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
, 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
, 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
, 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_is_testing(root
->fs_info
))
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).
1870 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1871 struct extent_buffer
*parent
, int slot
)
1873 int level
= btrfs_header_level(parent
);
1874 struct extent_buffer
*eb
;
1876 if (slot
< 0 || slot
>= btrfs_header_nritems(parent
))
1877 return ERR_PTR(-ENOENT
);
1881 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1882 btrfs_node_ptr_generation(parent
, slot
));
1883 if (!IS_ERR(eb
) && !extent_buffer_uptodate(eb
)) {
1884 free_extent_buffer(eb
);
1892 * node level balancing, used to make sure nodes are in proper order for
1893 * item deletion. We balance from the top down, so we have to make sure
1894 * that a deletion won't leave an node completely empty later on.
1896 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1897 struct btrfs_root
*root
,
1898 struct btrfs_path
*path
, int level
)
1900 struct extent_buffer
*right
= NULL
;
1901 struct extent_buffer
*mid
;
1902 struct extent_buffer
*left
= NULL
;
1903 struct extent_buffer
*parent
= NULL
;
1907 int orig_slot
= path
->slots
[level
];
1913 mid
= path
->nodes
[level
];
1915 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1916 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1917 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1919 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1921 if (level
< BTRFS_MAX_LEVEL
- 1) {
1922 parent
= path
->nodes
[level
+ 1];
1923 pslot
= path
->slots
[level
+ 1];
1927 * deal with the case where there is only one pointer in the root
1928 * by promoting the node below to a root
1931 struct extent_buffer
*child
;
1933 if (btrfs_header_nritems(mid
) != 1)
1936 /* promote the child to a root */
1937 child
= read_node_slot(root
, mid
, 0);
1938 if (IS_ERR(child
)) {
1939 ret
= PTR_ERR(child
);
1940 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
1944 btrfs_tree_lock(child
);
1945 btrfs_set_lock_blocking(child
);
1946 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1948 btrfs_tree_unlock(child
);
1949 free_extent_buffer(child
);
1953 tree_mod_log_set_root_pointer(root
, child
, 1);
1954 rcu_assign_pointer(root
->node
, child
);
1956 add_root_to_dirty_list(root
);
1957 btrfs_tree_unlock(child
);
1959 path
->locks
[level
] = 0;
1960 path
->nodes
[level
] = NULL
;
1961 clean_tree_block(trans
, root
->fs_info
, mid
);
1962 btrfs_tree_unlock(mid
);
1963 /* once for the path */
1964 free_extent_buffer(mid
);
1966 root_sub_used(root
, mid
->len
);
1967 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1968 /* once for the root ptr */
1969 free_extent_buffer_stale(mid
);
1972 if (btrfs_header_nritems(mid
) >
1973 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1976 left
= read_node_slot(root
, parent
, pslot
- 1);
1981 btrfs_tree_lock(left
);
1982 btrfs_set_lock_blocking(left
);
1983 wret
= btrfs_cow_block(trans
, root
, left
,
1984 parent
, pslot
- 1, &left
);
1991 right
= read_node_slot(root
, parent
, pslot
+ 1);
1996 btrfs_tree_lock(right
);
1997 btrfs_set_lock_blocking(right
);
1998 wret
= btrfs_cow_block(trans
, root
, right
,
1999 parent
, pslot
+ 1, &right
);
2006 /* first, try to make some room in the middle buffer */
2008 orig_slot
+= btrfs_header_nritems(left
);
2009 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2015 * then try to empty the right most buffer into the middle
2018 wret
= push_node_left(trans
, root
, mid
, right
, 1);
2019 if (wret
< 0 && wret
!= -ENOSPC
)
2021 if (btrfs_header_nritems(right
) == 0) {
2022 clean_tree_block(trans
, root
->fs_info
, right
);
2023 btrfs_tree_unlock(right
);
2024 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
2025 root_sub_used(root
, right
->len
);
2026 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
2027 free_extent_buffer_stale(right
);
2030 struct btrfs_disk_key right_key
;
2031 btrfs_node_key(right
, &right_key
, 0);
2032 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2034 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
2035 btrfs_mark_buffer_dirty(parent
);
2038 if (btrfs_header_nritems(mid
) == 1) {
2040 * we're not allowed to leave a node with one item in the
2041 * tree during a delete. A deletion from lower in the tree
2042 * could try to delete the only pointer in this node.
2043 * So, pull some keys from the left.
2044 * There has to be a left pointer at this point because
2045 * otherwise we would have pulled some pointers from the
2050 btrfs_handle_fs_error(root
->fs_info
, ret
, NULL
);
2053 wret
= balance_node_right(trans
, root
, mid
, left
);
2059 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2065 if (btrfs_header_nritems(mid
) == 0) {
2066 clean_tree_block(trans
, root
->fs_info
, mid
);
2067 btrfs_tree_unlock(mid
);
2068 del_ptr(root
, path
, level
+ 1, pslot
);
2069 root_sub_used(root
, mid
->len
);
2070 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
2071 free_extent_buffer_stale(mid
);
2074 /* update the parent key to reflect our changes */
2075 struct btrfs_disk_key mid_key
;
2076 btrfs_node_key(mid
, &mid_key
, 0);
2077 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2079 btrfs_set_node_key(parent
, &mid_key
, pslot
);
2080 btrfs_mark_buffer_dirty(parent
);
2083 /* update the path */
2085 if (btrfs_header_nritems(left
) > orig_slot
) {
2086 extent_buffer_get(left
);
2087 /* left was locked after cow */
2088 path
->nodes
[level
] = left
;
2089 path
->slots
[level
+ 1] -= 1;
2090 path
->slots
[level
] = orig_slot
;
2092 btrfs_tree_unlock(mid
);
2093 free_extent_buffer(mid
);
2096 orig_slot
-= btrfs_header_nritems(left
);
2097 path
->slots
[level
] = orig_slot
;
2100 /* double check we haven't messed things up */
2102 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2106 btrfs_tree_unlock(right
);
2107 free_extent_buffer(right
);
2110 if (path
->nodes
[level
] != left
)
2111 btrfs_tree_unlock(left
);
2112 free_extent_buffer(left
);
2117 /* Node balancing for insertion. Here we only split or push nodes around
2118 * when they are completely full. This is also done top down, so we
2119 * have to be pessimistic.
2121 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2122 struct btrfs_root
*root
,
2123 struct btrfs_path
*path
, int level
)
2125 struct extent_buffer
*right
= NULL
;
2126 struct extent_buffer
*mid
;
2127 struct extent_buffer
*left
= NULL
;
2128 struct extent_buffer
*parent
= NULL
;
2132 int orig_slot
= path
->slots
[level
];
2137 mid
= path
->nodes
[level
];
2138 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2140 if (level
< BTRFS_MAX_LEVEL
- 1) {
2141 parent
= path
->nodes
[level
+ 1];
2142 pslot
= path
->slots
[level
+ 1];
2148 left
= read_node_slot(root
, parent
, pslot
- 1);
2152 /* first, try to make some room in the middle buffer */
2156 btrfs_tree_lock(left
);
2157 btrfs_set_lock_blocking(left
);
2159 left_nr
= btrfs_header_nritems(left
);
2160 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2163 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2168 wret
= push_node_left(trans
, root
,
2175 struct btrfs_disk_key disk_key
;
2176 orig_slot
+= left_nr
;
2177 btrfs_node_key(mid
, &disk_key
, 0);
2178 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2180 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2181 btrfs_mark_buffer_dirty(parent
);
2182 if (btrfs_header_nritems(left
) > orig_slot
) {
2183 path
->nodes
[level
] = left
;
2184 path
->slots
[level
+ 1] -= 1;
2185 path
->slots
[level
] = orig_slot
;
2186 btrfs_tree_unlock(mid
);
2187 free_extent_buffer(mid
);
2190 btrfs_header_nritems(left
);
2191 path
->slots
[level
] = orig_slot
;
2192 btrfs_tree_unlock(left
);
2193 free_extent_buffer(left
);
2197 btrfs_tree_unlock(left
);
2198 free_extent_buffer(left
);
2200 right
= read_node_slot(root
, parent
, pslot
+ 1);
2205 * then try to empty the right most buffer into the middle
2210 btrfs_tree_lock(right
);
2211 btrfs_set_lock_blocking(right
);
2213 right_nr
= btrfs_header_nritems(right
);
2214 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2217 ret
= btrfs_cow_block(trans
, root
, right
,
2223 wret
= balance_node_right(trans
, root
,
2230 struct btrfs_disk_key disk_key
;
2232 btrfs_node_key(right
, &disk_key
, 0);
2233 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2235 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2236 btrfs_mark_buffer_dirty(parent
);
2238 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2239 path
->nodes
[level
] = right
;
2240 path
->slots
[level
+ 1] += 1;
2241 path
->slots
[level
] = orig_slot
-
2242 btrfs_header_nritems(mid
);
2243 btrfs_tree_unlock(mid
);
2244 free_extent_buffer(mid
);
2246 btrfs_tree_unlock(right
);
2247 free_extent_buffer(right
);
2251 btrfs_tree_unlock(right
);
2252 free_extent_buffer(right
);
2258 * readahead one full node of leaves, finding things that are close
2259 * to the block in 'slot', and triggering ra on them.
2261 static void reada_for_search(struct btrfs_root
*root
,
2262 struct btrfs_path
*path
,
2263 int level
, int slot
, u64 objectid
)
2265 struct extent_buffer
*node
;
2266 struct btrfs_disk_key disk_key
;
2271 struct extent_buffer
*eb
;
2279 if (!path
->nodes
[level
])
2282 node
= path
->nodes
[level
];
2284 search
= btrfs_node_blockptr(node
, slot
);
2285 blocksize
= root
->nodesize
;
2286 eb
= btrfs_find_tree_block(root
->fs_info
, search
);
2288 free_extent_buffer(eb
);
2294 nritems
= btrfs_header_nritems(node
);
2298 if (path
->reada
== READA_BACK
) {
2302 } else if (path
->reada
== READA_FORWARD
) {
2307 if (path
->reada
== READA_BACK
&& objectid
) {
2308 btrfs_node_key(node
, &disk_key
, nr
);
2309 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2312 search
= btrfs_node_blockptr(node
, nr
);
2313 if ((search
<= target
&& target
- search
<= 65536) ||
2314 (search
> target
&& search
- target
<= 65536)) {
2315 readahead_tree_block(root
, search
);
2319 if ((nread
> 65536 || nscan
> 32))
2324 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2325 struct btrfs_path
*path
, int level
)
2329 struct extent_buffer
*parent
;
2330 struct extent_buffer
*eb
;
2335 parent
= path
->nodes
[level
+ 1];
2339 nritems
= btrfs_header_nritems(parent
);
2340 slot
= path
->slots
[level
+ 1];
2343 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2344 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2345 eb
= btrfs_find_tree_block(root
->fs_info
, block1
);
2347 * if we get -eagain from btrfs_buffer_uptodate, we
2348 * don't want to return eagain here. That will loop
2351 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2353 free_extent_buffer(eb
);
2355 if (slot
+ 1 < nritems
) {
2356 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2357 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2358 eb
= btrfs_find_tree_block(root
->fs_info
, block2
);
2359 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2361 free_extent_buffer(eb
);
2365 readahead_tree_block(root
, block1
);
2367 readahead_tree_block(root
, block2
);
2372 * when we walk down the tree, it is usually safe to unlock the higher layers
2373 * in the tree. The exceptions are when our path goes through slot 0, because
2374 * operations on the tree might require changing key pointers higher up in the
2377 * callers might also have set path->keep_locks, which tells this code to keep
2378 * the lock if the path points to the last slot in the block. This is part of
2379 * walking through the tree, and selecting the next slot in the higher block.
2381 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2382 * if lowest_unlock is 1, level 0 won't be unlocked
2384 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2385 int lowest_unlock
, int min_write_lock_level
,
2386 int *write_lock_level
)
2389 int skip_level
= level
;
2391 struct extent_buffer
*t
;
2393 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2394 if (!path
->nodes
[i
])
2396 if (!path
->locks
[i
])
2398 if (!no_skips
&& path
->slots
[i
] == 0) {
2402 if (!no_skips
&& path
->keep_locks
) {
2405 nritems
= btrfs_header_nritems(t
);
2406 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2411 if (skip_level
< i
&& i
>= lowest_unlock
)
2415 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2416 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2418 if (write_lock_level
&&
2419 i
> min_write_lock_level
&&
2420 i
<= *write_lock_level
) {
2421 *write_lock_level
= i
- 1;
2428 * This releases any locks held in the path starting at level and
2429 * going all the way up to the root.
2431 * btrfs_search_slot will keep the lock held on higher nodes in a few
2432 * corner cases, such as COW of the block at slot zero in the node. This
2433 * ignores those rules, and it should only be called when there are no
2434 * more updates to be done higher up in the tree.
2436 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2440 if (path
->keep_locks
)
2443 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2444 if (!path
->nodes
[i
])
2446 if (!path
->locks
[i
])
2448 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2454 * helper function for btrfs_search_slot. The goal is to find a block
2455 * in cache without setting the path to blocking. If we find the block
2456 * we return zero and the path is unchanged.
2458 * If we can't find the block, we set the path blocking and do some
2459 * reada. -EAGAIN is returned and the search must be repeated.
2462 read_block_for_search(struct btrfs_trans_handle
*trans
,
2463 struct btrfs_root
*root
, struct btrfs_path
*p
,
2464 struct extent_buffer
**eb_ret
, int level
, int slot
,
2465 struct btrfs_key
*key
, u64 time_seq
)
2469 struct extent_buffer
*b
= *eb_ret
;
2470 struct extent_buffer
*tmp
;
2473 blocknr
= btrfs_node_blockptr(b
, slot
);
2474 gen
= btrfs_node_ptr_generation(b
, slot
);
2476 tmp
= btrfs_find_tree_block(root
->fs_info
, blocknr
);
2478 /* first we do an atomic uptodate check */
2479 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2484 /* the pages were up to date, but we failed
2485 * the generation number check. Do a full
2486 * read for the generation number that is correct.
2487 * We must do this without dropping locks so
2488 * we can trust our generation number
2490 btrfs_set_path_blocking(p
);
2492 /* now we're allowed to do a blocking uptodate check */
2493 ret
= btrfs_read_buffer(tmp
, gen
);
2498 free_extent_buffer(tmp
);
2499 btrfs_release_path(p
);
2504 * reduce lock contention at high levels
2505 * of the btree by dropping locks before
2506 * we read. Don't release the lock on the current
2507 * level because we need to walk this node to figure
2508 * out which blocks to read.
2510 btrfs_unlock_up_safe(p
, level
+ 1);
2511 btrfs_set_path_blocking(p
);
2513 free_extent_buffer(tmp
);
2514 if (p
->reada
!= READA_NONE
)
2515 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2517 btrfs_release_path(p
);
2520 tmp
= read_tree_block(root
, blocknr
, 0);
2523 * If the read above didn't mark this buffer up to date,
2524 * it will never end up being up to date. Set ret to EIO now
2525 * and give up so that our caller doesn't loop forever
2528 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2530 free_extent_buffer(tmp
);
2538 * helper function for btrfs_search_slot. This does all of the checks
2539 * for node-level blocks and does any balancing required based on
2542 * If no extra work was required, zero is returned. If we had to
2543 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2547 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2548 struct btrfs_root
*root
, struct btrfs_path
*p
,
2549 struct extent_buffer
*b
, int level
, int ins_len
,
2550 int *write_lock_level
)
2553 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2554 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2557 if (*write_lock_level
< level
+ 1) {
2558 *write_lock_level
= level
+ 1;
2559 btrfs_release_path(p
);
2563 btrfs_set_path_blocking(p
);
2564 reada_for_balance(root
, p
, level
);
2565 sret
= split_node(trans
, root
, p
, level
);
2566 btrfs_clear_path_blocking(p
, NULL
, 0);
2573 b
= p
->nodes
[level
];
2574 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2575 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2578 if (*write_lock_level
< level
+ 1) {
2579 *write_lock_level
= level
+ 1;
2580 btrfs_release_path(p
);
2584 btrfs_set_path_blocking(p
);
2585 reada_for_balance(root
, p
, level
);
2586 sret
= balance_level(trans
, root
, p
, level
);
2587 btrfs_clear_path_blocking(p
, NULL
, 0);
2593 b
= p
->nodes
[level
];
2595 btrfs_release_path(p
);
2598 BUG_ON(btrfs_header_nritems(b
) == 1);
2608 static void key_search_validate(struct extent_buffer
*b
,
2609 struct btrfs_key
*key
,
2612 #ifdef CONFIG_BTRFS_ASSERT
2613 struct btrfs_disk_key disk_key
;
2615 btrfs_cpu_key_to_disk(&disk_key
, key
);
2618 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2619 offsetof(struct btrfs_leaf
, items
[0].key
),
2622 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2623 offsetof(struct btrfs_node
, ptrs
[0].key
),
2628 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2629 int level
, int *prev_cmp
, int *slot
)
2631 if (*prev_cmp
!= 0) {
2632 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2636 key_search_validate(b
, key
, level
);
2642 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*path
,
2643 u64 iobjectid
, u64 ioff
, u8 key_type
,
2644 struct btrfs_key
*found_key
)
2647 struct btrfs_key key
;
2648 struct extent_buffer
*eb
;
2653 key
.type
= key_type
;
2654 key
.objectid
= iobjectid
;
2657 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2661 eb
= path
->nodes
[0];
2662 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2663 ret
= btrfs_next_leaf(fs_root
, path
);
2666 eb
= path
->nodes
[0];
2669 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2670 if (found_key
->type
!= key
.type
||
2671 found_key
->objectid
!= key
.objectid
)
2678 * look for key in the tree. path is filled in with nodes along the way
2679 * if key is found, we return zero and you can find the item in the leaf
2680 * level of the path (level 0)
2682 * If the key isn't found, the path points to the slot where it should
2683 * be inserted, and 1 is returned. If there are other errors during the
2684 * search a negative error number is returned.
2686 * if ins_len > 0, nodes and leaves will be split as we walk down the
2687 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2690 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2691 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2694 struct extent_buffer
*b
;
2699 int lowest_unlock
= 1;
2701 /* everything at write_lock_level or lower must be write locked */
2702 int write_lock_level
= 0;
2703 u8 lowest_level
= 0;
2704 int min_write_lock_level
;
2707 lowest_level
= p
->lowest_level
;
2708 WARN_ON(lowest_level
&& ins_len
> 0);
2709 WARN_ON(p
->nodes
[0] != NULL
);
2710 BUG_ON(!cow
&& ins_len
);
2715 /* when we are removing items, we might have to go up to level
2716 * two as we update tree pointers Make sure we keep write
2717 * for those levels as well
2719 write_lock_level
= 2;
2720 } else if (ins_len
> 0) {
2722 * for inserting items, make sure we have a write lock on
2723 * level 1 so we can update keys
2725 write_lock_level
= 1;
2729 write_lock_level
= -1;
2731 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2732 write_lock_level
= BTRFS_MAX_LEVEL
;
2734 min_write_lock_level
= write_lock_level
;
2739 * we try very hard to do read locks on the root
2741 root_lock
= BTRFS_READ_LOCK
;
2743 if (p
->search_commit_root
) {
2745 * the commit roots are read only
2746 * so we always do read locks
2748 if (p
->need_commit_sem
)
2749 down_read(&root
->fs_info
->commit_root_sem
);
2750 b
= root
->commit_root
;
2751 extent_buffer_get(b
);
2752 level
= btrfs_header_level(b
);
2753 if (p
->need_commit_sem
)
2754 up_read(&root
->fs_info
->commit_root_sem
);
2755 if (!p
->skip_locking
)
2756 btrfs_tree_read_lock(b
);
2758 if (p
->skip_locking
) {
2759 b
= btrfs_root_node(root
);
2760 level
= btrfs_header_level(b
);
2762 /* we don't know the level of the root node
2763 * until we actually have it read locked
2765 b
= btrfs_read_lock_root_node(root
);
2766 level
= btrfs_header_level(b
);
2767 if (level
<= write_lock_level
) {
2768 /* whoops, must trade for write lock */
2769 btrfs_tree_read_unlock(b
);
2770 free_extent_buffer(b
);
2771 b
= btrfs_lock_root_node(root
);
2772 root_lock
= BTRFS_WRITE_LOCK
;
2774 /* the level might have changed, check again */
2775 level
= btrfs_header_level(b
);
2779 p
->nodes
[level
] = b
;
2780 if (!p
->skip_locking
)
2781 p
->locks
[level
] = root_lock
;
2784 level
= btrfs_header_level(b
);
2787 * setup the path here so we can release it under lock
2788 * contention with the cow code
2792 * if we don't really need to cow this block
2793 * then we don't want to set the path blocking,
2794 * so we test it here
2796 if (!should_cow_block(trans
, root
, b
)) {
2797 trans
->dirty
= true;
2802 * must have write locks on this node and the
2805 if (level
> write_lock_level
||
2806 (level
+ 1 > write_lock_level
&&
2807 level
+ 1 < BTRFS_MAX_LEVEL
&&
2808 p
->nodes
[level
+ 1])) {
2809 write_lock_level
= level
+ 1;
2810 btrfs_release_path(p
);
2814 btrfs_set_path_blocking(p
);
2815 err
= btrfs_cow_block(trans
, root
, b
,
2816 p
->nodes
[level
+ 1],
2817 p
->slots
[level
+ 1], &b
);
2824 p
->nodes
[level
] = b
;
2825 btrfs_clear_path_blocking(p
, NULL
, 0);
2828 * we have a lock on b and as long as we aren't changing
2829 * the tree, there is no way to for the items in b to change.
2830 * It is safe to drop the lock on our parent before we
2831 * go through the expensive btree search on b.
2833 * If we're inserting or deleting (ins_len != 0), then we might
2834 * be changing slot zero, which may require changing the parent.
2835 * So, we can't drop the lock until after we know which slot
2836 * we're operating on.
2838 if (!ins_len
&& !p
->keep_locks
) {
2841 if (u
< BTRFS_MAX_LEVEL
&& p
->locks
[u
]) {
2842 btrfs_tree_unlock_rw(p
->nodes
[u
], p
->locks
[u
]);
2847 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2853 if (ret
&& slot
> 0) {
2857 p
->slots
[level
] = slot
;
2858 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2859 ins_len
, &write_lock_level
);
2866 b
= p
->nodes
[level
];
2867 slot
= p
->slots
[level
];
2870 * slot 0 is special, if we change the key
2871 * we have to update the parent pointer
2872 * which means we must have a write lock
2875 if (slot
== 0 && ins_len
&&
2876 write_lock_level
< level
+ 1) {
2877 write_lock_level
= level
+ 1;
2878 btrfs_release_path(p
);
2882 unlock_up(p
, level
, lowest_unlock
,
2883 min_write_lock_level
, &write_lock_level
);
2885 if (level
== lowest_level
) {
2891 err
= read_block_for_search(trans
, root
, p
,
2892 &b
, level
, slot
, key
, 0);
2900 if (!p
->skip_locking
) {
2901 level
= btrfs_header_level(b
);
2902 if (level
<= write_lock_level
) {
2903 err
= btrfs_try_tree_write_lock(b
);
2905 btrfs_set_path_blocking(p
);
2907 btrfs_clear_path_blocking(p
, b
,
2910 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2912 err
= btrfs_tree_read_lock_atomic(b
);
2914 btrfs_set_path_blocking(p
);
2915 btrfs_tree_read_lock(b
);
2916 btrfs_clear_path_blocking(p
, b
,
2919 p
->locks
[level
] = BTRFS_READ_LOCK
;
2921 p
->nodes
[level
] = b
;
2924 p
->slots
[level
] = slot
;
2926 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2927 if (write_lock_level
< 1) {
2928 write_lock_level
= 1;
2929 btrfs_release_path(p
);
2933 btrfs_set_path_blocking(p
);
2934 err
= split_leaf(trans
, root
, key
,
2935 p
, ins_len
, ret
== 0);
2936 btrfs_clear_path_blocking(p
, NULL
, 0);
2944 if (!p
->search_for_split
)
2945 unlock_up(p
, level
, lowest_unlock
,
2946 min_write_lock_level
, &write_lock_level
);
2953 * we don't really know what they plan on doing with the path
2954 * from here on, so for now just mark it as blocking
2956 if (!p
->leave_spinning
)
2957 btrfs_set_path_blocking(p
);
2958 if (ret
< 0 && !p
->skip_release_on_error
)
2959 btrfs_release_path(p
);
2964 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2965 * current state of the tree together with the operations recorded in the tree
2966 * modification log to search for the key in a previous version of this tree, as
2967 * denoted by the time_seq parameter.
2969 * Naturally, there is no support for insert, delete or cow operations.
2971 * The resulting path and return value will be set up as if we called
2972 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2974 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2975 struct btrfs_path
*p
, u64 time_seq
)
2977 struct extent_buffer
*b
;
2982 int lowest_unlock
= 1;
2983 u8 lowest_level
= 0;
2986 lowest_level
= p
->lowest_level
;
2987 WARN_ON(p
->nodes
[0] != NULL
);
2989 if (p
->search_commit_root
) {
2991 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2995 b
= get_old_root(root
, time_seq
);
2996 level
= btrfs_header_level(b
);
2997 p
->locks
[level
] = BTRFS_READ_LOCK
;
3000 level
= btrfs_header_level(b
);
3001 p
->nodes
[level
] = b
;
3002 btrfs_clear_path_blocking(p
, NULL
, 0);
3005 * we have a lock on b and as long as we aren't changing
3006 * the tree, there is no way to for the items in b to change.
3007 * It is safe to drop the lock on our parent before we
3008 * go through the expensive btree search on b.
3010 btrfs_unlock_up_safe(p
, level
+ 1);
3013 * Since we can unwind ebs we want to do a real search every
3017 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
3021 if (ret
&& slot
> 0) {
3025 p
->slots
[level
] = slot
;
3026 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3028 if (level
== lowest_level
) {
3034 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
3035 slot
, key
, time_seq
);
3043 level
= btrfs_header_level(b
);
3044 err
= btrfs_tree_read_lock_atomic(b
);
3046 btrfs_set_path_blocking(p
);
3047 btrfs_tree_read_lock(b
);
3048 btrfs_clear_path_blocking(p
, b
,
3051 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
3056 p
->locks
[level
] = BTRFS_READ_LOCK
;
3057 p
->nodes
[level
] = b
;
3059 p
->slots
[level
] = slot
;
3060 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3066 if (!p
->leave_spinning
)
3067 btrfs_set_path_blocking(p
);
3069 btrfs_release_path(p
);
3075 * helper to use instead of search slot if no exact match is needed but
3076 * instead the next or previous item should be returned.
3077 * When find_higher is true, the next higher item is returned, the next lower
3079 * When return_any and find_higher are both true, and no higher item is found,
3080 * return the next lower instead.
3081 * When return_any is true and find_higher is false, and no lower item is found,
3082 * return the next higher instead.
3083 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3086 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3087 struct btrfs_key
*key
, struct btrfs_path
*p
,
3088 int find_higher
, int return_any
)
3091 struct extent_buffer
*leaf
;
3094 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3098 * a return value of 1 means the path is at the position where the
3099 * item should be inserted. Normally this is the next bigger item,
3100 * but in case the previous item is the last in a leaf, path points
3101 * to the first free slot in the previous leaf, i.e. at an invalid
3107 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3108 ret
= btrfs_next_leaf(root
, p
);
3114 * no higher item found, return the next
3119 btrfs_release_path(p
);
3123 if (p
->slots
[0] == 0) {
3124 ret
= btrfs_prev_leaf(root
, p
);
3129 if (p
->slots
[0] == btrfs_header_nritems(leaf
))
3136 * no lower item found, return the next
3141 btrfs_release_path(p
);
3151 * adjust the pointers going up the tree, starting at level
3152 * making sure the right key of each node is points to 'key'.
3153 * This is used after shifting pointers to the left, so it stops
3154 * fixing up pointers when a given leaf/node is not in slot 0 of the
3158 static void fixup_low_keys(struct btrfs_fs_info
*fs_info
,
3159 struct btrfs_path
*path
,
3160 struct btrfs_disk_key
*key
, int level
)
3163 struct extent_buffer
*t
;
3165 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3166 int tslot
= path
->slots
[i
];
3167 if (!path
->nodes
[i
])
3170 tree_mod_log_set_node_key(fs_info
, t
, tslot
, 1);
3171 btrfs_set_node_key(t
, key
, tslot
);
3172 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3181 * This function isn't completely safe. It's the caller's responsibility
3182 * that the new key won't break the order
3184 void btrfs_set_item_key_safe(struct btrfs_fs_info
*fs_info
,
3185 struct btrfs_path
*path
,
3186 struct btrfs_key
*new_key
)
3188 struct btrfs_disk_key disk_key
;
3189 struct extent_buffer
*eb
;
3192 eb
= path
->nodes
[0];
3193 slot
= path
->slots
[0];
3195 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3196 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
3198 if (slot
< btrfs_header_nritems(eb
) - 1) {
3199 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3200 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
3203 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3204 btrfs_set_item_key(eb
, &disk_key
, slot
);
3205 btrfs_mark_buffer_dirty(eb
);
3207 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
3211 * try to push data from one node into the next node left in the
3214 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3215 * error, and > 0 if there was no room in the left hand block.
3217 static int push_node_left(struct btrfs_trans_handle
*trans
,
3218 struct btrfs_root
*root
, struct extent_buffer
*dst
,
3219 struct extent_buffer
*src
, int empty
)
3226 src_nritems
= btrfs_header_nritems(src
);
3227 dst_nritems
= btrfs_header_nritems(dst
);
3228 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3229 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3230 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3232 if (!empty
&& src_nritems
<= 8)
3235 if (push_items
<= 0)
3239 push_items
= min(src_nritems
, push_items
);
3240 if (push_items
< src_nritems
) {
3241 /* leave at least 8 pointers in the node if
3242 * we aren't going to empty it
3244 if (src_nritems
- push_items
< 8) {
3245 if (push_items
<= 8)
3251 push_items
= min(src_nritems
- 8, push_items
);
3253 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3256 btrfs_abort_transaction(trans
, ret
);
3259 copy_extent_buffer(dst
, src
,
3260 btrfs_node_key_ptr_offset(dst_nritems
),
3261 btrfs_node_key_ptr_offset(0),
3262 push_items
* sizeof(struct btrfs_key_ptr
));
3264 if (push_items
< src_nritems
) {
3266 * don't call tree_mod_log_eb_move here, key removal was already
3267 * fully logged by tree_mod_log_eb_copy above.
3269 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3270 btrfs_node_key_ptr_offset(push_items
),
3271 (src_nritems
- push_items
) *
3272 sizeof(struct btrfs_key_ptr
));
3274 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3275 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3276 btrfs_mark_buffer_dirty(src
);
3277 btrfs_mark_buffer_dirty(dst
);
3283 * try to push data from one node into the next node right in the
3286 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3287 * error, and > 0 if there was no room in the right hand block.
3289 * this will only push up to 1/2 the contents of the left node over
3291 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3292 struct btrfs_root
*root
,
3293 struct extent_buffer
*dst
,
3294 struct extent_buffer
*src
)
3302 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3303 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3305 src_nritems
= btrfs_header_nritems(src
);
3306 dst_nritems
= btrfs_header_nritems(dst
);
3307 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3308 if (push_items
<= 0)
3311 if (src_nritems
< 4)
3314 max_push
= src_nritems
/ 2 + 1;
3315 /* don't try to empty the node */
3316 if (max_push
>= src_nritems
)
3319 if (max_push
< push_items
)
3320 push_items
= max_push
;
3322 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3323 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3324 btrfs_node_key_ptr_offset(0),
3326 sizeof(struct btrfs_key_ptr
));
3328 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3329 src_nritems
- push_items
, push_items
);
3331 btrfs_abort_transaction(trans
, ret
);
3334 copy_extent_buffer(dst
, src
,
3335 btrfs_node_key_ptr_offset(0),
3336 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3337 push_items
* sizeof(struct btrfs_key_ptr
));
3339 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3340 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3342 btrfs_mark_buffer_dirty(src
);
3343 btrfs_mark_buffer_dirty(dst
);
3349 * helper function to insert a new root level in the tree.
3350 * A new node is allocated, and a single item is inserted to
3351 * point to the existing root
3353 * returns zero on success or < 0 on failure.
3355 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3356 struct btrfs_root
*root
,
3357 struct btrfs_path
*path
, int level
)
3360 struct extent_buffer
*lower
;
3361 struct extent_buffer
*c
;
3362 struct extent_buffer
*old
;
3363 struct btrfs_disk_key lower_key
;
3365 BUG_ON(path
->nodes
[level
]);
3366 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3368 lower
= path
->nodes
[level
-1];
3370 btrfs_item_key(lower
, &lower_key
, 0);
3372 btrfs_node_key(lower
, &lower_key
, 0);
3374 c
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3375 &lower_key
, level
, root
->node
->start
, 0);
3379 root_add_used(root
, root
->nodesize
);
3381 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3382 btrfs_set_header_nritems(c
, 1);
3383 btrfs_set_header_level(c
, level
);
3384 btrfs_set_header_bytenr(c
, c
->start
);
3385 btrfs_set_header_generation(c
, trans
->transid
);
3386 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3387 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3389 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3392 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3393 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3395 btrfs_set_node_key(c
, &lower_key
, 0);
3396 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3397 lower_gen
= btrfs_header_generation(lower
);
3398 WARN_ON(lower_gen
!= trans
->transid
);
3400 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3402 btrfs_mark_buffer_dirty(c
);
3405 tree_mod_log_set_root_pointer(root
, c
, 0);
3406 rcu_assign_pointer(root
->node
, c
);
3408 /* the super has an extra ref to root->node */
3409 free_extent_buffer(old
);
3411 add_root_to_dirty_list(root
);
3412 extent_buffer_get(c
);
3413 path
->nodes
[level
] = c
;
3414 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
3415 path
->slots
[level
] = 0;
3420 * worker function to insert a single pointer in a node.
3421 * the node should have enough room for the pointer already
3423 * slot and level indicate where you want the key to go, and
3424 * blocknr is the block the key points to.
3426 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3427 struct btrfs_root
*root
, struct btrfs_path
*path
,
3428 struct btrfs_disk_key
*key
, u64 bytenr
,
3429 int slot
, int level
)
3431 struct extent_buffer
*lower
;
3435 BUG_ON(!path
->nodes
[level
]);
3436 btrfs_assert_tree_locked(path
->nodes
[level
]);
3437 lower
= path
->nodes
[level
];
3438 nritems
= btrfs_header_nritems(lower
);
3439 BUG_ON(slot
> nritems
);
3440 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3441 if (slot
!= nritems
) {
3443 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3444 slot
, nritems
- slot
);
3445 memmove_extent_buffer(lower
,
3446 btrfs_node_key_ptr_offset(slot
+ 1),
3447 btrfs_node_key_ptr_offset(slot
),
3448 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3451 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3452 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3455 btrfs_set_node_key(lower
, key
, slot
);
3456 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3457 WARN_ON(trans
->transid
== 0);
3458 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3459 btrfs_set_header_nritems(lower
, nritems
+ 1);
3460 btrfs_mark_buffer_dirty(lower
);
3464 * split the node at the specified level in path in two.
3465 * The path is corrected to point to the appropriate node after the split
3467 * Before splitting this tries to make some room in the node by pushing
3468 * left and right, if either one works, it returns right away.
3470 * returns 0 on success and < 0 on failure
3472 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3473 struct btrfs_root
*root
,
3474 struct btrfs_path
*path
, int level
)
3476 struct extent_buffer
*c
;
3477 struct extent_buffer
*split
;
3478 struct btrfs_disk_key disk_key
;
3483 c
= path
->nodes
[level
];
3484 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3485 if (c
== root
->node
) {
3487 * trying to split the root, lets make a new one
3489 * tree mod log: We don't log_removal old root in
3490 * insert_new_root, because that root buffer will be kept as a
3491 * normal node. We are going to log removal of half of the
3492 * elements below with tree_mod_log_eb_copy. We're holding a
3493 * tree lock on the buffer, which is why we cannot race with
3494 * other tree_mod_log users.
3496 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3500 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3501 c
= path
->nodes
[level
];
3502 if (!ret
&& btrfs_header_nritems(c
) <
3503 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3509 c_nritems
= btrfs_header_nritems(c
);
3510 mid
= (c_nritems
+ 1) / 2;
3511 btrfs_node_key(c
, &disk_key
, mid
);
3513 split
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
3514 &disk_key
, level
, c
->start
, 0);
3516 return PTR_ERR(split
);
3518 root_add_used(root
, root
->nodesize
);
3520 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3521 btrfs_set_header_level(split
, btrfs_header_level(c
));
3522 btrfs_set_header_bytenr(split
, split
->start
);
3523 btrfs_set_header_generation(split
, trans
->transid
);
3524 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3525 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3526 write_extent_buffer(split
, root
->fs_info
->fsid
,
3527 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3528 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3529 btrfs_header_chunk_tree_uuid(split
),
3532 ret
= tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0,
3533 mid
, c_nritems
- mid
);
3535 btrfs_abort_transaction(trans
, ret
);
3538 copy_extent_buffer(split
, c
,
3539 btrfs_node_key_ptr_offset(0),
3540 btrfs_node_key_ptr_offset(mid
),
3541 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3542 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3543 btrfs_set_header_nritems(c
, mid
);
3546 btrfs_mark_buffer_dirty(c
);
3547 btrfs_mark_buffer_dirty(split
);
3549 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3550 path
->slots
[level
+ 1] + 1, level
+ 1);
3552 if (path
->slots
[level
] >= mid
) {
3553 path
->slots
[level
] -= mid
;
3554 btrfs_tree_unlock(c
);
3555 free_extent_buffer(c
);
3556 path
->nodes
[level
] = split
;
3557 path
->slots
[level
+ 1] += 1;
3559 btrfs_tree_unlock(split
);
3560 free_extent_buffer(split
);
3566 * how many bytes are required to store the items in a leaf. start
3567 * and nr indicate which items in the leaf to check. This totals up the
3568 * space used both by the item structs and the item data
3570 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3572 struct btrfs_item
*start_item
;
3573 struct btrfs_item
*end_item
;
3574 struct btrfs_map_token token
;
3576 int nritems
= btrfs_header_nritems(l
);
3577 int end
= min(nritems
, start
+ nr
) - 1;
3581 btrfs_init_map_token(&token
);
3582 start_item
= btrfs_item_nr(start
);
3583 end_item
= btrfs_item_nr(end
);
3584 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3585 btrfs_token_item_size(l
, start_item
, &token
);
3586 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3587 data_len
+= sizeof(struct btrfs_item
) * nr
;
3588 WARN_ON(data_len
< 0);
3593 * The space between the end of the leaf items and
3594 * the start of the leaf data. IOW, how much room
3595 * the leaf has left for both items and data
3597 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3598 struct extent_buffer
*leaf
)
3600 int nritems
= btrfs_header_nritems(leaf
);
3602 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3604 btrfs_crit(root
->fs_info
,
3605 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3606 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3607 leaf_space_used(leaf
, 0, nritems
), nritems
);
3613 * min slot controls the lowest index we're willing to push to the
3614 * right. We'll push up to and including min_slot, but no lower
3616 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3617 struct btrfs_root
*root
,
3618 struct btrfs_path
*path
,
3619 int data_size
, int empty
,
3620 struct extent_buffer
*right
,
3621 int free_space
, u32 left_nritems
,
3624 struct extent_buffer
*left
= path
->nodes
[0];
3625 struct extent_buffer
*upper
= path
->nodes
[1];
3626 struct btrfs_map_token token
;
3627 struct btrfs_disk_key disk_key
;
3632 struct btrfs_item
*item
;
3638 btrfs_init_map_token(&token
);
3643 nr
= max_t(u32
, 1, min_slot
);
3645 if (path
->slots
[0] >= left_nritems
)
3646 push_space
+= data_size
;
3648 slot
= path
->slots
[1];
3649 i
= left_nritems
- 1;
3651 item
= btrfs_item_nr(i
);
3653 if (!empty
&& push_items
> 0) {
3654 if (path
->slots
[0] > i
)
3656 if (path
->slots
[0] == i
) {
3657 int space
= btrfs_leaf_free_space(root
, left
);
3658 if (space
+ push_space
* 2 > free_space
)
3663 if (path
->slots
[0] == i
)
3664 push_space
+= data_size
;
3666 this_item_size
= btrfs_item_size(left
, item
);
3667 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3671 push_space
+= this_item_size
+ sizeof(*item
);
3677 if (push_items
== 0)
3680 WARN_ON(!empty
&& push_items
== left_nritems
);
3682 /* push left to right */
3683 right_nritems
= btrfs_header_nritems(right
);
3685 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3686 push_space
-= leaf_data_end(root
, left
);
3688 /* make room in the right data area */
3689 data_end
= leaf_data_end(root
, right
);
3690 memmove_extent_buffer(right
,
3691 btrfs_leaf_data(right
) + data_end
- push_space
,
3692 btrfs_leaf_data(right
) + data_end
,
3693 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3695 /* copy from the left data area */
3696 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3697 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3698 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3701 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3702 btrfs_item_nr_offset(0),
3703 right_nritems
* sizeof(struct btrfs_item
));
3705 /* copy the items from left to right */
3706 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3707 btrfs_item_nr_offset(left_nritems
- push_items
),
3708 push_items
* sizeof(struct btrfs_item
));
3710 /* update the item pointers */
3711 right_nritems
+= push_items
;
3712 btrfs_set_header_nritems(right
, right_nritems
);
3713 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3714 for (i
= 0; i
< right_nritems
; i
++) {
3715 item
= btrfs_item_nr(i
);
3716 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3717 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3720 left_nritems
-= push_items
;
3721 btrfs_set_header_nritems(left
, left_nritems
);
3724 btrfs_mark_buffer_dirty(left
);
3726 clean_tree_block(trans
, root
->fs_info
, left
);
3728 btrfs_mark_buffer_dirty(right
);
3730 btrfs_item_key(right
, &disk_key
, 0);
3731 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3732 btrfs_mark_buffer_dirty(upper
);
3734 /* then fixup the leaf pointer in the path */
3735 if (path
->slots
[0] >= left_nritems
) {
3736 path
->slots
[0] -= left_nritems
;
3737 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3738 clean_tree_block(trans
, root
->fs_info
, path
->nodes
[0]);
3739 btrfs_tree_unlock(path
->nodes
[0]);
3740 free_extent_buffer(path
->nodes
[0]);
3741 path
->nodes
[0] = right
;
3742 path
->slots
[1] += 1;
3744 btrfs_tree_unlock(right
);
3745 free_extent_buffer(right
);
3750 btrfs_tree_unlock(right
);
3751 free_extent_buffer(right
);
3756 * push some data in the path leaf to the right, trying to free up at
3757 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3759 * returns 1 if the push failed because the other node didn't have enough
3760 * room, 0 if everything worked out and < 0 if there were major errors.
3762 * this will push starting from min_slot to the end of the leaf. It won't
3763 * push any slot lower than min_slot
3765 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3766 *root
, struct btrfs_path
*path
,
3767 int min_data_size
, int data_size
,
3768 int empty
, u32 min_slot
)
3770 struct extent_buffer
*left
= path
->nodes
[0];
3771 struct extent_buffer
*right
;
3772 struct extent_buffer
*upper
;
3778 if (!path
->nodes
[1])
3781 slot
= path
->slots
[1];
3782 upper
= path
->nodes
[1];
3783 if (slot
>= btrfs_header_nritems(upper
) - 1)
3786 btrfs_assert_tree_locked(path
->nodes
[1]);
3788 right
= read_node_slot(root
, upper
, slot
+ 1);
3790 * slot + 1 is not valid or we fail to read the right node,
3791 * no big deal, just return.
3796 btrfs_tree_lock(right
);
3797 btrfs_set_lock_blocking(right
);
3799 free_space
= btrfs_leaf_free_space(root
, right
);
3800 if (free_space
< data_size
)
3803 /* cow and double check */
3804 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3809 free_space
= btrfs_leaf_free_space(root
, right
);
3810 if (free_space
< data_size
)
3813 left_nritems
= btrfs_header_nritems(left
);
3814 if (left_nritems
== 0)
3817 if (path
->slots
[0] == left_nritems
&& !empty
) {
3818 /* Key greater than all keys in the leaf, right neighbor has
3819 * enough room for it and we're not emptying our leaf to delete
3820 * it, therefore use right neighbor to insert the new item and
3821 * no need to touch/dirty our left leaft. */
3822 btrfs_tree_unlock(left
);
3823 free_extent_buffer(left
);
3824 path
->nodes
[0] = right
;
3830 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3831 right
, free_space
, left_nritems
, min_slot
);
3833 btrfs_tree_unlock(right
);
3834 free_extent_buffer(right
);
3839 * push some data in the path leaf to the left, trying to free up at
3840 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3842 * max_slot can put a limit on how far into the leaf we'll push items. The
3843 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3846 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3847 struct btrfs_root
*root
,
3848 struct btrfs_path
*path
, int data_size
,
3849 int empty
, struct extent_buffer
*left
,
3850 int free_space
, u32 right_nritems
,
3853 struct btrfs_disk_key disk_key
;
3854 struct extent_buffer
*right
= path
->nodes
[0];
3858 struct btrfs_item
*item
;
3859 u32 old_left_nritems
;
3863 u32 old_left_item_size
;
3864 struct btrfs_map_token token
;
3866 btrfs_init_map_token(&token
);
3869 nr
= min(right_nritems
, max_slot
);
3871 nr
= min(right_nritems
- 1, max_slot
);
3873 for (i
= 0; i
< nr
; i
++) {
3874 item
= btrfs_item_nr(i
);
3876 if (!empty
&& push_items
> 0) {
3877 if (path
->slots
[0] < i
)
3879 if (path
->slots
[0] == i
) {
3880 int space
= btrfs_leaf_free_space(root
, right
);
3881 if (space
+ push_space
* 2 > free_space
)
3886 if (path
->slots
[0] == i
)
3887 push_space
+= data_size
;
3889 this_item_size
= btrfs_item_size(right
, item
);
3890 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3894 push_space
+= this_item_size
+ sizeof(*item
);
3897 if (push_items
== 0) {
3901 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3903 /* push data from right to left */
3904 copy_extent_buffer(left
, right
,
3905 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3906 btrfs_item_nr_offset(0),
3907 push_items
* sizeof(struct btrfs_item
));
3909 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3910 btrfs_item_offset_nr(right
, push_items
- 1);
3912 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3913 leaf_data_end(root
, left
) - push_space
,
3914 btrfs_leaf_data(right
) +
3915 btrfs_item_offset_nr(right
, push_items
- 1),
3917 old_left_nritems
= btrfs_header_nritems(left
);
3918 BUG_ON(old_left_nritems
<= 0);
3920 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3921 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3924 item
= btrfs_item_nr(i
);
3926 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3927 btrfs_set_token_item_offset(left
, item
,
3928 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3931 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3933 /* fixup right node */
3934 if (push_items
> right_nritems
)
3935 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3938 if (push_items
< right_nritems
) {
3939 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3940 leaf_data_end(root
, right
);
3941 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3942 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3943 btrfs_leaf_data(right
) +
3944 leaf_data_end(root
, right
), push_space
);
3946 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3947 btrfs_item_nr_offset(push_items
),
3948 (btrfs_header_nritems(right
) - push_items
) *
3949 sizeof(struct btrfs_item
));
3951 right_nritems
-= push_items
;
3952 btrfs_set_header_nritems(right
, right_nritems
);
3953 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3954 for (i
= 0; i
< right_nritems
; i
++) {
3955 item
= btrfs_item_nr(i
);
3957 push_space
= push_space
- btrfs_token_item_size(right
,
3959 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3962 btrfs_mark_buffer_dirty(left
);
3964 btrfs_mark_buffer_dirty(right
);
3966 clean_tree_block(trans
, root
->fs_info
, right
);
3968 btrfs_item_key(right
, &disk_key
, 0);
3969 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
3971 /* then fixup the leaf pointer in the path */
3972 if (path
->slots
[0] < push_items
) {
3973 path
->slots
[0] += old_left_nritems
;
3974 btrfs_tree_unlock(path
->nodes
[0]);
3975 free_extent_buffer(path
->nodes
[0]);
3976 path
->nodes
[0] = left
;
3977 path
->slots
[1] -= 1;
3979 btrfs_tree_unlock(left
);
3980 free_extent_buffer(left
);
3981 path
->slots
[0] -= push_items
;
3983 BUG_ON(path
->slots
[0] < 0);
3986 btrfs_tree_unlock(left
);
3987 free_extent_buffer(left
);
3992 * push some data in the path leaf to the left, trying to free up at
3993 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3995 * max_slot can put a limit on how far into the leaf we'll push items. The
3996 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3999 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
4000 *root
, struct btrfs_path
*path
, int min_data_size
,
4001 int data_size
, int empty
, u32 max_slot
)
4003 struct extent_buffer
*right
= path
->nodes
[0];
4004 struct extent_buffer
*left
;
4010 slot
= path
->slots
[1];
4013 if (!path
->nodes
[1])
4016 right_nritems
= btrfs_header_nritems(right
);
4017 if (right_nritems
== 0)
4020 btrfs_assert_tree_locked(path
->nodes
[1]);
4022 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
4024 * slot - 1 is not valid or we fail to read the left node,
4025 * no big deal, just return.
4030 btrfs_tree_lock(left
);
4031 btrfs_set_lock_blocking(left
);
4033 free_space
= btrfs_leaf_free_space(root
, left
);
4034 if (free_space
< data_size
) {
4039 /* cow and double check */
4040 ret
= btrfs_cow_block(trans
, root
, left
,
4041 path
->nodes
[1], slot
- 1, &left
);
4043 /* we hit -ENOSPC, but it isn't fatal here */
4049 free_space
= btrfs_leaf_free_space(root
, left
);
4050 if (free_space
< data_size
) {
4055 return __push_leaf_left(trans
, root
, path
, min_data_size
,
4056 empty
, left
, free_space
, right_nritems
,
4059 btrfs_tree_unlock(left
);
4060 free_extent_buffer(left
);
4065 * split the path's leaf in two, making sure there is at least data_size
4066 * available for the resulting leaf level of the path.
4068 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4069 struct btrfs_root
*root
,
4070 struct btrfs_path
*path
,
4071 struct extent_buffer
*l
,
4072 struct extent_buffer
*right
,
4073 int slot
, int mid
, int nritems
)
4078 struct btrfs_disk_key disk_key
;
4079 struct btrfs_map_token token
;
4081 btrfs_init_map_token(&token
);
4083 nritems
= nritems
- mid
;
4084 btrfs_set_header_nritems(right
, nritems
);
4085 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
4087 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4088 btrfs_item_nr_offset(mid
),
4089 nritems
* sizeof(struct btrfs_item
));
4091 copy_extent_buffer(right
, l
,
4092 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
4093 data_copy_size
, btrfs_leaf_data(l
) +
4094 leaf_data_end(root
, l
), data_copy_size
);
4096 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
4097 btrfs_item_end_nr(l
, mid
);
4099 for (i
= 0; i
< nritems
; i
++) {
4100 struct btrfs_item
*item
= btrfs_item_nr(i
);
4103 ioff
= btrfs_token_item_offset(right
, item
, &token
);
4104 btrfs_set_token_item_offset(right
, item
,
4105 ioff
+ rt_data_off
, &token
);
4108 btrfs_set_header_nritems(l
, mid
);
4109 btrfs_item_key(right
, &disk_key
, 0);
4110 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4111 path
->slots
[1] + 1, 1);
4113 btrfs_mark_buffer_dirty(right
);
4114 btrfs_mark_buffer_dirty(l
);
4115 BUG_ON(path
->slots
[0] != slot
);
4118 btrfs_tree_unlock(path
->nodes
[0]);
4119 free_extent_buffer(path
->nodes
[0]);
4120 path
->nodes
[0] = right
;
4121 path
->slots
[0] -= mid
;
4122 path
->slots
[1] += 1;
4124 btrfs_tree_unlock(right
);
4125 free_extent_buffer(right
);
4128 BUG_ON(path
->slots
[0] < 0);
4132 * double splits happen when we need to insert a big item in the middle
4133 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4134 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4137 * We avoid this by trying to push the items on either side of our target
4138 * into the adjacent leaves. If all goes well we can avoid the double split
4141 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4142 struct btrfs_root
*root
,
4143 struct btrfs_path
*path
,
4150 int space_needed
= data_size
;
4152 slot
= path
->slots
[0];
4153 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4154 space_needed
-= btrfs_leaf_free_space(root
, path
->nodes
[0]);
4157 * try to push all the items after our slot into the
4160 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4167 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4169 * our goal is to get our slot at the start or end of a leaf. If
4170 * we've done so we're done
4172 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4175 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4178 /* try to push all the items before our slot into the next leaf */
4179 slot
= path
->slots
[0];
4180 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4193 * split the path's leaf in two, making sure there is at least data_size
4194 * available for the resulting leaf level of the path.
4196 * returns 0 if all went well and < 0 on failure.
4198 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4199 struct btrfs_root
*root
,
4200 struct btrfs_key
*ins_key
,
4201 struct btrfs_path
*path
, int data_size
,
4204 struct btrfs_disk_key disk_key
;
4205 struct extent_buffer
*l
;
4209 struct extent_buffer
*right
;
4210 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4214 int num_doubles
= 0;
4215 int tried_avoid_double
= 0;
4218 slot
= path
->slots
[0];
4219 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4220 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
4223 /* first try to make some room by pushing left and right */
4224 if (data_size
&& path
->nodes
[1]) {
4225 int space_needed
= data_size
;
4227 if (slot
< btrfs_header_nritems(l
))
4228 space_needed
-= btrfs_leaf_free_space(root
, l
);
4230 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4231 space_needed
, 0, 0);
4235 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4236 space_needed
, 0, (u32
)-1);
4242 /* did the pushes work? */
4243 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
4247 if (!path
->nodes
[1]) {
4248 ret
= insert_new_root(trans
, root
, path
, 1);
4255 slot
= path
->slots
[0];
4256 nritems
= btrfs_header_nritems(l
);
4257 mid
= (nritems
+ 1) / 2;
4261 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4262 BTRFS_LEAF_DATA_SIZE(root
)) {
4263 if (slot
>= nritems
) {
4267 if (mid
!= nritems
&&
4268 leaf_space_used(l
, mid
, nritems
- mid
) +
4269 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4270 if (data_size
&& !tried_avoid_double
)
4271 goto push_for_double
;
4277 if (leaf_space_used(l
, 0, mid
) + data_size
>
4278 BTRFS_LEAF_DATA_SIZE(root
)) {
4279 if (!extend
&& data_size
&& slot
== 0) {
4281 } else if ((extend
|| !data_size
) && slot
== 0) {
4285 if (mid
!= nritems
&&
4286 leaf_space_used(l
, mid
, nritems
- mid
) +
4287 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4288 if (data_size
&& !tried_avoid_double
)
4289 goto push_for_double
;
4297 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4299 btrfs_item_key(l
, &disk_key
, mid
);
4301 right
= btrfs_alloc_tree_block(trans
, root
, 0, root
->root_key
.objectid
,
4302 &disk_key
, 0, l
->start
, 0);
4304 return PTR_ERR(right
);
4306 root_add_used(root
, root
->nodesize
);
4308 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4309 btrfs_set_header_bytenr(right
, right
->start
);
4310 btrfs_set_header_generation(right
, trans
->transid
);
4311 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4312 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4313 btrfs_set_header_level(right
, 0);
4314 write_extent_buffer(right
, fs_info
->fsid
,
4315 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4317 write_extent_buffer(right
, fs_info
->chunk_tree_uuid
,
4318 btrfs_header_chunk_tree_uuid(right
),
4323 btrfs_set_header_nritems(right
, 0);
4324 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4325 path
->slots
[1] + 1, 1);
4326 btrfs_tree_unlock(path
->nodes
[0]);
4327 free_extent_buffer(path
->nodes
[0]);
4328 path
->nodes
[0] = right
;
4330 path
->slots
[1] += 1;
4332 btrfs_set_header_nritems(right
, 0);
4333 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4335 btrfs_tree_unlock(path
->nodes
[0]);
4336 free_extent_buffer(path
->nodes
[0]);
4337 path
->nodes
[0] = right
;
4339 if (path
->slots
[1] == 0)
4340 fixup_low_keys(fs_info
, path
, &disk_key
, 1);
4342 btrfs_mark_buffer_dirty(right
);
4346 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4349 BUG_ON(num_doubles
!= 0);
4357 push_for_double_split(trans
, root
, path
, data_size
);
4358 tried_avoid_double
= 1;
4359 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4364 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4365 struct btrfs_root
*root
,
4366 struct btrfs_path
*path
, int ins_len
)
4368 struct btrfs_key key
;
4369 struct extent_buffer
*leaf
;
4370 struct btrfs_file_extent_item
*fi
;
4375 leaf
= path
->nodes
[0];
4376 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4378 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4379 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4381 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4384 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4385 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4386 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4387 struct btrfs_file_extent_item
);
4388 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4390 btrfs_release_path(path
);
4392 path
->keep_locks
= 1;
4393 path
->search_for_split
= 1;
4394 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4395 path
->search_for_split
= 0;
4402 leaf
= path
->nodes
[0];
4403 /* if our item isn't there, return now */
4404 if (item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4407 /* the leaf has changed, it now has room. return now */
4408 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4411 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4412 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4413 struct btrfs_file_extent_item
);
4414 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4418 btrfs_set_path_blocking(path
);
4419 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4423 path
->keep_locks
= 0;
4424 btrfs_unlock_up_safe(path
, 1);
4427 path
->keep_locks
= 0;
4431 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4432 struct btrfs_root
*root
,
4433 struct btrfs_path
*path
,
4434 struct btrfs_key
*new_key
,
4435 unsigned long split_offset
)
4437 struct extent_buffer
*leaf
;
4438 struct btrfs_item
*item
;
4439 struct btrfs_item
*new_item
;
4445 struct btrfs_disk_key disk_key
;
4447 leaf
= path
->nodes
[0];
4448 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4450 btrfs_set_path_blocking(path
);
4452 item
= btrfs_item_nr(path
->slots
[0]);
4453 orig_offset
= btrfs_item_offset(leaf
, item
);
4454 item_size
= btrfs_item_size(leaf
, item
);
4456 buf
= kmalloc(item_size
, GFP_NOFS
);
4460 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4461 path
->slots
[0]), item_size
);
4463 slot
= path
->slots
[0] + 1;
4464 nritems
= btrfs_header_nritems(leaf
);
4465 if (slot
!= nritems
) {
4466 /* shift the items */
4467 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4468 btrfs_item_nr_offset(slot
),
4469 (nritems
- slot
) * sizeof(struct btrfs_item
));
4472 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4473 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4475 new_item
= btrfs_item_nr(slot
);
4477 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4478 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4480 btrfs_set_item_offset(leaf
, item
,
4481 orig_offset
+ item_size
- split_offset
);
4482 btrfs_set_item_size(leaf
, item
, split_offset
);
4484 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4486 /* write the data for the start of the original item */
4487 write_extent_buffer(leaf
, buf
,
4488 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4491 /* write the data for the new item */
4492 write_extent_buffer(leaf
, buf
+ split_offset
,
4493 btrfs_item_ptr_offset(leaf
, slot
),
4494 item_size
- split_offset
);
4495 btrfs_mark_buffer_dirty(leaf
);
4497 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4503 * This function splits a single item into two items,
4504 * giving 'new_key' to the new item and splitting the
4505 * old one at split_offset (from the start of the item).
4507 * The path may be released by this operation. After
4508 * the split, the path is pointing to the old item. The
4509 * new item is going to be in the same node as the old one.
4511 * Note, the item being split must be smaller enough to live alone on
4512 * a tree block with room for one extra struct btrfs_item
4514 * This allows us to split the item in place, keeping a lock on the
4515 * leaf the entire time.
4517 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4518 struct btrfs_root
*root
,
4519 struct btrfs_path
*path
,
4520 struct btrfs_key
*new_key
,
4521 unsigned long split_offset
)
4524 ret
= setup_leaf_for_split(trans
, root
, path
,
4525 sizeof(struct btrfs_item
));
4529 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4534 * This function duplicate a item, giving 'new_key' to the new item.
4535 * It guarantees both items live in the same tree leaf and the new item
4536 * is contiguous with the original item.
4538 * This allows us to split file extent in place, keeping a lock on the
4539 * leaf the entire time.
4541 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4542 struct btrfs_root
*root
,
4543 struct btrfs_path
*path
,
4544 struct btrfs_key
*new_key
)
4546 struct extent_buffer
*leaf
;
4550 leaf
= path
->nodes
[0];
4551 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4552 ret
= setup_leaf_for_split(trans
, root
, path
,
4553 item_size
+ sizeof(struct btrfs_item
));
4558 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4559 item_size
, item_size
+
4560 sizeof(struct btrfs_item
), 1);
4561 leaf
= path
->nodes
[0];
4562 memcpy_extent_buffer(leaf
,
4563 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4564 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4570 * make the item pointed to by the path smaller. new_size indicates
4571 * how small to make it, and from_end tells us if we just chop bytes
4572 * off the end of the item or if we shift the item to chop bytes off
4575 void btrfs_truncate_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4576 u32 new_size
, int from_end
)
4579 struct extent_buffer
*leaf
;
4580 struct btrfs_item
*item
;
4582 unsigned int data_end
;
4583 unsigned int old_data_start
;
4584 unsigned int old_size
;
4585 unsigned int size_diff
;
4587 struct btrfs_map_token token
;
4589 btrfs_init_map_token(&token
);
4591 leaf
= path
->nodes
[0];
4592 slot
= path
->slots
[0];
4594 old_size
= btrfs_item_size_nr(leaf
, slot
);
4595 if (old_size
== new_size
)
4598 nritems
= btrfs_header_nritems(leaf
);
4599 data_end
= leaf_data_end(root
, leaf
);
4601 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4603 size_diff
= old_size
- new_size
;
4606 BUG_ON(slot
>= nritems
);
4609 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4611 /* first correct the data pointers */
4612 for (i
= slot
; i
< nritems
; i
++) {
4614 item
= btrfs_item_nr(i
);
4616 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4617 btrfs_set_token_item_offset(leaf
, item
,
4618 ioff
+ size_diff
, &token
);
4621 /* shift the data */
4623 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4624 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4625 data_end
, old_data_start
+ new_size
- data_end
);
4627 struct btrfs_disk_key disk_key
;
4630 btrfs_item_key(leaf
, &disk_key
, slot
);
4632 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4634 struct btrfs_file_extent_item
*fi
;
4636 fi
= btrfs_item_ptr(leaf
, slot
,
4637 struct btrfs_file_extent_item
);
4638 fi
= (struct btrfs_file_extent_item
*)(
4639 (unsigned long)fi
- size_diff
);
4641 if (btrfs_file_extent_type(leaf
, fi
) ==
4642 BTRFS_FILE_EXTENT_INLINE
) {
4643 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4644 memmove_extent_buffer(leaf
, ptr
,
4646 BTRFS_FILE_EXTENT_INLINE_DATA_START
);
4650 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4651 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4652 data_end
, old_data_start
- data_end
);
4654 offset
= btrfs_disk_key_offset(&disk_key
);
4655 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4656 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4658 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4661 item
= btrfs_item_nr(slot
);
4662 btrfs_set_item_size(leaf
, item
, new_size
);
4663 btrfs_mark_buffer_dirty(leaf
);
4665 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4666 btrfs_print_leaf(root
, leaf
);
4672 * make the item pointed to by the path bigger, data_size is the added size.
4674 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4678 struct extent_buffer
*leaf
;
4679 struct btrfs_item
*item
;
4681 unsigned int data_end
;
4682 unsigned int old_data
;
4683 unsigned int old_size
;
4685 struct btrfs_map_token token
;
4687 btrfs_init_map_token(&token
);
4689 leaf
= path
->nodes
[0];
4691 nritems
= btrfs_header_nritems(leaf
);
4692 data_end
= leaf_data_end(root
, leaf
);
4694 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4695 btrfs_print_leaf(root
, leaf
);
4698 slot
= path
->slots
[0];
4699 old_data
= btrfs_item_end_nr(leaf
, slot
);
4702 if (slot
>= nritems
) {
4703 btrfs_print_leaf(root
, leaf
);
4704 btrfs_crit(root
->fs_info
, "slot %d too large, nritems %d",
4710 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4712 /* first correct the data pointers */
4713 for (i
= slot
; i
< nritems
; i
++) {
4715 item
= btrfs_item_nr(i
);
4717 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4718 btrfs_set_token_item_offset(leaf
, item
,
4719 ioff
- data_size
, &token
);
4722 /* shift the data */
4723 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4724 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4725 data_end
, old_data
- data_end
);
4727 data_end
= old_data
;
4728 old_size
= btrfs_item_size_nr(leaf
, slot
);
4729 item
= btrfs_item_nr(slot
);
4730 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4731 btrfs_mark_buffer_dirty(leaf
);
4733 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4734 btrfs_print_leaf(root
, leaf
);
4740 * this is a helper for btrfs_insert_empty_items, the main goal here is
4741 * to save stack depth by doing the bulk of the work in a function
4742 * that doesn't call btrfs_search_slot
4744 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4745 struct btrfs_key
*cpu_key
, u32
*data_size
,
4746 u32 total_data
, u32 total_size
, int nr
)
4748 struct btrfs_item
*item
;
4751 unsigned int data_end
;
4752 struct btrfs_disk_key disk_key
;
4753 struct extent_buffer
*leaf
;
4755 struct btrfs_map_token token
;
4757 if (path
->slots
[0] == 0) {
4758 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4759 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
4761 btrfs_unlock_up_safe(path
, 1);
4763 btrfs_init_map_token(&token
);
4765 leaf
= path
->nodes
[0];
4766 slot
= path
->slots
[0];
4768 nritems
= btrfs_header_nritems(leaf
);
4769 data_end
= leaf_data_end(root
, leaf
);
4771 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4772 btrfs_print_leaf(root
, leaf
);
4773 btrfs_crit(root
->fs_info
, "not enough freespace need %u have %d",
4774 total_size
, btrfs_leaf_free_space(root
, leaf
));
4778 if (slot
!= nritems
) {
4779 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4781 if (old_data
< data_end
) {
4782 btrfs_print_leaf(root
, leaf
);
4783 btrfs_crit(root
->fs_info
, "slot %d old_data %d data_end %d",
4784 slot
, old_data
, data_end
);
4788 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4790 /* first correct the data pointers */
4791 for (i
= slot
; i
< nritems
; i
++) {
4794 item
= btrfs_item_nr( i
);
4795 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4796 btrfs_set_token_item_offset(leaf
, item
,
4797 ioff
- total_data
, &token
);
4799 /* shift the items */
4800 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4801 btrfs_item_nr_offset(slot
),
4802 (nritems
- slot
) * sizeof(struct btrfs_item
));
4804 /* shift the data */
4805 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4806 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4807 data_end
, old_data
- data_end
);
4808 data_end
= old_data
;
4811 /* setup the item for the new data */
4812 for (i
= 0; i
< nr
; i
++) {
4813 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4814 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4815 item
= btrfs_item_nr(slot
+ i
);
4816 btrfs_set_token_item_offset(leaf
, item
,
4817 data_end
- data_size
[i
], &token
);
4818 data_end
-= data_size
[i
];
4819 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4822 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4823 btrfs_mark_buffer_dirty(leaf
);
4825 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4826 btrfs_print_leaf(root
, leaf
);
4832 * Given a key and some data, insert items into the tree.
4833 * This does all the path init required, making room in the tree if needed.
4835 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4836 struct btrfs_root
*root
,
4837 struct btrfs_path
*path
,
4838 struct btrfs_key
*cpu_key
, u32
*data_size
,
4847 for (i
= 0; i
< nr
; i
++)
4848 total_data
+= data_size
[i
];
4850 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4851 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4857 slot
= path
->slots
[0];
4860 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4861 total_data
, total_size
, nr
);
4866 * Given a key and some data, insert an item into the tree.
4867 * This does all the path init required, making room in the tree if needed.
4869 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4870 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4874 struct btrfs_path
*path
;
4875 struct extent_buffer
*leaf
;
4878 path
= btrfs_alloc_path();
4881 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4883 leaf
= path
->nodes
[0];
4884 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4885 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4886 btrfs_mark_buffer_dirty(leaf
);
4888 btrfs_free_path(path
);
4893 * delete the pointer from a given node.
4895 * the tree should have been previously balanced so the deletion does not
4898 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4899 int level
, int slot
)
4901 struct extent_buffer
*parent
= path
->nodes
[level
];
4905 nritems
= btrfs_header_nritems(parent
);
4906 if (slot
!= nritems
- 1) {
4908 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4909 slot
+ 1, nritems
- slot
- 1);
4910 memmove_extent_buffer(parent
,
4911 btrfs_node_key_ptr_offset(slot
),
4912 btrfs_node_key_ptr_offset(slot
+ 1),
4913 sizeof(struct btrfs_key_ptr
) *
4914 (nritems
- slot
- 1));
4916 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4917 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4922 btrfs_set_header_nritems(parent
, nritems
);
4923 if (nritems
== 0 && parent
== root
->node
) {
4924 BUG_ON(btrfs_header_level(root
->node
) != 1);
4925 /* just turn the root into a leaf and break */
4926 btrfs_set_header_level(root
->node
, 0);
4927 } else if (slot
== 0) {
4928 struct btrfs_disk_key disk_key
;
4930 btrfs_node_key(parent
, &disk_key
, 0);
4931 fixup_low_keys(root
->fs_info
, path
, &disk_key
, level
+ 1);
4933 btrfs_mark_buffer_dirty(parent
);
4937 * a helper function to delete the leaf pointed to by path->slots[1] and
4940 * This deletes the pointer in path->nodes[1] and frees the leaf
4941 * block extent. zero is returned if it all worked out, < 0 otherwise.
4943 * The path must have already been setup for deleting the leaf, including
4944 * all the proper balancing. path->nodes[1] must be locked.
4946 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4947 struct btrfs_root
*root
,
4948 struct btrfs_path
*path
,
4949 struct extent_buffer
*leaf
)
4951 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4952 del_ptr(root
, path
, 1, path
->slots
[1]);
4955 * btrfs_free_extent is expensive, we want to make sure we
4956 * aren't holding any locks when we call it
4958 btrfs_unlock_up_safe(path
, 0);
4960 root_sub_used(root
, leaf
->len
);
4962 extent_buffer_get(leaf
);
4963 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4964 free_extent_buffer_stale(leaf
);
4967 * delete the item at the leaf level in path. If that empties
4968 * the leaf, remove it from the tree
4970 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4971 struct btrfs_path
*path
, int slot
, int nr
)
4973 struct extent_buffer
*leaf
;
4974 struct btrfs_item
*item
;
4981 struct btrfs_map_token token
;
4983 btrfs_init_map_token(&token
);
4985 leaf
= path
->nodes
[0];
4986 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4988 for (i
= 0; i
< nr
; i
++)
4989 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4991 nritems
= btrfs_header_nritems(leaf
);
4993 if (slot
+ nr
!= nritems
) {
4994 int data_end
= leaf_data_end(root
, leaf
);
4996 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4998 btrfs_leaf_data(leaf
) + data_end
,
4999 last_off
- data_end
);
5001 for (i
= slot
+ nr
; i
< nritems
; i
++) {
5004 item
= btrfs_item_nr(i
);
5005 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
5006 btrfs_set_token_item_offset(leaf
, item
,
5007 ioff
+ dsize
, &token
);
5010 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
5011 btrfs_item_nr_offset(slot
+ nr
),
5012 sizeof(struct btrfs_item
) *
5013 (nritems
- slot
- nr
));
5015 btrfs_set_header_nritems(leaf
, nritems
- nr
);
5018 /* delete the leaf if we've emptied it */
5020 if (leaf
== root
->node
) {
5021 btrfs_set_header_level(leaf
, 0);
5023 btrfs_set_path_blocking(path
);
5024 clean_tree_block(trans
, root
->fs_info
, leaf
);
5025 btrfs_del_leaf(trans
, root
, path
, leaf
);
5028 int used
= leaf_space_used(leaf
, 0, nritems
);
5030 struct btrfs_disk_key disk_key
;
5032 btrfs_item_key(leaf
, &disk_key
, 0);
5033 fixup_low_keys(root
->fs_info
, path
, &disk_key
, 1);
5036 /* delete the leaf if it is mostly empty */
5037 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
5038 /* push_leaf_left fixes the path.
5039 * make sure the path still points to our leaf
5040 * for possible call to del_ptr below
5042 slot
= path
->slots
[1];
5043 extent_buffer_get(leaf
);
5045 btrfs_set_path_blocking(path
);
5046 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
5048 if (wret
< 0 && wret
!= -ENOSPC
)
5051 if (path
->nodes
[0] == leaf
&&
5052 btrfs_header_nritems(leaf
)) {
5053 wret
= push_leaf_right(trans
, root
, path
, 1,
5055 if (wret
< 0 && wret
!= -ENOSPC
)
5059 if (btrfs_header_nritems(leaf
) == 0) {
5060 path
->slots
[1] = slot
;
5061 btrfs_del_leaf(trans
, root
, path
, leaf
);
5062 free_extent_buffer(leaf
);
5065 /* if we're still in the path, make sure
5066 * we're dirty. Otherwise, one of the
5067 * push_leaf functions must have already
5068 * dirtied this buffer
5070 if (path
->nodes
[0] == leaf
)
5071 btrfs_mark_buffer_dirty(leaf
);
5072 free_extent_buffer(leaf
);
5075 btrfs_mark_buffer_dirty(leaf
);
5082 * search the tree again to find a leaf with lesser keys
5083 * returns 0 if it found something or 1 if there are no lesser leaves.
5084 * returns < 0 on io errors.
5086 * This may release the path, and so you may lose any locks held at the
5089 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5091 struct btrfs_key key
;
5092 struct btrfs_disk_key found_key
;
5095 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5097 if (key
.offset
> 0) {
5099 } else if (key
.type
> 0) {
5101 key
.offset
= (u64
)-1;
5102 } else if (key
.objectid
> 0) {
5105 key
.offset
= (u64
)-1;
5110 btrfs_release_path(path
);
5111 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5114 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5115 ret
= comp_keys(&found_key
, &key
);
5117 * We might have had an item with the previous key in the tree right
5118 * before we released our path. And after we released our path, that
5119 * item might have been pushed to the first slot (0) of the leaf we
5120 * were holding due to a tree balance. Alternatively, an item with the
5121 * previous key can exist as the only element of a leaf (big fat item).
5122 * Therefore account for these 2 cases, so that our callers (like
5123 * btrfs_previous_item) don't miss an existing item with a key matching
5124 * the previous key we computed above.
5132 * A helper function to walk down the tree starting at min_key, and looking
5133 * for nodes or leaves that are have a minimum transaction id.
5134 * This is used by the btree defrag code, and tree logging
5136 * This does not cow, but it does stuff the starting key it finds back
5137 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5138 * key and get a writable path.
5140 * This does lock as it descends, and path->keep_locks should be set
5141 * to 1 by the caller.
5143 * This honors path->lowest_level to prevent descent past a given level
5146 * min_trans indicates the oldest transaction that you are interested
5147 * in walking through. Any nodes or leaves older than min_trans are
5148 * skipped over (without reading them).
5150 * returns zero if something useful was found, < 0 on error and 1 if there
5151 * was nothing in the tree that matched the search criteria.
5153 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5154 struct btrfs_path
*path
,
5157 struct extent_buffer
*cur
;
5158 struct btrfs_key found_key
;
5164 int keep_locks
= path
->keep_locks
;
5166 path
->keep_locks
= 1;
5168 cur
= btrfs_read_lock_root_node(root
);
5169 level
= btrfs_header_level(cur
);
5170 WARN_ON(path
->nodes
[level
]);
5171 path
->nodes
[level
] = cur
;
5172 path
->locks
[level
] = BTRFS_READ_LOCK
;
5174 if (btrfs_header_generation(cur
) < min_trans
) {
5179 nritems
= btrfs_header_nritems(cur
);
5180 level
= btrfs_header_level(cur
);
5181 sret
= bin_search(cur
, min_key
, level
, &slot
);
5183 /* at the lowest level, we're done, setup the path and exit */
5184 if (level
== path
->lowest_level
) {
5185 if (slot
>= nritems
)
5188 path
->slots
[level
] = slot
;
5189 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5192 if (sret
&& slot
> 0)
5195 * check this node pointer against the min_trans parameters.
5196 * If it is too old, old, skip to the next one.
5198 while (slot
< nritems
) {
5201 gen
= btrfs_node_ptr_generation(cur
, slot
);
5202 if (gen
< min_trans
) {
5210 * we didn't find a candidate key in this node, walk forward
5211 * and find another one
5213 if (slot
>= nritems
) {
5214 path
->slots
[level
] = slot
;
5215 btrfs_set_path_blocking(path
);
5216 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5219 btrfs_release_path(path
);
5225 /* save our key for returning back */
5226 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5227 path
->slots
[level
] = slot
;
5228 if (level
== path
->lowest_level
) {
5232 btrfs_set_path_blocking(path
);
5233 cur
= read_node_slot(root
, cur
, slot
);
5239 btrfs_tree_read_lock(cur
);
5241 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5242 path
->nodes
[level
- 1] = cur
;
5243 unlock_up(path
, level
, 1, 0, NULL
);
5244 btrfs_clear_path_blocking(path
, NULL
, 0);
5247 path
->keep_locks
= keep_locks
;
5249 btrfs_unlock_up_safe(path
, path
->lowest_level
+ 1);
5250 btrfs_set_path_blocking(path
);
5251 memcpy(min_key
, &found_key
, sizeof(found_key
));
5256 static int tree_move_down(struct btrfs_root
*root
,
5257 struct btrfs_path
*path
,
5258 int *level
, int root_level
)
5260 struct extent_buffer
*eb
;
5262 BUG_ON(*level
== 0);
5263 eb
= read_node_slot(root
, path
->nodes
[*level
], path
->slots
[*level
]);
5267 path
->nodes
[*level
- 1] = eb
;
5268 path
->slots
[*level
- 1] = 0;
5273 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
5274 struct btrfs_path
*path
,
5275 int *level
, int root_level
)
5279 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5281 path
->slots
[*level
]++;
5283 while (path
->slots
[*level
] >= nritems
) {
5284 if (*level
== root_level
)
5288 path
->slots
[*level
] = 0;
5289 free_extent_buffer(path
->nodes
[*level
]);
5290 path
->nodes
[*level
] = NULL
;
5292 path
->slots
[*level
]++;
5294 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5301 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5304 static int tree_advance(struct btrfs_root
*root
,
5305 struct btrfs_path
*path
,
5306 int *level
, int root_level
,
5308 struct btrfs_key
*key
)
5312 if (*level
== 0 || !allow_down
) {
5313 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5315 ret
= tree_move_down(root
, path
, level
, root_level
);
5319 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5320 path
->slots
[*level
]);
5322 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5323 path
->slots
[*level
]);
5328 static int tree_compare_item(struct btrfs_root
*left_root
,
5329 struct btrfs_path
*left_path
,
5330 struct btrfs_path
*right_path
,
5335 unsigned long off1
, off2
;
5337 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5338 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5342 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5343 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5344 right_path
->slots
[0]);
5346 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5348 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5355 #define ADVANCE_ONLY_NEXT -1
5358 * This function compares two trees and calls the provided callback for
5359 * every changed/new/deleted item it finds.
5360 * If shared tree blocks are encountered, whole subtrees are skipped, making
5361 * the compare pretty fast on snapshotted subvolumes.
5363 * This currently works on commit roots only. As commit roots are read only,
5364 * we don't do any locking. The commit roots are protected with transactions.
5365 * Transactions are ended and rejoined when a commit is tried in between.
5367 * This function checks for modifications done to the trees while comparing.
5368 * If it detects a change, it aborts immediately.
5370 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5371 struct btrfs_root
*right_root
,
5372 btrfs_changed_cb_t changed_cb
, void *ctx
)
5376 struct btrfs_path
*left_path
= NULL
;
5377 struct btrfs_path
*right_path
= NULL
;
5378 struct btrfs_key left_key
;
5379 struct btrfs_key right_key
;
5380 char *tmp_buf
= NULL
;
5381 int left_root_level
;
5382 int right_root_level
;
5385 int left_end_reached
;
5386 int right_end_reached
;
5394 left_path
= btrfs_alloc_path();
5399 right_path
= btrfs_alloc_path();
5405 tmp_buf
= kmalloc(left_root
->nodesize
, GFP_KERNEL
| __GFP_NOWARN
);
5407 tmp_buf
= vmalloc(left_root
->nodesize
);
5414 left_path
->search_commit_root
= 1;
5415 left_path
->skip_locking
= 1;
5416 right_path
->search_commit_root
= 1;
5417 right_path
->skip_locking
= 1;
5420 * Strategy: Go to the first items of both trees. Then do
5422 * If both trees are at level 0
5423 * Compare keys of current items
5424 * If left < right treat left item as new, advance left tree
5426 * If left > right treat right item as deleted, advance right tree
5428 * If left == right do deep compare of items, treat as changed if
5429 * needed, advance both trees and repeat
5430 * If both trees are at the same level but not at level 0
5431 * Compare keys of current nodes/leafs
5432 * If left < right advance left tree and repeat
5433 * If left > right advance right tree and repeat
5434 * If left == right compare blockptrs of the next nodes/leafs
5435 * If they match advance both trees but stay at the same level
5437 * If they don't match advance both trees while allowing to go
5439 * If tree levels are different
5440 * Advance the tree that needs it and repeat
5442 * Advancing a tree means:
5443 * If we are at level 0, try to go to the next slot. If that's not
5444 * possible, go one level up and repeat. Stop when we found a level
5445 * where we could go to the next slot. We may at this point be on a
5448 * If we are not at level 0 and not on shared tree blocks, go one
5451 * If we are not at level 0 and on shared tree blocks, go one slot to
5452 * the right if possible or go up and right.
5455 down_read(&left_root
->fs_info
->commit_root_sem
);
5456 left_level
= btrfs_header_level(left_root
->commit_root
);
5457 left_root_level
= left_level
;
5458 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5459 extent_buffer_get(left_path
->nodes
[left_level
]);
5461 right_level
= btrfs_header_level(right_root
->commit_root
);
5462 right_root_level
= right_level
;
5463 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5464 extent_buffer_get(right_path
->nodes
[right_level
]);
5465 up_read(&left_root
->fs_info
->commit_root_sem
);
5467 if (left_level
== 0)
5468 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5469 &left_key
, left_path
->slots
[left_level
]);
5471 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5472 &left_key
, left_path
->slots
[left_level
]);
5473 if (right_level
== 0)
5474 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5475 &right_key
, right_path
->slots
[right_level
]);
5477 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5478 &right_key
, right_path
->slots
[right_level
]);
5480 left_end_reached
= right_end_reached
= 0;
5481 advance_left
= advance_right
= 0;
5484 if (advance_left
&& !left_end_reached
) {
5485 ret
= tree_advance(left_root
, left_path
, &left_level
,
5487 advance_left
!= ADVANCE_ONLY_NEXT
,
5490 left_end_reached
= ADVANCE
;
5495 if (advance_right
&& !right_end_reached
) {
5496 ret
= tree_advance(right_root
, right_path
, &right_level
,
5498 advance_right
!= ADVANCE_ONLY_NEXT
,
5501 right_end_reached
= ADVANCE
;
5507 if (left_end_reached
&& right_end_reached
) {
5510 } else if (left_end_reached
) {
5511 if (right_level
== 0) {
5512 ret
= changed_cb(left_root
, right_root
,
5513 left_path
, right_path
,
5515 BTRFS_COMPARE_TREE_DELETED
,
5520 advance_right
= ADVANCE
;
5522 } else if (right_end_reached
) {
5523 if (left_level
== 0) {
5524 ret
= changed_cb(left_root
, right_root
,
5525 left_path
, right_path
,
5527 BTRFS_COMPARE_TREE_NEW
,
5532 advance_left
= ADVANCE
;
5536 if (left_level
== 0 && right_level
== 0) {
5537 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5539 ret
= changed_cb(left_root
, right_root
,
5540 left_path
, right_path
,
5542 BTRFS_COMPARE_TREE_NEW
,
5546 advance_left
= ADVANCE
;
5547 } else if (cmp
> 0) {
5548 ret
= changed_cb(left_root
, right_root
,
5549 left_path
, right_path
,
5551 BTRFS_COMPARE_TREE_DELETED
,
5555 advance_right
= ADVANCE
;
5557 enum btrfs_compare_tree_result result
;
5559 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5560 ret
= tree_compare_item(left_root
, left_path
,
5561 right_path
, tmp_buf
);
5563 result
= BTRFS_COMPARE_TREE_CHANGED
;
5565 result
= BTRFS_COMPARE_TREE_SAME
;
5566 ret
= changed_cb(left_root
, right_root
,
5567 left_path
, right_path
,
5568 &left_key
, result
, ctx
);
5571 advance_left
= ADVANCE
;
5572 advance_right
= ADVANCE
;
5574 } else if (left_level
== right_level
) {
5575 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5577 advance_left
= ADVANCE
;
5578 } else if (cmp
> 0) {
5579 advance_right
= ADVANCE
;
5581 left_blockptr
= btrfs_node_blockptr(
5582 left_path
->nodes
[left_level
],
5583 left_path
->slots
[left_level
]);
5584 right_blockptr
= btrfs_node_blockptr(
5585 right_path
->nodes
[right_level
],
5586 right_path
->slots
[right_level
]);
5587 left_gen
= btrfs_node_ptr_generation(
5588 left_path
->nodes
[left_level
],
5589 left_path
->slots
[left_level
]);
5590 right_gen
= btrfs_node_ptr_generation(
5591 right_path
->nodes
[right_level
],
5592 right_path
->slots
[right_level
]);
5593 if (left_blockptr
== right_blockptr
&&
5594 left_gen
== right_gen
) {
5596 * As we're on a shared block, don't
5597 * allow to go deeper.
5599 advance_left
= ADVANCE_ONLY_NEXT
;
5600 advance_right
= ADVANCE_ONLY_NEXT
;
5602 advance_left
= ADVANCE
;
5603 advance_right
= ADVANCE
;
5606 } else if (left_level
< right_level
) {
5607 advance_right
= ADVANCE
;
5609 advance_left
= ADVANCE
;
5614 btrfs_free_path(left_path
);
5615 btrfs_free_path(right_path
);
5621 * this is similar to btrfs_next_leaf, but does not try to preserve
5622 * and fixup the path. It looks for and returns the next key in the
5623 * tree based on the current path and the min_trans parameters.
5625 * 0 is returned if another key is found, < 0 if there are any errors
5626 * and 1 is returned if there are no higher keys in the tree
5628 * path->keep_locks should be set to 1 on the search made before
5629 * calling this function.
5631 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5632 struct btrfs_key
*key
, int level
, u64 min_trans
)
5635 struct extent_buffer
*c
;
5637 WARN_ON(!path
->keep_locks
);
5638 while (level
< BTRFS_MAX_LEVEL
) {
5639 if (!path
->nodes
[level
])
5642 slot
= path
->slots
[level
] + 1;
5643 c
= path
->nodes
[level
];
5645 if (slot
>= btrfs_header_nritems(c
)) {
5648 struct btrfs_key cur_key
;
5649 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5650 !path
->nodes
[level
+ 1])
5653 if (path
->locks
[level
+ 1]) {
5658 slot
= btrfs_header_nritems(c
) - 1;
5660 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5662 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5664 orig_lowest
= path
->lowest_level
;
5665 btrfs_release_path(path
);
5666 path
->lowest_level
= level
;
5667 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5669 path
->lowest_level
= orig_lowest
;
5673 c
= path
->nodes
[level
];
5674 slot
= path
->slots
[level
];
5681 btrfs_item_key_to_cpu(c
, key
, slot
);
5683 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5685 if (gen
< min_trans
) {
5689 btrfs_node_key_to_cpu(c
, key
, slot
);
5697 * search the tree again to find a leaf with greater keys
5698 * returns 0 if it found something or 1 if there are no greater leaves.
5699 * returns < 0 on io errors.
5701 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5703 return btrfs_next_old_leaf(root
, path
, 0);
5706 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5711 struct extent_buffer
*c
;
5712 struct extent_buffer
*next
;
5713 struct btrfs_key key
;
5716 int old_spinning
= path
->leave_spinning
;
5717 int next_rw_lock
= 0;
5719 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5723 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5728 btrfs_release_path(path
);
5730 path
->keep_locks
= 1;
5731 path
->leave_spinning
= 1;
5734 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5736 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5737 path
->keep_locks
= 0;
5742 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5744 * by releasing the path above we dropped all our locks. A balance
5745 * could have added more items next to the key that used to be
5746 * at the very end of the block. So, check again here and
5747 * advance the path if there are now more items available.
5749 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5756 * So the above check misses one case:
5757 * - after releasing the path above, someone has removed the item that
5758 * used to be at the very end of the block, and balance between leafs
5759 * gets another one with bigger key.offset to replace it.
5761 * This one should be returned as well, or we can get leaf corruption
5762 * later(esp. in __btrfs_drop_extents()).
5764 * And a bit more explanation about this check,
5765 * with ret > 0, the key isn't found, the path points to the slot
5766 * where it should be inserted, so the path->slots[0] item must be the
5769 if (nritems
> 0 && ret
> 0 && path
->slots
[0] == nritems
- 1) {
5774 while (level
< BTRFS_MAX_LEVEL
) {
5775 if (!path
->nodes
[level
]) {
5780 slot
= path
->slots
[level
] + 1;
5781 c
= path
->nodes
[level
];
5782 if (slot
>= btrfs_header_nritems(c
)) {
5784 if (level
== BTRFS_MAX_LEVEL
) {
5792 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5793 free_extent_buffer(next
);
5797 next_rw_lock
= path
->locks
[level
];
5798 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5804 btrfs_release_path(path
);
5808 if (!path
->skip_locking
) {
5809 ret
= btrfs_try_tree_read_lock(next
);
5810 if (!ret
&& time_seq
) {
5812 * If we don't get the lock, we may be racing
5813 * with push_leaf_left, holding that lock while
5814 * itself waiting for the leaf we've currently
5815 * locked. To solve this situation, we give up
5816 * on our lock and cycle.
5818 free_extent_buffer(next
);
5819 btrfs_release_path(path
);
5824 btrfs_set_path_blocking(path
);
5825 btrfs_tree_read_lock(next
);
5826 btrfs_clear_path_blocking(path
, next
,
5829 next_rw_lock
= BTRFS_READ_LOCK
;
5833 path
->slots
[level
] = slot
;
5836 c
= path
->nodes
[level
];
5837 if (path
->locks
[level
])
5838 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5840 free_extent_buffer(c
);
5841 path
->nodes
[level
] = next
;
5842 path
->slots
[level
] = 0;
5843 if (!path
->skip_locking
)
5844 path
->locks
[level
] = next_rw_lock
;
5848 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5854 btrfs_release_path(path
);
5858 if (!path
->skip_locking
) {
5859 ret
= btrfs_try_tree_read_lock(next
);
5861 btrfs_set_path_blocking(path
);
5862 btrfs_tree_read_lock(next
);
5863 btrfs_clear_path_blocking(path
, next
,
5866 next_rw_lock
= BTRFS_READ_LOCK
;
5871 unlock_up(path
, 0, 1, 0, NULL
);
5872 path
->leave_spinning
= old_spinning
;
5874 btrfs_set_path_blocking(path
);
5880 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5881 * searching until it gets past min_objectid or finds an item of 'type'
5883 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5885 int btrfs_previous_item(struct btrfs_root
*root
,
5886 struct btrfs_path
*path
, u64 min_objectid
,
5889 struct btrfs_key found_key
;
5890 struct extent_buffer
*leaf
;
5895 if (path
->slots
[0] == 0) {
5896 btrfs_set_path_blocking(path
);
5897 ret
= btrfs_prev_leaf(root
, path
);
5903 leaf
= path
->nodes
[0];
5904 nritems
= btrfs_header_nritems(leaf
);
5907 if (path
->slots
[0] == nritems
)
5910 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5911 if (found_key
.objectid
< min_objectid
)
5913 if (found_key
.type
== type
)
5915 if (found_key
.objectid
== min_objectid
&&
5916 found_key
.type
< type
)
5923 * search in extent tree to find a previous Metadata/Data extent item with
5926 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5928 int btrfs_previous_extent_item(struct btrfs_root
*root
,
5929 struct btrfs_path
*path
, u64 min_objectid
)
5931 struct btrfs_key found_key
;
5932 struct extent_buffer
*leaf
;
5937 if (path
->slots
[0] == 0) {
5938 btrfs_set_path_blocking(path
);
5939 ret
= btrfs_prev_leaf(root
, path
);
5945 leaf
= path
->nodes
[0];
5946 nritems
= btrfs_header_nritems(leaf
);
5949 if (path
->slots
[0] == nritems
)
5952 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5953 if (found_key
.objectid
< min_objectid
)
5955 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
||
5956 found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
5958 if (found_key
.objectid
== min_objectid
&&
5959 found_key
.type
< BTRFS_EXTENT_ITEM_KEY
)