2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
29 *root
, struct btrfs_path
*path
, int level
);
30 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
31 *root
, struct btrfs_key
*ins_key
,
32 struct btrfs_path
*path
, int data_size
, int extend
);
33 static int push_node_left(struct btrfs_trans_handle
*trans
,
34 struct btrfs_root
*root
, struct extent_buffer
*dst
,
35 struct extent_buffer
*src
, int empty
);
36 static int balance_node_right(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 struct extent_buffer
*dst_buf
,
39 struct extent_buffer
*src_buf
);
40 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
42 static int tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
43 struct extent_buffer
*eb
);
45 struct btrfs_path
*btrfs_alloc_path(void)
47 struct btrfs_path
*path
;
48 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
59 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
60 if (!p
->nodes
[i
] || !p
->locks
[i
])
62 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
63 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
64 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
65 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
66 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
79 struct extent_buffer
*held
, int held_rw
)
83 #ifdef CONFIG_DEBUG_LOCK_ALLOC
84 /* lockdep really cares that we take all of these spinlocks
85 * in the right order. If any of the locks in the path are not
86 * currently blocking, it is going to complain. So, make really
87 * really sure by forcing the path to blocking before we clear
91 btrfs_set_lock_blocking_rw(held
, held_rw
);
92 if (held_rw
== BTRFS_WRITE_LOCK
)
93 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
94 else if (held_rw
== BTRFS_READ_LOCK
)
95 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
97 btrfs_set_path_blocking(p
);
100 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
101 if (p
->nodes
[i
] && p
->locks
[i
]) {
102 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
103 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
104 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
105 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
106 p
->locks
[i
] = BTRFS_READ_LOCK
;
110 #ifdef CONFIG_DEBUG_LOCK_ALLOC
112 btrfs_clear_lock_blocking_rw(held
, held_rw
);
116 /* this also releases the path */
117 void btrfs_free_path(struct btrfs_path
*p
)
121 btrfs_release_path(p
);
122 kmem_cache_free(btrfs_path_cachep
, p
);
126 * path release drops references on the extent buffers in the path
127 * and it drops any locks held by this path
129 * It is safe to call this on paths that no locks or extent buffers held.
131 noinline
void btrfs_release_path(struct btrfs_path
*p
)
135 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
140 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
143 free_extent_buffer(p
->nodes
[i
]);
149 * safely gets a reference on the root node of a tree. A lock
150 * is not taken, so a concurrent writer may put a different node
151 * at the root of the tree. See btrfs_lock_root_node for the
154 * The extent buffer returned by this has a reference taken, so
155 * it won't disappear. It may stop being the root of the tree
156 * at any time because there are no locks held.
158 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
160 struct extent_buffer
*eb
;
164 eb
= rcu_dereference(root
->node
);
167 * RCU really hurts here, we could free up the root node because
168 * it was cow'ed but we may not get the new root node yet so do
169 * the inc_not_zero dance and if it doesn't work then
170 * synchronize_rcu and try again.
172 if (atomic_inc_not_zero(&eb
->refs
)) {
182 /* loop around taking references on and locking the root node of the
183 * tree until you end up with a lock on the root. A locked buffer
184 * is returned, with a reference held.
186 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
188 struct extent_buffer
*eb
;
191 eb
= btrfs_root_node(root
);
193 if (eb
== root
->node
)
195 btrfs_tree_unlock(eb
);
196 free_extent_buffer(eb
);
201 /* loop around taking references on and locking the root node of the
202 * tree until you end up with a lock on the root. A locked buffer
203 * is returned, with a reference held.
205 static struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
207 struct extent_buffer
*eb
;
210 eb
= btrfs_root_node(root
);
211 btrfs_tree_read_lock(eb
);
212 if (eb
== root
->node
)
214 btrfs_tree_read_unlock(eb
);
215 free_extent_buffer(eb
);
220 /* cowonly root (everything not a reference counted cow subvolume), just get
221 * put onto a simple dirty list. transaction.c walks this to make sure they
222 * get properly updated on disk.
224 static void add_root_to_dirty_list(struct btrfs_root
*root
)
226 spin_lock(&root
->fs_info
->trans_lock
);
227 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
228 list_add(&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(root
->ref_cows
&& trans
->transid
!=
250 root
->fs_info
->running_transaction
->transid
);
251 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
253 level
= btrfs_header_level(buf
);
255 btrfs_item_key(buf
, &disk_key
, 0);
257 btrfs_node_key(buf
, &disk_key
, 0);
259 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, 0,
260 new_root_objectid
, &disk_key
, level
,
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, 1);
283 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
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
{
315 u64 index
; /* shifted logical */
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 * Increment the upper half of tree_mod_seq, set lower half zero.
359 * Must be called with fs_info->tree_mod_seq_lock held.
361 static inline u64
btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info
*fs_info
)
363 u64 seq
= atomic64_read(&fs_info
->tree_mod_seq
);
364 seq
&= 0xffffffff00000000ull
;
366 atomic64_set(&fs_info
->tree_mod_seq
, seq
);
371 * Increment the lower half of tree_mod_seq.
373 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
374 * are generated should not technically require a spin lock here. (Rationale:
375 * incrementing the minor while incrementing the major seq number is between its
376 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
377 * just returns a unique sequence number as usual.) We have decided to leave
378 * that requirement in here and rethink it once we notice it really imposes a
379 * problem on some workload.
381 static inline u64
btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info
*fs_info
)
383 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
387 * return the last minor in the previous major tree_mod_seq number
389 u64
btrfs_tree_mod_seq_prev(u64 seq
)
391 return (seq
& 0xffffffff00000000ull
) - 1ull;
395 * This adds a new blocker to the tree mod log's blocker list if the @elem
396 * passed does not already have a sequence number set. So when a caller expects
397 * to record tree modifications, it should ensure to set elem->seq to zero
398 * before calling btrfs_get_tree_mod_seq.
399 * Returns a fresh, unused tree log modification sequence number, even if no new
402 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
403 struct seq_list
*elem
)
407 tree_mod_log_write_lock(fs_info
);
408 spin_lock(&fs_info
->tree_mod_seq_lock
);
410 elem
->seq
= btrfs_inc_tree_mod_seq_major(fs_info
);
411 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
413 seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
414 spin_unlock(&fs_info
->tree_mod_seq_lock
);
415 tree_mod_log_write_unlock(fs_info
);
420 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
421 struct seq_list
*elem
)
423 struct rb_root
*tm_root
;
424 struct rb_node
*node
;
425 struct rb_node
*next
;
426 struct seq_list
*cur_elem
;
427 struct tree_mod_elem
*tm
;
428 u64 min_seq
= (u64
)-1;
429 u64 seq_putting
= elem
->seq
;
434 spin_lock(&fs_info
->tree_mod_seq_lock
);
435 list_del(&elem
->list
);
438 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
439 if (cur_elem
->seq
< min_seq
) {
440 if (seq_putting
> cur_elem
->seq
) {
442 * blocker with lower sequence number exists, we
443 * cannot remove anything from the log
445 spin_unlock(&fs_info
->tree_mod_seq_lock
);
448 min_seq
= cur_elem
->seq
;
451 spin_unlock(&fs_info
->tree_mod_seq_lock
);
454 * anything that's lower than the lowest existing (read: blocked)
455 * sequence number can be removed from the tree.
457 tree_mod_log_write_lock(fs_info
);
458 tm_root
= &fs_info
->tree_mod_log
;
459 for (node
= rb_first(tm_root
); node
; node
= next
) {
460 next
= rb_next(node
);
461 tm
= container_of(node
, struct tree_mod_elem
, node
);
462 if (tm
->seq
> min_seq
)
464 rb_erase(node
, tm_root
);
467 tree_mod_log_write_unlock(fs_info
);
471 * key order of the log:
474 * the index is the shifted logical of the *new* root node for root replace
475 * operations, or the shifted logical of the affected block for all other
478 * Note: must be called with write lock (tree_mod_log_write_lock).
481 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
483 struct rb_root
*tm_root
;
484 struct rb_node
**new;
485 struct rb_node
*parent
= NULL
;
486 struct tree_mod_elem
*cur
;
490 spin_lock(&fs_info
->tree_mod_seq_lock
);
491 tm
->seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
492 spin_unlock(&fs_info
->tree_mod_seq_lock
);
494 tm_root
= &fs_info
->tree_mod_log
;
495 new = &tm_root
->rb_node
;
497 cur
= container_of(*new, struct tree_mod_elem
, node
);
499 if (cur
->index
< tm
->index
)
500 new = &((*new)->rb_left
);
501 else if (cur
->index
> tm
->index
)
502 new = &((*new)->rb_right
);
503 else if (cur
->seq
< tm
->seq
)
504 new = &((*new)->rb_left
);
505 else if (cur
->seq
> tm
->seq
)
506 new = &((*new)->rb_right
);
511 rb_link_node(&tm
->node
, parent
, new);
512 rb_insert_color(&tm
->node
, tm_root
);
517 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
518 * returns zero with the tree_mod_log_lock acquired. The caller must hold
519 * this until all tree mod log insertions are recorded in the rb tree and then
520 * call tree_mod_log_write_unlock() to release.
522 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
523 struct extent_buffer
*eb
) {
525 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
527 if (eb
&& btrfs_header_level(eb
) == 0)
530 tree_mod_log_write_lock(fs_info
);
531 if (list_empty(&(fs_info
)->tree_mod_seq_list
)) {
532 tree_mod_log_write_unlock(fs_info
);
539 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
540 static inline int tree_mod_need_log(const struct btrfs_fs_info
*fs_info
,
541 struct extent_buffer
*eb
)
544 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
546 if (eb
&& btrfs_header_level(eb
) == 0)
552 static struct tree_mod_elem
*
553 alloc_tree_mod_elem(struct extent_buffer
*eb
, int slot
,
554 enum mod_log_op op
, gfp_t flags
)
556 struct tree_mod_elem
*tm
;
558 tm
= kzalloc(sizeof(*tm
), flags
);
562 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
563 if (op
!= MOD_LOG_KEY_ADD
) {
564 btrfs_node_key(eb
, &tm
->key
, slot
);
565 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
569 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
570 RB_CLEAR_NODE(&tm
->node
);
576 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
577 struct extent_buffer
*eb
, int slot
,
578 enum mod_log_op op
, gfp_t flags
)
580 struct tree_mod_elem
*tm
;
583 if (!tree_mod_need_log(fs_info
, eb
))
586 tm
= alloc_tree_mod_elem(eb
, slot
, op
, flags
);
590 if (tree_mod_dont_log(fs_info
, eb
)) {
595 ret
= __tree_mod_log_insert(fs_info
, tm
);
596 tree_mod_log_write_unlock(fs_info
);
604 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
605 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
606 int nr_items
, gfp_t flags
)
608 struct tree_mod_elem
*tm
= NULL
;
609 struct tree_mod_elem
**tm_list
= NULL
;
614 if (!tree_mod_need_log(fs_info
, eb
))
617 tm_list
= kzalloc(nr_items
* sizeof(struct tree_mod_elem
*), flags
);
621 tm
= kzalloc(sizeof(*tm
), flags
);
627 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
629 tm
->move
.dst_slot
= dst_slot
;
630 tm
->move
.nr_items
= nr_items
;
631 tm
->op
= MOD_LOG_MOVE_KEYS
;
633 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
634 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
+ dst_slot
,
635 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, flags
);
642 if (tree_mod_dont_log(fs_info
, eb
))
647 * When we override something during the move, we log these removals.
648 * This can only happen when we move towards the beginning of the
649 * buffer, i.e. dst_slot < src_slot.
651 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
652 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
657 ret
= __tree_mod_log_insert(fs_info
, tm
);
660 tree_mod_log_write_unlock(fs_info
);
665 for (i
= 0; i
< nr_items
; i
++) {
666 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
667 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
671 tree_mod_log_write_unlock(fs_info
);
679 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
680 struct tree_mod_elem
**tm_list
,
686 for (i
= nritems
- 1; i
>= 0; i
--) {
687 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
689 for (j
= nritems
- 1; j
> i
; j
--)
690 rb_erase(&tm_list
[j
]->node
,
691 &fs_info
->tree_mod_log
);
700 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
701 struct extent_buffer
*old_root
,
702 struct extent_buffer
*new_root
, gfp_t flags
,
705 struct tree_mod_elem
*tm
= NULL
;
706 struct tree_mod_elem
**tm_list
= NULL
;
711 if (!tree_mod_need_log(fs_info
, NULL
))
714 if (log_removal
&& btrfs_header_level(old_root
) > 0) {
715 nritems
= btrfs_header_nritems(old_root
);
716 tm_list
= kzalloc(nritems
* sizeof(struct tree_mod_elem
*),
722 for (i
= 0; i
< nritems
; i
++) {
723 tm_list
[i
] = alloc_tree_mod_elem(old_root
, i
,
724 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, flags
);
732 tm
= kzalloc(sizeof(*tm
), flags
);
738 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
739 tm
->old_root
.logical
= old_root
->start
;
740 tm
->old_root
.level
= btrfs_header_level(old_root
);
741 tm
->generation
= btrfs_header_generation(old_root
);
742 tm
->op
= MOD_LOG_ROOT_REPLACE
;
744 if (tree_mod_dont_log(fs_info
, NULL
))
748 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
750 ret
= __tree_mod_log_insert(fs_info
, tm
);
752 tree_mod_log_write_unlock(fs_info
);
761 for (i
= 0; i
< nritems
; i
++)
770 static struct tree_mod_elem
*
771 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
774 struct rb_root
*tm_root
;
775 struct rb_node
*node
;
776 struct tree_mod_elem
*cur
= NULL
;
777 struct tree_mod_elem
*found
= NULL
;
778 u64 index
= start
>> PAGE_CACHE_SHIFT
;
780 tree_mod_log_read_lock(fs_info
);
781 tm_root
= &fs_info
->tree_mod_log
;
782 node
= tm_root
->rb_node
;
784 cur
= container_of(node
, struct tree_mod_elem
, node
);
785 if (cur
->index
< index
) {
786 node
= node
->rb_left
;
787 } else if (cur
->index
> index
) {
788 node
= node
->rb_right
;
789 } else if (cur
->seq
< min_seq
) {
790 node
= node
->rb_left
;
791 } else if (!smallest
) {
792 /* we want the node with the highest seq */
794 BUG_ON(found
->seq
> cur
->seq
);
796 node
= node
->rb_left
;
797 } else if (cur
->seq
> min_seq
) {
798 /* we want the node with the smallest seq */
800 BUG_ON(found
->seq
< cur
->seq
);
802 node
= node
->rb_right
;
808 tree_mod_log_read_unlock(fs_info
);
814 * this returns the element from the log with the smallest time sequence
815 * value that's in the log (the oldest log item). any element with a time
816 * sequence lower than min_seq will be ignored.
818 static struct tree_mod_elem
*
819 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
822 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
826 * this returns the element from the log with the largest time sequence
827 * value that's in the log (the most recent log item). any element with
828 * a time sequence lower than min_seq will be ignored.
830 static struct tree_mod_elem
*
831 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
833 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
837 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
838 struct extent_buffer
*src
, unsigned long dst_offset
,
839 unsigned long src_offset
, int nr_items
)
842 struct tree_mod_elem
**tm_list
= NULL
;
843 struct tree_mod_elem
**tm_list_add
, **tm_list_rem
;
847 if (!tree_mod_need_log(fs_info
, NULL
))
850 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
853 tm_list
= kzalloc(nr_items
* 2 * sizeof(struct tree_mod_elem
*),
858 tm_list_add
= tm_list
;
859 tm_list_rem
= tm_list
+ nr_items
;
860 for (i
= 0; i
< nr_items
; i
++) {
861 tm_list_rem
[i
] = alloc_tree_mod_elem(src
, i
+ src_offset
,
862 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
863 if (!tm_list_rem
[i
]) {
868 tm_list_add
[i
] = alloc_tree_mod_elem(dst
, i
+ dst_offset
,
869 MOD_LOG_KEY_ADD
, GFP_NOFS
);
870 if (!tm_list_add
[i
]) {
876 if (tree_mod_dont_log(fs_info
, NULL
))
880 for (i
= 0; i
< nr_items
; i
++) {
881 ret
= __tree_mod_log_insert(fs_info
, tm_list_rem
[i
]);
884 ret
= __tree_mod_log_insert(fs_info
, tm_list_add
[i
]);
889 tree_mod_log_write_unlock(fs_info
);
895 for (i
= 0; i
< nr_items
* 2; i
++) {
896 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
897 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
901 tree_mod_log_write_unlock(fs_info
);
908 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
909 int dst_offset
, int src_offset
, int nr_items
)
912 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
918 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
919 struct extent_buffer
*eb
, int slot
, int atomic
)
923 ret
= tree_mod_log_insert_key(fs_info
, eb
, slot
,
925 atomic
? GFP_ATOMIC
: GFP_NOFS
);
930 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
932 struct tree_mod_elem
**tm_list
= NULL
;
937 if (btrfs_header_level(eb
) == 0)
940 if (!tree_mod_need_log(fs_info
, NULL
))
943 nritems
= btrfs_header_nritems(eb
);
944 tm_list
= kzalloc(nritems
* sizeof(struct tree_mod_elem
*),
949 for (i
= 0; i
< nritems
; i
++) {
950 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
,
951 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
958 if (tree_mod_dont_log(fs_info
, eb
))
961 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
962 tree_mod_log_write_unlock(fs_info
);
970 for (i
= 0; i
< nritems
; i
++)
978 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
979 struct extent_buffer
*new_root_node
,
983 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
984 new_root_node
, GFP_NOFS
, log_removal
);
989 * check if the tree block can be shared by multiple trees
991 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
992 struct extent_buffer
*buf
)
995 * Tree blocks not in refernece counted trees and tree roots
996 * are never shared. If a block was allocated after the last
997 * snapshot and the block was not allocated by tree relocation,
998 * we know the block is not shared.
1000 if (root
->ref_cows
&&
1001 buf
!= root
->node
&& buf
!= root
->commit_root
&&
1002 (btrfs_header_generation(buf
) <=
1003 btrfs_root_last_snapshot(&root
->root_item
) ||
1004 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
1006 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1007 if (root
->ref_cows
&&
1008 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1014 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
1015 struct btrfs_root
*root
,
1016 struct extent_buffer
*buf
,
1017 struct extent_buffer
*cow
,
1027 * Backrefs update rules:
1029 * Always use full backrefs for extent pointers in tree block
1030 * allocated by tree relocation.
1032 * If a shared tree block is no longer referenced by its owner
1033 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
1034 * use full backrefs for extent pointers in tree block.
1036 * If a tree block is been relocating
1037 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1038 * use full backrefs for extent pointers in tree block.
1039 * The reason for this is some operations (such as drop tree)
1040 * are only allowed for blocks use full backrefs.
1043 if (btrfs_block_can_be_shared(root
, buf
)) {
1044 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
1045 btrfs_header_level(buf
), 1,
1051 btrfs_std_error(root
->fs_info
, ret
);
1056 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1057 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1058 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1063 owner
= btrfs_header_owner(buf
);
1064 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
1065 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
1068 if ((owner
== root
->root_key
.objectid
||
1069 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
1070 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
1071 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
1072 BUG_ON(ret
); /* -ENOMEM */
1074 if (root
->root_key
.objectid
==
1075 BTRFS_TREE_RELOC_OBJECTID
) {
1076 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
1077 BUG_ON(ret
); /* -ENOMEM */
1078 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1079 BUG_ON(ret
); /* -ENOMEM */
1081 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1084 if (root
->root_key
.objectid
==
1085 BTRFS_TREE_RELOC_OBJECTID
)
1086 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1088 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
1089 BUG_ON(ret
); /* -ENOMEM */
1091 if (new_flags
!= 0) {
1092 int level
= btrfs_header_level(buf
);
1094 ret
= btrfs_set_disk_extent_flags(trans
, root
,
1097 new_flags
, level
, 0);
1102 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
1103 if (root
->root_key
.objectid
==
1104 BTRFS_TREE_RELOC_OBJECTID
)
1105 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1107 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
1108 BUG_ON(ret
); /* -ENOMEM */
1109 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
1110 BUG_ON(ret
); /* -ENOMEM */
1112 clean_tree_block(trans
, root
, buf
);
1119 * does the dirty work in cow of a single block. The parent block (if
1120 * supplied) is updated to point to the new cow copy. The new buffer is marked
1121 * dirty and returned locked. If you modify the block it needs to be marked
1124 * search_start -- an allocation hint for the new block
1126 * empty_size -- a hint that you plan on doing more cow. This is the size in
1127 * bytes the allocator should try to find free next to the block it returns.
1128 * This is just a hint and may be ignored by the allocator.
1130 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1131 struct btrfs_root
*root
,
1132 struct extent_buffer
*buf
,
1133 struct extent_buffer
*parent
, int parent_slot
,
1134 struct extent_buffer
**cow_ret
,
1135 u64 search_start
, u64 empty_size
)
1137 struct btrfs_disk_key disk_key
;
1138 struct extent_buffer
*cow
;
1141 int unlock_orig
= 0;
1144 if (*cow_ret
== buf
)
1147 btrfs_assert_tree_locked(buf
);
1149 WARN_ON(root
->ref_cows
&& trans
->transid
!=
1150 root
->fs_info
->running_transaction
->transid
);
1151 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
1153 level
= btrfs_header_level(buf
);
1156 btrfs_item_key(buf
, &disk_key
, 0);
1158 btrfs_node_key(buf
, &disk_key
, 0);
1160 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
1162 parent_start
= parent
->start
;
1168 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
1169 root
->root_key
.objectid
, &disk_key
,
1170 level
, search_start
, empty_size
);
1172 return PTR_ERR(cow
);
1174 /* cow is set to blocking by btrfs_init_new_buffer */
1176 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
1177 btrfs_set_header_bytenr(cow
, cow
->start
);
1178 btrfs_set_header_generation(cow
, trans
->transid
);
1179 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1180 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1181 BTRFS_HEADER_FLAG_RELOC
);
1182 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1183 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1185 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1187 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
1190 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1192 btrfs_abort_transaction(trans
, root
, ret
);
1196 if (root
->ref_cows
) {
1197 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1202 if (buf
== root
->node
) {
1203 WARN_ON(parent
&& parent
!= buf
);
1204 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1205 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1206 parent_start
= buf
->start
;
1210 extent_buffer_get(cow
);
1211 tree_mod_log_set_root_pointer(root
, cow
, 1);
1212 rcu_assign_pointer(root
->node
, cow
);
1214 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1216 free_extent_buffer(buf
);
1217 add_root_to_dirty_list(root
);
1219 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1220 parent_start
= parent
->start
;
1224 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1225 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1226 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1227 btrfs_set_node_blockptr(parent
, parent_slot
,
1229 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1231 btrfs_mark_buffer_dirty(parent
);
1233 ret
= tree_mod_log_free_eb(root
->fs_info
, buf
);
1235 btrfs_abort_transaction(trans
, root
, ret
);
1239 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1243 btrfs_tree_unlock(buf
);
1244 free_extent_buffer_stale(buf
);
1245 btrfs_mark_buffer_dirty(cow
);
1251 * returns the logical address of the oldest predecessor of the given root.
1252 * entries older than time_seq are ignored.
1254 static struct tree_mod_elem
*
1255 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1256 struct extent_buffer
*eb_root
, u64 time_seq
)
1258 struct tree_mod_elem
*tm
;
1259 struct tree_mod_elem
*found
= NULL
;
1260 u64 root_logical
= eb_root
->start
;
1267 * the very last operation that's logged for a root is the replacement
1268 * operation (if it is replaced at all). this has the index of the *new*
1269 * root, making it the very first operation that's logged for this root.
1272 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1277 * if there are no tree operation for the oldest root, we simply
1278 * return it. this should only happen if that (old) root is at
1285 * if there's an operation that's not a root replacement, we
1286 * found the oldest version of our root. normally, we'll find a
1287 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1289 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1293 root_logical
= tm
->old_root
.logical
;
1297 /* if there's no old root to return, return what we found instead */
1305 * tm is a pointer to the first operation to rewind within eb. then, all
1306 * previous operations will be rewinded (until we reach something older than
1310 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1311 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1314 struct rb_node
*next
;
1315 struct tree_mod_elem
*tm
= first_tm
;
1316 unsigned long o_dst
;
1317 unsigned long o_src
;
1318 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1320 n
= btrfs_header_nritems(eb
);
1321 tree_mod_log_read_lock(fs_info
);
1322 while (tm
&& tm
->seq
>= time_seq
) {
1324 * all the operations are recorded with the operator used for
1325 * the modification. as we're going backwards, we do the
1326 * opposite of each operation here.
1329 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1330 BUG_ON(tm
->slot
< n
);
1332 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1333 case MOD_LOG_KEY_REMOVE
:
1334 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1335 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1336 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1340 case MOD_LOG_KEY_REPLACE
:
1341 BUG_ON(tm
->slot
>= n
);
1342 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1343 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1344 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1347 case MOD_LOG_KEY_ADD
:
1348 /* if a move operation is needed it's in the log */
1351 case MOD_LOG_MOVE_KEYS
:
1352 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1353 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1354 memmove_extent_buffer(eb
, o_dst
, o_src
,
1355 tm
->move
.nr_items
* p_size
);
1357 case MOD_LOG_ROOT_REPLACE
:
1359 * this operation is special. for roots, this must be
1360 * handled explicitly before rewinding.
1361 * for non-roots, this operation may exist if the node
1362 * was a root: root A -> child B; then A gets empty and
1363 * B is promoted to the new root. in the mod log, we'll
1364 * have a root-replace operation for B, a tree block
1365 * that is no root. we simply ignore that operation.
1369 next
= rb_next(&tm
->node
);
1372 tm
= container_of(next
, struct tree_mod_elem
, node
);
1373 if (tm
->index
!= first_tm
->index
)
1376 tree_mod_log_read_unlock(fs_info
);
1377 btrfs_set_header_nritems(eb
, n
);
1381 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1382 * is returned. If rewind operations happen, a fresh buffer is returned. The
1383 * returned buffer is always read-locked. If the returned buffer is not the
1384 * input buffer, the lock on the input buffer is released and the input buffer
1385 * is freed (its refcount is decremented).
1387 static struct extent_buffer
*
1388 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1389 struct extent_buffer
*eb
, u64 time_seq
)
1391 struct extent_buffer
*eb_rewin
;
1392 struct tree_mod_elem
*tm
;
1397 if (btrfs_header_level(eb
) == 0)
1400 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1404 btrfs_set_path_blocking(path
);
1405 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1407 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1408 BUG_ON(tm
->slot
!= 0);
1409 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1410 fs_info
->tree_root
->nodesize
);
1412 btrfs_tree_read_unlock_blocking(eb
);
1413 free_extent_buffer(eb
);
1416 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1417 btrfs_set_header_backref_rev(eb_rewin
,
1418 btrfs_header_backref_rev(eb
));
1419 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1420 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1422 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1424 btrfs_tree_read_unlock_blocking(eb
);
1425 free_extent_buffer(eb
);
1430 btrfs_clear_path_blocking(path
, NULL
, BTRFS_READ_LOCK
);
1431 btrfs_tree_read_unlock_blocking(eb
);
1432 free_extent_buffer(eb
);
1434 extent_buffer_get(eb_rewin
);
1435 btrfs_tree_read_lock(eb_rewin
);
1436 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1437 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1438 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1444 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1445 * value. If there are no changes, the current root->root_node is returned. If
1446 * anything changed in between, there's a fresh buffer allocated on which the
1447 * rewind operations are done. In any case, the returned buffer is read locked.
1448 * Returns NULL on error (with no locks held).
1450 static inline struct extent_buffer
*
1451 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1453 struct tree_mod_elem
*tm
;
1454 struct extent_buffer
*eb
= NULL
;
1455 struct extent_buffer
*eb_root
;
1456 struct extent_buffer
*old
;
1457 struct tree_mod_root
*old_root
= NULL
;
1458 u64 old_generation
= 0;
1462 eb_root
= btrfs_read_lock_root_node(root
);
1463 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1467 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1468 old_root
= &tm
->old_root
;
1469 old_generation
= tm
->generation
;
1470 logical
= old_root
->logical
;
1472 logical
= eb_root
->start
;
1475 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1476 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1477 btrfs_tree_read_unlock(eb_root
);
1478 free_extent_buffer(eb_root
);
1479 blocksize
= btrfs_level_size(root
, old_root
->level
);
1480 old
= read_tree_block(root
, logical
, blocksize
, 0);
1481 if (WARN_ON(!old
|| !extent_buffer_uptodate(old
))) {
1482 free_extent_buffer(old
);
1483 btrfs_warn(root
->fs_info
,
1484 "failed to read tree block %llu from get_old_root", logical
);
1486 eb
= btrfs_clone_extent_buffer(old
);
1487 free_extent_buffer(old
);
1489 } else if (old_root
) {
1490 btrfs_tree_read_unlock(eb_root
);
1491 free_extent_buffer(eb_root
);
1492 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1494 btrfs_set_lock_blocking_rw(eb_root
, BTRFS_READ_LOCK
);
1495 eb
= btrfs_clone_extent_buffer(eb_root
);
1496 btrfs_tree_read_unlock_blocking(eb_root
);
1497 free_extent_buffer(eb_root
);
1502 extent_buffer_get(eb
);
1503 btrfs_tree_read_lock(eb
);
1505 btrfs_set_header_bytenr(eb
, eb
->start
);
1506 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1507 btrfs_set_header_owner(eb
, btrfs_header_owner(eb_root
));
1508 btrfs_set_header_level(eb
, old_root
->level
);
1509 btrfs_set_header_generation(eb
, old_generation
);
1512 __tree_mod_log_rewind(root
->fs_info
, eb
, time_seq
, tm
);
1514 WARN_ON(btrfs_header_level(eb
) != 0);
1515 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1520 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1522 struct tree_mod_elem
*tm
;
1524 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1526 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1527 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1528 level
= tm
->old_root
.level
;
1530 level
= btrfs_header_level(eb_root
);
1532 free_extent_buffer(eb_root
);
1537 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1538 struct btrfs_root
*root
,
1539 struct extent_buffer
*buf
)
1541 /* ensure we can see the force_cow */
1545 * We do not need to cow a block if
1546 * 1) this block is not created or changed in this transaction;
1547 * 2) this block does not belong to TREE_RELOC tree;
1548 * 3) the root is not forced COW.
1550 * What is forced COW:
1551 * when we create snapshot during commiting the transaction,
1552 * after we've finished coping src root, we must COW the shared
1553 * block to ensure the metadata consistency.
1555 if (btrfs_header_generation(buf
) == trans
->transid
&&
1556 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1557 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1558 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1565 * cows a single block, see __btrfs_cow_block for the real work.
1566 * This version of it has extra checks so that a block isn't cow'd more than
1567 * once per transaction, as long as it hasn't been written yet
1569 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1570 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1571 struct extent_buffer
*parent
, int parent_slot
,
1572 struct extent_buffer
**cow_ret
)
1577 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1578 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1580 root
->fs_info
->running_transaction
->transid
);
1582 if (trans
->transid
!= root
->fs_info
->generation
)
1583 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1584 trans
->transid
, root
->fs_info
->generation
);
1586 if (!should_cow_block(trans
, root
, buf
)) {
1591 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1594 btrfs_set_lock_blocking(parent
);
1595 btrfs_set_lock_blocking(buf
);
1597 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1598 parent_slot
, cow_ret
, search_start
, 0);
1600 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1606 * helper function for defrag to decide if two blocks pointed to by a
1607 * node are actually close by
1609 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1611 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1613 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1619 * compare two keys in a memcmp fashion
1621 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1623 struct btrfs_key k1
;
1625 btrfs_disk_key_to_cpu(&k1
, disk
);
1627 return btrfs_comp_cpu_keys(&k1
, k2
);
1631 * same as comp_keys only with two btrfs_key's
1633 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1635 if (k1
->objectid
> k2
->objectid
)
1637 if (k1
->objectid
< k2
->objectid
)
1639 if (k1
->type
> k2
->type
)
1641 if (k1
->type
< k2
->type
)
1643 if (k1
->offset
> k2
->offset
)
1645 if (k1
->offset
< k2
->offset
)
1651 * this is used by the defrag code to go through all the
1652 * leaves pointed to by a node and reallocate them so that
1653 * disk order is close to key order
1655 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1656 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1657 int start_slot
, u64
*last_ret
,
1658 struct btrfs_key
*progress
)
1660 struct extent_buffer
*cur
;
1663 u64 search_start
= *last_ret
;
1673 int progress_passed
= 0;
1674 struct btrfs_disk_key disk_key
;
1676 parent_level
= btrfs_header_level(parent
);
1678 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1679 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1681 parent_nritems
= btrfs_header_nritems(parent
);
1682 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1683 end_slot
= parent_nritems
;
1685 if (parent_nritems
== 1)
1688 btrfs_set_lock_blocking(parent
);
1690 for (i
= start_slot
; i
< end_slot
; i
++) {
1693 btrfs_node_key(parent
, &disk_key
, i
);
1694 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1697 progress_passed
= 1;
1698 blocknr
= btrfs_node_blockptr(parent
, i
);
1699 gen
= btrfs_node_ptr_generation(parent
, i
);
1700 if (last_block
== 0)
1701 last_block
= blocknr
;
1704 other
= btrfs_node_blockptr(parent
, i
- 1);
1705 close
= close_blocks(blocknr
, other
, blocksize
);
1707 if (!close
&& i
< end_slot
- 2) {
1708 other
= btrfs_node_blockptr(parent
, i
+ 1);
1709 close
= close_blocks(blocknr
, other
, blocksize
);
1712 last_block
= blocknr
;
1716 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1718 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1721 if (!cur
|| !uptodate
) {
1723 cur
= read_tree_block(root
, blocknr
,
1725 if (!cur
|| !extent_buffer_uptodate(cur
)) {
1726 free_extent_buffer(cur
);
1729 } else if (!uptodate
) {
1730 err
= btrfs_read_buffer(cur
, gen
);
1732 free_extent_buffer(cur
);
1737 if (search_start
== 0)
1738 search_start
= last_block
;
1740 btrfs_tree_lock(cur
);
1741 btrfs_set_lock_blocking(cur
);
1742 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1745 (end_slot
- i
) * blocksize
));
1747 btrfs_tree_unlock(cur
);
1748 free_extent_buffer(cur
);
1751 search_start
= cur
->start
;
1752 last_block
= cur
->start
;
1753 *last_ret
= search_start
;
1754 btrfs_tree_unlock(cur
);
1755 free_extent_buffer(cur
);
1761 * The leaf data grows from end-to-front in the node.
1762 * this returns the address of the start of the last item,
1763 * which is the stop of the leaf data stack
1765 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1766 struct extent_buffer
*leaf
)
1768 u32 nr
= btrfs_header_nritems(leaf
);
1770 return BTRFS_LEAF_DATA_SIZE(root
);
1771 return btrfs_item_offset_nr(leaf
, nr
- 1);
1776 * search for key in the extent_buffer. The items start at offset p,
1777 * and they are item_size apart. There are 'max' items in p.
1779 * the slot in the array is returned via slot, and it points to
1780 * the place where you would insert key if it is not found in
1783 * slot may point to max if the key is bigger than all of the keys
1785 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1787 int item_size
, struct btrfs_key
*key
,
1794 struct btrfs_disk_key
*tmp
= NULL
;
1795 struct btrfs_disk_key unaligned
;
1796 unsigned long offset
;
1798 unsigned long map_start
= 0;
1799 unsigned long map_len
= 0;
1802 while (low
< high
) {
1803 mid
= (low
+ high
) / 2;
1804 offset
= p
+ mid
* item_size
;
1806 if (!kaddr
|| offset
< map_start
||
1807 (offset
+ sizeof(struct btrfs_disk_key
)) >
1808 map_start
+ map_len
) {
1810 err
= map_private_extent_buffer(eb
, offset
,
1811 sizeof(struct btrfs_disk_key
),
1812 &kaddr
, &map_start
, &map_len
);
1815 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1818 read_extent_buffer(eb
, &unaligned
,
1819 offset
, sizeof(unaligned
));
1824 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1827 ret
= comp_keys(tmp
, key
);
1843 * simple bin_search frontend that does the right thing for
1846 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1847 int level
, int *slot
)
1850 return generic_bin_search(eb
,
1851 offsetof(struct btrfs_leaf
, items
),
1852 sizeof(struct btrfs_item
),
1853 key
, btrfs_header_nritems(eb
),
1856 return generic_bin_search(eb
,
1857 offsetof(struct btrfs_node
, ptrs
),
1858 sizeof(struct btrfs_key_ptr
),
1859 key
, btrfs_header_nritems(eb
),
1863 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1864 int level
, int *slot
)
1866 return bin_search(eb
, key
, level
, slot
);
1869 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1871 spin_lock(&root
->accounting_lock
);
1872 btrfs_set_root_used(&root
->root_item
,
1873 btrfs_root_used(&root
->root_item
) + size
);
1874 spin_unlock(&root
->accounting_lock
);
1877 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1879 spin_lock(&root
->accounting_lock
);
1880 btrfs_set_root_used(&root
->root_item
,
1881 btrfs_root_used(&root
->root_item
) - size
);
1882 spin_unlock(&root
->accounting_lock
);
1885 /* given a node and slot number, this reads the blocks it points to. The
1886 * extent buffer is returned with a reference taken (but unlocked).
1887 * NULL is returned on error.
1889 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1890 struct extent_buffer
*parent
, int slot
)
1892 int level
= btrfs_header_level(parent
);
1893 struct extent_buffer
*eb
;
1897 if (slot
>= btrfs_header_nritems(parent
))
1902 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1903 btrfs_level_size(root
, level
- 1),
1904 btrfs_node_ptr_generation(parent
, slot
));
1905 if (eb
&& !extent_buffer_uptodate(eb
)) {
1906 free_extent_buffer(eb
);
1914 * node level balancing, used to make sure nodes are in proper order for
1915 * item deletion. We balance from the top down, so we have to make sure
1916 * that a deletion won't leave an node completely empty later on.
1918 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1919 struct btrfs_root
*root
,
1920 struct btrfs_path
*path
, int level
)
1922 struct extent_buffer
*right
= NULL
;
1923 struct extent_buffer
*mid
;
1924 struct extent_buffer
*left
= NULL
;
1925 struct extent_buffer
*parent
= NULL
;
1929 int orig_slot
= path
->slots
[level
];
1935 mid
= path
->nodes
[level
];
1937 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1938 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1939 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1941 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1943 if (level
< BTRFS_MAX_LEVEL
- 1) {
1944 parent
= path
->nodes
[level
+ 1];
1945 pslot
= path
->slots
[level
+ 1];
1949 * deal with the case where there is only one pointer in the root
1950 * by promoting the node below to a root
1953 struct extent_buffer
*child
;
1955 if (btrfs_header_nritems(mid
) != 1)
1958 /* promote the child to a root */
1959 child
= read_node_slot(root
, mid
, 0);
1962 btrfs_std_error(root
->fs_info
, ret
);
1966 btrfs_tree_lock(child
);
1967 btrfs_set_lock_blocking(child
);
1968 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1970 btrfs_tree_unlock(child
);
1971 free_extent_buffer(child
);
1975 tree_mod_log_set_root_pointer(root
, child
, 1);
1976 rcu_assign_pointer(root
->node
, child
);
1978 add_root_to_dirty_list(root
);
1979 btrfs_tree_unlock(child
);
1981 path
->locks
[level
] = 0;
1982 path
->nodes
[level
] = NULL
;
1983 clean_tree_block(trans
, root
, mid
);
1984 btrfs_tree_unlock(mid
);
1985 /* once for the path */
1986 free_extent_buffer(mid
);
1988 root_sub_used(root
, mid
->len
);
1989 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1990 /* once for the root ptr */
1991 free_extent_buffer_stale(mid
);
1994 if (btrfs_header_nritems(mid
) >
1995 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1998 left
= read_node_slot(root
, parent
, pslot
- 1);
2000 btrfs_tree_lock(left
);
2001 btrfs_set_lock_blocking(left
);
2002 wret
= btrfs_cow_block(trans
, root
, left
,
2003 parent
, pslot
- 1, &left
);
2009 right
= read_node_slot(root
, parent
, pslot
+ 1);
2011 btrfs_tree_lock(right
);
2012 btrfs_set_lock_blocking(right
);
2013 wret
= btrfs_cow_block(trans
, root
, right
,
2014 parent
, pslot
+ 1, &right
);
2021 /* first, try to make some room in the middle buffer */
2023 orig_slot
+= btrfs_header_nritems(left
);
2024 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2030 * then try to empty the right most buffer into the middle
2033 wret
= push_node_left(trans
, root
, mid
, right
, 1);
2034 if (wret
< 0 && wret
!= -ENOSPC
)
2036 if (btrfs_header_nritems(right
) == 0) {
2037 clean_tree_block(trans
, root
, right
);
2038 btrfs_tree_unlock(right
);
2039 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
2040 root_sub_used(root
, right
->len
);
2041 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
2042 free_extent_buffer_stale(right
);
2045 struct btrfs_disk_key right_key
;
2046 btrfs_node_key(right
, &right_key
, 0);
2047 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2049 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
2050 btrfs_mark_buffer_dirty(parent
);
2053 if (btrfs_header_nritems(mid
) == 1) {
2055 * we're not allowed to leave a node with one item in the
2056 * tree during a delete. A deletion from lower in the tree
2057 * could try to delete the only pointer in this node.
2058 * So, pull some keys from the left.
2059 * There has to be a left pointer at this point because
2060 * otherwise we would have pulled some pointers from the
2065 btrfs_std_error(root
->fs_info
, ret
);
2068 wret
= balance_node_right(trans
, root
, mid
, left
);
2074 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2080 if (btrfs_header_nritems(mid
) == 0) {
2081 clean_tree_block(trans
, root
, mid
);
2082 btrfs_tree_unlock(mid
);
2083 del_ptr(root
, path
, level
+ 1, pslot
);
2084 root_sub_used(root
, mid
->len
);
2085 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
2086 free_extent_buffer_stale(mid
);
2089 /* update the parent key to reflect our changes */
2090 struct btrfs_disk_key mid_key
;
2091 btrfs_node_key(mid
, &mid_key
, 0);
2092 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2094 btrfs_set_node_key(parent
, &mid_key
, pslot
);
2095 btrfs_mark_buffer_dirty(parent
);
2098 /* update the path */
2100 if (btrfs_header_nritems(left
) > orig_slot
) {
2101 extent_buffer_get(left
);
2102 /* left was locked after cow */
2103 path
->nodes
[level
] = left
;
2104 path
->slots
[level
+ 1] -= 1;
2105 path
->slots
[level
] = orig_slot
;
2107 btrfs_tree_unlock(mid
);
2108 free_extent_buffer(mid
);
2111 orig_slot
-= btrfs_header_nritems(left
);
2112 path
->slots
[level
] = orig_slot
;
2115 /* double check we haven't messed things up */
2117 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2121 btrfs_tree_unlock(right
);
2122 free_extent_buffer(right
);
2125 if (path
->nodes
[level
] != left
)
2126 btrfs_tree_unlock(left
);
2127 free_extent_buffer(left
);
2132 /* Node balancing for insertion. Here we only split or push nodes around
2133 * when they are completely full. This is also done top down, so we
2134 * have to be pessimistic.
2136 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2137 struct btrfs_root
*root
,
2138 struct btrfs_path
*path
, int level
)
2140 struct extent_buffer
*right
= NULL
;
2141 struct extent_buffer
*mid
;
2142 struct extent_buffer
*left
= NULL
;
2143 struct extent_buffer
*parent
= NULL
;
2147 int orig_slot
= path
->slots
[level
];
2152 mid
= path
->nodes
[level
];
2153 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2155 if (level
< BTRFS_MAX_LEVEL
- 1) {
2156 parent
= path
->nodes
[level
+ 1];
2157 pslot
= path
->slots
[level
+ 1];
2163 left
= read_node_slot(root
, parent
, pslot
- 1);
2165 /* first, try to make some room in the middle buffer */
2169 btrfs_tree_lock(left
);
2170 btrfs_set_lock_blocking(left
);
2172 left_nr
= btrfs_header_nritems(left
);
2173 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2176 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2181 wret
= push_node_left(trans
, root
,
2188 struct btrfs_disk_key disk_key
;
2189 orig_slot
+= left_nr
;
2190 btrfs_node_key(mid
, &disk_key
, 0);
2191 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2193 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2194 btrfs_mark_buffer_dirty(parent
);
2195 if (btrfs_header_nritems(left
) > orig_slot
) {
2196 path
->nodes
[level
] = left
;
2197 path
->slots
[level
+ 1] -= 1;
2198 path
->slots
[level
] = orig_slot
;
2199 btrfs_tree_unlock(mid
);
2200 free_extent_buffer(mid
);
2203 btrfs_header_nritems(left
);
2204 path
->slots
[level
] = orig_slot
;
2205 btrfs_tree_unlock(left
);
2206 free_extent_buffer(left
);
2210 btrfs_tree_unlock(left
);
2211 free_extent_buffer(left
);
2213 right
= read_node_slot(root
, parent
, pslot
+ 1);
2216 * then try to empty the right most buffer into the middle
2221 btrfs_tree_lock(right
);
2222 btrfs_set_lock_blocking(right
);
2224 right_nr
= btrfs_header_nritems(right
);
2225 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2228 ret
= btrfs_cow_block(trans
, root
, right
,
2234 wret
= balance_node_right(trans
, root
,
2241 struct btrfs_disk_key disk_key
;
2243 btrfs_node_key(right
, &disk_key
, 0);
2244 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2246 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2247 btrfs_mark_buffer_dirty(parent
);
2249 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2250 path
->nodes
[level
] = right
;
2251 path
->slots
[level
+ 1] += 1;
2252 path
->slots
[level
] = orig_slot
-
2253 btrfs_header_nritems(mid
);
2254 btrfs_tree_unlock(mid
);
2255 free_extent_buffer(mid
);
2257 btrfs_tree_unlock(right
);
2258 free_extent_buffer(right
);
2262 btrfs_tree_unlock(right
);
2263 free_extent_buffer(right
);
2269 * readahead one full node of leaves, finding things that are close
2270 * to the block in 'slot', and triggering ra on them.
2272 static void reada_for_search(struct btrfs_root
*root
,
2273 struct btrfs_path
*path
,
2274 int level
, int slot
, u64 objectid
)
2276 struct extent_buffer
*node
;
2277 struct btrfs_disk_key disk_key
;
2283 int direction
= path
->reada
;
2284 struct extent_buffer
*eb
;
2292 if (!path
->nodes
[level
])
2295 node
= path
->nodes
[level
];
2297 search
= btrfs_node_blockptr(node
, slot
);
2298 blocksize
= btrfs_level_size(root
, level
- 1);
2299 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
2301 free_extent_buffer(eb
);
2307 nritems
= btrfs_header_nritems(node
);
2311 if (direction
< 0) {
2315 } else if (direction
> 0) {
2320 if (path
->reada
< 0 && objectid
) {
2321 btrfs_node_key(node
, &disk_key
, nr
);
2322 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2325 search
= btrfs_node_blockptr(node
, nr
);
2326 if ((search
<= target
&& target
- search
<= 65536) ||
2327 (search
> target
&& search
- target
<= 65536)) {
2328 gen
= btrfs_node_ptr_generation(node
, nr
);
2329 readahead_tree_block(root
, search
, blocksize
, gen
);
2333 if ((nread
> 65536 || nscan
> 32))
2338 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2339 struct btrfs_path
*path
, int level
)
2343 struct extent_buffer
*parent
;
2344 struct extent_buffer
*eb
;
2350 parent
= path
->nodes
[level
+ 1];
2354 nritems
= btrfs_header_nritems(parent
);
2355 slot
= path
->slots
[level
+ 1];
2356 blocksize
= btrfs_level_size(root
, level
);
2359 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2360 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2361 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2363 * if we get -eagain from btrfs_buffer_uptodate, we
2364 * don't want to return eagain here. That will loop
2367 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2369 free_extent_buffer(eb
);
2371 if (slot
+ 1 < nritems
) {
2372 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2373 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2374 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2375 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2377 free_extent_buffer(eb
);
2381 readahead_tree_block(root
, block1
, blocksize
, 0);
2383 readahead_tree_block(root
, block2
, blocksize
, 0);
2388 * when we walk down the tree, it is usually safe to unlock the higher layers
2389 * in the tree. The exceptions are when our path goes through slot 0, because
2390 * operations on the tree might require changing key pointers higher up in the
2393 * callers might also have set path->keep_locks, which tells this code to keep
2394 * the lock if the path points to the last slot in the block. This is part of
2395 * walking through the tree, and selecting the next slot in the higher block.
2397 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2398 * if lowest_unlock is 1, level 0 won't be unlocked
2400 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2401 int lowest_unlock
, int min_write_lock_level
,
2402 int *write_lock_level
)
2405 int skip_level
= level
;
2407 struct extent_buffer
*t
;
2409 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2410 if (!path
->nodes
[i
])
2412 if (!path
->locks
[i
])
2414 if (!no_skips
&& path
->slots
[i
] == 0) {
2418 if (!no_skips
&& path
->keep_locks
) {
2421 nritems
= btrfs_header_nritems(t
);
2422 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2427 if (skip_level
< i
&& i
>= lowest_unlock
)
2431 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2432 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2434 if (write_lock_level
&&
2435 i
> min_write_lock_level
&&
2436 i
<= *write_lock_level
) {
2437 *write_lock_level
= i
- 1;
2444 * This releases any locks held in the path starting at level and
2445 * going all the way up to the root.
2447 * btrfs_search_slot will keep the lock held on higher nodes in a few
2448 * corner cases, such as COW of the block at slot zero in the node. This
2449 * ignores those rules, and it should only be called when there are no
2450 * more updates to be done higher up in the tree.
2452 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2456 if (path
->keep_locks
)
2459 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2460 if (!path
->nodes
[i
])
2462 if (!path
->locks
[i
])
2464 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2470 * helper function for btrfs_search_slot. The goal is to find a block
2471 * in cache without setting the path to blocking. If we find the block
2472 * we return zero and the path is unchanged.
2474 * If we can't find the block, we set the path blocking and do some
2475 * reada. -EAGAIN is returned and the search must be repeated.
2478 read_block_for_search(struct btrfs_trans_handle
*trans
,
2479 struct btrfs_root
*root
, struct btrfs_path
*p
,
2480 struct extent_buffer
**eb_ret
, int level
, int slot
,
2481 struct btrfs_key
*key
, u64 time_seq
)
2486 struct extent_buffer
*b
= *eb_ret
;
2487 struct extent_buffer
*tmp
;
2490 blocknr
= btrfs_node_blockptr(b
, slot
);
2491 gen
= btrfs_node_ptr_generation(b
, slot
);
2492 blocksize
= btrfs_level_size(root
, level
- 1);
2494 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2496 /* first we do an atomic uptodate check */
2497 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2502 /* the pages were up to date, but we failed
2503 * the generation number check. Do a full
2504 * read for the generation number that is correct.
2505 * We must do this without dropping locks so
2506 * we can trust our generation number
2508 btrfs_set_path_blocking(p
);
2510 /* now we're allowed to do a blocking uptodate check */
2511 ret
= btrfs_read_buffer(tmp
, gen
);
2516 free_extent_buffer(tmp
);
2517 btrfs_release_path(p
);
2522 * reduce lock contention at high levels
2523 * of the btree by dropping locks before
2524 * we read. Don't release the lock on the current
2525 * level because we need to walk this node to figure
2526 * out which blocks to read.
2528 btrfs_unlock_up_safe(p
, level
+ 1);
2529 btrfs_set_path_blocking(p
);
2531 free_extent_buffer(tmp
);
2533 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2535 btrfs_release_path(p
);
2538 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2541 * If the read above didn't mark this buffer up to date,
2542 * it will never end up being up to date. Set ret to EIO now
2543 * and give up so that our caller doesn't loop forever
2546 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2548 free_extent_buffer(tmp
);
2554 * helper function for btrfs_search_slot. This does all of the checks
2555 * for node-level blocks and does any balancing required based on
2558 * If no extra work was required, zero is returned. If we had to
2559 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2563 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2564 struct btrfs_root
*root
, struct btrfs_path
*p
,
2565 struct extent_buffer
*b
, int level
, int ins_len
,
2566 int *write_lock_level
)
2569 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2570 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2573 if (*write_lock_level
< level
+ 1) {
2574 *write_lock_level
= level
+ 1;
2575 btrfs_release_path(p
);
2579 btrfs_set_path_blocking(p
);
2580 reada_for_balance(root
, p
, level
);
2581 sret
= split_node(trans
, root
, p
, level
);
2582 btrfs_clear_path_blocking(p
, NULL
, 0);
2589 b
= p
->nodes
[level
];
2590 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2591 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2594 if (*write_lock_level
< level
+ 1) {
2595 *write_lock_level
= level
+ 1;
2596 btrfs_release_path(p
);
2600 btrfs_set_path_blocking(p
);
2601 reada_for_balance(root
, p
, level
);
2602 sret
= balance_level(trans
, root
, p
, level
);
2603 btrfs_clear_path_blocking(p
, NULL
, 0);
2609 b
= p
->nodes
[level
];
2611 btrfs_release_path(p
);
2614 BUG_ON(btrfs_header_nritems(b
) == 1);
2624 static void key_search_validate(struct extent_buffer
*b
,
2625 struct btrfs_key
*key
,
2628 #ifdef CONFIG_BTRFS_ASSERT
2629 struct btrfs_disk_key disk_key
;
2631 btrfs_cpu_key_to_disk(&disk_key
, key
);
2634 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2635 offsetof(struct btrfs_leaf
, items
[0].key
),
2638 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2639 offsetof(struct btrfs_node
, ptrs
[0].key
),
2644 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2645 int level
, int *prev_cmp
, int *slot
)
2647 if (*prev_cmp
!= 0) {
2648 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2652 key_search_validate(b
, key
, level
);
2658 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*found_path
,
2659 u64 iobjectid
, u64 ioff
, u8 key_type
,
2660 struct btrfs_key
*found_key
)
2663 struct btrfs_key key
;
2664 struct extent_buffer
*eb
;
2665 struct btrfs_path
*path
;
2667 key
.type
= key_type
;
2668 key
.objectid
= iobjectid
;
2671 if (found_path
== NULL
) {
2672 path
= btrfs_alloc_path();
2678 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2679 if ((ret
< 0) || (found_key
== NULL
)) {
2680 if (path
!= found_path
)
2681 btrfs_free_path(path
);
2685 eb
= path
->nodes
[0];
2686 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2687 ret
= btrfs_next_leaf(fs_root
, path
);
2690 eb
= path
->nodes
[0];
2693 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2694 if (found_key
->type
!= key
.type
||
2695 found_key
->objectid
!= key
.objectid
)
2702 * look for key in the tree. path is filled in with nodes along the way
2703 * if key is found, we return zero and you can find the item in the leaf
2704 * level of the path (level 0)
2706 * If the key isn't found, the path points to the slot where it should
2707 * be inserted, and 1 is returned. If there are other errors during the
2708 * search a negative error number is returned.
2710 * if ins_len > 0, nodes and leaves will be split as we walk down the
2711 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2714 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2715 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2718 struct extent_buffer
*b
;
2723 int lowest_unlock
= 1;
2725 /* everything at write_lock_level or lower must be write locked */
2726 int write_lock_level
= 0;
2727 u8 lowest_level
= 0;
2728 int min_write_lock_level
;
2731 lowest_level
= p
->lowest_level
;
2732 WARN_ON(lowest_level
&& ins_len
> 0);
2733 WARN_ON(p
->nodes
[0] != NULL
);
2738 /* when we are removing items, we might have to go up to level
2739 * two as we update tree pointers Make sure we keep write
2740 * for those levels as well
2742 write_lock_level
= 2;
2743 } else if (ins_len
> 0) {
2745 * for inserting items, make sure we have a write lock on
2746 * level 1 so we can update keys
2748 write_lock_level
= 1;
2752 write_lock_level
= -1;
2754 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2755 write_lock_level
= BTRFS_MAX_LEVEL
;
2757 min_write_lock_level
= write_lock_level
;
2762 * we try very hard to do read locks on the root
2764 root_lock
= BTRFS_READ_LOCK
;
2766 if (p
->search_commit_root
) {
2768 * the commit roots are read only
2769 * so we always do read locks
2771 b
= root
->commit_root
;
2772 extent_buffer_get(b
);
2773 level
= btrfs_header_level(b
);
2774 if (!p
->skip_locking
)
2775 btrfs_tree_read_lock(b
);
2777 if (p
->skip_locking
) {
2778 b
= btrfs_root_node(root
);
2779 level
= btrfs_header_level(b
);
2781 /* we don't know the level of the root node
2782 * until we actually have it read locked
2784 b
= btrfs_read_lock_root_node(root
);
2785 level
= btrfs_header_level(b
);
2786 if (level
<= write_lock_level
) {
2787 /* whoops, must trade for write lock */
2788 btrfs_tree_read_unlock(b
);
2789 free_extent_buffer(b
);
2790 b
= btrfs_lock_root_node(root
);
2791 root_lock
= BTRFS_WRITE_LOCK
;
2793 /* the level might have changed, check again */
2794 level
= btrfs_header_level(b
);
2798 p
->nodes
[level
] = b
;
2799 if (!p
->skip_locking
)
2800 p
->locks
[level
] = root_lock
;
2803 level
= btrfs_header_level(b
);
2806 * setup the path here so we can release it under lock
2807 * contention with the cow code
2811 * if we don't really need to cow this block
2812 * then we don't want to set the path blocking,
2813 * so we test it here
2815 if (!should_cow_block(trans
, root
, b
))
2818 btrfs_set_path_blocking(p
);
2821 * must have write locks on this node and the
2824 if (level
> write_lock_level
||
2825 (level
+ 1 > write_lock_level
&&
2826 level
+ 1 < BTRFS_MAX_LEVEL
&&
2827 p
->nodes
[level
+ 1])) {
2828 write_lock_level
= level
+ 1;
2829 btrfs_release_path(p
);
2833 err
= btrfs_cow_block(trans
, root
, b
,
2834 p
->nodes
[level
+ 1],
2835 p
->slots
[level
+ 1], &b
);
2842 BUG_ON(!cow
&& ins_len
);
2844 p
->nodes
[level
] = b
;
2845 btrfs_clear_path_blocking(p
, NULL
, 0);
2848 * we have a lock on b and as long as we aren't changing
2849 * the tree, there is no way to for the items in b to change.
2850 * It is safe to drop the lock on our parent before we
2851 * go through the expensive btree search on b.
2853 * If cow is true, then we might be changing slot zero,
2854 * which may require changing the parent. So, we can't
2855 * drop the lock until after we know which slot we're
2859 btrfs_unlock_up_safe(p
, level
+ 1);
2861 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2865 if (ret
&& slot
> 0) {
2869 p
->slots
[level
] = slot
;
2870 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2871 ins_len
, &write_lock_level
);
2878 b
= p
->nodes
[level
];
2879 slot
= p
->slots
[level
];
2882 * slot 0 is special, if we change the key
2883 * we have to update the parent pointer
2884 * which means we must have a write lock
2887 if (slot
== 0 && cow
&&
2888 write_lock_level
< level
+ 1) {
2889 write_lock_level
= level
+ 1;
2890 btrfs_release_path(p
);
2894 unlock_up(p
, level
, lowest_unlock
,
2895 min_write_lock_level
, &write_lock_level
);
2897 if (level
== lowest_level
) {
2903 err
= read_block_for_search(trans
, root
, p
,
2904 &b
, level
, slot
, key
, 0);
2912 if (!p
->skip_locking
) {
2913 level
= btrfs_header_level(b
);
2914 if (level
<= write_lock_level
) {
2915 err
= btrfs_try_tree_write_lock(b
);
2917 btrfs_set_path_blocking(p
);
2919 btrfs_clear_path_blocking(p
, b
,
2922 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2924 err
= btrfs_try_tree_read_lock(b
);
2926 btrfs_set_path_blocking(p
);
2927 btrfs_tree_read_lock(b
);
2928 btrfs_clear_path_blocking(p
, b
,
2931 p
->locks
[level
] = BTRFS_READ_LOCK
;
2933 p
->nodes
[level
] = b
;
2936 p
->slots
[level
] = slot
;
2938 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2939 if (write_lock_level
< 1) {
2940 write_lock_level
= 1;
2941 btrfs_release_path(p
);
2945 btrfs_set_path_blocking(p
);
2946 err
= split_leaf(trans
, root
, key
,
2947 p
, ins_len
, ret
== 0);
2948 btrfs_clear_path_blocking(p
, NULL
, 0);
2956 if (!p
->search_for_split
)
2957 unlock_up(p
, level
, lowest_unlock
,
2958 min_write_lock_level
, &write_lock_level
);
2965 * we don't really know what they plan on doing with the path
2966 * from here on, so for now just mark it as blocking
2968 if (!p
->leave_spinning
)
2969 btrfs_set_path_blocking(p
);
2971 btrfs_release_path(p
);
2976 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2977 * current state of the tree together with the operations recorded in the tree
2978 * modification log to search for the key in a previous version of this tree, as
2979 * denoted by the time_seq parameter.
2981 * Naturally, there is no support for insert, delete or cow operations.
2983 * The resulting path and return value will be set up as if we called
2984 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2986 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2987 struct btrfs_path
*p
, u64 time_seq
)
2989 struct extent_buffer
*b
;
2994 int lowest_unlock
= 1;
2995 u8 lowest_level
= 0;
2998 lowest_level
= p
->lowest_level
;
2999 WARN_ON(p
->nodes
[0] != NULL
);
3001 if (p
->search_commit_root
) {
3003 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3007 b
= get_old_root(root
, time_seq
);
3008 level
= btrfs_header_level(b
);
3009 p
->locks
[level
] = BTRFS_READ_LOCK
;
3012 level
= btrfs_header_level(b
);
3013 p
->nodes
[level
] = b
;
3014 btrfs_clear_path_blocking(p
, NULL
, 0);
3017 * we have a lock on b and as long as we aren't changing
3018 * the tree, there is no way to for the items in b to change.
3019 * It is safe to drop the lock on our parent before we
3020 * go through the expensive btree search on b.
3022 btrfs_unlock_up_safe(p
, level
+ 1);
3025 * Since we can unwind eb's we want to do a real search every
3029 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
3033 if (ret
&& slot
> 0) {
3037 p
->slots
[level
] = slot
;
3038 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3040 if (level
== lowest_level
) {
3046 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
3047 slot
, key
, time_seq
);
3055 level
= btrfs_header_level(b
);
3056 err
= btrfs_try_tree_read_lock(b
);
3058 btrfs_set_path_blocking(p
);
3059 btrfs_tree_read_lock(b
);
3060 btrfs_clear_path_blocking(p
, b
,
3063 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
3068 p
->locks
[level
] = BTRFS_READ_LOCK
;
3069 p
->nodes
[level
] = b
;
3071 p
->slots
[level
] = slot
;
3072 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3078 if (!p
->leave_spinning
)
3079 btrfs_set_path_blocking(p
);
3081 btrfs_release_path(p
);
3087 * helper to use instead of search slot if no exact match is needed but
3088 * instead the next or previous item should be returned.
3089 * When find_higher is true, the next higher item is returned, the next lower
3091 * When return_any and find_higher are both true, and no higher item is found,
3092 * return the next lower instead.
3093 * When return_any is true and find_higher is false, and no lower item is found,
3094 * return the next higher instead.
3095 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3098 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3099 struct btrfs_key
*key
, struct btrfs_path
*p
,
3100 int find_higher
, int return_any
)
3103 struct extent_buffer
*leaf
;
3106 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3110 * a return value of 1 means the path is at the position where the
3111 * item should be inserted. Normally this is the next bigger item,
3112 * but in case the previous item is the last in a leaf, path points
3113 * to the first free slot in the previous leaf, i.e. at an invalid
3119 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3120 ret
= btrfs_next_leaf(root
, p
);
3126 * no higher item found, return the next
3131 btrfs_release_path(p
);
3135 if (p
->slots
[0] == 0) {
3136 ret
= btrfs_prev_leaf(root
, p
);
3140 p
->slots
[0] = btrfs_header_nritems(leaf
) - 1;
3146 * no lower item found, return the next
3151 btrfs_release_path(p
);
3161 * adjust the pointers going up the tree, starting at level
3162 * making sure the right key of each node is points to 'key'.
3163 * This is used after shifting pointers to the left, so it stops
3164 * fixing up pointers when a given leaf/node is not in slot 0 of the
3168 static void fixup_low_keys(struct btrfs_root
*root
, struct btrfs_path
*path
,
3169 struct btrfs_disk_key
*key
, int level
)
3172 struct extent_buffer
*t
;
3174 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3175 int tslot
= path
->slots
[i
];
3176 if (!path
->nodes
[i
])
3179 tree_mod_log_set_node_key(root
->fs_info
, t
, tslot
, 1);
3180 btrfs_set_node_key(t
, key
, tslot
);
3181 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3190 * This function isn't completely safe. It's the caller's responsibility
3191 * that the new key won't break the order
3193 void btrfs_set_item_key_safe(struct btrfs_root
*root
, struct btrfs_path
*path
,
3194 struct btrfs_key
*new_key
)
3196 struct btrfs_disk_key disk_key
;
3197 struct extent_buffer
*eb
;
3200 eb
= path
->nodes
[0];
3201 slot
= path
->slots
[0];
3203 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3204 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
3206 if (slot
< btrfs_header_nritems(eb
) - 1) {
3207 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3208 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
3211 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3212 btrfs_set_item_key(eb
, &disk_key
, slot
);
3213 btrfs_mark_buffer_dirty(eb
);
3215 fixup_low_keys(root
, path
, &disk_key
, 1);
3219 * try to push data from one node into the next node left in the
3222 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3223 * error, and > 0 if there was no room in the left hand block.
3225 static int push_node_left(struct btrfs_trans_handle
*trans
,
3226 struct btrfs_root
*root
, struct extent_buffer
*dst
,
3227 struct extent_buffer
*src
, int empty
)
3234 src_nritems
= btrfs_header_nritems(src
);
3235 dst_nritems
= btrfs_header_nritems(dst
);
3236 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3237 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3238 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3240 if (!empty
&& src_nritems
<= 8)
3243 if (push_items
<= 0)
3247 push_items
= min(src_nritems
, push_items
);
3248 if (push_items
< src_nritems
) {
3249 /* leave at least 8 pointers in the node if
3250 * we aren't going to empty it
3252 if (src_nritems
- push_items
< 8) {
3253 if (push_items
<= 8)
3259 push_items
= min(src_nritems
- 8, push_items
);
3261 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3264 btrfs_abort_transaction(trans
, root
, ret
);
3267 copy_extent_buffer(dst
, src
,
3268 btrfs_node_key_ptr_offset(dst_nritems
),
3269 btrfs_node_key_ptr_offset(0),
3270 push_items
* sizeof(struct btrfs_key_ptr
));
3272 if (push_items
< src_nritems
) {
3274 * don't call tree_mod_log_eb_move here, key removal was already
3275 * fully logged by tree_mod_log_eb_copy above.
3277 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3278 btrfs_node_key_ptr_offset(push_items
),
3279 (src_nritems
- push_items
) *
3280 sizeof(struct btrfs_key_ptr
));
3282 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3283 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3284 btrfs_mark_buffer_dirty(src
);
3285 btrfs_mark_buffer_dirty(dst
);
3291 * try to push data from one node into the next node right in the
3294 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3295 * error, and > 0 if there was no room in the right hand block.
3297 * this will only push up to 1/2 the contents of the left node over
3299 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3300 struct btrfs_root
*root
,
3301 struct extent_buffer
*dst
,
3302 struct extent_buffer
*src
)
3310 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3311 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3313 src_nritems
= btrfs_header_nritems(src
);
3314 dst_nritems
= btrfs_header_nritems(dst
);
3315 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3316 if (push_items
<= 0)
3319 if (src_nritems
< 4)
3322 max_push
= src_nritems
/ 2 + 1;
3323 /* don't try to empty the node */
3324 if (max_push
>= src_nritems
)
3327 if (max_push
< push_items
)
3328 push_items
= max_push
;
3330 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3331 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3332 btrfs_node_key_ptr_offset(0),
3334 sizeof(struct btrfs_key_ptr
));
3336 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3337 src_nritems
- push_items
, push_items
);
3339 btrfs_abort_transaction(trans
, root
, ret
);
3342 copy_extent_buffer(dst
, src
,
3343 btrfs_node_key_ptr_offset(0),
3344 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3345 push_items
* sizeof(struct btrfs_key_ptr
));
3347 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3348 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3350 btrfs_mark_buffer_dirty(src
);
3351 btrfs_mark_buffer_dirty(dst
);
3357 * helper function to insert a new root level in the tree.
3358 * A new node is allocated, and a single item is inserted to
3359 * point to the existing root
3361 * returns zero on success or < 0 on failure.
3363 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3364 struct btrfs_root
*root
,
3365 struct btrfs_path
*path
, int level
)
3368 struct extent_buffer
*lower
;
3369 struct extent_buffer
*c
;
3370 struct extent_buffer
*old
;
3371 struct btrfs_disk_key lower_key
;
3373 BUG_ON(path
->nodes
[level
]);
3374 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3376 lower
= path
->nodes
[level
-1];
3378 btrfs_item_key(lower
, &lower_key
, 0);
3380 btrfs_node_key(lower
, &lower_key
, 0);
3382 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3383 root
->root_key
.objectid
, &lower_key
,
3384 level
, root
->node
->start
, 0);
3388 root_add_used(root
, root
->nodesize
);
3390 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3391 btrfs_set_header_nritems(c
, 1);
3392 btrfs_set_header_level(c
, level
);
3393 btrfs_set_header_bytenr(c
, c
->start
);
3394 btrfs_set_header_generation(c
, trans
->transid
);
3395 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3396 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3398 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3401 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3402 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3404 btrfs_set_node_key(c
, &lower_key
, 0);
3405 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3406 lower_gen
= btrfs_header_generation(lower
);
3407 WARN_ON(lower_gen
!= trans
->transid
);
3409 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3411 btrfs_mark_buffer_dirty(c
);
3414 tree_mod_log_set_root_pointer(root
, c
, 0);
3415 rcu_assign_pointer(root
->node
, c
);
3417 /* the super has an extra ref to root->node */
3418 free_extent_buffer(old
);
3420 add_root_to_dirty_list(root
);
3421 extent_buffer_get(c
);
3422 path
->nodes
[level
] = c
;
3423 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
3424 path
->slots
[level
] = 0;
3429 * worker function to insert a single pointer in a node.
3430 * the node should have enough room for the pointer already
3432 * slot and level indicate where you want the key to go, and
3433 * blocknr is the block the key points to.
3435 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3436 struct btrfs_root
*root
, struct btrfs_path
*path
,
3437 struct btrfs_disk_key
*key
, u64 bytenr
,
3438 int slot
, int level
)
3440 struct extent_buffer
*lower
;
3444 BUG_ON(!path
->nodes
[level
]);
3445 btrfs_assert_tree_locked(path
->nodes
[level
]);
3446 lower
= path
->nodes
[level
];
3447 nritems
= btrfs_header_nritems(lower
);
3448 BUG_ON(slot
> nritems
);
3449 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3450 if (slot
!= nritems
) {
3452 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3453 slot
, nritems
- slot
);
3454 memmove_extent_buffer(lower
,
3455 btrfs_node_key_ptr_offset(slot
+ 1),
3456 btrfs_node_key_ptr_offset(slot
),
3457 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3460 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3461 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3464 btrfs_set_node_key(lower
, key
, slot
);
3465 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3466 WARN_ON(trans
->transid
== 0);
3467 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3468 btrfs_set_header_nritems(lower
, nritems
+ 1);
3469 btrfs_mark_buffer_dirty(lower
);
3473 * split the node at the specified level in path in two.
3474 * The path is corrected to point to the appropriate node after the split
3476 * Before splitting this tries to make some room in the node by pushing
3477 * left and right, if either one works, it returns right away.
3479 * returns 0 on success and < 0 on failure
3481 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3482 struct btrfs_root
*root
,
3483 struct btrfs_path
*path
, int level
)
3485 struct extent_buffer
*c
;
3486 struct extent_buffer
*split
;
3487 struct btrfs_disk_key disk_key
;
3492 c
= path
->nodes
[level
];
3493 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3494 if (c
== root
->node
) {
3496 * trying to split the root, lets make a new one
3498 * tree mod log: We don't log_removal old root in
3499 * insert_new_root, because that root buffer will be kept as a
3500 * normal node. We are going to log removal of half of the
3501 * elements below with tree_mod_log_eb_copy. We're holding a
3502 * tree lock on the buffer, which is why we cannot race with
3503 * other tree_mod_log users.
3505 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3509 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3510 c
= path
->nodes
[level
];
3511 if (!ret
&& btrfs_header_nritems(c
) <
3512 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3518 c_nritems
= btrfs_header_nritems(c
);
3519 mid
= (c_nritems
+ 1) / 2;
3520 btrfs_node_key(c
, &disk_key
, mid
);
3522 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3523 root
->root_key
.objectid
,
3524 &disk_key
, level
, c
->start
, 0);
3526 return PTR_ERR(split
);
3528 root_add_used(root
, root
->nodesize
);
3530 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3531 btrfs_set_header_level(split
, btrfs_header_level(c
));
3532 btrfs_set_header_bytenr(split
, split
->start
);
3533 btrfs_set_header_generation(split
, trans
->transid
);
3534 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3535 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3536 write_extent_buffer(split
, root
->fs_info
->fsid
,
3537 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3538 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3539 btrfs_header_chunk_tree_uuid(split
),
3542 ret
= tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0,
3543 mid
, c_nritems
- mid
);
3545 btrfs_abort_transaction(trans
, root
, ret
);
3548 copy_extent_buffer(split
, c
,
3549 btrfs_node_key_ptr_offset(0),
3550 btrfs_node_key_ptr_offset(mid
),
3551 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3552 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3553 btrfs_set_header_nritems(c
, mid
);
3556 btrfs_mark_buffer_dirty(c
);
3557 btrfs_mark_buffer_dirty(split
);
3559 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3560 path
->slots
[level
+ 1] + 1, level
+ 1);
3562 if (path
->slots
[level
] >= mid
) {
3563 path
->slots
[level
] -= mid
;
3564 btrfs_tree_unlock(c
);
3565 free_extent_buffer(c
);
3566 path
->nodes
[level
] = split
;
3567 path
->slots
[level
+ 1] += 1;
3569 btrfs_tree_unlock(split
);
3570 free_extent_buffer(split
);
3576 * how many bytes are required to store the items in a leaf. start
3577 * and nr indicate which items in the leaf to check. This totals up the
3578 * space used both by the item structs and the item data
3580 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3582 struct btrfs_item
*start_item
;
3583 struct btrfs_item
*end_item
;
3584 struct btrfs_map_token token
;
3586 int nritems
= btrfs_header_nritems(l
);
3587 int end
= min(nritems
, start
+ nr
) - 1;
3591 btrfs_init_map_token(&token
);
3592 start_item
= btrfs_item_nr(start
);
3593 end_item
= btrfs_item_nr(end
);
3594 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3595 btrfs_token_item_size(l
, start_item
, &token
);
3596 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3597 data_len
+= sizeof(struct btrfs_item
) * nr
;
3598 WARN_ON(data_len
< 0);
3603 * The space between the end of the leaf items and
3604 * the start of the leaf data. IOW, how much room
3605 * the leaf has left for both items and data
3607 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3608 struct extent_buffer
*leaf
)
3610 int nritems
= btrfs_header_nritems(leaf
);
3612 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3614 btrfs_crit(root
->fs_info
,
3615 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3616 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3617 leaf_space_used(leaf
, 0, nritems
), nritems
);
3623 * min slot controls the lowest index we're willing to push to the
3624 * right. We'll push up to and including min_slot, but no lower
3626 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3627 struct btrfs_root
*root
,
3628 struct btrfs_path
*path
,
3629 int data_size
, int empty
,
3630 struct extent_buffer
*right
,
3631 int free_space
, u32 left_nritems
,
3634 struct extent_buffer
*left
= path
->nodes
[0];
3635 struct extent_buffer
*upper
= path
->nodes
[1];
3636 struct btrfs_map_token token
;
3637 struct btrfs_disk_key disk_key
;
3642 struct btrfs_item
*item
;
3648 btrfs_init_map_token(&token
);
3653 nr
= max_t(u32
, 1, min_slot
);
3655 if (path
->slots
[0] >= left_nritems
)
3656 push_space
+= data_size
;
3658 slot
= path
->slots
[1];
3659 i
= left_nritems
- 1;
3661 item
= btrfs_item_nr(i
);
3663 if (!empty
&& push_items
> 0) {
3664 if (path
->slots
[0] > i
)
3666 if (path
->slots
[0] == i
) {
3667 int space
= btrfs_leaf_free_space(root
, left
);
3668 if (space
+ push_space
* 2 > free_space
)
3673 if (path
->slots
[0] == i
)
3674 push_space
+= data_size
;
3676 this_item_size
= btrfs_item_size(left
, item
);
3677 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3681 push_space
+= this_item_size
+ sizeof(*item
);
3687 if (push_items
== 0)
3690 WARN_ON(!empty
&& push_items
== left_nritems
);
3692 /* push left to right */
3693 right_nritems
= btrfs_header_nritems(right
);
3695 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3696 push_space
-= leaf_data_end(root
, left
);
3698 /* make room in the right data area */
3699 data_end
= leaf_data_end(root
, right
);
3700 memmove_extent_buffer(right
,
3701 btrfs_leaf_data(right
) + data_end
- push_space
,
3702 btrfs_leaf_data(right
) + data_end
,
3703 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3705 /* copy from the left data area */
3706 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3707 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3708 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3711 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3712 btrfs_item_nr_offset(0),
3713 right_nritems
* sizeof(struct btrfs_item
));
3715 /* copy the items from left to right */
3716 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3717 btrfs_item_nr_offset(left_nritems
- push_items
),
3718 push_items
* sizeof(struct btrfs_item
));
3720 /* update the item pointers */
3721 right_nritems
+= push_items
;
3722 btrfs_set_header_nritems(right
, right_nritems
);
3723 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3724 for (i
= 0; i
< right_nritems
; i
++) {
3725 item
= btrfs_item_nr(i
);
3726 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3727 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3730 left_nritems
-= push_items
;
3731 btrfs_set_header_nritems(left
, left_nritems
);
3734 btrfs_mark_buffer_dirty(left
);
3736 clean_tree_block(trans
, root
, left
);
3738 btrfs_mark_buffer_dirty(right
);
3740 btrfs_item_key(right
, &disk_key
, 0);
3741 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3742 btrfs_mark_buffer_dirty(upper
);
3744 /* then fixup the leaf pointer in the path */
3745 if (path
->slots
[0] >= left_nritems
) {
3746 path
->slots
[0] -= left_nritems
;
3747 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3748 clean_tree_block(trans
, root
, path
->nodes
[0]);
3749 btrfs_tree_unlock(path
->nodes
[0]);
3750 free_extent_buffer(path
->nodes
[0]);
3751 path
->nodes
[0] = right
;
3752 path
->slots
[1] += 1;
3754 btrfs_tree_unlock(right
);
3755 free_extent_buffer(right
);
3760 btrfs_tree_unlock(right
);
3761 free_extent_buffer(right
);
3766 * push some data in the path leaf to the right, trying to free up at
3767 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3769 * returns 1 if the push failed because the other node didn't have enough
3770 * room, 0 if everything worked out and < 0 if there were major errors.
3772 * this will push starting from min_slot to the end of the leaf. It won't
3773 * push any slot lower than min_slot
3775 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3776 *root
, struct btrfs_path
*path
,
3777 int min_data_size
, int data_size
,
3778 int empty
, u32 min_slot
)
3780 struct extent_buffer
*left
= path
->nodes
[0];
3781 struct extent_buffer
*right
;
3782 struct extent_buffer
*upper
;
3788 if (!path
->nodes
[1])
3791 slot
= path
->slots
[1];
3792 upper
= path
->nodes
[1];
3793 if (slot
>= btrfs_header_nritems(upper
) - 1)
3796 btrfs_assert_tree_locked(path
->nodes
[1]);
3798 right
= read_node_slot(root
, upper
, slot
+ 1);
3802 btrfs_tree_lock(right
);
3803 btrfs_set_lock_blocking(right
);
3805 free_space
= btrfs_leaf_free_space(root
, right
);
3806 if (free_space
< data_size
)
3809 /* cow and double check */
3810 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3815 free_space
= btrfs_leaf_free_space(root
, right
);
3816 if (free_space
< data_size
)
3819 left_nritems
= btrfs_header_nritems(left
);
3820 if (left_nritems
== 0)
3823 if (path
->slots
[0] == left_nritems
&& !empty
) {
3824 /* Key greater than all keys in the leaf, right neighbor has
3825 * enough room for it and we're not emptying our leaf to delete
3826 * it, therefore use right neighbor to insert the new item and
3827 * no need to touch/dirty our left leaft. */
3828 btrfs_tree_unlock(left
);
3829 free_extent_buffer(left
);
3830 path
->nodes
[0] = right
;
3836 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3837 right
, free_space
, left_nritems
, min_slot
);
3839 btrfs_tree_unlock(right
);
3840 free_extent_buffer(right
);
3845 * push some data in the path leaf to the left, trying to free up at
3846 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3848 * max_slot can put a limit on how far into the leaf we'll push items. The
3849 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3852 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3853 struct btrfs_root
*root
,
3854 struct btrfs_path
*path
, int data_size
,
3855 int empty
, struct extent_buffer
*left
,
3856 int free_space
, u32 right_nritems
,
3859 struct btrfs_disk_key disk_key
;
3860 struct extent_buffer
*right
= path
->nodes
[0];
3864 struct btrfs_item
*item
;
3865 u32 old_left_nritems
;
3869 u32 old_left_item_size
;
3870 struct btrfs_map_token token
;
3872 btrfs_init_map_token(&token
);
3875 nr
= min(right_nritems
, max_slot
);
3877 nr
= min(right_nritems
- 1, max_slot
);
3879 for (i
= 0; i
< nr
; i
++) {
3880 item
= btrfs_item_nr(i
);
3882 if (!empty
&& push_items
> 0) {
3883 if (path
->slots
[0] < i
)
3885 if (path
->slots
[0] == i
) {
3886 int space
= btrfs_leaf_free_space(root
, right
);
3887 if (space
+ push_space
* 2 > free_space
)
3892 if (path
->slots
[0] == i
)
3893 push_space
+= data_size
;
3895 this_item_size
= btrfs_item_size(right
, item
);
3896 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3900 push_space
+= this_item_size
+ sizeof(*item
);
3903 if (push_items
== 0) {
3907 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3909 /* push data from right to left */
3910 copy_extent_buffer(left
, right
,
3911 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3912 btrfs_item_nr_offset(0),
3913 push_items
* sizeof(struct btrfs_item
));
3915 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3916 btrfs_item_offset_nr(right
, push_items
- 1);
3918 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3919 leaf_data_end(root
, left
) - push_space
,
3920 btrfs_leaf_data(right
) +
3921 btrfs_item_offset_nr(right
, push_items
- 1),
3923 old_left_nritems
= btrfs_header_nritems(left
);
3924 BUG_ON(old_left_nritems
<= 0);
3926 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3927 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3930 item
= btrfs_item_nr(i
);
3932 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3933 btrfs_set_token_item_offset(left
, item
,
3934 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3937 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3939 /* fixup right node */
3940 if (push_items
> right_nritems
)
3941 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3944 if (push_items
< right_nritems
) {
3945 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3946 leaf_data_end(root
, right
);
3947 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3948 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3949 btrfs_leaf_data(right
) +
3950 leaf_data_end(root
, right
), push_space
);
3952 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3953 btrfs_item_nr_offset(push_items
),
3954 (btrfs_header_nritems(right
) - push_items
) *
3955 sizeof(struct btrfs_item
));
3957 right_nritems
-= push_items
;
3958 btrfs_set_header_nritems(right
, right_nritems
);
3959 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3960 for (i
= 0; i
< right_nritems
; i
++) {
3961 item
= btrfs_item_nr(i
);
3963 push_space
= push_space
- btrfs_token_item_size(right
,
3965 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3968 btrfs_mark_buffer_dirty(left
);
3970 btrfs_mark_buffer_dirty(right
);
3972 clean_tree_block(trans
, root
, right
);
3974 btrfs_item_key(right
, &disk_key
, 0);
3975 fixup_low_keys(root
, path
, &disk_key
, 1);
3977 /* then fixup the leaf pointer in the path */
3978 if (path
->slots
[0] < push_items
) {
3979 path
->slots
[0] += old_left_nritems
;
3980 btrfs_tree_unlock(path
->nodes
[0]);
3981 free_extent_buffer(path
->nodes
[0]);
3982 path
->nodes
[0] = left
;
3983 path
->slots
[1] -= 1;
3985 btrfs_tree_unlock(left
);
3986 free_extent_buffer(left
);
3987 path
->slots
[0] -= push_items
;
3989 BUG_ON(path
->slots
[0] < 0);
3992 btrfs_tree_unlock(left
);
3993 free_extent_buffer(left
);
3998 * push some data in the path leaf to the left, trying to free up at
3999 * least data_size bytes. returns zero if the push worked, nonzero otherwise
4001 * max_slot can put a limit on how far into the leaf we'll push items. The
4002 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
4005 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
4006 *root
, struct btrfs_path
*path
, int min_data_size
,
4007 int data_size
, int empty
, u32 max_slot
)
4009 struct extent_buffer
*right
= path
->nodes
[0];
4010 struct extent_buffer
*left
;
4016 slot
= path
->slots
[1];
4019 if (!path
->nodes
[1])
4022 right_nritems
= btrfs_header_nritems(right
);
4023 if (right_nritems
== 0)
4026 btrfs_assert_tree_locked(path
->nodes
[1]);
4028 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
4032 btrfs_tree_lock(left
);
4033 btrfs_set_lock_blocking(left
);
4035 free_space
= btrfs_leaf_free_space(root
, left
);
4036 if (free_space
< data_size
) {
4041 /* cow and double check */
4042 ret
= btrfs_cow_block(trans
, root
, left
,
4043 path
->nodes
[1], slot
- 1, &left
);
4045 /* we hit -ENOSPC, but it isn't fatal here */
4051 free_space
= btrfs_leaf_free_space(root
, left
);
4052 if (free_space
< data_size
) {
4057 return __push_leaf_left(trans
, root
, path
, min_data_size
,
4058 empty
, left
, free_space
, right_nritems
,
4061 btrfs_tree_unlock(left
);
4062 free_extent_buffer(left
);
4067 * split the path's leaf in two, making sure there is at least data_size
4068 * available for the resulting leaf level of the path.
4070 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4071 struct btrfs_root
*root
,
4072 struct btrfs_path
*path
,
4073 struct extent_buffer
*l
,
4074 struct extent_buffer
*right
,
4075 int slot
, int mid
, int nritems
)
4080 struct btrfs_disk_key disk_key
;
4081 struct btrfs_map_token token
;
4083 btrfs_init_map_token(&token
);
4085 nritems
= nritems
- mid
;
4086 btrfs_set_header_nritems(right
, nritems
);
4087 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
4089 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4090 btrfs_item_nr_offset(mid
),
4091 nritems
* sizeof(struct btrfs_item
));
4093 copy_extent_buffer(right
, l
,
4094 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
4095 data_copy_size
, btrfs_leaf_data(l
) +
4096 leaf_data_end(root
, l
), data_copy_size
);
4098 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
4099 btrfs_item_end_nr(l
, mid
);
4101 for (i
= 0; i
< nritems
; i
++) {
4102 struct btrfs_item
*item
= btrfs_item_nr(i
);
4105 ioff
= btrfs_token_item_offset(right
, item
, &token
);
4106 btrfs_set_token_item_offset(right
, item
,
4107 ioff
+ rt_data_off
, &token
);
4110 btrfs_set_header_nritems(l
, mid
);
4111 btrfs_item_key(right
, &disk_key
, 0);
4112 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4113 path
->slots
[1] + 1, 1);
4115 btrfs_mark_buffer_dirty(right
);
4116 btrfs_mark_buffer_dirty(l
);
4117 BUG_ON(path
->slots
[0] != slot
);
4120 btrfs_tree_unlock(path
->nodes
[0]);
4121 free_extent_buffer(path
->nodes
[0]);
4122 path
->nodes
[0] = right
;
4123 path
->slots
[0] -= mid
;
4124 path
->slots
[1] += 1;
4126 btrfs_tree_unlock(right
);
4127 free_extent_buffer(right
);
4130 BUG_ON(path
->slots
[0] < 0);
4134 * double splits happen when we need to insert a big item in the middle
4135 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4136 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4139 * We avoid this by trying to push the items on either side of our target
4140 * into the adjacent leaves. If all goes well we can avoid the double split
4143 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4144 struct btrfs_root
*root
,
4145 struct btrfs_path
*path
,
4152 int space_needed
= data_size
;
4154 slot
= path
->slots
[0];
4155 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4156 space_needed
-= btrfs_leaf_free_space(root
, path
->nodes
[0]);
4159 * try to push all the items after our slot into the
4162 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4169 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4171 * our goal is to get our slot at the start or end of a leaf. If
4172 * we've done so we're done
4174 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4177 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4180 /* try to push all the items before our slot into the next leaf */
4181 slot
= path
->slots
[0];
4182 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4195 * split the path's leaf in two, making sure there is at least data_size
4196 * available for the resulting leaf level of the path.
4198 * returns 0 if all went well and < 0 on failure.
4200 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4201 struct btrfs_root
*root
,
4202 struct btrfs_key
*ins_key
,
4203 struct btrfs_path
*path
, int data_size
,
4206 struct btrfs_disk_key disk_key
;
4207 struct extent_buffer
*l
;
4211 struct extent_buffer
*right
;
4215 int num_doubles
= 0;
4216 int tried_avoid_double
= 0;
4219 slot
= path
->slots
[0];
4220 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4221 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
4224 /* first try to make some room by pushing left and right */
4225 if (data_size
&& path
->nodes
[1]) {
4226 int space_needed
= data_size
;
4228 if (slot
< btrfs_header_nritems(l
))
4229 space_needed
-= btrfs_leaf_free_space(root
, l
);
4231 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4232 space_needed
, 0, 0);
4236 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4237 space_needed
, 0, (u32
)-1);
4243 /* did the pushes work? */
4244 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
4248 if (!path
->nodes
[1]) {
4249 ret
= insert_new_root(trans
, root
, path
, 1);
4256 slot
= path
->slots
[0];
4257 nritems
= btrfs_header_nritems(l
);
4258 mid
= (nritems
+ 1) / 2;
4262 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4263 BTRFS_LEAF_DATA_SIZE(root
)) {
4264 if (slot
>= nritems
) {
4268 if (mid
!= nritems
&&
4269 leaf_space_used(l
, mid
, nritems
- mid
) +
4270 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4271 if (data_size
&& !tried_avoid_double
)
4272 goto push_for_double
;
4278 if (leaf_space_used(l
, 0, mid
) + data_size
>
4279 BTRFS_LEAF_DATA_SIZE(root
)) {
4280 if (!extend
&& data_size
&& slot
== 0) {
4282 } else if ((extend
|| !data_size
) && slot
== 0) {
4286 if (mid
!= nritems
&&
4287 leaf_space_used(l
, mid
, nritems
- mid
) +
4288 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4289 if (data_size
&& !tried_avoid_double
)
4290 goto push_for_double
;
4298 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4300 btrfs_item_key(l
, &disk_key
, mid
);
4302 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
4303 root
->root_key
.objectid
,
4304 &disk_key
, 0, l
->start
, 0);
4306 return PTR_ERR(right
);
4308 root_add_used(root
, root
->leafsize
);
4310 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4311 btrfs_set_header_bytenr(right
, right
->start
);
4312 btrfs_set_header_generation(right
, trans
->transid
);
4313 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4314 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4315 btrfs_set_header_level(right
, 0);
4316 write_extent_buffer(right
, root
->fs_info
->fsid
,
4317 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4319 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
4320 btrfs_header_chunk_tree_uuid(right
),
4325 btrfs_set_header_nritems(right
, 0);
4326 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4327 path
->slots
[1] + 1, 1);
4328 btrfs_tree_unlock(path
->nodes
[0]);
4329 free_extent_buffer(path
->nodes
[0]);
4330 path
->nodes
[0] = right
;
4332 path
->slots
[1] += 1;
4334 btrfs_set_header_nritems(right
, 0);
4335 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4337 btrfs_tree_unlock(path
->nodes
[0]);
4338 free_extent_buffer(path
->nodes
[0]);
4339 path
->nodes
[0] = right
;
4341 if (path
->slots
[1] == 0)
4342 fixup_low_keys(root
, path
, &disk_key
, 1);
4344 btrfs_mark_buffer_dirty(right
);
4348 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4351 BUG_ON(num_doubles
!= 0);
4359 push_for_double_split(trans
, root
, path
, data_size
);
4360 tried_avoid_double
= 1;
4361 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4366 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4367 struct btrfs_root
*root
,
4368 struct btrfs_path
*path
, int ins_len
)
4370 struct btrfs_key key
;
4371 struct extent_buffer
*leaf
;
4372 struct btrfs_file_extent_item
*fi
;
4377 leaf
= path
->nodes
[0];
4378 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4380 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4381 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4383 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4386 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4387 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4388 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4389 struct btrfs_file_extent_item
);
4390 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4392 btrfs_release_path(path
);
4394 path
->keep_locks
= 1;
4395 path
->search_for_split
= 1;
4396 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4397 path
->search_for_split
= 0;
4402 leaf
= path
->nodes
[0];
4403 /* if our item isn't there or got smaller, return now */
4404 if (ret
> 0 || 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 offsetof(struct btrfs_file_extent_item
,
4651 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4652 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4653 data_end
, old_data_start
- data_end
);
4655 offset
= btrfs_disk_key_offset(&disk_key
);
4656 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4657 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4659 fixup_low_keys(root
, path
, &disk_key
, 1);
4662 item
= btrfs_item_nr(slot
);
4663 btrfs_set_item_size(leaf
, item
, new_size
);
4664 btrfs_mark_buffer_dirty(leaf
);
4666 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4667 btrfs_print_leaf(root
, leaf
);
4673 * make the item pointed to by the path bigger, data_size is the added size.
4675 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4679 struct extent_buffer
*leaf
;
4680 struct btrfs_item
*item
;
4682 unsigned int data_end
;
4683 unsigned int old_data
;
4684 unsigned int old_size
;
4686 struct btrfs_map_token token
;
4688 btrfs_init_map_token(&token
);
4690 leaf
= path
->nodes
[0];
4692 nritems
= btrfs_header_nritems(leaf
);
4693 data_end
= leaf_data_end(root
, leaf
);
4695 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4696 btrfs_print_leaf(root
, leaf
);
4699 slot
= path
->slots
[0];
4700 old_data
= btrfs_item_end_nr(leaf
, slot
);
4703 if (slot
>= nritems
) {
4704 btrfs_print_leaf(root
, leaf
);
4705 btrfs_crit(root
->fs_info
, "slot %d too large, nritems %d",
4711 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4713 /* first correct the data pointers */
4714 for (i
= slot
; i
< nritems
; i
++) {
4716 item
= btrfs_item_nr(i
);
4718 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4719 btrfs_set_token_item_offset(leaf
, item
,
4720 ioff
- data_size
, &token
);
4723 /* shift the data */
4724 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4725 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4726 data_end
, old_data
- data_end
);
4728 data_end
= old_data
;
4729 old_size
= btrfs_item_size_nr(leaf
, slot
);
4730 item
= btrfs_item_nr(slot
);
4731 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4732 btrfs_mark_buffer_dirty(leaf
);
4734 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4735 btrfs_print_leaf(root
, leaf
);
4741 * this is a helper for btrfs_insert_empty_items, the main goal here is
4742 * to save stack depth by doing the bulk of the work in a function
4743 * that doesn't call btrfs_search_slot
4745 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4746 struct btrfs_key
*cpu_key
, u32
*data_size
,
4747 u32 total_data
, u32 total_size
, int nr
)
4749 struct btrfs_item
*item
;
4752 unsigned int data_end
;
4753 struct btrfs_disk_key disk_key
;
4754 struct extent_buffer
*leaf
;
4756 struct btrfs_map_token token
;
4758 btrfs_init_map_token(&token
);
4760 leaf
= path
->nodes
[0];
4761 slot
= path
->slots
[0];
4763 nritems
= btrfs_header_nritems(leaf
);
4764 data_end
= leaf_data_end(root
, leaf
);
4766 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4767 btrfs_print_leaf(root
, leaf
);
4768 btrfs_crit(root
->fs_info
, "not enough freespace need %u have %d",
4769 total_size
, btrfs_leaf_free_space(root
, leaf
));
4773 if (slot
!= nritems
) {
4774 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4776 if (old_data
< data_end
) {
4777 btrfs_print_leaf(root
, leaf
);
4778 btrfs_crit(root
->fs_info
, "slot %d old_data %d data_end %d",
4779 slot
, old_data
, data_end
);
4783 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4785 /* first correct the data pointers */
4786 for (i
= slot
; i
< nritems
; i
++) {
4789 item
= btrfs_item_nr( i
);
4790 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4791 btrfs_set_token_item_offset(leaf
, item
,
4792 ioff
- total_data
, &token
);
4794 /* shift the items */
4795 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4796 btrfs_item_nr_offset(slot
),
4797 (nritems
- slot
) * sizeof(struct btrfs_item
));
4799 /* shift the data */
4800 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4801 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4802 data_end
, old_data
- data_end
);
4803 data_end
= old_data
;
4806 /* setup the item for the new data */
4807 for (i
= 0; i
< nr
; i
++) {
4808 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4809 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4810 item
= btrfs_item_nr(slot
+ i
);
4811 btrfs_set_token_item_offset(leaf
, item
,
4812 data_end
- data_size
[i
], &token
);
4813 data_end
-= data_size
[i
];
4814 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4817 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4820 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4821 fixup_low_keys(root
, path
, &disk_key
, 1);
4823 btrfs_unlock_up_safe(path
, 1);
4824 btrfs_mark_buffer_dirty(leaf
);
4826 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4827 btrfs_print_leaf(root
, leaf
);
4833 * Given a key and some data, insert items into the tree.
4834 * This does all the path init required, making room in the tree if needed.
4836 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4837 struct btrfs_root
*root
,
4838 struct btrfs_path
*path
,
4839 struct btrfs_key
*cpu_key
, u32
*data_size
,
4848 for (i
= 0; i
< nr
; i
++)
4849 total_data
+= data_size
[i
];
4851 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4852 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4858 slot
= path
->slots
[0];
4861 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4862 total_data
, total_size
, nr
);
4867 * Given a key and some data, insert an item into the tree.
4868 * This does all the path init required, making room in the tree if needed.
4870 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4871 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4875 struct btrfs_path
*path
;
4876 struct extent_buffer
*leaf
;
4879 path
= btrfs_alloc_path();
4882 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4884 leaf
= path
->nodes
[0];
4885 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4886 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4887 btrfs_mark_buffer_dirty(leaf
);
4889 btrfs_free_path(path
);
4894 * delete the pointer from a given node.
4896 * the tree should have been previously balanced so the deletion does not
4899 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4900 int level
, int slot
)
4902 struct extent_buffer
*parent
= path
->nodes
[level
];
4906 nritems
= btrfs_header_nritems(parent
);
4907 if (slot
!= nritems
- 1) {
4909 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4910 slot
+ 1, nritems
- slot
- 1);
4911 memmove_extent_buffer(parent
,
4912 btrfs_node_key_ptr_offset(slot
),
4913 btrfs_node_key_ptr_offset(slot
+ 1),
4914 sizeof(struct btrfs_key_ptr
) *
4915 (nritems
- slot
- 1));
4917 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4918 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4923 btrfs_set_header_nritems(parent
, nritems
);
4924 if (nritems
== 0 && parent
== root
->node
) {
4925 BUG_ON(btrfs_header_level(root
->node
) != 1);
4926 /* just turn the root into a leaf and break */
4927 btrfs_set_header_level(root
->node
, 0);
4928 } else if (slot
== 0) {
4929 struct btrfs_disk_key disk_key
;
4931 btrfs_node_key(parent
, &disk_key
, 0);
4932 fixup_low_keys(root
, path
, &disk_key
, level
+ 1);
4934 btrfs_mark_buffer_dirty(parent
);
4938 * a helper function to delete the leaf pointed to by path->slots[1] and
4941 * This deletes the pointer in path->nodes[1] and frees the leaf
4942 * block extent. zero is returned if it all worked out, < 0 otherwise.
4944 * The path must have already been setup for deleting the leaf, including
4945 * all the proper balancing. path->nodes[1] must be locked.
4947 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4948 struct btrfs_root
*root
,
4949 struct btrfs_path
*path
,
4950 struct extent_buffer
*leaf
)
4952 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4953 del_ptr(root
, path
, 1, path
->slots
[1]);
4956 * btrfs_free_extent is expensive, we want to make sure we
4957 * aren't holding any locks when we call it
4959 btrfs_unlock_up_safe(path
, 0);
4961 root_sub_used(root
, leaf
->len
);
4963 extent_buffer_get(leaf
);
4964 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4965 free_extent_buffer_stale(leaf
);
4968 * delete the item at the leaf level in path. If that empties
4969 * the leaf, remove it from the tree
4971 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4972 struct btrfs_path
*path
, int slot
, int nr
)
4974 struct extent_buffer
*leaf
;
4975 struct btrfs_item
*item
;
4982 struct btrfs_map_token token
;
4984 btrfs_init_map_token(&token
);
4986 leaf
= path
->nodes
[0];
4987 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4989 for (i
= 0; i
< nr
; i
++)
4990 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4992 nritems
= btrfs_header_nritems(leaf
);
4994 if (slot
+ nr
!= nritems
) {
4995 int data_end
= leaf_data_end(root
, leaf
);
4997 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4999 btrfs_leaf_data(leaf
) + data_end
,
5000 last_off
- data_end
);
5002 for (i
= slot
+ nr
; i
< nritems
; i
++) {
5005 item
= btrfs_item_nr(i
);
5006 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
5007 btrfs_set_token_item_offset(leaf
, item
,
5008 ioff
+ dsize
, &token
);
5011 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
5012 btrfs_item_nr_offset(slot
+ nr
),
5013 sizeof(struct btrfs_item
) *
5014 (nritems
- slot
- nr
));
5016 btrfs_set_header_nritems(leaf
, nritems
- nr
);
5019 /* delete the leaf if we've emptied it */
5021 if (leaf
== root
->node
) {
5022 btrfs_set_header_level(leaf
, 0);
5024 btrfs_set_path_blocking(path
);
5025 clean_tree_block(trans
, root
, leaf
);
5026 btrfs_del_leaf(trans
, root
, path
, leaf
);
5029 int used
= leaf_space_used(leaf
, 0, nritems
);
5031 struct btrfs_disk_key disk_key
;
5033 btrfs_item_key(leaf
, &disk_key
, 0);
5034 fixup_low_keys(root
, path
, &disk_key
, 1);
5037 /* delete the leaf if it is mostly empty */
5038 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
5039 /* push_leaf_left fixes the path.
5040 * make sure the path still points to our leaf
5041 * for possible call to del_ptr below
5043 slot
= path
->slots
[1];
5044 extent_buffer_get(leaf
);
5046 btrfs_set_path_blocking(path
);
5047 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
5049 if (wret
< 0 && wret
!= -ENOSPC
)
5052 if (path
->nodes
[0] == leaf
&&
5053 btrfs_header_nritems(leaf
)) {
5054 wret
= push_leaf_right(trans
, root
, path
, 1,
5056 if (wret
< 0 && wret
!= -ENOSPC
)
5060 if (btrfs_header_nritems(leaf
) == 0) {
5061 path
->slots
[1] = slot
;
5062 btrfs_del_leaf(trans
, root
, path
, leaf
);
5063 free_extent_buffer(leaf
);
5066 /* if we're still in the path, make sure
5067 * we're dirty. Otherwise, one of the
5068 * push_leaf functions must have already
5069 * dirtied this buffer
5071 if (path
->nodes
[0] == leaf
)
5072 btrfs_mark_buffer_dirty(leaf
);
5073 free_extent_buffer(leaf
);
5076 btrfs_mark_buffer_dirty(leaf
);
5083 * search the tree again to find a leaf with lesser keys
5084 * returns 0 if it found something or 1 if there are no lesser leaves.
5085 * returns < 0 on io errors.
5087 * This may release the path, and so you may lose any locks held at the
5090 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5092 struct btrfs_key key
;
5093 struct btrfs_disk_key found_key
;
5096 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5098 if (key
.offset
> 0) {
5100 } else if (key
.type
> 0) {
5102 key
.offset
= (u64
)-1;
5103 } else if (key
.objectid
> 0) {
5106 key
.offset
= (u64
)-1;
5111 btrfs_release_path(path
);
5112 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5115 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5116 ret
= comp_keys(&found_key
, &key
);
5123 * A helper function to walk down the tree starting at min_key, and looking
5124 * for nodes or leaves that are have a minimum transaction id.
5125 * This is used by the btree defrag code, and tree logging
5127 * This does not cow, but it does stuff the starting key it finds back
5128 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5129 * key and get a writable path.
5131 * This does lock as it descends, and path->keep_locks should be set
5132 * to 1 by the caller.
5134 * This honors path->lowest_level to prevent descent past a given level
5137 * min_trans indicates the oldest transaction that you are interested
5138 * in walking through. Any nodes or leaves older than min_trans are
5139 * skipped over (without reading them).
5141 * returns zero if something useful was found, < 0 on error and 1 if there
5142 * was nothing in the tree that matched the search criteria.
5144 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5145 struct btrfs_path
*path
,
5148 struct extent_buffer
*cur
;
5149 struct btrfs_key found_key
;
5156 WARN_ON(!path
->keep_locks
);
5158 cur
= btrfs_read_lock_root_node(root
);
5159 level
= btrfs_header_level(cur
);
5160 WARN_ON(path
->nodes
[level
]);
5161 path
->nodes
[level
] = cur
;
5162 path
->locks
[level
] = BTRFS_READ_LOCK
;
5164 if (btrfs_header_generation(cur
) < min_trans
) {
5169 nritems
= btrfs_header_nritems(cur
);
5170 level
= btrfs_header_level(cur
);
5171 sret
= bin_search(cur
, min_key
, level
, &slot
);
5173 /* at the lowest level, we're done, setup the path and exit */
5174 if (level
== path
->lowest_level
) {
5175 if (slot
>= nritems
)
5178 path
->slots
[level
] = slot
;
5179 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5182 if (sret
&& slot
> 0)
5185 * check this node pointer against the min_trans parameters.
5186 * If it is too old, old, skip to the next one.
5188 while (slot
< nritems
) {
5191 gen
= btrfs_node_ptr_generation(cur
, slot
);
5192 if (gen
< min_trans
) {
5200 * we didn't find a candidate key in this node, walk forward
5201 * and find another one
5203 if (slot
>= nritems
) {
5204 path
->slots
[level
] = slot
;
5205 btrfs_set_path_blocking(path
);
5206 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5209 btrfs_release_path(path
);
5215 /* save our key for returning back */
5216 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5217 path
->slots
[level
] = slot
;
5218 if (level
== path
->lowest_level
) {
5220 unlock_up(path
, level
, 1, 0, NULL
);
5223 btrfs_set_path_blocking(path
);
5224 cur
= read_node_slot(root
, cur
, slot
);
5225 BUG_ON(!cur
); /* -ENOMEM */
5227 btrfs_tree_read_lock(cur
);
5229 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5230 path
->nodes
[level
- 1] = cur
;
5231 unlock_up(path
, level
, 1, 0, NULL
);
5232 btrfs_clear_path_blocking(path
, NULL
, 0);
5236 memcpy(min_key
, &found_key
, sizeof(found_key
));
5237 btrfs_set_path_blocking(path
);
5241 static void tree_move_down(struct btrfs_root
*root
,
5242 struct btrfs_path
*path
,
5243 int *level
, int root_level
)
5245 BUG_ON(*level
== 0);
5246 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
5247 path
->slots
[*level
]);
5248 path
->slots
[*level
- 1] = 0;
5252 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
5253 struct btrfs_path
*path
,
5254 int *level
, int root_level
)
5258 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5260 path
->slots
[*level
]++;
5262 while (path
->slots
[*level
] >= nritems
) {
5263 if (*level
== root_level
)
5267 path
->slots
[*level
] = 0;
5268 free_extent_buffer(path
->nodes
[*level
]);
5269 path
->nodes
[*level
] = NULL
;
5271 path
->slots
[*level
]++;
5273 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5280 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5283 static int tree_advance(struct btrfs_root
*root
,
5284 struct btrfs_path
*path
,
5285 int *level
, int root_level
,
5287 struct btrfs_key
*key
)
5291 if (*level
== 0 || !allow_down
) {
5292 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5294 tree_move_down(root
, path
, level
, root_level
);
5299 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5300 path
->slots
[*level
]);
5302 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5303 path
->slots
[*level
]);
5308 static int tree_compare_item(struct btrfs_root
*left_root
,
5309 struct btrfs_path
*left_path
,
5310 struct btrfs_path
*right_path
,
5315 unsigned long off1
, off2
;
5317 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5318 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5322 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5323 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5324 right_path
->slots
[0]);
5326 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5328 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5335 #define ADVANCE_ONLY_NEXT -1
5338 * This function compares two trees and calls the provided callback for
5339 * every changed/new/deleted item it finds.
5340 * If shared tree blocks are encountered, whole subtrees are skipped, making
5341 * the compare pretty fast on snapshotted subvolumes.
5343 * This currently works on commit roots only. As commit roots are read only,
5344 * we don't do any locking. The commit roots are protected with transactions.
5345 * Transactions are ended and rejoined when a commit is tried in between.
5347 * This function checks for modifications done to the trees while comparing.
5348 * If it detects a change, it aborts immediately.
5350 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5351 struct btrfs_root
*right_root
,
5352 btrfs_changed_cb_t changed_cb
, void *ctx
)
5356 struct btrfs_trans_handle
*trans
= NULL
;
5357 struct btrfs_path
*left_path
= NULL
;
5358 struct btrfs_path
*right_path
= NULL
;
5359 struct btrfs_key left_key
;
5360 struct btrfs_key right_key
;
5361 char *tmp_buf
= NULL
;
5362 int left_root_level
;
5363 int right_root_level
;
5366 int left_end_reached
;
5367 int right_end_reached
;
5372 u64 left_start_ctransid
;
5373 u64 right_start_ctransid
;
5376 left_path
= btrfs_alloc_path();
5381 right_path
= btrfs_alloc_path();
5387 tmp_buf
= kmalloc(left_root
->leafsize
, GFP_NOFS
);
5393 left_path
->search_commit_root
= 1;
5394 left_path
->skip_locking
= 1;
5395 right_path
->search_commit_root
= 1;
5396 right_path
->skip_locking
= 1;
5398 spin_lock(&left_root
->root_item_lock
);
5399 left_start_ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5400 spin_unlock(&left_root
->root_item_lock
);
5402 spin_lock(&right_root
->root_item_lock
);
5403 right_start_ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5404 spin_unlock(&right_root
->root_item_lock
);
5406 trans
= btrfs_join_transaction(left_root
);
5407 if (IS_ERR(trans
)) {
5408 ret
= PTR_ERR(trans
);
5414 * Strategy: Go to the first items of both trees. Then do
5416 * If both trees are at level 0
5417 * Compare keys of current items
5418 * If left < right treat left item as new, advance left tree
5420 * If left > right treat right item as deleted, advance right tree
5422 * If left == right do deep compare of items, treat as changed if
5423 * needed, advance both trees and repeat
5424 * If both trees are at the same level but not at level 0
5425 * Compare keys of current nodes/leafs
5426 * If left < right advance left tree and repeat
5427 * If left > right advance right tree and repeat
5428 * If left == right compare blockptrs of the next nodes/leafs
5429 * If they match advance both trees but stay at the same level
5431 * If they don't match advance both trees while allowing to go
5433 * If tree levels are different
5434 * Advance the tree that needs it and repeat
5436 * Advancing a tree means:
5437 * If we are at level 0, try to go to the next slot. If that's not
5438 * possible, go one level up and repeat. Stop when we found a level
5439 * where we could go to the next slot. We may at this point be on a
5442 * If we are not at level 0 and not on shared tree blocks, go one
5445 * If we are not at level 0 and on shared tree blocks, go one slot to
5446 * the right if possible or go up and right.
5449 left_level
= btrfs_header_level(left_root
->commit_root
);
5450 left_root_level
= left_level
;
5451 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5452 extent_buffer_get(left_path
->nodes
[left_level
]);
5454 right_level
= btrfs_header_level(right_root
->commit_root
);
5455 right_root_level
= right_level
;
5456 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5457 extent_buffer_get(right_path
->nodes
[right_level
]);
5459 if (left_level
== 0)
5460 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5461 &left_key
, left_path
->slots
[left_level
]);
5463 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5464 &left_key
, left_path
->slots
[left_level
]);
5465 if (right_level
== 0)
5466 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5467 &right_key
, right_path
->slots
[right_level
]);
5469 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5470 &right_key
, right_path
->slots
[right_level
]);
5472 left_end_reached
= right_end_reached
= 0;
5473 advance_left
= advance_right
= 0;
5477 * We need to make sure the transaction does not get committed
5478 * while we do anything on commit roots. This means, we need to
5479 * join and leave transactions for every item that we process.
5481 if (trans
&& btrfs_should_end_transaction(trans
, left_root
)) {
5482 btrfs_release_path(left_path
);
5483 btrfs_release_path(right_path
);
5485 ret
= btrfs_end_transaction(trans
, left_root
);
5490 /* now rejoin the transaction */
5492 trans
= btrfs_join_transaction(left_root
);
5493 if (IS_ERR(trans
)) {
5494 ret
= PTR_ERR(trans
);
5499 spin_lock(&left_root
->root_item_lock
);
5500 ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5501 spin_unlock(&left_root
->root_item_lock
);
5502 if (ctransid
!= left_start_ctransid
)
5503 left_start_ctransid
= 0;
5505 spin_lock(&right_root
->root_item_lock
);
5506 ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5507 spin_unlock(&right_root
->root_item_lock
);
5508 if (ctransid
!= right_start_ctransid
)
5509 right_start_ctransid
= 0;
5511 if (!left_start_ctransid
|| !right_start_ctransid
) {
5512 WARN(1, KERN_WARNING
5513 "BTRFS: btrfs_compare_tree detected "
5514 "a change in one of the trees while "
5515 "iterating. This is probably a "
5522 * the commit root may have changed, so start again
5525 left_path
->lowest_level
= left_level
;
5526 right_path
->lowest_level
= right_level
;
5527 ret
= btrfs_search_slot(NULL
, left_root
,
5528 &left_key
, left_path
, 0, 0);
5531 ret
= btrfs_search_slot(NULL
, right_root
,
5532 &right_key
, right_path
, 0, 0);
5537 if (advance_left
&& !left_end_reached
) {
5538 ret
= tree_advance(left_root
, left_path
, &left_level
,
5540 advance_left
!= ADVANCE_ONLY_NEXT
,
5543 left_end_reached
= ADVANCE
;
5546 if (advance_right
&& !right_end_reached
) {
5547 ret
= tree_advance(right_root
, right_path
, &right_level
,
5549 advance_right
!= ADVANCE_ONLY_NEXT
,
5552 right_end_reached
= ADVANCE
;
5556 if (left_end_reached
&& right_end_reached
) {
5559 } else if (left_end_reached
) {
5560 if (right_level
== 0) {
5561 ret
= changed_cb(left_root
, right_root
,
5562 left_path
, right_path
,
5564 BTRFS_COMPARE_TREE_DELETED
,
5569 advance_right
= ADVANCE
;
5571 } else if (right_end_reached
) {
5572 if (left_level
== 0) {
5573 ret
= changed_cb(left_root
, right_root
,
5574 left_path
, right_path
,
5576 BTRFS_COMPARE_TREE_NEW
,
5581 advance_left
= ADVANCE
;
5585 if (left_level
== 0 && right_level
== 0) {
5586 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5588 ret
= changed_cb(left_root
, right_root
,
5589 left_path
, right_path
,
5591 BTRFS_COMPARE_TREE_NEW
,
5595 advance_left
= ADVANCE
;
5596 } else if (cmp
> 0) {
5597 ret
= changed_cb(left_root
, right_root
,
5598 left_path
, right_path
,
5600 BTRFS_COMPARE_TREE_DELETED
,
5604 advance_right
= ADVANCE
;
5606 enum btrfs_compare_tree_result cmp
;
5608 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5609 ret
= tree_compare_item(left_root
, left_path
,
5610 right_path
, tmp_buf
);
5612 cmp
= BTRFS_COMPARE_TREE_CHANGED
;
5614 cmp
= BTRFS_COMPARE_TREE_SAME
;
5615 ret
= changed_cb(left_root
, right_root
,
5616 left_path
, right_path
,
5617 &left_key
, cmp
, ctx
);
5620 advance_left
= ADVANCE
;
5621 advance_right
= ADVANCE
;
5623 } else if (left_level
== right_level
) {
5624 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5626 advance_left
= ADVANCE
;
5627 } else if (cmp
> 0) {
5628 advance_right
= ADVANCE
;
5630 left_blockptr
= btrfs_node_blockptr(
5631 left_path
->nodes
[left_level
],
5632 left_path
->slots
[left_level
]);
5633 right_blockptr
= btrfs_node_blockptr(
5634 right_path
->nodes
[right_level
],
5635 right_path
->slots
[right_level
]);
5636 if (left_blockptr
== right_blockptr
) {
5638 * As we're on a shared block, don't
5639 * allow to go deeper.
5641 advance_left
= ADVANCE_ONLY_NEXT
;
5642 advance_right
= ADVANCE_ONLY_NEXT
;
5644 advance_left
= ADVANCE
;
5645 advance_right
= ADVANCE
;
5648 } else if (left_level
< right_level
) {
5649 advance_right
= ADVANCE
;
5651 advance_left
= ADVANCE
;
5656 btrfs_free_path(left_path
);
5657 btrfs_free_path(right_path
);
5662 ret
= btrfs_end_transaction(trans
, left_root
);
5664 btrfs_end_transaction(trans
, left_root
);
5671 * this is similar to btrfs_next_leaf, but does not try to preserve
5672 * and fixup the path. It looks for and returns the next key in the
5673 * tree based on the current path and the min_trans parameters.
5675 * 0 is returned if another key is found, < 0 if there are any errors
5676 * and 1 is returned if there are no higher keys in the tree
5678 * path->keep_locks should be set to 1 on the search made before
5679 * calling this function.
5681 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5682 struct btrfs_key
*key
, int level
, u64 min_trans
)
5685 struct extent_buffer
*c
;
5687 WARN_ON(!path
->keep_locks
);
5688 while (level
< BTRFS_MAX_LEVEL
) {
5689 if (!path
->nodes
[level
])
5692 slot
= path
->slots
[level
] + 1;
5693 c
= path
->nodes
[level
];
5695 if (slot
>= btrfs_header_nritems(c
)) {
5698 struct btrfs_key cur_key
;
5699 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5700 !path
->nodes
[level
+ 1])
5703 if (path
->locks
[level
+ 1]) {
5708 slot
= btrfs_header_nritems(c
) - 1;
5710 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5712 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5714 orig_lowest
= path
->lowest_level
;
5715 btrfs_release_path(path
);
5716 path
->lowest_level
= level
;
5717 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5719 path
->lowest_level
= orig_lowest
;
5723 c
= path
->nodes
[level
];
5724 slot
= path
->slots
[level
];
5731 btrfs_item_key_to_cpu(c
, key
, slot
);
5733 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5735 if (gen
< min_trans
) {
5739 btrfs_node_key_to_cpu(c
, key
, slot
);
5747 * search the tree again to find a leaf with greater keys
5748 * returns 0 if it found something or 1 if there are no greater leaves.
5749 * returns < 0 on io errors.
5751 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5753 return btrfs_next_old_leaf(root
, path
, 0);
5756 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5761 struct extent_buffer
*c
;
5762 struct extent_buffer
*next
;
5763 struct btrfs_key key
;
5766 int old_spinning
= path
->leave_spinning
;
5767 int next_rw_lock
= 0;
5769 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5773 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5778 btrfs_release_path(path
);
5780 path
->keep_locks
= 1;
5781 path
->leave_spinning
= 1;
5784 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5786 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5787 path
->keep_locks
= 0;
5792 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5794 * by releasing the path above we dropped all our locks. A balance
5795 * could have added more items next to the key that used to be
5796 * at the very end of the block. So, check again here and
5797 * advance the path if there are now more items available.
5799 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5806 while (level
< BTRFS_MAX_LEVEL
) {
5807 if (!path
->nodes
[level
]) {
5812 slot
= path
->slots
[level
] + 1;
5813 c
= path
->nodes
[level
];
5814 if (slot
>= btrfs_header_nritems(c
)) {
5816 if (level
== BTRFS_MAX_LEVEL
) {
5824 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5825 free_extent_buffer(next
);
5829 next_rw_lock
= path
->locks
[level
];
5830 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5836 btrfs_release_path(path
);
5840 if (!path
->skip_locking
) {
5841 ret
= btrfs_try_tree_read_lock(next
);
5842 if (!ret
&& time_seq
) {
5844 * If we don't get the lock, we may be racing
5845 * with push_leaf_left, holding that lock while
5846 * itself waiting for the leaf we've currently
5847 * locked. To solve this situation, we give up
5848 * on our lock and cycle.
5850 free_extent_buffer(next
);
5851 btrfs_release_path(path
);
5856 btrfs_set_path_blocking(path
);
5857 btrfs_tree_read_lock(next
);
5858 btrfs_clear_path_blocking(path
, next
,
5861 next_rw_lock
= BTRFS_READ_LOCK
;
5865 path
->slots
[level
] = slot
;
5868 c
= path
->nodes
[level
];
5869 if (path
->locks
[level
])
5870 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5872 free_extent_buffer(c
);
5873 path
->nodes
[level
] = next
;
5874 path
->slots
[level
] = 0;
5875 if (!path
->skip_locking
)
5876 path
->locks
[level
] = next_rw_lock
;
5880 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5886 btrfs_release_path(path
);
5890 if (!path
->skip_locking
) {
5891 ret
= btrfs_try_tree_read_lock(next
);
5893 btrfs_set_path_blocking(path
);
5894 btrfs_tree_read_lock(next
);
5895 btrfs_clear_path_blocking(path
, next
,
5898 next_rw_lock
= BTRFS_READ_LOCK
;
5903 unlock_up(path
, 0, 1, 0, NULL
);
5904 path
->leave_spinning
= old_spinning
;
5906 btrfs_set_path_blocking(path
);
5912 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5913 * searching until it gets past min_objectid or finds an item of 'type'
5915 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5917 int btrfs_previous_item(struct btrfs_root
*root
,
5918 struct btrfs_path
*path
, u64 min_objectid
,
5921 struct btrfs_key found_key
;
5922 struct extent_buffer
*leaf
;
5927 if (path
->slots
[0] == 0) {
5928 btrfs_set_path_blocking(path
);
5929 ret
= btrfs_prev_leaf(root
, path
);
5935 leaf
= path
->nodes
[0];
5936 nritems
= btrfs_header_nritems(leaf
);
5939 if (path
->slots
[0] == nritems
)
5942 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5943 if (found_key
.objectid
< min_objectid
)
5945 if (found_key
.type
== type
)
5947 if (found_key
.objectid
== min_objectid
&&
5948 found_key
.type
< type
)