log_removal
);
952 * check if the tree block can be shared by multiple trees
954 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
955 struct extent_buffer
*buf
)
958 * Tree blocks not in reference counted trees and tree roots
959 * are never shared. If a block was allocated after the last
960 * snapshot and the block was not allocated by tree relocation,
961 * we know the block is not shared.
963 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
964 buf
!= root
->node
&& buf
!= root
->commit_root
&&
965 (btrfs_header_generation(buf
) <=
966 btrfs_root_last_snapshot(&root
->root_item
) ||
967 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
969 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
970 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
971 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
977 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
978 struct btrfs_root
*root
,
979 struct extent_buffer
*buf
,
980 struct extent_buffer
*cow
,
990 * Backrefs update rules:
992 * Always use full backrefs for extent pointers in tree block
993 * allocated by tree relocation.
995 * If a shared tree block is no longer referenced by its owner
996 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
997 * use full backrefs for extent pointers in tree block.
999 * If a tree block is been relocating
1000 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1001 * use full backrefs for extent pointers in tree block.
1002 * The reason for this is some operations (such as drop tree)
1003 * are only allowed for blocks use full backrefs.
1006 if (btrfs_block_can_be_shared(root
, buf
)) {
1007 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
1008 btrfs_header_level(buf
), 1,
1014 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
1019 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1020 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1021 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1026 owner
= btrfs_header_owner(buf
);
1027 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
1028 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
1031 if ((owner
== root
->root_key
.objectid
||
1032 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
1033 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
1034 ret
= btrfs_inc_ref(trans
, root
, buf
, 1);
1035 BUG_ON(ret
); /* -ENOMEM */
1037 if (root
->root_key
.objectid
==
1038 BTRFS_TREE_RELOC_OBJECTID
) {
1039 ret
= btrfs_dec_ref(trans
, root
, buf
, 0);
1040 BUG_ON(ret
); /* -ENOMEM */
1041 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1042 BUG_ON(ret
); /* -ENOMEM */
1044 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1047 if (root
->root_key
.objectid
==
1048 BTRFS_TREE_RELOC_OBJECTID
)
1049 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1051 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
1052 BUG_ON(ret
); /* -ENOMEM */
1054 if (new_flags
!= 0) {
1055 int level
= btrfs_header_level(buf
);
1057 ret
= btrfs_set_disk_extent_flags(trans
, root
,
1060 new_flags
, level
, 0);
1065 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
1066 if (root
->root_key
.objectid
==
1067 BTRFS_TREE_RELOC_OBJECTID
)
1068 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
1070 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
1071 BUG_ON(ret
); /* -ENOMEM */
1072 ret
= btrfs_dec_ref(trans
, root
, buf
, 1);
1073 BUG_ON(ret
); /* -ENOMEM */
1075 clean_tree_block(trans
, root
->fs_info
, buf
);
1082 * does the dirty work in cow of a single block. The parent block (if
1083 * supplied) is updated to point to the new cow copy. The new buffer is marked
1084 * dirty and returned locked. If you modify the block it needs to be marked
1087 * search_start -- an allocation hint for the new block
1089 * empty_size -- a hint that you plan on doing more cow. This is the size in
1090 * bytes the allocator should try to find free next to the block it returns.
1091 * This is just a hint and may be ignored by the allocator.
1093 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1094 struct btrfs_root
*root
,
1095 struct extent_buffer
*buf
,
1096 struct extent_buffer
*parent
, int parent_slot
,
1097 struct extent_buffer
**cow_ret
,
1098 u64 search_start
, u64 empty_size
)
1100 struct btrfs_disk_key disk_key
;
1101 struct extent_buffer
*cow
;
1104 int unlock_orig
= 0;
1107 if (*cow_ret
== buf
)
1110 btrfs_assert_tree_locked(buf
);
1112 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
1113 trans
->transid
!= root
->fs_info
->running_transaction
->transid
);
1114 WARN_ON(test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
) &&
1115 trans
->transid
!= root
->last_trans
);
1117 level
= btrfs_header_level(buf
);
1120 btrfs_item_key(buf
, &disk_key
, 0);
1122 btrfs_node_key(buf
, &disk_key
, 0);
1124 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
1126 parent_start
= parent
->start
;
1132 cow
= btrfs_alloc_tree_block(trans
, root
, parent_start
,
1133 root
->root_key
.objectid
, &disk_key
, level
,
1134 search_start
, empty_size
);
1136 return PTR_ERR(cow
);
1138 /* cow is set to blocking by btrfs_init_new_buffer */
1140 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
1141 btrfs_set_header_bytenr(cow
, cow
->start
);
1142 btrfs_set_header_generation(cow
, trans
->transid
);
1143 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1144 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1145 BTRFS_HEADER_FLAG_RELOC
);
1146 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1147 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1149 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1151 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
1154 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1156 btrfs_abort_transaction(trans
, root
, ret
);
1160 if (test_bit(BTRFS_ROOT_REF_COWS
, &root
->state
)) {
1161 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1163 btrfs_abort_transaction(trans
, root
, ret
);
1168 if (buf
== root
->node
) {
1169 WARN_ON(parent
&& parent
!= buf
);
1170 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1171 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1172 parent_start
= buf
->start
;
1176 extent_buffer_get(cow
);
1177 tree_mod_log_set_root_pointer(root
, cow
, 1);
1178 rcu_assign_pointer(root
->node
, cow
);
1180 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1182 free_extent_buffer(buf
);
1183 add_root_to_dirty_list(root
);
1185 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1186 parent_start
= parent
->start
;
1190 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1191 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1192 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1193 btrfs_set_node_blockptr(parent
, parent_slot
,
1195 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1197 btrfs_mark_buffer_dirty(parent
);
1199 ret
= tree_mod_log_free_eb(root
->fs_info
, buf
);
1201 btrfs_abort_transaction(trans
, root
, ret
);
1205 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1209 btrfs_tree_unlock(buf
);
1210 free_extent_buffer_stale(buf
);
1211 btrfs_mark_buffer_dirty(cow
);
1217 * returns the logical address of the oldest predecessor of the given root.
1218 * entries older than time_seq are ignored.
1220 static struct tree_mod_elem
*
1221 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1222 struct extent_buffer
*eb_root
, u64 time_seq
)
1224 struct tree_mod_elem
*tm
;
1225 struct tree_mod_elem
*found
= NULL
;
1226 u64 root_logical
= eb_root
->start
;
1233 * the very last operation that's logged for a root is the
1234 * replacement operation (if it is replaced at all). this has
1235 * the logical address of the *new* root, making it the very
1236 * first operation that's logged for this root.
1239 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1244 * if there are no tree operation for the oldest root, we simply
1245 * return it. this should only happen if that (old) root is at
1252 * if there's an operation that's not a root replacement, we
1253 * found the oldest version of our root. normally, we'll find a
1254 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1256 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1260 root_logical
= tm
->old_root
.logical
;
1264 /* if there's no old root to return, return what we found instead */
1272 * tm is a pointer to the first operation to rewind within eb. then, all
1273 * previous operations will be rewound (until we reach something older than
1277 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1278 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1281 struct rb_node
*next
;
1282 struct tree_mod_elem
*tm
= first_tm
;
1283 unsigned long o_dst
;
1284 unsigned long o_src
;
1285 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1287 n
= btrfs_header_nritems(eb
);
1288 tree_mod_log_read_lock(fs_info
);
1289 while (tm
&& tm
->seq
>= time_seq
) {
1291 * all the operations are recorded with the operator used for
1292 * the modification. as we're going backwards, we do the
1293 * opposite of each operation here.
1296 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1297 BUG_ON(tm
->slot
< n
);
1299 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1300 case MOD_LOG_KEY_REMOVE
:
1301 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1302 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1303 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1307 case MOD_LOG_KEY_REPLACE
:
1308 BUG_ON(tm
->slot
>= n
);
1309 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1310 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1311 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1314 case MOD_LOG_KEY_ADD
:
1315 /* if a move operation is needed it's in the log */
1318 case MOD_LOG_MOVE_KEYS
:
1319 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1320 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1321 memmove_extent_buffer(eb
, o_dst
, o_src
,
1322 tm
->move
.nr_items
* p_size
);
1324 case MOD_LOG_ROOT_REPLACE
:
1326 * this operation is special. for roots, this must be
1327 * handled explicitly before rewinding.
1328 * for non-roots, this operation may exist if the node
1329 * was a root: root A -> child B; then A gets empty and
1330 * B is promoted to the new root. in the mod log, we'll
1331 * have a root-replace operation for B, a tree block
1332 * that is no root. we simply ignore that operation.
1336 next
= rb_next(&tm
->node
);
1339 tm
= container_of(next
, struct tree_mod_elem
, node
);
1340 if (tm
->logical
!= first_tm
->logical
)
1343 tree_mod_log_read_unlock(fs_info
);
1344 btrfs_set_header_nritems(eb
, n
);
1348 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1349 * is returned. If rewind operations happen, a fresh buffer is returned. The
1350 * returned buffer is always read-locked. If the returned buffer is not the
1351 * input buffer, the lock on the input buffer is released and the input buffer
1352 * is freed (its refcount is decremented).
1354 static struct extent_buffer
*
1355 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1356 struct extent_buffer
*eb
, u64 time_seq
)
1358 struct extent_buffer
*eb_rewin
;
1359 struct tree_mod_elem
*tm
;
1364 if (btrfs_header_level(eb
) == 0)
1367 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1371 btrfs_set_path_blocking(path
);
1372 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1374 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1375 BUG_ON(tm
->slot
!= 0);
1376 eb_rewin
= alloc_dummy_extent_buffer(fs_info
, eb
->start
,
1379 btrfs_tree_read_unlock_blocking(eb
);
1380 free_extent_buffer(eb
);
1383 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1384 btrfs_set_header_backref_rev(eb_rewin
,
1385 btrfs_header_backref_rev(eb
));
1386 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1387 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1389 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1391 btrfs_tree_read_unlock_blocking(eb
);
1392 free_extent_buffer(eb
);
1397 btrfs_clear_path_blocking(path
, NULL
, BTRFS_READ_LOCK
);
1398 btrfs_tree_read_unlock_blocking(eb
);
1399 free_extent_buffer(eb
);
1401 extent_buffer_get(eb_rewin
);
1402 btrfs_tree_read_lock(eb_rewin
);
1403 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1404 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1405 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1411 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1412 * value. If there are no changes, the current root->root_node is returned. If
1413 * anything changed in between, there's a fresh buffer allocated on which the
1414 * rewind operations are done. In any case, the returned buffer is read locked.
1415 * Returns NULL on error (with no locks held).
1417 static inline struct extent_buffer
*
1418 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1420 struct tree_mod_elem
*tm
;
1421 struct extent_buffer
*eb
= NULL
;
1422 struct extent_buffer
*eb_root
;
1423 struct extent_buffer
*old
;
1424 struct tree_mod_root
*old_root
= NULL
;
1425 u64 old_generation
= 0;
1428 eb_root
= btrfs_read_lock_root_node(root
);
1429 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1433 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1434 old_root
= &tm
->old_root
;
1435 old_generation
= tm
->generation
;
1436 logical
= old_root
->logical
;
1438 logical
= eb_root
->start
;
1441 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1442 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1443 btrfs_tree_read_unlock(eb_root
);
1444 free_extent_buffer(eb_root
);
1445 old
= read_tree_block(root
, logical
, 0);
1446 if (WARN_ON(IS_ERR(old
) || !extent_buffer_uptodate(old
))) {
1448 free_extent_buffer(old
);
1449 btrfs_warn(root
->fs_info
,
1450 "failed to read tree block %llu from get_old_root", logical
);
1452 eb
= btrfs_clone_extent_buffer(old
);
1453 free_extent_buffer(old
);
1455 } else if (old_root
) {
1456 btrfs_tree_read_unlock(eb_root
);
1457 free_extent_buffer(eb_root
);
1458 eb
= alloc_dummy_extent_buffer(root
->fs_info
, logical
,
1461 btrfs_set_lock_blocking_rw(eb_root
, BTRFS_READ_LOCK
);
1462 eb
= btrfs_clone_extent_buffer(eb_root
);
1463 btrfs_tree_read_unlock_blocking(eb_root
);
1464 free_extent_buffer(eb_root
);
1469 extent_buffer_get(eb
);
1470 btrfs_tree_read_lock(eb
);
1472 btrfs_set_header_bytenr(eb
, eb
->start
);
1473 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1474 btrfs_set_header_owner(eb
, btrfs_header_owner(eb_root
));
1475 btrfs_set_header_level(eb
, old_root
->level
);
1476 btrfs_set_header_generation(eb
, old_generation
);
1479 __tree_mod_log_rewind(root
->fs_info
, eb
, time_seq
, tm
);
1481 WARN_ON(btrfs_header_level(eb
) != 0);
1482 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1487 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1489 struct tree_mod_elem
*tm
;
1491 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1493 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1494 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1495 level
= tm
->old_root
.level
;
1497 level
= btrfs_header_level(eb_root
);
1499 free_extent_buffer(eb_root
);
1504 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1505 struct btrfs_root
*root
,
1506 struct extent_buffer
*buf
)
1508 if (btrfs_test_is_dummy_root(root
))
1511 /* ensure we can see the force_cow */
1515 * We do not need to cow a block if
1516 * 1) this block is not created or changed in this transaction;
1517 * 2) this block does not belong to TREE_RELOC tree;
1518 * 3) the root is not forced COW.
1520 * What is forced COW:
1521 * when we create snapshot during committing the transaction,
1522 * after we've finished coping src root, we must COW the shared
1523 * block to ensure the metadata consistency.
1525 if (btrfs_header_generation(buf
) == trans
->transid
&&
1526 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1527 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1528 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1529 !test_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
))
1535 * cows a single block, see __btrfs_cow_block for the real work.
1536 * This version of it has extra checks so that a block isn't COWed more than
1537 * once per transaction, as long as it hasn't been written yet
1539 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1540 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1541 struct extent_buffer
*parent
, int parent_slot
,
1542 struct extent_buffer
**cow_ret
)
1547 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1548 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1550 root
->fs_info
->running_transaction
->transid
);
1552 if (trans
->transid
!= root
->fs_info
->generation
)
1553 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1554 trans
->transid
, root
->fs_info
->generation
);
1556 if (!should_cow_block(trans
, root
, buf
)) {
1557 trans
->dirty
= true;
1562 search_start
= buf
->start
& ~((u64
)SZ_1G
- 1);
1565 btrfs_set_lock_blocking(parent
);
1566 btrfs_set_lock_blocking(buf
);
1568 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1569 parent_slot
, cow_ret
, search_start
, 0);
1571 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1577 * helper function for defrag to decide if two blocks pointed to by a
1578 * node are actually close by
1580 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1582 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1584 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1590 * compare two keys in a memcmp fashion
1592 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1594 struct btrfs_key k1
;
1596 btrfs_disk_key_to_cpu(&k1
, disk
);
1598 return btrfs_comp_cpu_keys(&k1
, k2
);
1602 * same as comp_keys only with two btrfs_key's
1604 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1606 if (k1
->objectid
> k2
->objectid
)
1608 if (k1
->objectid
< k2
->objectid
)
1610 if (k1
->type
> k2
->type
)
1612 if (k1
->type
< k2
->type
)
1614 if (k1
->offset
> k2
->offset
)
1616 if (k1
->offset
< k2
->offset
)
1622 * this is used by the defrag code to go through all the
1623 * leaves pointed to by a node and reallocate them so that
1624 * disk order is close to key order
1626 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1627 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1628 int start_slot
, u64
*last_ret
,
1629 struct btrfs_key
*progress
)
1631 struct extent_buffer
*cur
;
1634 u64 search_start
= *last_ret
;
1644 int progress_passed
= 0;
1645 struct btrfs_disk_key disk_key
;
1647 parent_level
= btrfs_header_level(parent
);
1649 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1650 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1652 parent_nritems
= btrfs_header_nritems(parent
);
1653 blocksize
= root
->nodesize
;
1654 end_slot
= parent_nritems
- 1;
1656 if (parent_nritems
<= 1)
1659 btrfs_set_lock_blocking(parent
);
1661 for (i
= start_slot
; i
<= end_slot
; i
++) {
1664 btrfs_node_key(parent
, &disk_key
, i
);
1665 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1668 progress_passed
= 1;
1669 blocknr
= btrfs_node_blockptr(parent
, i
);
1670 gen
= btrfs_node_ptr_generation(parent
, i
);
1671 if (last_block
== 0)
1672 last_block
= blocknr
;
1675 other
= btrfs_node_blockptr(parent
, i
- 1);
1676 close
= close_blocks(blocknr
, other
, blocksize
);
1678 if (!close
&& i
< end_slot
) {
1679 other
= btrfs_node_blockptr(parent
, i
+ 1);
1680 close
= close_blocks(blocknr
, other
, blocksize
);
1683 last_block
= blocknr
;
1687 cur
= btrfs_find_tree_block(root
->fs_info
, blocknr
);
1689 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1692 if (!cur
|| !uptodate
) {
1694 cur
= read_tree_block(root
, blocknr
, gen
);
1696 return PTR_ERR(cur
);
1697 } else if (!extent_buffer_uptodate(cur
)) {
1698 free_extent_buffer(cur
);
1701 } else if (!uptodate
) {
1702 err
= btrfs_read_buffer(cur
, gen
);
1704 free_extent_buffer(cur
);
1709 if (search_start
== 0)
1710 search_start
= last_block
;
1712 btrfs_tree_lock(cur
);
1713 btrfs_set_lock_blocking(cur
);
1714 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1717 (end_slot
- i
) * blocksize
));
1719 btrfs_tree_unlock(cur
);
1720 free_extent_buffer(cur
);
1723 search_start
= cur
->start
;
1724 last_block
= cur
->start
;
1725 *last_ret
= search_start
;
1726 btrfs_tree_unlock(cur
);
1727 free_extent_buffer(cur
);
1733 * The leaf data grows from end-to-front in the node.
1734 * this returns the address of the start of the last item,
1735 * which is the stop of the leaf data stack
1737 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1738 struct extent_buffer
*leaf
)
1740 u32 nr
= btrfs_header_nritems(leaf
);
1742 return BTRFS_LEAF_DATA_SIZE(root
);
1743 return btrfs_item_offset_nr(leaf
, nr
- 1);
1748 * search for key in the extent_buffer. The items start at offset p,
1749 * and they are item_size apart. There are 'max' items in p.
1751 * the slot in the array is returned via slot, and it points to
1752 * the place where you would insert key if it is not found in
1755 * slot may point to max if the key is bigger than all of the keys
1757 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1759 int item_size
, struct btrfs_key
*key
,
1766 struct btrfs_disk_key
*tmp
= NULL
;
1767 struct btrfs_disk_key unaligned
;
1768 unsigned long offset
;
1770 unsigned long map_start
= 0;
1771 unsigned long map_len
= 0;
1775 btrfs_err(eb
->fs_info
,
1776 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1777 __func__
, low
, high
, eb
->start
,
1778 btrfs_header_owner(eb
), btrfs_header_level(eb
));
1782 while (low
< high
) {
1783 mid
= (low
+ high
) / 2;
1784 offset
= p
+ mid
* item_size
;
1786 if (!kaddr
|| offset
< map_start
||
1787 (offset
+ sizeof(struct btrfs_disk_key
)) >
1788 map_start
+ map_len
) {
1790 err
= map_private_extent_buffer(eb
, offset
,
1791 sizeof(struct btrfs_disk_key
),
1792 &kaddr
, &map_start
, &map_len
);
1795 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1797 } else if (err
== 1) {
1798 read_extent_buffer(eb
, &unaligned
,
1799 offset
, sizeof(unaligned
));
1806 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1809 ret
= comp_keys(tmp
, key
);
1825 * simple bin_search frontend that does the right thing for
1828 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1829 int level
, int *slot
)
1832 return generic_bin_search(eb
,
1833 offsetof(struct btrfs_leaf
, items
),
1834 sizeof(struct btrfs_item
),
1835 key
, btrfs_header_nritems(eb
),
1838 return generic_bin_search(eb
,
1839 offsetof(struct btrfs_node
, ptrs
),
1840 sizeof(struct btrfs_key_ptr
),
1841 key
, btrfs_header_nritems(eb
),
1845 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1846 int level
, int *slot
)
1848 return bin_search(eb
, key
, level
, slot
);
1851 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1853 spin_lock(&root
->accounting_lock
);
1854 btrfs_set_root_used(&root
->root_item
,
1855 btrfs_root_used(&root
->root_item
) + size
);
1856 spin_unlock(&root
->accounting_lock
);
1859 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1861 spin_lock(&root
->accounting_lock
);
1862 btrfs_set_root_used(&root
->root_item
,
1863 btrfs_root_used(&root
->root_item
) - size
);
1864 spin_unlock(&root
->accounting_lock
);
1867 /* given a node and slot number, this reads the blocks it points to. The
1868 * extent buffer is returned with a reference taken (but unlocked).
1869 * NULL is returned on error.
1871 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1872 struct extent_buffer
*parent
, int slot
)
1874 int level
= btrfs_header_level(parent
);
1875 struct extent_buffer
*eb
;
1879 if (slot
>= btrfs_header_nritems(parent
))
1884 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1885 btrfs_node_ptr_generation(parent
, slot
));
1886 if (IS_ERR(eb
) || !extent_buffer_uptodate(eb
)) {
1888 free_extent_buffer(eb
);
1896 * node level balancing, used to make sure nodes are in proper order for
1897 * item deletion. We balance from the top down, so we have to make sure
1898 * that a deletion won't leave an node completely empty later on.
1900 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1901 struct btrfs_root
*root
,
1902 struct btrfs_path
*path
, int level
)
1904 struct extent_buffer
*right
= NULL
;
1905 struct extent_buffer
*mid
;
1906 struct extent_buffer
*left
= NULL
;
1907 struct extent_buffer
*parent
= NULL
;
1911 int orig_slot
= path
->slots
[level
];
1917 mid
= path
->nodes
[level
];
1919 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1920 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1921 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1923 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1925 if (level
< BTRFS_MAX_LEVEL
- 1) {
1926 parent
= path
->nodes
[level
+ 1];
1927 pslot
= path
->slots
[level
+ 1];
1931 * deal with the case where there is only one pointer in the root
1932 * by promoting the node below to a root
1935 struct extent_buffer
*child
;
1937 if (btrfs_header_nritems(mid
) != 1)
1940 /* promote the child to a root */
1941 child
= read_node_slot(root
, mid
, 0);
1944 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
1948 btrfs_tree_lock(child
);
1949 btrfs_set_lock_blocking(child
);
1950 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1952 btrfs_tree_unlock(child
);
1953 free_extent_buffer(child
);
1957 tree_mod_log_set_root_pointer(root
, child
, 1);
1958 rcu_assign_pointer(root
->node
, child
);
1960 add_root_to_dirty_list(root
);
1961 btrfs_tree_unlock(child
);
1963 path
->locks
[level
] = 0;
1964 path
->nodes
[level
] = NULL
;
1965 clean_tree_block(trans
, root
->fs_info
, mid
);
1966 btrfs_tree_unlock(mid
);
1967 /* once for the path */
1968 free_extent_buffer(mid
);
1970 root_sub_used(root
, mid
->len
);
1971 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1972 /* once for the root ptr */
1973 free_extent_buffer_stale(mid
);
1976 if (btrfs_header_nritems(mid
) >
1977 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1980 left
= read_node_slot(root
, parent
, pslot
- 1);
1982 btrfs_tree_lock(left
);
1983 btrfs_set_lock_blocking(left
);
1984 wret
= btrfs_cow_block(trans
, root
, left
,
1985 parent
, pslot
- 1, &left
);
1991 right
= read_node_slot(root
, parent
, pslot
+ 1);
1993 btrfs_tree_lock(right
);
1994 btrfs_set_lock_blocking(right
);
1995 wret
= btrfs_cow_block(trans
, root
, right
,
1996 parent
, pslot
+ 1, &right
);
2003 /* first, try to make some room in the middle buffer */
2005 orig_slot
+= btrfs_header_nritems(left
);
2006 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2012 * then try to empty the right most buffer into the middle
2015 wret
= push_node_left(trans
, root
, mid
, right
, 1);
2016 if (wret
< 0 && wret
!= -ENOSPC
)
2018 if (btrfs_header_nritems(right
) == 0) {
2019 clean_tree_block(trans
, root
->fs_info
, right
);
2020 btrfs_tree_unlock(right
);
2021 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
2022 root_sub_used(root
, right
->len
);
2023 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
2024 free_extent_buffer_stale(right
);
2027 struct btrfs_disk_key right_key
;
2028 btrfs_node_key(right
, &right_key
, 0);
2029 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2031 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
2032 btrfs_mark_buffer_dirty(parent
);
2035 if (btrfs_header_nritems(mid
) == 1) {
2037 * we're not allowed to leave a node with one item in the
2038 * tree during a delete. A deletion from lower in the tree
2039 * could try to delete the only pointer in this node.
2040 * So, pull some keys from the left.
2041 * There has to be a left pointer at this point because
2042 * otherwise we would have pulled some pointers from the
2047 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2050 wret
= balance_node_right(trans
, root
, mid
, left
);
2056 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2062 if (btrfs_header_nritems(mid
) == 0) {
2063 clean_tree_block(trans
, root
->fs_info
, mid
);
2064 btrfs_tree_unlock(mid
);
2065 del_ptr(root
, path
, level
+ 1, pslot
);
2066 root_sub_used(root
, mid
->len
);
2067 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
2068 free_extent_buffer_stale(mid
);
2071 /* update the parent key to reflect our changes */
2072 struct btrfs_disk_key mid_key
;
2073 btrfs_node_key(mid
, &mid_key
, 0);
2074 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2076 btrfs_set_node_key(parent
, &mid_key
, pslot
);
2077 btrfs_mark_buffer_dirty(parent
);
2080 /* update the path */
2082 if (btrfs_header_nritems(left
) > orig_slot
) {
2083 extent_buffer_get(left
);
2084 /* left was locked after cow */
2085 path
->nodes
[level
] = left
;
2086 path
->slots
[level
+ 1] -= 1;
2087 path
->slots
[level
] = orig_slot
;
2089 btrfs_tree_unlock(mid
);
2090 free_extent_buffer(mid
);
2093 orig_slot
-= btrfs_header_nritems(left
);
2094 path
->slots
[level
] = orig_slot
;
2097 /* double check we haven't messed things up */
2099 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2103 btrfs_tree_unlock(right
);
2104 free_extent_buffer(right
);
2107 if (path
->nodes
[level
] != left
)
2108 btrfs_tree_unlock(left
);
2109 free_extent_buffer(left
);
2114 /* Node balancing for insertion. Here we only split or push nodes around
2115 * when they are completely full. This is also done top down, so we
2116 * have to be pessimistic.
2118 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2119 struct btrfs_root
*root
,
2120 struct btrfs_path
*path
, int level
)
2122 struct extent_buffer
*right
= NULL
;
2123 struct extent_buffer
*mid
;
2124 struct extent_buffer
*left
= NULL
;
2125 struct extent_buffer
*parent
= NULL
;
2129 int orig_slot
= path
->slots
[level
];
2134 mid
= path
->nodes
[level
];
2135 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2137 if (level
< BTRFS_MAX_LEVEL
- 1) {
2138 parent
= path
->nodes
[level
+ 1];
2139 pslot
= path
->slots
[level
+ 1];
2145 left
= read_node_slot(root
, parent
, pslot
- 1);
2147 /* first, try to make some room in the middle buffer */
2151 btrfs_tree_lock(left
);
2152 btrfs_set_lock_blocking(left
);
2154 left_nr
= btrfs_header_nritems(left
);
2155 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2158 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2163 wret
= push_node_left(trans
, root
,
2170 struct btrfs_disk_key disk_key
;
2171 orig_slot
+= left_nr
;
2172 btrfs_node_key(mid
, &disk_key
, 0);
2173 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2175 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2176 btrfs_mark_buffer_dirty(parent
);
2177 if (btrfs_header_nritems(left
) > orig_slot
) {
2178 path
->nodes
[level
] = left
;
2179 path
->slots
[level
+ 1] -= 1;
2180 path
->slots
[level
] = orig_slot
;
2181 btrfs_tree_unlock(mid
);
2182 free_extent_buffer(mid
);
2185 btrfs_header_nritems(left
);
2186 path
->slots
[level
] = orig_slot
;
2187 btrfs_tree_unlock(left
);
2188 free_extent_buffer(left
);
2192 btrfs_tree_unlock(left
);
2193 free_extent_buffer(left
);
2195 right
= read_node_slot(root
, parent
, pslot
+ 1);
2198 * then try to empty the right most buffer into the middle
2203 btrfs_tree_lock(right
);
2204 btrfs_set_lock_blocking(right
);
2206 right_nr
= btrfs_header_nritems(right
);
2207 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2210 ret
= btrfs_cow_block(trans
, root
, right
,
2216 wret
= balance_node_right(trans
, root
,
2223 struct btrfs_disk_key disk_key
;
2225 btrfs_node_key(right
, &disk_key
, 0);
2226 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2228 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2229 btrfs_mark_buffer_dirty(parent
);
2231 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2232 path
->nodes
[level
] = right
;
2233 path
->slots
[level
+ 1] += 1;
2234 path
->slots
[level
] = orig_slot
-
2235 btrfs_header_nritems(mid
);
2236 btrfs_tree_unlock(mid
);
2237 free_extent_buffer(mid
);
2239 btrfs_tree_unlock(right
);
2240 free_extent_buffer(right
);
2244 btrfs_tree_unlock(right
);
2245 free_extent_buffer(right
);
2251 * readahead one full node of leaves, finding things that are close
2252 * to the block in 'slot', and triggering ra on them.
2254 static void reada_for_search(struct btrfs_root
*root
,
2255 struct btrfs_path
*path
,
2256 int level
, int slot
, u64 objectid
)
2258 struct extent_buffer
*node
;
2259 struct btrfs_disk_key disk_key
;
2265 struct extent_buffer
*eb
;
2273 if (!path
->nodes
[level
])
2276 node
= path
->nodes
[level
];
2278 search
= btrfs_node_blockptr(node
, slot
);
2279 blocksize
= root
->nodesize
;
2280 eb
= btrfs_find_tree_block(root
->fs_info
, search
);
2282 free_extent_buffer(eb
);
2288 nritems
= btrfs_header_nritems(node
);
2292 if (path
->reada
== READA_BACK
) {
2296 } else if (path
->reada
== READA_FORWARD
) {
2301 if (path
->reada
== READA_BACK
&& objectid
) {
2302 btrfs_node_key(node
, &disk_key
, nr
);
2303 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2306 search
= btrfs_node_blockptr(node
, nr
);
2307 if ((search
<= target
&& target
- search
<= 65536) ||
2308 (search
> target
&& search
- target
<= 65536)) {
2309 gen
= btrfs_node_ptr_generation(node
, nr
);
2310 readahead_tree_block(root
, search
);
2314 if ((nread
> 65536 || nscan
> 32))
2319 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2320 struct btrfs_path
*path
, int level
)
2324 struct extent_buffer
*parent
;
2325 struct extent_buffer
*eb
;
2330 parent
= path
->nodes
[level
+ 1];
2334 nritems
= btrfs_header_nritems(parent
);
2335 slot
= path
->slots
[level
+ 1];
2338 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2339 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2340 eb
= btrfs_find_tree_block(root
->fs_info
, block1
);
2342 * if we get -eagain from btrfs_buffer_uptodate, we
2343 * don't want to return eagain here. That will loop
2346 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2348 free_extent_buffer(eb
);
2350 if (slot
+ 1 < nritems
) {
2351 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2352 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2353 eb
= btrfs_find_tree_block(root
->fs_info
, block2
);
2354 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2356 free_extent_buffer(eb
);
2360 readahead_tree_block(root
, block1
);
2362 readahead_tree_block(root
, block2
);
2367 * when we walk down the tree, it is usually safe to unlock the higher layers
2368 * in the tree. The exceptions are when our path goes through slot 0, because
2369 * operations on the tree might require changing key pointers higher up in the
2372 * callers might also have set path->keep_locks, which tells this code to keep
2373 * the lock if the path points to the last slot in the block. This is part of
2374 * walking through the tree, and selecting the next slot in the higher block.
2376 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2377 * if lowest_unlock is 1, level 0 won't be unlocked
2379 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2380 int lowest_unlock
, int min_write_lock_level
,
2381 int *write_lock_level
)
2384 int skip_level
= level
;
2386 struct extent_buffer
*t
;
2388 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2389 if (!path
->nodes
[i
])
2391 if (!path
->locks
[i
])
2393 if (!no_skips
&& path
->slots
[i
] == 0) {
2397 if (!no_skips
&& path
->keep_locks
) {
2400 nritems
= btrfs_header_nritems(t
);
2401 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2406 if (skip_level
< i
&& i
>= lowest_unlock
)
2410 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2411 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2413 if (write_lock_level
&&
2414 i
> min_write_lock_level
&&
2415 i
<= *write_lock_level
) {
2416 *write_lock_level
= i
- 1;
2423 * This releases any locks held in the path starting at level and
2424 * going all the way up to the root.
2426 * btrfs_search_slot will keep the lock held on higher nodes in a few
2427 * corner cases, such as COW of the block at slot zero in the node. This
2428 * ignores those rules, and it should only be called when there are no
2429 * more updates to be done higher up in the tree.
2431 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2435 if (path
->keep_locks
)
2438 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2439 if (!path
->nodes
[i
])
2441 if (!path
->locks
[i
])
2443 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2449 * helper function for btrfs_search_slot. The goal is to find a block
2450 * in cache without setting the path to blocking. If we find the block
2451 * we return zero and the path is unchanged.
2453 * If we can't find the block, we set the path blocking and do some
2454 * reada. -EAGAIN is returned and the search must be repeated.
2457 read_block_for_search(struct btrfs_trans_handle
*trans
,
2458 struct btrfs_root
*root
, struct btrfs_path
*p
,
2459 struct extent_buffer
**eb_ret
, int level
, int slot
,
2460 struct btrfs_key
*key
, u64 time_seq
)
2464 struct extent_buffer
*b
= *eb_ret
;
2465 struct extent_buffer
*tmp
;
2468 blocknr
= btrfs_node_blockptr(b
, slot
);
2469 gen
= btrfs_node_ptr_generation(b
, slot
);
2471 tmp
= btrfs_find_tree_block(root
->fs_info
, blocknr
);
2473 /* first we do an atomic uptodate check */
2474 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2479 /* the pages were up to date, but we failed
2480 * the generation number check. Do a full
2481 * read for the generation number that is correct.
2482 * We must do this without dropping locks so
2483 * we can trust our generation number
2485 btrfs_set_path_blocking(p
);
2487 /* now we're allowed to do a blocking uptodate check */
2488 ret
= btrfs_read_buffer(tmp
, gen
);
2493 free_extent_buffer(tmp
);
2494 btrfs_release_path(p
);
2499 * reduce lock contention at high levels
2500 * of the btree by dropping locks before
2501 * we read. Don't release the lock on the current
2502 * level because we need to walk this node to figure
2503 * out which blocks to read.
2505 btrfs_unlock_up_safe(p
, level
+ 1);
2506 btrfs_set_path_blocking(p
);
2508 free_extent_buffer(tmp
);
2509 if (p
->reada
!= READA_NONE
)
2510 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2512 btrfs_release_path(p
);
2515 tmp
= read_tree_block(root
, blocknr
, 0);
2518 * If the read above didn't mark this buffer up to date,
2519 * it will never end up being up to date. Set ret to EIO now
2520 * and give up so that our caller doesn't loop forever
2523 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2525 free_extent_buffer(tmp
);
2533 * helper function for btrfs_search_slot. This does all of the checks
2534 * for node-level blocks and does any balancing required based on
2537 * If no extra work was required, zero is returned. If we had to
2538 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2542 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2543 struct btrfs_root
*root
, struct btrfs_path
*p
,
2544 struct extent_buffer
*b
, int level
, int ins_len
,
2545 int *write_lock_level
)
2548 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2549 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2552 if (*write_lock_level
< level
+ 1) {
2553 *write_lock_level
= level
+ 1;
2554 btrfs_release_path(p
);
2558 btrfs_set_path_blocking(p
);
2559 reada_for_balance(root
, p
, level
);
2560 sret
= split_node(trans
, root
, p
, level
);
2561 btrfs_clear_path_blocking(p
, NULL
, 0);
2568 b
= p
->nodes
[level
];
2569 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2570 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2573 if (*write_lock_level
< level
+ 1) {
2574 *write_lock_level
= level
+ 1;
2575 btrfs_release_path(p
);
2579 btrfs_set_path_blocking(p
);
2580 reada_for_balance(root
, p
, level
);
2581 sret
= balance_level(trans
, root
, p
, level
);
2582 btrfs_clear_path_blocking(p
, NULL
, 0);
2588 b
= p
->nodes
[level
];
2590 btrfs_release_path(p
);
2593 BUG_ON(btrfs_header_nritems(b
) == 1);
2603 static void key_search_validate(struct extent_buffer
*b
,
2604 struct btrfs_key
*key
,
2607 #ifdef CONFIG_BTRFS_ASSERT
2608 struct btrfs_disk_key disk_key
;
2610 btrfs_cpu_key_to_disk(&disk_key
, key
);
2613 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2614 offsetof(struct btrfs_leaf
, items
[0].key
),
2617 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2618 offsetof(struct btrfs_node
, ptrs
[0].key
),
2623 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2624 int level
, int *prev_cmp
, int *slot
)
2626 if (*prev_cmp
!= 0) {
2627 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2631 key_search_validate(b
, key
, level
);
2637 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*path
,
2638 u64 iobjectid
, u64 ioff
, u8 key_type
,
2639 struct btrfs_key
*found_key
)
2642 struct btrfs_key key
;
2643 struct extent_buffer
*eb
;
2648 key
.type
= key_type
;
2649 key
.objectid
= iobjectid
;
2652 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2656 eb
= path
->nodes
[0];
2657 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2658 ret
= btrfs_next_leaf(fs_root
, path
);
2661 eb
= path
->nodes
[0];
2664 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2665 if (found_key
->type
!= key
.type
||
2666 found_key
->objectid
!= key
.objectid
)
2673 * look for key in the tree. path is filled in with nodes along the way
2674 * if key is found, we return zero and you can find the item in the leaf
2675 * level of the path (level 0)
2677 * If the key isn't found, the path points to the slot where it should
2678 * be inserted, and 1 is returned. If there are other errors during the
2679 * search a negative error number is returned.
2681 * if ins_len > 0, nodes and leaves will be split as we walk down the
2682 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2685 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2686 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2689 struct extent_buffer
*b
;
2694 int lowest_unlock
= 1;
2696 /* everything at write_lock_level or lower must be write locked */
2697 int write_lock_level
= 0;
2698 u8 lowest_level
= 0;
2699 int min_write_lock_level
;
2702 lowest_level
= p
->lowest_level
;
2703 WARN_ON(lowest_level
&& ins_len
> 0);
2704 WARN_ON(p
->nodes
[0] != NULL
);
2705 BUG_ON(!cow
&& ins_len
);
2710 /* when we are removing items, we might have to go up to level
2711 * two as we update tree pointers Make sure we keep write
2712 * for those levels as well
2714 write_lock_level
= 2;
2715 } else if (ins_len
> 0) {
2717 * for inserting items, make sure we have a write lock on
2718 * level 1 so we can update keys
2720 write_lock_level
= 1;
2724 write_lock_level
= -1;
2726 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2727 write_lock_level
= BTRFS_MAX_LEVEL
;
2729 min_write_lock_level
= write_lock_level
;
2734 * we try very hard to do read locks on the root
2736 root_lock
= BTRFS_READ_LOCK
;
2738 if (p
->search_commit_root
) {
2740 * the commit roots are read only
2741 * so we always do read locks
2743 if (p
->need_commit_sem
)
2744 down_read(&root
->fs_info
->commit_root_sem
);
2745 b
= root
->commit_root
;
2746 extent_buffer_get(b
);
2747 level
= btrfs_header_level(b
);
2748 if (p
->need_commit_sem
)
2749 up_read(&root
->fs_info
->commit_root_sem
);
2750 if (!p
->skip_locking
)
2751 btrfs_tree_read_lock(b
);
2753 if (p
->skip_locking
) {
2754 b
= btrfs_root_node(root
);
2755 level
= btrfs_header_level(b
);
2757 /* we don't know the level of the root node
2758 * until we actually have it read locked
2760 b
= btrfs_read_lock_root_node(root
);
2761 level
= btrfs_header_level(b
);
2762 if (level
<= write_lock_level
) {
2763 /* whoops, must trade for write lock */
2764 btrfs_tree_read_unlock(b
);
2765 free_extent_buffer(b
);
2766 b
= btrfs_lock_root_node(root
);
2767 root_lock
= BTRFS_WRITE_LOCK
;
2769 /* the level might have changed, check again */
2770 level
= btrfs_header_level(b
);
2774 p
->nodes
[level
] = b
;
2775 if (!p
->skip_locking
)
2776 p
->locks
[level
] = root_lock
;
2779 level
= btrfs_header_level(b
);
2782 * setup the path here so we can release it under lock
2783 * contention with the cow code
2787 * if we don't really need to cow this block
2788 * then we don't want to set the path blocking,
2789 * so we test it here
2791 if (!should_cow_block(trans
, root
, b
)) {
2792 trans
->dirty
= true;
2797 * must have write locks on this node and the
2800 if (level
> write_lock_level
||
2801 (level
+ 1 > write_lock_level
&&
2802 level
+ 1 < BTRFS_MAX_LEVEL
&&
2803 p
->nodes
[level
+ 1])) {
2804 write_lock_level
= level
+ 1;
2805 btrfs_release_path(p
);
2809 btrfs_set_path_blocking(p
);
2810 err
= btrfs_cow_block(trans
, root
, b
,
2811 p
->nodes
[level
+ 1],
2812 p
->slots
[level
+ 1], &b
);
2819 p
->nodes
[level
] = b
;
2820 btrfs_clear_path_blocking(p
, NULL
, 0);
2823 * we have a lock on b and as long as we aren't changing
2824 * the tree, there is no way to for the items in b to change.
2825 * It is safe to drop the lock on our parent before we
2826 * go through the expensive btree search on b.
2828 * If we're inserting or deleting (ins_len != 0), then we might
2829 * be changing slot zero, which may require changing the parent.
2830 * So, we can't drop the lock until after we know which slot
2831 * we're operating on.
2833 if (!ins_len
&& !p
->keep_locks
) {
2836 if (u
< BTRFS_MAX_LEVEL
&& p
->locks
[u
]) {
2837 btrfs_tree_unlock_rw(p
->nodes
[u
], p
->locks
[u
]);
2842 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2848 if (ret
&& slot
> 0) {
2852 p
->slots
[level
] = slot
;
2853 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2854 ins_len
, &write_lock_level
);
2861 b
= p
->nodes
[level
];
2862 slot
= p
->slots
[level
];
2865 * slot 0 is special, if we change the key
2866 * we have to update the parent pointer
2867 * which means we must have a write lock
2870 if (slot
== 0 && ins_len
&&
2871 write_lock_level
< level
+ 1) {
2872 write_lock_level
= level
+ 1;
2873 btrfs_release_path(p
);
2877 unlock_up(p
, level
, lowest_unlock
,
2878 min_write_lock_level
, &write_lock_level
);
2880 if (level
== lowest_level
) {
2886 err
= read_block_for_search(trans
, root
, p
,
2887 &b
, level
, slot
, key
, 0);
2895 if (!p
->skip_locking
) {
2896 level
= btrfs_header_level(b
);
2897 if (level
<= write_lock_level
) {
2898 err
= btrfs_try_tree_write_lock(b
);
2900 btrfs_set_path_blocking(p
);
2902 btrfs_clear_path_blocking(p
, b
,
2905 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2907 err
= btrfs_tree_read_lock_atomic(b
);
2909 btrfs_set_path_blocking(p
);
2910 btrfs_tree_read_lock(b
);
2911 btrfs_clear_path_blocking(p
, b
,
2914 p
->locks
[level
] = BTRFS_READ_LOCK
;
2916 p
->nodes
[level
] = b
;
2919 p
->slots
[level
] = slot
;
2921 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2922 if (write_lock_level
< 1) {
2923 write_lock_level
= 1;
2924 btrfs_release_path(p
);
2928 btrfs_set_path_blocking(p
);
2929 err
= split_leaf(trans
, root
, key
,
2930 p
, ins_len
, ret
== 0);
2931 btrfs_clear_path_blocking(p
, NULL
, 0);
2939 if (!p
->search_for_split
)
2940 unlock_up(p
, level
, lowest_unlock
,
2941 min_write_lock_level
, &write_lock_level
);
2948 * we don't really know what they plan on doing with the path
2949 * from here on, so for now just mark it as blocking
2951 if (!p
->leave_spinning
)
2952 btrfs_set_path_blocking(p
);
2953 if (ret
< 0 && !p
->skip_release_on_error
)
2954 btrfs_release_path(p
);
2959 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2960 * current state of the tree together with the operations recorded in the tree
2961 * modification log to search for the key in a previous version of this tree, as
2962 * denoted by the time_seq parameter.
2964 * Naturally, there is no support for insert, delete or cow operations.
2966 * The resulting path and return value will be set up as if we called
2967 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2969 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2970 struct btrfs_path
*p
, u64 time_seq
)
2972 struct extent_buffer
*b
;
2977 int lowest_unlock
= 1;
2978 u8 lowest_level
= 0;
2981 lowest_level
= p
->lowest_level
;
2982 WARN_ON(p
->nodes
[0] != NULL
);
2984 if (p
->search_commit_root
) {
2986 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2990 b
= get_old_root(root
, time_seq
);
2991 level
= btrfs_header_level(b
);
2992 p
->locks
[level
] = BTRFS_READ_LOCK
;
2995 level
= btrfs_header_level(b
);
2996 p
->nodes
[level
] = b
;
2997 btrfs_clear_path_blocking(p
, NULL
, 0);
3000 * we have a lock on b and as long as we aren't changing
3001 * the tree, there is no way to for the items in b to change.
3002 * It is safe to drop the lock on our parent before we
3003 * go through the expensive btree search on b.
3005 btrfs_unlock_up_safe(p
, level
+ 1);
3008 * Since we can unwind ebs we want to do a real search every
3012 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
3016 if (ret
&& slot
> 0) {
3020 p
->slots
[level
] = slot
;
3021 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3023 if (level
== lowest_level
) {
3029 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
3030 slot
, key
, time_seq
);
3038 level
= btrfs_header_level(b
);
3039 err
= btrfs_tree_read_lock_atomic(b
);
3041 btrfs_set_path_blocking(p
);
3042 btrfs_tree_read_lock(b
);
3043 btrfs_clear_path_blocking(p
, b
,
3046 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
3051 p
->locks
[level
] = BTRFS_READ_LOCK
;
3052 p
->nodes
[level
] = b
;
3054 p
->slots
[level
] = slot
;
3055 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3061 if (!p
->leave_spinning
)
3062 btrfs_set_path_blocking(p
);
3064 btrfs_release_path(p
);
3070 * helper to use instead of search slot if no exact match is needed but
3071 * instead the next or previous item should be returned.
3072 * When find_higher is true, the next higher item is returned, the next lower
3074 * When return_any and find_higher are both true, and no higher item is found,
3075 * return the next lower instead.
3076 * When return_any is true and find_higher is false, and no lower item is found,
3077 * return the next higher instead.
3078 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3081 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3082 struct btrfs_key
*key
, struct btrfs_path
*p
,
3083 int find_higher
, int return_any
)
3086 struct extent_buffer
*leaf
;
3089 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3093 * a return value of 1 means the path is at the position where the
3094 * item should be inserted. Normally this is the next bigger item,
3095 * but in case the previous item is the last in a leaf, path points
3096 * to the first free slot in the previous leaf, i.e. at an invalid
3102 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3103 ret
= btrfs_next_leaf(root
, p
);
3109 * no higher item found, return the next
3114 btrfs_release_path(p
);
3118 if (p
->slots
[0] == 0) {
3119 ret
= btrfs_prev_leaf(root
, p
);
3124 if (p
->slots
[0] == btrfs_header_nritems(leaf
))
3131 * no lower item found, return the next
3136 btrfs_release_path(p
);
3146 * adjust the pointers going up the tree, starting at level
3147 * making sure the right key of each node is points to 'key'.
3148 * This is used after shifting pointers to the left, so it stops
3149 * fixing up pointers when a given leaf/node is not in slot 0 of the
3153 static void fixup_low_keys(struct btrfs_fs_info
*fs_info
,
3154 struct btrfs_path
*path
,
3155 struct btrfs_disk_key
*key
, int level
)
3158 struct extent_buffer
*t
;
3160 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3161 int tslot
= path
->slots
[i
];
3162 if (!path
->nodes
[i
])
3165 tree_mod_log_set_node_key(fs_info
, t
, tslot
, 1);
3166 btrfs_set_node_key(t
, key
, tslot
);
3167 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3176 * This function isn't completely safe. It's the caller's responsibility
3177 * that the new key won't break the order
3179 void btrfs_set_item_key_safe(struct btrfs_fs_info
*fs_info
,
3180 struct btrfs_path
*path
,
3181 struct btrfs_key
*new_key
)
3183 struct btrfs_disk_key disk_key
;
3184 struct extent_buffer
*eb
;
3187 eb
= path
->nodes
[0];
3188 slot
= path
->slots
[0];
3190 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3191 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
3193 if (slot
< btrfs_header_nritems(eb
) - 1) {
3194 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3195 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
3198 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3199 btrfs_set_item_key(eb
, &disk_key
, slot
);
3200 btrfs_mark_buffer_dirty(eb
);
3202 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
3206 * try to push data from one node into the next node left in the
3209 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3210 * error, and > 0 if there was no room in the left hand block.
3212 static int push_node_left(struct btrfs_trans_handle
*trans
,
3213 struct btrfs_root
*root
, struct extent_buffer
*dst
,
3214 struct extent_buffer
*src
, int empty
)
3221 src_nritems
= btrfs_header_nritems(src
);
3222 dst_nritems
= btrfs_header_nritems(dst
);
3223 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3224 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3225 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3227 if (!empty
&& src_nritems
<= 8)
3230 if (push_items
<= 0)
3234 push_items
= min(src_nritems
, push_items
);
3235 if (push_items
< src_nritems
) {
3236 /* leave at least 8 pointers in the node if
3237 * we aren't going to empty it
3239 if (src_nritems
- push_items
< 8) {
3240 if (push_items
<= 8)
3246 push_items
= min(src_nritems
- 8, push_items
);
3248 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3251 btrfs_abort_transaction(trans
, root
, ret
);
3254 copy_extent_buffer(dst
, src
,
3255 btrfs_node_key_ptr_offset(dst_nritems
),
3256 btrfs_node_key_ptr_offset(0),
3257 push_items
* sizeof(struct btrfs_key_ptr
));
3259 if (push_items
< src_nritems
) {
3261 * don't call tree_mod_log_eb_move here, key removal was already
3262 * fully logged by tree_mod_log_eb_copy above.
3264 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3265 btrfs_node_key_ptr_offset(push_items
),
3266 (src_nritems
- push_items
) *
3267 sizeof(struct btrfs_key_ptr
));
3269 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3270 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3271 btrfs_mark_buffer_dirty(src
);
3272 btrfs_mark_buffer_dirty(dst
);
3278 * try to push data from one node into the next node right in the
3281 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3282 * error, and > 0 if there was no room in the right hand block.
3284 * this will only push up to 1/2 the contents of the left node over
3286 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3287 struct btrfs_root
*root
,
3288 struct extent_buffer
*dst
,
3289 struct extent_buffer
*src
)
3297 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3298 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3300 src_nritems
= btrfs_header_nritems(src
);
3301 dst_nritems
= btrfs_header_nritems(dst
);
3302 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3303 if (push_items
<= 0)
3306 if (src_nritems
< 4)
3309 max_push
= src_nritems
/ 2 + 1;
3310 /* don't try to empty the node */
3311 if (max_push
>= src_nritems
)
3314 if (max_push
< push_items
)
3315 push_items
= max_push
;
3317 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3318 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3319 btrfs_node_key_ptr_offset(0),
3321 sizeof(struct btrfs_key_ptr
));
3323 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3324 src_nritems
- push_items
, push_items
);
3326 btrfs_abort_transaction(trans
, root
, ret
);
3329 copy_extent_buffer(dst
, src
,
3330 btrfs_node_key_ptr_offset(0),
3331 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3332 push_items
* sizeof(struct btrfs_key_ptr
));
3334 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3335 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3337 btrfs_mark_buffer_dirty(src
);
3338 btrfs_mark_buffer_dirty(dst
);
3344 * helper function to insert a new root level in the tree.
3345 * A new node is allocated, and a single item is inserted to
3346 * point to the existing root
3348 * returns zero on success or < 0 on failure.
3350 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3351 struct btrfs_root
*root
,
3352 struct btrfs_path
*path
, int level
)
3355 struct extent_buffer
*lower
;
3356 struct extent_buffer
*c
;
3357 struct extent_buffer
*old
;
3358 struct btrfs_disk_key lower_key
;
3360 BUG_ON(path
->nodes
[level
]);
3361 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3363 lower
= path
->nodes
[level
-1];
3365 btrfs_item_key(lower
, &lower_key
, 0);
3367 btrfs_node_key(lower
, &lower_key
, 0);
3369 c
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3370 &lower_key
, level
, root
->node
->start
, 0);
3374 root_add_used(root
, root
->nodesize
);
3376 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3377 btrfs_set_header_nritems(c
, 1);
3378 btrfs_set_header_level(c
, level
);
3379 btrfs_set_header_bytenr(c
, c
->start
);
3380 btrfs_set_header_generation(c
, trans
->transid
);
3381 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3382 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3384 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3387 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3388 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3390 btrfs_set_node_key(c
, &lower_key
, 0);
3391 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3392 lower_gen
= btrfs_header_generation(lower
);
3393 WARN_ON(lower_gen
!= trans
->transid
);
3395 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3397 btrfs_mark_buffer_dirty(c
);
3400 tree_mod_log_set_root_pointer(root
, c
, 0);
3401 rcu_assign_pointer(root
->node
, c
);
3403 /* the super has an extra ref to root->node */
3404 free_extent_buffer(old
);
3406 add_root_to_dirty_list(root
);
3407 extent_buffer_get(c
);
3408 path
->nodes
[level
] = c
;
3409 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
3410 path
->slots
[level
] = 0;
3415 * worker function to insert a single pointer in a node.
3416 * the node should have enough room for the pointer already
3418 * slot and level indicate where you want the key to go, and
3419 * blocknr is the block the key points to.
3421 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3422 struct btrfs_root
*root
, struct btrfs_path
*path
,
3423 struct btrfs_disk_key
*key
, u64 bytenr
,
3424 int slot
, int level
)
3426 struct extent_buffer
*lower
;
3430 BUG_ON(!path
->nodes
[level
]);
3431 btrfs_assert_tree_locked(path
->nodes
[level
]);
3432 lower
= path
->nodes
[level
];
3433 nritems
= btrfs_header_nritems(lower
);
3434 BUG_ON(slot
> nritems
);
3435 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3436 if (slot
!= nritems
) {
3438 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3439 slot
, nritems
- slot
);
3440 memmove_extent_buffer(lower
,
3441 btrfs_node_key_ptr_offset(slot
+ 1),
3442 btrfs_node_key_ptr_offset(slot
),
3443 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3446 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3447 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3450 btrfs_set_node_key(lower
, key
, slot
);
3451 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3452 WARN_ON(trans
->transid
== 0);
3453 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3454 btrfs_set_header_nritems(lower
, nritems
+ 1);
3455 btrfs_mark_buffer_dirty(lower
);
3459 * split the node at the specified level in path in two.
3460 * The path is corrected to point to the appropriate node after the split
3462 * Before splitting this tries to make some room in the node by pushing
3463 * left and right, if either one works, it returns right away.
3465 * returns 0 on success and < 0 on failure
3467 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3468 struct btrfs_root
*root
,
3469 struct btrfs_path
*path
, int level
)
3471 struct extent_buffer
*c
;
3472 struct extent_buffer
*split
;
3473 struct btrfs_disk_key disk_key
;
3478 c
= path
->nodes
[level
];
3479 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3480 if (c
== root
->node
) {
3482 * trying to split the root, lets make a new one
3484 * tree mod log: We don't log_removal old root in
3485 * insert_new_root, because that root buffer will be kept as a
3486 * normal node. We are going to log removal of half of the
3487 * elements below with tree_mod_log_eb_copy. We're holding a
3488 * tree lock on the buffer, which is why we cannot race with
3489 * other tree_mod_log users.
3491 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3495 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3496 c
= path
->nodes
[level
];
3497 if (!ret
&& btrfs_header_nritems(c
) <
3498 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3504 c_nritems
= btrfs_header_nritems(c
);
3505 mid
= (c_nritems
+ 1) / 2;
3506 btrfs_node_key(c
, &disk_key
, mid
);
3508 split
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3509 &disk_key
, level
, c
->start
, 0);
3511 return PTR_ERR(split
);
3513 root_add_used(root
, root
->nodesize
);
3515 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3516 btrfs_set_header_level(split
, btrfs_header_level(c
));
3517 btrfs_set_header_bytenr(split
, split
->start
);
3518 btrfs_set_header_generation(split
, trans
->transid
);
3519 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3520 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3521 write_extent_buffer(split
, root
->fs_info
->fsid
,
3522 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3523 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3524 btrfs_header_chunk_tree_uuid(split
),
3527 ret
= tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0,
3528 mid
, c_nritems
- mid
);
3530 btrfs_abort_transaction(trans
, root
, ret
);
3533 copy_extent_buffer(split
, c
,
3534 btrfs_node_key_ptr_offset(0),
3535 btrfs_node_key_ptr_offset(mid
),
3536 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3537 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3538 btrfs_set_header_nritems(c
, mid
);
3541 btrfs_mark_buffer_dirty(c
);
3542 btrfs_mark_buffer_dirty(split
);
3544 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3545 path
->slots
[level
+ 1] + 1, level
+ 1);
3547 if (path
->slots
[level
] >= mid
) {
3548 path
->slots
[level
] -= mid
;
3549 btrfs_tree_unlock(c
);
3550 free_extent_buffer(c
);
3551 path
->nodes
[level
] = split
;
3552 path
->slots
[level
+ 1] += 1;
3554 btrfs_tree_unlock(split
);
3555 free_extent_buffer(split
);
3561 * how many bytes are required to store the items in a leaf. start
3562 * and nr indicate which items in the leaf to check. This totals up the
3563 * space used both by the item structs and the item data
3565 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3567 struct btrfs_item
*start_item
;
3568 struct btrfs_item
*end_item
;
3569 struct btrfs_map_token token
;
3571 int nritems
= btrfs_header_nritems(l
);
3572 int end
= min(nritems
, start
+ nr
) - 1;
3576 btrfs_init_map_token(&token
);
3577 start_item
= btrfs_item_nr(start
);
3578 end_item
= btrfs_item_nr(end
);
3579 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3580 btrfs_token_item_size(l
, start_item
, &token
);
3581 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3582 data_len
+= sizeof(struct btrfs_item
) * nr
;
3583 WARN_ON(data_len
< 0);
3588 * The space between the end of the leaf items and
3589 * the start of the leaf data. IOW, how much room
3590 * the leaf has left for both items and data
3592 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3593 struct extent_buffer
*leaf
)
3595 int nritems
= btrfs_header_nritems(leaf
);
3597 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3599 btrfs_crit(root
->fs_info
,
3600 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3601 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3602 leaf_space_used(leaf
, 0, nritems
), nritems
);
3608 * min slot controls the lowest index we're willing to push to the
3609 * right. We'll push up to and including min_slot, but no lower
3611 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3612 struct btrfs_root
*root
,
3613 struct btrfs_path
*path
,
3614 int data_size
, int empty
,
3615 struct extent_buffer
*right
,
3616 int free_space
, u32 left_nritems
,
3619 struct extent_buffer
*left
= path
->nodes
[0];
3620 struct extent_buffer
*upper
= path
->nodes
[1];
3621 struct btrfs_map_token token
;
3622 struct btrfs_disk_key disk_key
;
3627 struct btrfs_item
*item
;
3633 btrfs_init_map_token(&token
);
3638 nr
= max_t(u32
, 1, min_slot
);
3640 if (path
->slots
[0] >= left_nritems
)
3641 push_space
+= data_size
;
3643 slot
= path
->slots
[1];
3644 i
= left_nritems
- 1;
3646 item
= btrfs_item_nr(i
);
3648 if (!empty
&& push_items
> 0) {
3649 if (path
->slots
[0] > i
)
3651 if (path
->slots
[0] == i
) {
3652 int space
= btrfs_leaf_free_space(root
, left
);
3653 if (space
+ push_space
* 2 > free_space
)
3658 if (path
->slots
[0] == i
)
3659 push_space
+= data_size
;
3661 this_item_size
= btrfs_item_size(left
, item
);
3662 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3666 push_space
+= this_item_size
+ sizeof(*item
);
3672 if (push_items
== 0)
3675 WARN_ON(!empty
&& push_items
== left_nritems
);
3677 /* push left to right */
3678 right_nritems
= btrfs_header_nritems(right
);
3680 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3681 push_space
-= leaf_data_end(root
, left
);
3683 /* make room in the right data area */
3684 data_end
= leaf_data_end(root
, right
);
3685 memmove_extent_buffer(right
,
3686 btrfs_leaf_data(right
) + data_end
- push_space
,
3687 btrfs_leaf_data(right
) + data_end
,
3688 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3690 /* copy from the left data area */
3691 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3692 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3693 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3696 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3697 btrfs_item_nr_offset(0),
3698 right_nritems
* sizeof(struct btrfs_item
));
3700 /* copy the items from left to right */
3701 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3702 btrfs_item_nr_offset(left_nritems
- push_items
),
3703 push_items
* sizeof(struct btrfs_item
));
3705 /* update the item pointers */
3706 right_nritems
+= push_items
;
3707 btrfs_set_header_nritems(right
, right_nritems
);
3708 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3709 for (i
= 0; i
< right_nritems
; i
++) {
3710 item
= btrfs_item_nr(i
);
3711 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3712 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3715 left_nritems
-= push_items
;
3716 btrfs_set_header_nritems(left
, left_nritems
);
3719 btrfs_mark_buffer_dirty(left
);
3721 clean_tree_block(trans
, root
->fs_info
, left
);
3723 btrfs_mark_buffer_dirty(right
);
3725 btrfs_item_key(right
, &disk_key
, 0);
3726 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3727 btrfs_mark_buffer_dirty(upper
);
3729 /* then fixup the leaf pointer in the path */
3730 if (path
->slots
[0] >= left_nritems
) {
3731 path
->slots
[0] -= left_nritems
;
3732 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3733 clean_tree_block(trans
, root
->fs_info
, path
->nodes
[0]);
3734 btrfs_tree_unlock(path
->nodes
[0]);
3735 free_extent_buffer(path
->nodes
[0]);
3736 path
->nodes
[0] = right
;
3737 path
->slots
[1] += 1;
3739 btrfs_tree_unlock(right
);
3740 free_extent_buffer(right
);
3745 btrfs_tree_unlock(right
);
3746 free_extent_buffer(right
);
3751 * push some data in the path leaf to the right, trying to free up at
3752 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3754 * returns 1 if the push failed because the other node didn't have enough
3755 * room, 0 if everything worked out and < 0 if there were major errors.
3757 * this will push starting from min_slot to the end of the leaf. It won't
3758 * push any slot lower than min_slot
3760 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3761 *root
, struct btrfs_path
*path
,
3762 int min_data_size
, int data_size
,
3763 int empty
, u32 min_slot
)
3765 struct extent_buffer
*left
= path
->nodes
[0];
3766 struct extent_buffer
*right
;
3767 struct extent_buffer
*upper
;
3773 if (!path
->nodes
[1])
3776 slot
= path
->slots
[1];
3777 upper
= path
->nodes
[1];
3778 if (slot
>= btrfs_header_nritems(upper
) - 1)
3781 btrfs_assert_tree_locked(path
->nodes
[1]);
3783 right
= read_node_slot(root
, upper
, slot
+ 1);
3787 btrfs_tree_lock(right
);
3788 btrfs_set_lock_blocking(right
);
3790 free_space
= btrfs_leaf_free_space(root
, right
);
3791 if (free_space
< data_size
)
3794 /* cow and double check */
3795 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3800 free_space
= btrfs_leaf_free_space(root
, right
);
3801 if (free_space
< data_size
)
3804 left_nritems
= btrfs_header_nritems(left
);
3805 if (left_nritems
== 0)
3808 if (path
->slots
[0] == left_nritems
&& !empty
) {
3809 /* Key greater than all keys in the leaf, right neighbor has
3810 * enough room for it and we're not emptying our leaf to delete
3811 * it, therefore use right neighbor to insert the new item and
3812 * no need to touch/dirty our left leaft. */
3813 btrfs_tree_unlock(left
);
3814 free_extent_buffer(left
);
3815 path
->nodes
[0] = right
;
3821 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3822 right
, free_space
, left_nritems
, min_slot
);
3824 btrfs_tree_unlock(right
);
3825 free_extent_buffer(right
);
3830 * push some data in the path leaf to the left, trying to free up at
3831 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3833 * max_slot can put a limit on how far into the leaf we'll push items. The
3834 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3837 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3838 struct btrfs_root
*root
,
3839 struct btrfs_path
*path
, int data_size
,
3840 int empty
, struct extent_buffer
*left
,
3841 int free_space
, u32 right_nritems
,
3844 struct btrfs_disk_key disk_key
;
3845 struct extent_buffer
*right
= path
->nodes
[0];
3849 struct btrfs_item
*item
;
3850 u32 old_left_nritems
;
3854 u32 old_left_item_size
;
3855 struct btrfs_map_token token
;
3857 btrfs_init_map_token(&token
);
3860 nr
= min(right_nritems
, max_slot
);
3862 nr
= min(right_nritems
- 1, max_slot
);
3864 for (i
= 0; i
< nr
; i
++) {
3865 item
= btrfs_item_nr(i
);
3867 if (!empty
&& push_items
> 0) {
3868 if (path
->slots
[0] < i
)
3870 if (path
->slots
[0] == i
) {
3871 int space
= btrfs_leaf_free_space(root
, right
);
3872 if (space
+ push_space
* 2 > free_space
)
3877 if (path
->slots
[0] == i
)
3878 push_space
+= data_size
;
3880 this_item_size
= btrfs_item_size(right
, item
);
3881 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3885 push_space
+= this_item_size
+ sizeof(*item
);
3888 if (push_items
== 0) {
3892 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3894 /* push data from right to left */
3895 copy_extent_buffer(left
, right
,
3896 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3897 btrfs_item_nr_offset(0),
3898 push_items
* sizeof(struct btrfs_item
));
3900 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3901 btrfs_item_offset_nr(right
, push_items
- 1);
3903 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3904 leaf_data_end(root
, left
) - push_space
,
3905 btrfs_leaf_data(right
) +
3906 btrfs_item_offset_nr(right
, push_items
- 1),
3908 old_left_nritems
= btrfs_header_nritems(left
);
3909 BUG_ON(old_left_nritems
<= 0);
3911 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3912 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3915 item
= btrfs_item_nr(i
);
3917 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3918 btrfs_set_token_item_offset(left
, item
,
3919 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3922 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3924 /* fixup right node */
3925 if (push_items
> right_nritems
)
3926 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3929 if (push_items
< right_nritems
) {
3930 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3931 leaf_data_end(root
, right
);
3932 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3933 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3934 btrfs_leaf_data(right
) +
3935 leaf_data_end(root
, right
), push_space
);
3937 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3938 btrfs_item_nr_offset(push_items
),
3939 (btrfs_header_nritems(right
) - push_items
) *
3940 sizeof(struct btrfs_item
));
3942 right_nritems
-= push_items
;
3943 btrfs_set_header_nritems(right
, right_nritems
);
3944 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3945 for (i
= 0; i
< right_nritems
; i
++) {
3946 item
= btrfs_item_nr(i
);
3948 push_space
= push_space
- btrfs_token_item_size(right
,
3950 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3953 btrfs_mark_buffer_dirty(left
);
3955 btrfs_mark_buffer_dirty(right
);
3957 clean_tree_block(trans
, root
->fs_info
, right
);
3959 btrfs_item_key(right
, &disk_key
, 0);
3960 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
3962 /* then fixup the leaf pointer in the path */
3963 if (path
->slots
[0] < push_items
) {
3964 path
->slots
[0] += old_left_nritems
;
3965 btrfs_tree_unlock(path
->nodes
[0]);
3966 free_extent_buffer(path
->nodes
[0]);
3967 path
->nodes
[0] = left
;
3968 path
->slots
[1] -= 1;
3970 btrfs_tree_unlock(left
);
3971 free_extent_buffer(left
);
3972 path
->slots
[0] -= push_items
;
3974 BUG_ON(path
->slots
[0] < 0);
3977 btrfs_tree_unlock(left
);
3978 free_extent_buffer(left
);
3983 * push some data in the path leaf to the left, trying to free up at
3984 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3986 * max_slot can put a limit on how far into the leaf we'll push items. The
3987 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3990 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3991 *root
, struct btrfs_path
*path
, int min_data_size
,
3992 int data_size
, int empty
, u32 max_slot
)
3994 struct extent_buffer
*right
= path
->nodes
[0];
3995 struct extent_buffer
*left
;
4001 slot
= path
->slots
[1];
4004 if (!path
->nodes
[1])
4007 right_nritems
= btrfs_header_nritems(right
);
4008 if (right_nritems
== 0)
4011 btrfs_assert_tree_locked(path
->nodes
[1]);
4013 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
4017 btrfs_tree_lock(left
);
4018 btrfs_set_lock_blocking(left
);
4020 free_space
= btrfs_leaf_free_space(root
, left
);
4021 if (free_space
< data_size
) {
4026 /* cow and double check */
4027 ret
= btrfs_cow_block(trans
, root
, left
,
4028 path
->nodes
[1], slot
- 1, &left
);
4030 /* we hit -ENOSPC, but it isn't fatal here */
4036 free_space
= btrfs_leaf_free_space(root
, left
);
4037 if (free_space
< data_size
) {
4042 return __push_leaf_left(trans
, root
, path
, min_data_size
,
4043 empty
, left
, free_space
, right_nritems
,
4046 btrfs_tree_unlock(left
);
4047 free_extent_buffer(left
);
4052 * split the path's leaf in two, making sure there is at least data_size
4053 * available for the resulting leaf level of the path.
4055 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4056 struct btrfs_root
*root
,
4057 struct btrfs_path
*path
,
4058 struct extent_buffer
*l
,
4059 struct extent_buffer
*right
,
4060 int slot
, int mid
, int nritems
)
4065 struct btrfs_disk_key disk_key
;
4066 struct btrfs_map_token token
;
4068 btrfs_init_map_token(&token
);
4070 nritems
= nritems
- mid
;
4071 btrfs_set_header_nritems(right
, nritems
);
4072 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
4074 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4075 btrfs_item_nr_offset(mid
),
4076 nritems
* sizeof(struct btrfs_item
));
4078 copy_extent_buffer(right
, l
,
4079 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
4080 data_copy_size
, btrfs_leaf_data(l
) +
4081 leaf_data_end(root
, l
), data_copy_size
);
4083 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
4084 btrfs_item_end_nr(l
, mid
);
4086 for (i
= 0; i
< nritems
; i
++) {
4087 struct btrfs_item
*item
= btrfs_item_nr(i
);
4090 ioff
= btrfs_token_item_offset(right
, item
, &token
);
4091 btrfs_set_token_item_offset(right
, item
,
4092 ioff
+ rt_data_off
, &token
);
4095 btrfs_set_header_nritems(l
, mid
);
4096 btrfs_item_key(right
, &disk_key
, 0);
4097 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4098 path
->slots
[1] + 1, 1);
4100 btrfs_mark_buffer_dirty(right
);
4101 btrfs_mark_buffer_dirty(l
);
4102 BUG_ON(path
->slots
[0] != slot
);
4105 btrfs_tree_unlock(path
->nodes
[0]);
4106 free_extent_buffer(path
->nodes
[0]);
4107 path
->nodes
[0] = right
;
4108 path
->slots
[0] -= mid
;
4109 path
->slots
[1] += 1;
4111 btrfs_tree_unlock(right
);
4112 free_extent_buffer(right
);
4115 BUG_ON(path
->slots
[0] < 0);
4119 * double splits happen when we need to insert a big item in the middle
4120 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4121 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4124 * We avoid this by trying to push the items on either side of our target
4125 * into the adjacent leaves. If all goes well we can avoid the double split
4128 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4129 struct btrfs_root
*root
,
4130 struct btrfs_path
*path
,
4137 int space_needed
= data_size
;
4139 slot
= path
->slots
[0];
4140 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4141 space_needed
-= btrfs_leaf_free_space(root
, path
->nodes
[0]);
4144 * try to push all the items after our slot into the
4147 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4154 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4156 * our goal is to get our slot at the start or end of a leaf. If
4157 * we've done so we're done
4159 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4162 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4165 /* try to push all the items before our slot into the next leaf */
4166 slot
= path
->slots
[0];
4167 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4180 * split the path's leaf in two, making sure there is at least data_size
4181 * available for the resulting leaf level of the path.
4183 * returns 0 if all went well and < 0 on failure.
4185 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4186 struct btrfs_root
*root
,
4187 struct btrfs_key
*ins_key
,
4188 struct btrfs_path
*path
, int data_size
,
4191 struct btrfs_disk_key disk_key
;
4192 struct extent_buffer
*l
;
4196 struct extent_buffer
*right
;
4197 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4201 int num_doubles
= 0;
4202 int tried_avoid_double
= 0;
4205 slot
= path
->slots
[0];
4206 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4207 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
4210 /* first try to make some room by pushing left and right */
4211 if (data_size
&& path
->nodes
[1]) {
4212 int space_needed
= data_size
;
4214 if (slot
< btrfs_header_nritems(l
))
4215 space_needed
-= btrfs_leaf_free_space(root
, l
);
4217 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4218 space_needed
, 0, 0);
4222 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4223 space_needed
, 0, (u32
)-1);
4229 /* did the pushes work? */
4230 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
4234 if (!path
->nodes
[1]) {
4235 ret
= insert_new_root(trans
, root
, path
, 1);
4242 slot
= path
->slots
[0];
4243 nritems
= btrfs_header_nritems(l
);
4244 mid
= (nritems
+ 1) / 2;
4248 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4249 BTRFS_LEAF_DATA_SIZE(root
)) {
4250 if (slot
>= nritems
) {
4254 if (mid
!= nritems
&&
4255 leaf_space_used(l
, mid
, nritems
- mid
) +
4256 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4257 if (data_size
&& !tried_avoid_double
)
4258 goto push_for_double
;
4264 if (leaf_space_used(l
, 0, mid
) + data_size
>
4265 BTRFS_LEAF_DATA_SIZE(root
)) {
4266 if (!extend
&& data_size
&& slot
== 0) {
4268 } else if ((extend
|| !data_size
) && slot
== 0) {
4272 if (mid
!= nritems
&&
4273 leaf_space_used(l
, mid
, nritems
- mid
) +
4274 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4275 if (data_size
&& !tried_avoid_double
)
4276 goto push_for_double
;
4284 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4286 btrfs_item_key(l
, &disk_key
, mid
);
4288 right
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
4289 &disk_key
, 0, l
->start
, 0);
4291 return PTR_ERR(right
);
4293 root_add_used(root
, root
->nodesize
);
4295 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4296 btrfs_set_header_bytenr(right
, right
->start
);
4297 btrfs_set_header_generation(right
, trans
->transid
);
4298 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4299 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4300 btrfs_set_header_level(right
, 0);
4301 write_extent_buffer(right
, fs_info
->fsid
,
4302 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4304 write_extent_buffer(right
, fs_info
->chunk_tree_uuid
,
4305 btrfs_header_chunk_tree_uuid(right
),
4310 btrfs_set_header_nritems(right
, 0);
4311 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4312 path
->slots
[1] + 1, 1);
4313 btrfs_tree_unlock(path
->nodes
[0]);
4314 free_extent_buffer(path
->nodes
[0]);
4315 path
->nodes
[0] = right
;
4317 path
->slots
[1] += 1;
4319 btrfs_set_header_nritems(right
, 0);
4320 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4322 btrfs_tree_unlock(path
->nodes
[0]);
4323 free_extent_buffer(path
->nodes
[0]);
4324 path
->nodes
[0] = right
;
4326 if (path
->slots
[1] == 0)
4327 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
4329 btrfs_mark_buffer_dirty(right
);
4333 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4336 BUG_ON(num_doubles
!= 0);
4344 push_for_double_split(trans
, root
, path
, data_size
);
4345 tried_avoid_double
= 1;
4346 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4351 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4352 struct btrfs_root
*root
,
4353 struct btrfs_path
*path
, int ins_len
)
4355 struct btrfs_key key
;
4356 struct extent_buffer
*leaf
;
4357 struct btrfs_file_extent_item
*fi
;
4362 leaf
= path
->nodes
[0];
4363 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4365 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4366 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4368 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4371 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4372 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4373 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4374 struct btrfs_file_extent_item
);
4375 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4377 btrfs_release_path(path
);
4379 path
->keep_locks
= 1;
4380 path
->search_for_split
= 1;
4381 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4382 path
->search_for_split
= 0;
4389 leaf
= path
->nodes
[0];
4390 /* if our item isn't there, return now */
4391 if (item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4394 /* the leaf has changed, it now has room. return now */
4395 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4398 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4399 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4400 struct btrfs_file_extent_item
);
4401 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4405 btrfs_set_path_blocking(path
);
4406 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4410 path
->keep_locks
= 0;
4411 btrfs_unlock_up_safe(path
, 1);
4414 path
->keep_locks
= 0;
4418 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4419 struct btrfs_root
*root
,
4420 struct btrfs_path
*path
,
4421 struct btrfs_key
*new_key
,
4422 unsigned long split_offset
)
4424 struct extent_buffer
*leaf
;
4425 struct btrfs_item
*item
;
4426 struct btrfs_item
*new_item
;
4432 struct btrfs_disk_key disk_key
;
4434 leaf
= path
->nodes
[0];
4435 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4437 btrfs_set_path_blocking(path
);
4439 item
= btrfs_item_nr(path
->slots
[0]);
4440 orig_offset
= btrfs_item_offset(leaf
, item
);
4441 item_size
= btrfs_item_size(leaf
, item
);
4443 buf
= kmalloc(item_size
, GFP_NOFS
);
4447 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4448 path
->slots
[0]), item_size
);
4450 slot
= path
->slots
[0] + 1;
4451 nritems
= btrfs_header_nritems(leaf
);
4452 if (slot
!= nritems
) {
4453 /* shift the items */
4454 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4455 btrfs_item_nr_offset(slot
),
4456 (nritems
- slot
) * sizeof(struct btrfs_item
));
4459 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4460 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4462 new_item
= btrfs_item_nr(slot
);
4464 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4465 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4467 btrfs_set_item_offset(leaf
, item
,
4468 orig_offset
+ item_size
- split_offset
);
4469 btrfs_set_item_size(leaf
, item
, split_offset
);
4471 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4473 /* write the data for the start of the original item */
4474 write_extent_buffer(leaf
, buf
,
4475 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4478 /* write the data for the new item */
4479 write_extent_buffer(leaf
, buf
+ split_offset
,
4480 btrfs_item_ptr_offset(leaf
, slot
),
4481 item_size
- split_offset
);
4482 btrfs_mark_buffer_dirty(leaf
);
4484 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4490 * This function splits a single item into two items,
4491 * giving 'new_key' to the new item and splitting the
4492 * old one at split_offset (from the start of the item).
4494 * The path may be released by this operation. After
4495 * the split, the path is pointing to the old item. The
4496 * new item is going to be in the same node as the old one.
4498 * Note, the item being split must be smaller enough to live alone on
4499 * a tree block with room for one extra struct btrfs_item
4501 * This allows us to split the item in place, keeping a lock on the
4502 * leaf the entire time.
4504 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4505 struct btrfs_root
*root
,
4506 struct btrfs_path
*path
,
4507 struct btrfs_key
*new_key
,
4508 unsigned long split_offset
)
4511 ret
= setup_leaf_for_split(trans
, root
, path
,
4512 sizeof(struct btrfs_item
));
4516 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4521 * This function duplicate a item, giving 'new_key' to the new item.
4522 * It guarantees both items live in the same tree leaf and the new item
4523 * is contiguous with the original item.
4525 * This allows us to split file extent in place, keeping a lock on the
4526 * leaf the entire time.
4528 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4529 struct btrfs_root
*root
,
4530 struct btrfs_path
*path
,
4531 struct btrfs_key
*new_key
)
4533 struct extent_buffer
*leaf
;
4537 leaf
= path
->nodes
[0];
4538 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4539 ret
= setup_leaf_for_split(trans
, root
, path
,
4540 item_size
+ sizeof(struct btrfs_item
));
4545 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4546 item_size
, item_size
+
4547 sizeof(struct btrfs_item
), 1);
4548 leaf
= path
->nodes
[0];
4549 memcpy_extent_buffer(leaf
,
4550 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4551 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4557 * make the item pointed to by the path smaller. new_size indicates
4558 * how small to make it, and from_end tells us if we just chop bytes
4559 * off the end of the item or if we shift the item to chop bytes off
4562 void btrfs_truncate_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4563 u32 new_size
, int from_end
)
4566 struct extent_buffer
*leaf
;
4567 struct btrfs_item
*item
;
4569 unsigned int data_end
;
4570 unsigned int old_data_start
;
4571 unsigned int old_size
;
4572 unsigned int size_diff
;
4574 struct btrfs_map_token token
;
4576 btrfs_init_map_token(&token
);
4578 leaf
= path
->nodes
[0];
4579 slot
= path
->slots
[0];
4581 old_size
= btrfs_item_size_nr(leaf
, slot
);
4582 if (old_size
== new_size
)
4585 nritems
= btrfs_header_nritems(leaf
);
4586 data_end
= leaf_data_end(root
, leaf
);
4588 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4590 size_diff
= old_size
- new_size
;
4593 BUG_ON(slot
>= nritems
);
4596 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4598 /* first correct the data pointers */
4599 for (i
= slot
; i
< nritems
; i
++) {
4601 item
= btrfs_item_nr(i
);
4603 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4604 btrfs_set_token_item_offset(leaf
, item
,
4605 ioff
+ size_diff
, &token
);
4608 /* shift the data */
4610 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4611 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4612 data_end
, old_data_start
+ new_size
- data_end
);
4614 struct btrfs_disk_key disk_key
;
4617 btrfs_item_key(leaf
, &disk_key
, slot
);
4619 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4621 struct btrfs_file_extent_item
*fi
;
4623 fi
= btrfs_item_ptr(leaf
, slot
,
4624 struct btrfs_file_extent_item
);
4625 fi
= (struct btrfs_file_extent_item
*)(
4626 (unsigned long)fi
- size_diff
);
4628 if (btrfs_file_extent_type(leaf
, fi
) ==
4629 BTRFS_FILE_EXTENT_INLINE
) {
4630 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4631 memmove_extent_buffer(leaf
, ptr
,
4633 BTRFS_FILE_EXTENT_INLINE_DATA_START
);
4637 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4638 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4639 data_end
, old_data_start
- data_end
);
4641 offset
= btrfs_disk_key_offset(&disk_key
);
4642 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4643 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4645 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4648 item
= btrfs_item_nr(slot
);
4649 btrfs_set_item_size(leaf
, item
, new_size
);
4650 btrfs_mark_buffer_dirty(leaf
);
4652 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4653 btrfs_print_leaf(root
, leaf
);
4659 * make the item pointed to by the path bigger, data_size is the added size.
4661 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4665 struct extent_buffer
*leaf
;
4666 struct btrfs_item
*item
;
4668 unsigned int data_end
;
4669 unsigned int old_data
;
4670 unsigned int old_size
;
4672 struct btrfs_map_token token
;
4674 btrfs_init_map_token(&token
);
4676 leaf
= path
->nodes
[0];
4678 nritems
= btrfs_header_nritems(leaf
);
4679 data_end
= leaf_data_end(root
, leaf
);
4681 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4682 btrfs_print_leaf(root
, leaf
);
4685 slot
= path
->slots
[0];
4686 old_data
= btrfs_item_end_nr(leaf
, slot
);
4689 if (slot
>= nritems
) {
4690 btrfs_print_leaf(root
, leaf
);
4691 btrfs_crit(root
->fs_info
, "slot %d too large, nritems %d",
4697 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4699 /* first correct the data pointers */
4700 for (i
= slot
; i
< nritems
; i
++) {
4702 item
= btrfs_item_nr(i
);
4704 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4705 btrfs_set_token_item_offset(leaf
, item
,
4706 ioff
- data_size
, &token
);
4709 /* shift the data */
4710 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4711 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4712 data_end
, old_data
- data_end
);
4714 data_end
= old_data
;
4715 old_size
= btrfs_item_size_nr(leaf
, slot
);
4716 item
= btrfs_item_nr(slot
);
4717 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4718 btrfs_mark_buffer_dirty(leaf
);
4720 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4721 btrfs_print_leaf(root
, leaf
);
4727 * this is a helper for btrfs_insert_empty_items, the main goal here is
4728 * to save stack depth by doing the bulk of the work in a function
4729 * that doesn't call btrfs_search_slot
4731 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4732 struct btrfs_key
*cpu_key
, u32
*data_size
,
4733 u32 total_data
, u32 total_size
, int nr
)
4735 struct btrfs_item
*item
;
4738 unsigned int data_end
;
4739 struct btrfs_disk_key disk_key
;
4740 struct extent_buffer
*leaf
;
4742 struct btrfs_map_token token
;
4744 if (path
->slots
[0] == 0) {
4745 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4746 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4748 btrfs_unlock_up_safe(path
, 1);
4750 btrfs_init_map_token(&token
);
4752 leaf
= path
->nodes
[0];
4753 slot
= path
->slots
[0];
4755 nritems
= btrfs_header_nritems(leaf
);
4756 data_end
= leaf_data_end(root
, leaf
);
4758 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4759 btrfs_print_leaf(root
, leaf
);
4760 btrfs_crit(root
->fs_info
, "not enough freespace need %u have %d",
4761 total_size
, btrfs_leaf_free_space(root
, leaf
));
4765 if (slot
!= nritems
) {
4766 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4768 if (old_data
< data_end
) {
4769 btrfs_print_leaf(root
, leaf
);
4770 btrfs_crit(root
->fs_info
, "slot %d old_data %d data_end %d",
4771 slot
, old_data
, data_end
);
4775 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4777 /* first correct the data pointers */
4778 for (i
= slot
; i
< nritems
; i
++) {
4781 item
= btrfs_item_nr( i
);
4782 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4783 btrfs_set_token_item_offset(leaf
, item
,
4784 ioff
- total_data
, &token
);
4786 /* shift the items */
4787 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4788 btrfs_item_nr_offset(slot
),
4789 (nritems
- slot
) * sizeof(struct btrfs_item
));
4791 /* shift the data */
4792 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4793 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4794 data_end
, old_data
- data_end
);
4795 data_end
= old_data
;
4798 /* setup the item for the new data */
4799 for (i
= 0; i
< nr
; i
++) {
4800 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4801 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4802 item
= btrfs_item_nr(slot
+ i
);
4803 btrfs_set_token_item_offset(leaf
, item
,
4804 data_end
- data_size
[i
], &token
);
4805 data_end
-= data_size
[i
];
4806 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4809 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4810 btrfs_mark_buffer_dirty(leaf
);
4812 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4813 btrfs_print_leaf(root
, leaf
);
4819 * Given a key and some data, insert items into the tree.
4820 * This does all the path init required, making room in the tree if needed.
4822 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4823 struct btrfs_root
*root
,
4824 struct btrfs_path
*path
,
4825 struct btrfs_key
*cpu_key
, u32
*data_size
,
4834 for (i
= 0; i
< nr
; i
++)
4835 total_data
+= data_size
[i
];
4837 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4838 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4844 slot
= path
->slots
[0];
4847 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4848 total_data
, total_size
, nr
);
4853 * Given a key and some data, insert an item into the tree.
4854 * This does all the path init required, making room in the tree if needed.
4856 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4857 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4861 struct btrfs_path
*path
;
4862 struct extent_buffer
*leaf
;
4865 path
= btrfs_alloc_path();
4868 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4870 leaf
= path
->nodes
[0];
4871 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4872 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4873 btrfs_mark_buffer_dirty(leaf
);
4875 btrfs_free_path(path
);
4880 * delete the pointer from a given node.
4882 * the tree should have been previously balanced so the deletion does not
4885 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4886 int level
, int slot
)
4888 struct extent_buffer
*parent
= path
->nodes
[level
];
4892 nritems
= btrfs_header_nritems(parent
);
4893 if (slot
!= nritems
- 1) {
4895 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4896 slot
+ 1, nritems
- slot
- 1);
4897 memmove_extent_buffer(parent
,
4898 btrfs_node_key_ptr_offset(slot
),
4899 btrfs_node_key_ptr_offset(slot
+ 1),
4900 sizeof(struct btrfs_key_ptr
) *
4901 (nritems
- slot
- 1));
4903 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4904 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4909 btrfs_set_header_nritems(parent
, nritems
);
4910 if (nritems
== 0 && parent
== root
->node
) {
4911 BUG_ON(btrfs_header_level(root
->node
) != 1);
4912 /* just turn the root into a leaf and break */
4913 btrfs_set_header_level(root
->node
, 0);
4914 } else if (slot
== 0) {
4915 struct btrfs_disk_key disk_key
;
4917 btrfs_node_key(parent
, &disk_key
, 0);
4918 fixup_low_keys(root
->fs_info
, path
, &disk_key
, level
+ 1);
4920 btrfs_mark_buffer_dirty(parent
);
4924 * a helper function to delete the leaf pointed to by path->slots[1] and
4927 * This deletes the pointer in path->nodes[1] and frees the leaf
4928 * block extent. zero is returned if it all worked out, < 0 otherwise.
4930 * The path must have already been setup for deleting the leaf, including
4931 * all the proper balancing. path->nodes[1] must be locked.
4933 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4934 struct btrfs_root
*root
,
4935 struct btrfs_path
*path
,
4936 struct extent_buffer
*leaf
)
4938 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4939 del_ptr(root
, path
, 1, path
->slots
[1]);
4942 * btrfs_free_extent is expensive, we want to make sure we
4943 * aren't holding any locks when we call it
4945 btrfs_unlock_up_safe(path
, 0);
4947 root_sub_used(root
, leaf
->len
);
4949 extent_buffer_get(leaf
);
4950 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4951 free_extent_buffer_stale(leaf
);
4954 * delete the item at the leaf level in path. If that empties
4955 * the leaf, remove it from the tree
4957 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4958 struct btrfs_path
*path
, int slot
, int nr
)
4960 struct extent_buffer
*leaf
;
4961 struct btrfs_item
*item
;
4968 struct btrfs_map_token token
;
4970 btrfs_init_map_token(&token
);
4972 leaf
= path
->nodes
[0];
4973 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4975 for (i
= 0; i
< nr
; i
++)
4976 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4978 nritems
= btrfs_header_nritems(leaf
);
4980 if (slot
+ nr
!= nritems
) {
4981 int data_end
= leaf_data_end(root
, leaf
);
4983 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4985 btrfs_leaf_data(leaf
) + data_end
,
4986 last_off
- data_end
);
4988 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4991 item
= btrfs_item_nr(i
);
4992 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4993 btrfs_set_token_item_offset(leaf
, item
,
4994 ioff
+ dsize
, &token
);
4997 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4998 btrfs_item_nr_offset(slot
+ nr
),
4999 sizeof(struct btrfs_item
) *
5000 (nritems
- slot
- nr
));
5002 btrfs_set_header_nritems(leaf
, nritems
- nr
);
5005 /* delete the leaf if we've emptied it */
5007 if (leaf
== root
->node
) {
5008 btrfs_set_header_level(leaf
, 0);
5010 btrfs_set_path_blocking(path
);
5011 clean_tree_block(trans
, root
->fs_info
, leaf
);
5012 btrfs_del_leaf(trans
, root
, path
, leaf
);
5015 int used
= leaf_space_used(leaf
, 0, nritems
);
5017 struct btrfs_disk_key disk_key
;
5019 btrfs_item_key(leaf
, &disk_key
, 0);
5020 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
5023 /* delete the leaf if it is mostly empty */
5024 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
5025 /* push_leaf_left fixes the path.
5026 * make sure the path still points to our leaf
5027 * for possible call to del_ptr below
5029 slot
= path
->slots
[1];
5030 extent_buffer_get(leaf
);
5032 btrfs_set_path_blocking(path
);
5033 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
5035 if (wret
< 0 && wret
!= -ENOSPC
)
5038 if (path
->nodes
[0] == leaf
&&
5039 btrfs_header_nritems(leaf
)) {
5040 wret
= push_leaf_right(trans
, root
, path
, 1,
5042 if (wret
< 0 && wret
!= -ENOSPC
)
5046 if (btrfs_header_nritems(leaf
) == 0) {
5047 path
->slots
[1] = slot
;
5048 btrfs_del_leaf(trans
, root
, path
, leaf
);
5049 free_extent_buffer(leaf
);
5052 /* if we're still in the path, make sure
5053 * we're dirty. Otherwise, one of the
5054 * push_leaf functions must have already
5055 * dirtied this buffer
5057 if (path
->nodes
[0] == leaf
)
5058 btrfs_mark_buffer_dirty(leaf
);
5059 free_extent_buffer(leaf
);
5062 btrfs_mark_buffer_dirty(leaf
);
5069 * search the tree again to find a leaf with lesser keys
5070 * returns 0 if it found something or 1 if there are no lesser leaves.
5071 * returns < 0 on io errors.
5073 * This may release the path, and so you may lose any locks held at the
5076 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5078 struct btrfs_key key
;
5079 struct btrfs_disk_key found_key
;
5082 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5084 if (key
.offset
> 0) {
5086 } else if (key
.type
> 0) {
5088 key
.offset
= (u64
)-1;
5089 } else if (key
.objectid
> 0) {
5092 key
.offset
= (u64
)-1;
5097 btrfs_release_path(path
);
5098 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5101 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5102 ret
= comp_keys(&found_key
, &key
);
5104 * We might have had an item with the previous key in the tree right
5105 * before we released our path. And after we released our path, that
5106 * item might have been pushed to the first slot (0) of the leaf we
5107 * were holding due to a tree balance. Alternatively, an item with the
5108 * previous key can exist as the only element of a leaf (big fat item).
5109 * Therefore account for these 2 cases, so that our callers (like
5110 * btrfs_previous_item) don't miss an existing item with a key matching
5111 * the previous key we computed above.
5119 * A helper function to walk down the tree starting at min_key, and looking
5120 * for nodes or leaves that are have a minimum transaction id.
5121 * This is used by the btree defrag code, and tree logging
5123 * This does not cow, but it does stuff the starting key it finds back
5124 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5125 * key and get a writable path.
5127 * This does lock as it descends, and path->keep_locks should be set
5128 * to 1 by the caller.
5130 * This honors path->lowest_level to prevent descent past a given level
5133 * min_trans indicates the oldest transaction that you are interested
5134 * in walking through. Any nodes or leaves older than min_trans are
5135 * skipped over (without reading them).
5137 * returns zero if something useful was found, < 0 on error and 1 if there
5138 * was nothing in the tree that matched the search criteria.
5140 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5141 struct btrfs_path
*path
,
5144 struct extent_buffer
*cur
;
5145 struct btrfs_key found_key
;
5151 int keep_locks
= path
->keep_locks
;
5153 path
->keep_locks
= 1;
5155 cur
= btrfs_read_lock_root_node(root
);
5156 level
= btrfs_header_level(cur
);
5157 WARN_ON(path
->nodes
[level
]);
5158 path
->nodes
[level
] = cur
;
5159 path
->locks
[level
] = BTRFS_READ_LOCK
;
5161 if (btrfs_header_generation(cur
) < min_trans
) {
5166 nritems
= btrfs_header_nritems(cur
);
5167 level
= btrfs_header_level(cur
);
5168 sret
= bin_search(cur
, min_key
, level
, &slot
);
5170 /* at the lowest level, we're done, setup the path and exit */
5171 if (level
== path
->lowest_level
) {
5172 if (slot
>= nritems
)
5175 path
->slots
[level
] = slot
;
5176 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5179 if (sret
&& slot
> 0)
5182 * check this node pointer against the min_trans parameters.
5183 * If it is too old, old, skip to the next one.
5185 while (slot
< nritems
) {
5188 gen
= btrfs_node_ptr_generation(cur
, slot
);
5189 if (gen
< min_trans
) {
5197 * we didn't find a candidate key in this node, walk forward
5198 * and find another one
5200 if (slot
>= nritems
) {
5201 path
->slots
[level
] = slot
;
5202 btrfs_set_path_blocking(path
);
5203 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5206 btrfs_release_path(path
);
5212 /* save our key for returning back */
5213 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5214 path
->slots
[level
] = slot
;
5215 if (level
== path
->lowest_level
) {
5219 btrfs_set_path_blocking(path
);
5220 cur
= read_node_slot(root
, cur
, slot
);
5221 BUG_ON(!cur
); /* -ENOMEM */
5223 btrfs_tree_read_lock(cur
);
5225 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5226 path
->nodes
[level
- 1] = cur
;
5227 unlock_up(path
, level
, 1, 0, NULL
);
5228 btrfs_clear_path_blocking(path
, NULL
, 0);
5231 path
->keep_locks
= keep_locks
;
5233 btrfs_unlock_up_safe(path
, path
->lowest_level
+ 1);
5234 btrfs_set_path_blocking(path
);
5235 memcpy(min_key
, &found_key
, sizeof(found_key
));
5240 static void tree_move_down(struct btrfs_root
*root
,
5241 struct btrfs_path
*path
,
5242 int *level
, int root_level
)
5244 BUG_ON(*level
== 0);
5245 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
5246 path
->slots
[*level
]);
5247 path
->slots
[*level
- 1] = 0;
5251 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
5252 struct btrfs_path
*path
,
5253 int *level
, int root_level
)
5257 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5259 path
->slots
[*level
]++;
5261 while (path
->slots
[*level
] >= nritems
) {
5262 if (*level
== root_level
)
5266 path
->slots
[*level
] = 0;
5267 free_extent_buffer(path
->nodes
[*level
]);
5268 path
->nodes
[*level
] = NULL
;
5270 path
->slots
[*level
]++;
5272 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5279 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5282 static int tree_advance(struct btrfs_root
*root
,
5283 struct btrfs_path
*path
,
5284 int *level
, int root_level
,
5286 struct btrfs_key
*key
)
5290 if (*level
== 0 || !allow_down
) {
5291 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5293 tree_move_down(root
, path
, level
, root_level
);
5298 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5299 path
->slots
[*level
]);
5301 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5302 path
->slots
[*level
]);
5307 static int tree_compare_item(struct btrfs_root
*left_root
,
5308 struct btrfs_path
*left_path
,
5309 struct btrfs_path
*right_path
,
5314 unsigned long off1
, off2
;
5316 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5317 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5321 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5322 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5323 right_path
->slots
[0]);
5325 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5327 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5334 #define ADVANCE_ONLY_NEXT -1
5337 * This function compares two trees and calls the provided callback for
5338 * every changed/new/deleted item it finds.
5339 * If shared tree blocks are encountered, whole subtrees are skipped, making
5340 * the compare pretty fast on snapshotted subvolumes.
5342 * This currently works on commit roots only. As commit roots are read only,
5343 * we don't do any locking. The commit roots are protected with transactions.
5344 * Transactions are ended and rejoined when a commit is tried in between.
5346 * This function checks for modifications done to the trees while comparing.
5347 * If it detects a change, it aborts immediately.
5349 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5350 struct btrfs_root
*right_root
,
5351 btrfs_changed_cb_t changed_cb
, void *ctx
)
5355 struct btrfs_path
*left_path
= NULL
;
5356 struct btrfs_path
*right_path
= NULL
;
5357 struct btrfs_key left_key
;
5358 struct btrfs_key right_key
;
5359 char *tmp_buf
= NULL
;
5360 int left_root_level
;
5361 int right_root_level
;
5364 int left_end_reached
;
5365 int right_end_reached
;
5373 left_path
= btrfs_alloc_path();
5378 right_path
= btrfs_alloc_path();
5384 tmp_buf
= kmalloc(left_root
->nodesize
, GFP_KERNEL
| __GFP_NOWARN
);
5386 tmp_buf
= vmalloc(left_root
->nodesize
);
5393 left_path
->search_commit_root
= 1;
5394 left_path
->skip_locking
= 1;
5395 right_path
->search_commit_root
= 1;
5396 right_path
->skip_locking
= 1;
5399 * Strategy: Go to the first items of both trees. Then do
5401 * If both trees are at level 0
5402 * Compare keys of current items
5403 * If left < right treat left item as new, advance left tree
5405 * If left > right treat right item as deleted, advance right tree
5407 * If left == right do deep compare of items, treat as changed if
5408 * needed, advance both trees and repeat
5409 * If both trees are at the same level but not at level 0
5410 * Compare keys of current nodes/leafs
5411 * If left < right advance left tree and repeat
5412 * If left > right advance right tree and repeat
5413 * If left == right compare blockptrs of the next nodes/leafs
5414 * If they match advance both trees but stay at the same level
5416 * If they don't match advance both trees while allowing to go
5418 * If tree levels are different
5419 * Advance the tree that needs it and repeat
5421 * Advancing a tree means:
5422 * If we are at level 0, try to go to the next slot. If that's not
5423 * possible, go one level up and repeat. Stop when we found a level
5424 * where we could go to the next slot. We may at this point be on a
5427 * If we are not at level 0 and not on shared tree blocks, go one
5430 * If we are not at level 0 and on shared tree blocks, go one slot to
5431 * the right if possible or go up and right.
5434 down_read(&left_root
->fs_info
->commit_root_sem
);
5435 left_level
= btrfs_header_level(left_root
->commit_root
);
5436 left_root_level
= left_level
;
5437 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5438 extent_buffer_get(left_path
->nodes
[left_level
]);
5440 right_level
= btrfs_header_level(right_root
->commit_root
);
5441 right_root_level
= right_level
;
5442 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5443 extent_buffer_get(right_path
->nodes
[right_level
]);
5444 up_read(&left_root
->fs_info
->commit_root_sem
);
5446 if (left_level
== 0)
5447 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5448 &left_key
, left_path
->slots
[left_level
]);
5450 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5451 &left_key
, left_path
->slots
[left_level
]);
5452 if (right_level
== 0)
5453 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5454 &right_key
, right_path
->slots
[right_level
]);
5456 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5457 &right_key
, right_path
->slots
[right_level
]);
5459 left_end_reached
= right_end_reached
= 0;
5460 advance_left
= advance_right
= 0;
5463 if (advance_left
&& !left_end_reached
) {
5464 ret
= tree_advance(left_root
, left_path
, &left_level
,
5466 advance_left
!= ADVANCE_ONLY_NEXT
,
5469 left_end_reached
= ADVANCE
;
5472 if (advance_right
&& !right_end_reached
) {
5473 ret
= tree_advance(right_root
, right_path
, &right_level
,
5475 advance_right
!= ADVANCE_ONLY_NEXT
,
5478 right_end_reached
= ADVANCE
;
5482 if (left_end_reached
&& right_end_reached
) {
5485 } else if (left_end_reached
) {
5486 if (right_level
== 0) {
5487 ret
= changed_cb(left_root
, right_root
,
5488 left_path
, right_path
,
5490 BTRFS_COMPARE_TREE_DELETED
,
5495 advance_right
= ADVANCE
;
5497 } else if (right_end_reached
) {
5498 if (left_level
== 0) {
5499 ret
= changed_cb(left_root
, right_root
,
5500 left_path
, right_path
,
5502 BTRFS_COMPARE_TREE_NEW
,
5507 advance_left
= ADVANCE
;
5511 if (left_level
== 0 && right_level
== 0) {
5512 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5514 ret
= changed_cb(left_root
, right_root
,
5515 left_path
, right_path
,
5517 BTRFS_COMPARE_TREE_NEW
,
5521 advance_left
= ADVANCE
;
5522 } else if (cmp
> 0) {
5523 ret
= changed_cb(left_root
, right_root
,
5524 left_path
, right_path
,
5526 BTRFS_COMPARE_TREE_DELETED
,
5530 advance_right
= ADVANCE
;
5532 enum btrfs_compare_tree_result result
;
5534 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5535 ret
= tree_compare_item(left_root
, left_path
,
5536 right_path
, tmp_buf
);
5538 result
= BTRFS_COMPARE_TREE_CHANGED
;
5540 result
= BTRFS_COMPARE_TREE_SAME
;
5541 ret
= changed_cb(left_root
, right_root
,
5542 left_path
, right_path
,
5543 &left_key
, result
, ctx
);
5546 advance_left
= ADVANCE
;
5547 advance_right
= ADVANCE
;
5549 } else if (left_level
== right_level
) {
5550 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5552 advance_left
= ADVANCE
;
5553 } else if (cmp
> 0) {
5554 advance_right
= ADVANCE
;
5556 left_blockptr
= btrfs_node_blockptr(
5557 left_path
->nodes
[left_level
],
5558 left_path
->slots
[left_level
]);
5559 right_blockptr
= btrfs_node_blockptr(
5560 right_path
->nodes
[right_level
],
5561 right_path
->slots
[right_level
]);
5562 left_gen
= btrfs_node_ptr_generation(
5563 left_path
->nodes
[left_level
],
5564 left_path
->slots
[left_level
]);
5565 right_gen
= btrfs_node_ptr_generation(
5566 right_path
->nodes
[right_level
],
5567 right_path
->slots
[right_level
]);
5568 if (left_blockptr
== right_blockptr
&&
5569 left_gen
== right_gen
) {
5571 * As we're on a shared block, don't
5572 * allow to go deeper.
5574 advance_left
= ADVANCE_ONLY_NEXT
;
5575 advance_right
= ADVANCE_ONLY_NEXT
;
5577 advance_left
= ADVANCE
;
5578 advance_right
= ADVANCE
;
5581 } else if (left_level
< right_level
) {
5582 advance_right
= ADVANCE
;
5584 advance_left
= ADVANCE
;
5589 btrfs_free_path(left_path
);
5590 btrfs_free_path(right_path
);
5596 * this is similar to btrfs_next_leaf, but does not try to preserve
5597 * and fixup the path. It looks for and returns the next key in the
5598 * tree based on the current path and the min_trans parameters.
5600 * 0 is returned if another key is found, < 0 if there are any errors
5601 * and 1 is returned if there are no higher keys in the tree
5603 * path->keep_locks should be set to 1 on the search made before
5604 * calling this function.
5606 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5607 struct btrfs_key
*key
, int level
, u64 min_trans
)
5610 struct extent_buffer
*c
;
5612 WARN_ON(!path
->keep_locks
);
5613 while (level
< BTRFS_MAX_LEVEL
) {
5614 if (!path
->nodes
[level
])
5617 slot
= path
->slots
[level
] + 1;
5618 c
= path
->nodes
[level
];
5620 if (slot
>= btrfs_header_nritems(c
)) {
5623 struct btrfs_key cur_key
;
5624 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5625 !path
->nodes
[level
+ 1])
5628 if (path
->locks
[level
+ 1]) {
5633 slot
= btrfs_header_nritems(c
) - 1;
5635 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5637 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5639 orig_lowest
= path
->lowest_level
;
5640 btrfs_release_path(path
);
5641 path
->lowest_level
= level
;
5642 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5644 path
->lowest_level
= orig_lowest
;
5648 c
= path
->nodes
[level
];
5649 slot
= path
->slots
[level
];
5656 btrfs_item_key_to_cpu(c
, key
, slot
);
5658 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5660 if (gen
< min_trans
) {
5664 btrfs_node_key_to_cpu(c
, key
, slot
);
5672 * search the tree again to find a leaf with greater keys
5673 * returns 0 if it found something or 1 if there are no greater leaves.
5674 * returns < 0 on io errors.
5676 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5678 return btrfs_next_old_leaf(root
, path
, 0);
5681 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5686 struct extent_buffer
*c
;
5687 struct extent_buffer
*next
;
5688 struct btrfs_key key
;
5691 int old_spinning
= path
->leave_spinning
;
5692 int next_rw_lock
= 0;
5694 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5698 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5703 btrfs_release_path(path
);
5705 path
->keep_locks
= 1;
5706 path
->leave_spinning
= 1;
5709 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5711 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5712 path
->keep_locks
= 0;
5717 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5719 * by releasing the path above we dropped all our locks. A balance
5720 * could have added more items next to the key that used to be
5721 * at the very end of the block. So, check again here and
5722 * advance the path if there are now more items available.
5724 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5731 * So the above check misses one case:
5732 * - after releasing the path above, someone has removed the item that
5733 * used to be at the very end of the block, and balance between leafs
5734 * gets another one with bigger key.offset to replace it.
5736 * This one should be returned as well, or we can get leaf corruption
5737 * later(esp. in __btrfs_drop_extents()).
5739 * And a bit more explanation about this check,
5740 * with ret > 0, the key isn't found, the path points to the slot
5741 * where it should be inserted, so the path->slots[0] item must be the
5744 if (nritems
> 0 && ret
> 0 && path
->slots
[0] == nritems
- 1) {
5749 while (level
< BTRFS_MAX_LEVEL
) {
5750 if (!path
->nodes
[level
]) {
5755 slot
= path
->slots
[level
] + 1;
5756 c
= path
->nodes
[level
];
5757 if (slot
>= btrfs_header_nritems(c
)) {
5759 if (level
== BTRFS_MAX_LEVEL
) {
5767 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5768 free_extent_buffer(next
);
5772 next_rw_lock
= path
->locks
[level
];
5773 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5779 btrfs_release_path(path
);
5783 if (!path
->skip_locking
) {
5784 ret
= btrfs_try_tree_read_lock(next
);
5785 if (!ret
&& time_seq
) {
5787 * If we don't get the lock, we may be racing
5788 * with push_leaf_left, holding that lock while
5789 * itself waiting for the leaf we've currently
5790 * locked. To solve this situation, we give up
5791 * on our lock and cycle.
5793 free_extent_buffer(next
);
5794 btrfs_release_path(path
);
5799 btrfs_set_path_blocking(path
);
5800 btrfs_tree_read_lock(next
);
5801 btrfs_clear_path_blocking(path
, next
,
5804 next_rw_lock
= BTRFS_READ_LOCK
;
5808 path
->slots
[level
] = slot
;
5811 c
= path
->nodes
[level
];
5812 if (path
->locks
[level
])
5813 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5815 free_extent_buffer(c
);
5816 path
->nodes
[level
] = next
;
5817 path
->slots
[level
] = 0;
5818 if (!path
->skip_locking
)
5819 path
->locks
[level
] = next_rw_lock
;
5823 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5829 btrfs_release_path(path
);
5833 if (!path
->skip_locking
) {
5834 ret
= btrfs_try_tree_read_lock(next
);
5836 btrfs_set_path_blocking(path
);
5837 btrfs_tree_read_lock(next
);
5838 btrfs_clear_path_blocking(path
, next
,
5841 next_rw_lock
= BTRFS_READ_LOCK
;
5846 unlock_up(path
, 0, 1, 0, NULL
);
5847 path
->leave_spinning
= old_spinning
;
5849 btrfs_set_path_blocking(path
);
5855 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5856 * searching until it gets past min_objectid or finds an item of 'type'
5858 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5860 int btrfs_previous_item(struct btrfs_root
*root
,
5861 struct btrfs_path
*path
, u64 min_objectid
,
5864 struct btrfs_key found_key
;
5865 struct extent_buffer
*leaf
;
5870 if (path
->slots
[0] == 0) {
5871 btrfs_set_path_blocking(path
);
5872 ret
= btrfs_prev_leaf(root
, path
);
5878 leaf
= path
->nodes
[0];
5879 nritems
= btrfs_header_nritems(leaf
);
5882 if (path
->slots
[0] == nritems
)
5885 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5886 if (found_key
.objectid
< min_objectid
)
5888 if (found_key
.type
== type
)
5890 if (found_key
.objectid
== min_objectid
&&
5891 found_key
.type
< type
)
5898 * search in extent tree to find a previous Metadata/Data extent item with
5901 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5903 int btrfs_previous_extent_item(struct btrfs_root
*root
,
5904 struct btrfs_path
*path
, u64 min_objectid
)
5906 struct btrfs_key found_key
;
5907 struct extent_buffer
*leaf
;
5912 if (path
->slots
[0] == 0) {
5913 btrfs_set_path_blocking(path
);
5914 ret
= btrfs_prev_leaf(root
, path
);
5920 leaf
= path
->nodes
[0];
5921 nritems
= btrfs_header_nritems(leaf
);
5924 if (path
->slots
[0] == nritems
)
5927 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5928 if (found_key
.objectid
< min_objectid
)
5930 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
||
5931 found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
5933 if (found_key
.objectid
== min_objectid
&&
5934 found_key
.type
< BTRFS_EXTENT_ITEM_KEY
)