2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
29 *root
, struct btrfs_path
*path
, int level
);
30 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
31 *root
, struct btrfs_key
*ins_key
,
32 struct btrfs_path
*path
, int data_size
, int extend
);
33 static int push_node_left(struct btrfs_trans_handle
*trans
,
34 struct btrfs_root
*root
, struct extent_buffer
*dst
,
35 struct extent_buffer
*src
, int empty
);
36 static int balance_node_right(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 struct extent_buffer
*dst_buf
,
39 struct extent_buffer
*src_buf
);
40 static void del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
41 struct btrfs_path
*path
, int level
, int slot
);
42 static void tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
43 struct extent_buffer
*eb
);
44 struct extent_buffer
*read_old_tree_block(struct btrfs_root
*root
, u64 bytenr
,
45 u32 blocksize
, u64 parent_transid
,
47 struct extent_buffer
*btrfs_find_old_tree_block(struct btrfs_root
*root
,
48 u64 bytenr
, u32 blocksize
,
51 struct btrfs_path
*btrfs_alloc_path(void)
53 struct btrfs_path
*path
;
54 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
59 * set all locked nodes in the path to blocking locks. This should
60 * be done before scheduling
62 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
65 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
66 if (!p
->nodes
[i
] || !p
->locks
[i
])
68 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
69 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
70 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
71 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
72 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
77 * reset all the locked nodes in the patch to spinning locks.
79 * held is used to keep lockdep happy, when lockdep is enabled
80 * we set held to a blocking lock before we go around and
81 * retake all the spinlocks in the path. You can safely use NULL
84 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
85 struct extent_buffer
*held
, int held_rw
)
89 #ifdef CONFIG_DEBUG_LOCK_ALLOC
90 /* lockdep really cares that we take all of these spinlocks
91 * in the right order. If any of the locks in the path are not
92 * currently blocking, it is going to complain. So, make really
93 * really sure by forcing the path to blocking before we clear
97 btrfs_set_lock_blocking_rw(held
, held_rw
);
98 if (held_rw
== BTRFS_WRITE_LOCK
)
99 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
100 else if (held_rw
== BTRFS_READ_LOCK
)
101 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
103 btrfs_set_path_blocking(p
);
106 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
107 if (p
->nodes
[i
] && p
->locks
[i
]) {
108 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
109 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
110 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
111 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
112 p
->locks
[i
] = BTRFS_READ_LOCK
;
116 #ifdef CONFIG_DEBUG_LOCK_ALLOC
118 btrfs_clear_lock_blocking_rw(held
, held_rw
);
122 /* this also releases the path */
123 void btrfs_free_path(struct btrfs_path
*p
)
127 btrfs_release_path(p
);
128 kmem_cache_free(btrfs_path_cachep
, p
);
132 * path release drops references on the extent buffers in the path
133 * and it drops any locks held by this path
135 * It is safe to call this on paths that no locks or extent buffers held.
137 noinline
void btrfs_release_path(struct btrfs_path
*p
)
141 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
146 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
149 free_extent_buffer(p
->nodes
[i
]);
155 * safely gets a reference on the root node of a tree. A lock
156 * is not taken, so a concurrent writer may put a different node
157 * at the root of the tree. See btrfs_lock_root_node for the
160 * The extent buffer returned by this has a reference taken, so
161 * it won't disappear. It may stop being the root of the tree
162 * at any time because there are no locks held.
164 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
166 struct extent_buffer
*eb
;
170 eb
= rcu_dereference(root
->node
);
173 * RCU really hurts here, we could free up the root node because
174 * it was cow'ed but we may not get the new root node yet so do
175 * the inc_not_zero dance and if it doesn't work then
176 * synchronize_rcu and try again.
178 if (atomic_inc_not_zero(&eb
->refs
)) {
188 /* loop around taking references on and locking the root node of the
189 * tree until you end up with a lock on the root. A locked buffer
190 * is returned, with a reference held.
192 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
194 struct extent_buffer
*eb
;
197 eb
= btrfs_root_node(root
);
199 if (eb
== root
->node
)
201 btrfs_tree_unlock(eb
);
202 free_extent_buffer(eb
);
207 /* loop around taking references on and locking the root node of the
208 * tree until you end up with a lock on the root. A locked buffer
209 * is returned, with a reference held.
211 struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
213 struct extent_buffer
*eb
;
216 eb
= btrfs_root_node(root
);
217 btrfs_tree_read_lock(eb
);
218 if (eb
== root
->node
)
220 btrfs_tree_read_unlock(eb
);
221 free_extent_buffer(eb
);
226 /* cowonly root (everything not a reference counted cow subvolume), just get
227 * put onto a simple dirty list. transaction.c walks this to make sure they
228 * get properly updated on disk.
230 static void add_root_to_dirty_list(struct btrfs_root
*root
)
232 spin_lock(&root
->fs_info
->trans_lock
);
233 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
234 list_add(&root
->dirty_list
,
235 &root
->fs_info
->dirty_cowonly_roots
);
237 spin_unlock(&root
->fs_info
->trans_lock
);
241 * used by snapshot creation to make a copy of a root for a tree with
242 * a given objectid. The buffer with the new root node is returned in
243 * cow_ret, and this func returns zero on success or a negative error code.
245 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
246 struct btrfs_root
*root
,
247 struct extent_buffer
*buf
,
248 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
250 struct extent_buffer
*cow
;
253 struct btrfs_disk_key disk_key
;
255 WARN_ON(root
->ref_cows
&& trans
->transid
!=
256 root
->fs_info
->running_transaction
->transid
);
257 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
259 level
= btrfs_header_level(buf
);
261 btrfs_item_key(buf
, &disk_key
, 0);
263 btrfs_node_key(buf
, &disk_key
, 0);
265 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, 0,
266 new_root_objectid
, &disk_key
, level
,
271 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
272 btrfs_set_header_bytenr(cow
, cow
->start
);
273 btrfs_set_header_generation(cow
, trans
->transid
);
274 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
275 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
276 BTRFS_HEADER_FLAG_RELOC
);
277 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
278 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
280 btrfs_set_header_owner(cow
, new_root_objectid
);
282 write_extent_buffer(cow
, root
->fs_info
->fsid
,
283 (unsigned long)btrfs_header_fsid(cow
),
286 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
287 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
288 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
290 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
295 btrfs_mark_buffer_dirty(cow
);
304 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
305 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
307 MOD_LOG_ROOT_REPLACE
,
310 struct tree_mod_move
{
315 struct tree_mod_root
{
320 struct tree_mod_elem
{
322 u64 index
; /* shifted logical */
326 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
329 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
332 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
333 struct btrfs_disk_key key
;
336 /* this is used for op == MOD_LOG_MOVE_KEYS */
337 struct tree_mod_move move
;
339 /* this is used for op == MOD_LOG_ROOT_REPLACE */
340 struct tree_mod_root old_root
;
343 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
345 read_lock(&fs_info
->tree_mod_log_lock
);
348 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
350 read_unlock(&fs_info
->tree_mod_log_lock
);
353 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
355 write_lock(&fs_info
->tree_mod_log_lock
);
358 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
360 write_unlock(&fs_info
->tree_mod_log_lock
);
364 * This adds a new blocker to the tree mod log's blocker list if the @elem
365 * passed does not already have a sequence number set. So when a caller expects
366 * to record tree modifications, it should ensure to set elem->seq to zero
367 * before calling btrfs_get_tree_mod_seq.
368 * Returns a fresh, unused tree log modification sequence number, even if no new
371 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
372 struct seq_list
*elem
)
376 tree_mod_log_write_lock(fs_info
);
377 spin_lock(&fs_info
->tree_mod_seq_lock
);
379 elem
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
380 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
382 seq
= btrfs_inc_tree_mod_seq(fs_info
);
383 spin_unlock(&fs_info
->tree_mod_seq_lock
);
384 tree_mod_log_write_unlock(fs_info
);
389 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
390 struct seq_list
*elem
)
392 struct rb_root
*tm_root
;
393 struct rb_node
*node
;
394 struct rb_node
*next
;
395 struct seq_list
*cur_elem
;
396 struct tree_mod_elem
*tm
;
397 u64 min_seq
= (u64
)-1;
398 u64 seq_putting
= elem
->seq
;
403 spin_lock(&fs_info
->tree_mod_seq_lock
);
404 list_del(&elem
->list
);
407 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
408 if (cur_elem
->seq
< min_seq
) {
409 if (seq_putting
> cur_elem
->seq
) {
411 * blocker with lower sequence number exists, we
412 * cannot remove anything from the log
414 spin_unlock(&fs_info
->tree_mod_seq_lock
);
417 min_seq
= cur_elem
->seq
;
420 spin_unlock(&fs_info
->tree_mod_seq_lock
);
423 * anything that's lower than the lowest existing (read: blocked)
424 * sequence number can be removed from the tree.
426 tree_mod_log_write_lock(fs_info
);
427 tm_root
= &fs_info
->tree_mod_log
;
428 for (node
= rb_first(tm_root
); node
; node
= next
) {
429 next
= rb_next(node
);
430 tm
= container_of(node
, struct tree_mod_elem
, node
);
431 if (tm
->seq
> min_seq
)
433 rb_erase(node
, tm_root
);
436 tree_mod_log_write_unlock(fs_info
);
440 * key order of the log:
443 * the index is the shifted logical of the *new* root node for root replace
444 * operations, or the shifted logical of the affected block for all other
448 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
450 struct rb_root
*tm_root
;
451 struct rb_node
**new;
452 struct rb_node
*parent
= NULL
;
453 struct tree_mod_elem
*cur
;
455 BUG_ON(!tm
|| !tm
->seq
);
457 tm_root
= &fs_info
->tree_mod_log
;
458 new = &tm_root
->rb_node
;
460 cur
= container_of(*new, struct tree_mod_elem
, node
);
462 if (cur
->index
< tm
->index
)
463 new = &((*new)->rb_left
);
464 else if (cur
->index
> tm
->index
)
465 new = &((*new)->rb_right
);
466 else if (cur
->seq
< tm
->seq
)
467 new = &((*new)->rb_left
);
468 else if (cur
->seq
> tm
->seq
)
469 new = &((*new)->rb_right
);
476 rb_link_node(&tm
->node
, parent
, new);
477 rb_insert_color(&tm
->node
, tm_root
);
482 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
483 * returns zero with the tree_mod_log_lock acquired. The caller must hold
484 * this until all tree mod log insertions are recorded in the rb tree and then
485 * call tree_mod_log_write_unlock() to release.
487 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
488 struct extent_buffer
*eb
) {
490 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
492 if (eb
&& btrfs_header_level(eb
) == 0)
495 tree_mod_log_write_lock(fs_info
);
496 if (list_empty(&fs_info
->tree_mod_seq_list
)) {
498 * someone emptied the list while we were waiting for the lock.
499 * we must not add to the list when no blocker exists.
501 tree_mod_log_write_unlock(fs_info
);
509 * This allocates memory and gets a tree modification sequence number.
511 * Returns <0 on error.
512 * Returns >0 (the added sequence number) on success.
514 static inline int tree_mod_alloc(struct btrfs_fs_info
*fs_info
, gfp_t flags
,
515 struct tree_mod_elem
**tm_ret
)
517 struct tree_mod_elem
*tm
;
520 * once we switch from spin locks to something different, we should
521 * honor the flags parameter here.
523 tm
= *tm_ret
= kzalloc(sizeof(*tm
), GFP_ATOMIC
);
527 tm
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
532 __tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
533 struct extent_buffer
*eb
, int slot
,
534 enum mod_log_op op
, gfp_t flags
)
537 struct tree_mod_elem
*tm
;
539 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
543 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
544 if (op
!= MOD_LOG_KEY_ADD
) {
545 btrfs_node_key(eb
, &tm
->key
, slot
);
546 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
550 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
552 return __tree_mod_log_insert(fs_info
, tm
);
556 tree_mod_log_insert_key_mask(struct btrfs_fs_info
*fs_info
,
557 struct extent_buffer
*eb
, int slot
,
558 enum mod_log_op op
, gfp_t flags
)
562 if (tree_mod_dont_log(fs_info
, eb
))
565 ret
= __tree_mod_log_insert_key(fs_info
, eb
, slot
, op
, flags
);
567 tree_mod_log_write_unlock(fs_info
);
572 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
573 int slot
, enum mod_log_op op
)
575 return tree_mod_log_insert_key_mask(fs_info
, eb
, slot
, op
, GFP_NOFS
);
579 tree_mod_log_insert_key_locked(struct btrfs_fs_info
*fs_info
,
580 struct extent_buffer
*eb
, int slot
,
583 return __tree_mod_log_insert_key(fs_info
, eb
, slot
, op
, GFP_NOFS
);
587 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
588 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
589 int nr_items
, gfp_t flags
)
591 struct tree_mod_elem
*tm
;
595 if (tree_mod_dont_log(fs_info
, eb
))
599 * When we override something during the move, we log these removals.
600 * This can only happen when we move towards the beginning of the
601 * buffer, i.e. dst_slot < src_slot.
603 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
604 ret
= tree_mod_log_insert_key_locked(fs_info
, eb
, i
+ dst_slot
,
605 MOD_LOG_KEY_REMOVE_WHILE_MOVING
);
609 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
613 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
615 tm
->move
.dst_slot
= dst_slot
;
616 tm
->move
.nr_items
= nr_items
;
617 tm
->op
= MOD_LOG_MOVE_KEYS
;
619 ret
= __tree_mod_log_insert(fs_info
, tm
);
621 tree_mod_log_write_unlock(fs_info
);
626 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
632 if (btrfs_header_level(eb
) == 0)
635 nritems
= btrfs_header_nritems(eb
);
636 for (i
= nritems
- 1; i
>= 0; i
--) {
637 ret
= tree_mod_log_insert_key_locked(fs_info
, eb
, i
,
638 MOD_LOG_KEY_REMOVE_WHILE_FREEING
);
644 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
645 struct extent_buffer
*old_root
,
646 struct extent_buffer
*new_root
, gfp_t flags
)
648 struct tree_mod_elem
*tm
;
651 if (tree_mod_dont_log(fs_info
, NULL
))
654 __tree_mod_log_free_eb(fs_info
, old_root
);
656 ret
= tree_mod_alloc(fs_info
, flags
, &tm
);
660 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
661 tm
->old_root
.logical
= old_root
->start
;
662 tm
->old_root
.level
= btrfs_header_level(old_root
);
663 tm
->generation
= btrfs_header_generation(old_root
);
664 tm
->op
= MOD_LOG_ROOT_REPLACE
;
666 ret
= __tree_mod_log_insert(fs_info
, tm
);
668 tree_mod_log_write_unlock(fs_info
);
672 static struct tree_mod_elem
*
673 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
676 struct rb_root
*tm_root
;
677 struct rb_node
*node
;
678 struct tree_mod_elem
*cur
= NULL
;
679 struct tree_mod_elem
*found
= NULL
;
680 u64 index
= start
>> PAGE_CACHE_SHIFT
;
682 tree_mod_log_read_lock(fs_info
);
683 tm_root
= &fs_info
->tree_mod_log
;
684 node
= tm_root
->rb_node
;
686 cur
= container_of(node
, struct tree_mod_elem
, node
);
687 if (cur
->index
< index
) {
688 node
= node
->rb_left
;
689 } else if (cur
->index
> index
) {
690 node
= node
->rb_right
;
691 } else if (cur
->seq
< min_seq
) {
692 node
= node
->rb_left
;
693 } else if (!smallest
) {
694 /* we want the node with the highest seq */
696 BUG_ON(found
->seq
> cur
->seq
);
698 node
= node
->rb_left
;
699 } else if (cur
->seq
> min_seq
) {
700 /* we want the node with the smallest seq */
702 BUG_ON(found
->seq
< cur
->seq
);
704 node
= node
->rb_right
;
710 tree_mod_log_read_unlock(fs_info
);
716 * this returns the element from the log with the smallest time sequence
717 * value that's in the log (the oldest log item). any element with a time
718 * sequence lower than min_seq will be ignored.
720 static struct tree_mod_elem
*
721 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
724 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
728 * this returns the element from the log with the largest time sequence
729 * value that's in the log (the most recent log item). any element with
730 * a time sequence lower than min_seq will be ignored.
732 static struct tree_mod_elem
*
733 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
735 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
739 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
740 struct extent_buffer
*src
, unsigned long dst_offset
,
741 unsigned long src_offset
, int nr_items
, int log_removal
)
746 if (tree_mod_dont_log(fs_info
, NULL
))
749 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0) {
750 tree_mod_log_write_unlock(fs_info
);
754 for (i
= 0; i
< nr_items
; i
++) {
756 ret
= tree_mod_log_insert_key_locked(fs_info
, src
,
761 ret
= tree_mod_log_insert_key_locked(fs_info
, dst
,
767 tree_mod_log_write_unlock(fs_info
);
771 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
772 int dst_offset
, int src_offset
, int nr_items
)
775 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
781 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
782 struct extent_buffer
*eb
, int slot
, int atomic
)
786 ret
= tree_mod_log_insert_key_mask(fs_info
, eb
, slot
,
788 atomic
? GFP_ATOMIC
: GFP_NOFS
);
793 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
795 if (tree_mod_dont_log(fs_info
, eb
))
798 __tree_mod_log_free_eb(fs_info
, eb
);
800 tree_mod_log_write_unlock(fs_info
);
804 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
805 struct extent_buffer
*new_root_node
)
808 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
809 new_root_node
, GFP_NOFS
);
814 * check if the tree block can be shared by multiple trees
816 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
817 struct extent_buffer
*buf
)
820 * Tree blocks not in refernece counted trees and tree roots
821 * are never shared. If a block was allocated after the last
822 * snapshot and the block was not allocated by tree relocation,
823 * we know the block is not shared.
825 if (root
->ref_cows
&&
826 buf
!= root
->node
&& buf
!= root
->commit_root
&&
827 (btrfs_header_generation(buf
) <=
828 btrfs_root_last_snapshot(&root
->root_item
) ||
829 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
831 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
832 if (root
->ref_cows
&&
833 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
839 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
840 struct btrfs_root
*root
,
841 struct extent_buffer
*buf
,
842 struct extent_buffer
*cow
,
852 * Backrefs update rules:
854 * Always use full backrefs for extent pointers in tree block
855 * allocated by tree relocation.
857 * If a shared tree block is no longer referenced by its owner
858 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
859 * use full backrefs for extent pointers in tree block.
861 * If a tree block is been relocating
862 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
863 * use full backrefs for extent pointers in tree block.
864 * The reason for this is some operations (such as drop tree)
865 * are only allowed for blocks use full backrefs.
868 if (btrfs_block_can_be_shared(root
, buf
)) {
869 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
870 btrfs_header_level(buf
), 1,
876 btrfs_std_error(root
->fs_info
, ret
);
881 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
882 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
883 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
888 owner
= btrfs_header_owner(buf
);
889 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
890 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
893 if ((owner
== root
->root_key
.objectid
||
894 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
895 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
896 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
897 BUG_ON(ret
); /* -ENOMEM */
899 if (root
->root_key
.objectid
==
900 BTRFS_TREE_RELOC_OBJECTID
) {
901 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
902 BUG_ON(ret
); /* -ENOMEM */
903 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
904 BUG_ON(ret
); /* -ENOMEM */
906 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
909 if (root
->root_key
.objectid
==
910 BTRFS_TREE_RELOC_OBJECTID
)
911 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
913 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
914 BUG_ON(ret
); /* -ENOMEM */
916 if (new_flags
!= 0) {
917 ret
= btrfs_set_disk_extent_flags(trans
, root
,
925 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
926 if (root
->root_key
.objectid
==
927 BTRFS_TREE_RELOC_OBJECTID
)
928 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
930 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
931 BUG_ON(ret
); /* -ENOMEM */
932 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
933 BUG_ON(ret
); /* -ENOMEM */
935 clean_tree_block(trans
, root
, buf
);
942 * does the dirty work in cow of a single block. The parent block (if
943 * supplied) is updated to point to the new cow copy. The new buffer is marked
944 * dirty and returned locked. If you modify the block it needs to be marked
947 * search_start -- an allocation hint for the new block
949 * empty_size -- a hint that you plan on doing more cow. This is the size in
950 * bytes the allocator should try to find free next to the block it returns.
951 * This is just a hint and may be ignored by the allocator.
953 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
954 struct btrfs_root
*root
,
955 struct extent_buffer
*buf
,
956 struct extent_buffer
*parent
, int parent_slot
,
957 struct extent_buffer
**cow_ret
,
958 u64 search_start
, u64 empty_size
)
960 struct btrfs_disk_key disk_key
;
961 struct extent_buffer
*cow
;
970 btrfs_assert_tree_locked(buf
);
972 WARN_ON(root
->ref_cows
&& trans
->transid
!=
973 root
->fs_info
->running_transaction
->transid
);
974 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
976 level
= btrfs_header_level(buf
);
979 btrfs_item_key(buf
, &disk_key
, 0);
981 btrfs_node_key(buf
, &disk_key
, 0);
983 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
985 parent_start
= parent
->start
;
991 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
992 root
->root_key
.objectid
, &disk_key
,
993 level
, search_start
, empty_size
);
997 /* cow is set to blocking by btrfs_init_new_buffer */
999 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
1000 btrfs_set_header_bytenr(cow
, cow
->start
);
1001 btrfs_set_header_generation(cow
, trans
->transid
);
1002 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1003 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1004 BTRFS_HEADER_FLAG_RELOC
);
1005 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1006 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1008 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1010 write_extent_buffer(cow
, root
->fs_info
->fsid
,
1011 (unsigned long)btrfs_header_fsid(cow
),
1014 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1016 btrfs_abort_transaction(trans
, root
, ret
);
1021 btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1023 if (buf
== root
->node
) {
1024 WARN_ON(parent
&& parent
!= buf
);
1025 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1026 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1027 parent_start
= buf
->start
;
1031 extent_buffer_get(cow
);
1032 tree_mod_log_set_root_pointer(root
, cow
);
1033 rcu_assign_pointer(root
->node
, cow
);
1035 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1037 free_extent_buffer(buf
);
1038 add_root_to_dirty_list(root
);
1040 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1041 parent_start
= parent
->start
;
1045 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1046 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1047 MOD_LOG_KEY_REPLACE
);
1048 btrfs_set_node_blockptr(parent
, parent_slot
,
1050 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1052 btrfs_mark_buffer_dirty(parent
);
1053 tree_mod_log_free_eb(root
->fs_info
, buf
);
1054 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1058 btrfs_tree_unlock(buf
);
1059 free_extent_buffer_stale(buf
);
1060 btrfs_mark_buffer_dirty(cow
);
1066 * returns the logical address of the oldest predecessor of the given root.
1067 * entries older than time_seq are ignored.
1069 static struct tree_mod_elem
*
1070 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1071 struct btrfs_root
*root
, u64 time_seq
)
1073 struct tree_mod_elem
*tm
;
1074 struct tree_mod_elem
*found
= NULL
;
1075 u64 root_logical
= root
->node
->start
;
1082 * the very last operation that's logged for a root is the replacement
1083 * operation (if it is replaced at all). this has the index of the *new*
1084 * root, making it the very first operation that's logged for this root.
1087 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1092 * if there are no tree operation for the oldest root, we simply
1093 * return it. this should only happen if that (old) root is at
1100 * if there's an operation that's not a root replacement, we
1101 * found the oldest version of our root. normally, we'll find a
1102 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1104 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1108 root_logical
= tm
->old_root
.logical
;
1109 BUG_ON(root_logical
== root
->node
->start
);
1113 /* if there's no old root to return, return what we found instead */
1121 * tm is a pointer to the first operation to rewind within eb. then, all
1122 * previous operations will be rewinded (until we reach something older than
1126 __tree_mod_log_rewind(struct extent_buffer
*eb
, u64 time_seq
,
1127 struct tree_mod_elem
*first_tm
)
1130 struct rb_node
*next
;
1131 struct tree_mod_elem
*tm
= first_tm
;
1132 unsigned long o_dst
;
1133 unsigned long o_src
;
1134 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1136 n
= btrfs_header_nritems(eb
);
1137 while (tm
&& tm
->seq
>= time_seq
) {
1139 * all the operations are recorded with the operator used for
1140 * the modification. as we're going backwards, we do the
1141 * opposite of each operation here.
1144 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1145 BUG_ON(tm
->slot
< n
);
1147 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1148 case MOD_LOG_KEY_REMOVE
:
1149 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1150 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1151 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1155 case MOD_LOG_KEY_REPLACE
:
1156 BUG_ON(tm
->slot
>= n
);
1157 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1158 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1159 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1162 case MOD_LOG_KEY_ADD
:
1163 /* if a move operation is needed it's in the log */
1166 case MOD_LOG_MOVE_KEYS
:
1167 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1168 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1169 memmove_extent_buffer(eb
, o_dst
, o_src
,
1170 tm
->move
.nr_items
* p_size
);
1172 case MOD_LOG_ROOT_REPLACE
:
1174 * this operation is special. for roots, this must be
1175 * handled explicitly before rewinding.
1176 * for non-roots, this operation may exist if the node
1177 * was a root: root A -> child B; then A gets empty and
1178 * B is promoted to the new root. in the mod log, we'll
1179 * have a root-replace operation for B, a tree block
1180 * that is no root. we simply ignore that operation.
1184 next
= rb_next(&tm
->node
);
1187 tm
= container_of(next
, struct tree_mod_elem
, node
);
1188 if (tm
->index
!= first_tm
->index
)
1191 btrfs_set_header_nritems(eb
, n
);
1194 static struct extent_buffer
*
1195 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1198 struct extent_buffer
*eb_rewin
;
1199 struct tree_mod_elem
*tm
;
1204 if (btrfs_header_level(eb
) == 0)
1207 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1211 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1212 BUG_ON(tm
->slot
!= 0);
1213 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1214 fs_info
->tree_root
->nodesize
);
1216 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1217 btrfs_set_header_backref_rev(eb_rewin
,
1218 btrfs_header_backref_rev(eb
));
1219 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1220 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1222 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1226 extent_buffer_get(eb_rewin
);
1227 free_extent_buffer(eb
);
1229 __tree_mod_log_rewind(eb_rewin
, time_seq
, tm
);
1230 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1231 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1237 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1238 * value. If there are no changes, the current root->root_node is returned. If
1239 * anything changed in between, there's a fresh buffer allocated on which the
1240 * rewind operations are done. In any case, the returned buffer is read locked.
1241 * Returns NULL on error (with no locks held).
1243 static inline struct extent_buffer
*
1244 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1246 struct tree_mod_elem
*tm
;
1247 struct extent_buffer
*eb
;
1248 struct extent_buffer
*old
;
1249 struct tree_mod_root
*old_root
= NULL
;
1250 u64 old_generation
= 0;
1254 eb
= btrfs_read_lock_root_node(root
);
1255 tm
= __tree_mod_log_oldest_root(root
->fs_info
, root
, time_seq
);
1259 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1260 old_root
= &tm
->old_root
;
1261 old_generation
= tm
->generation
;
1262 logical
= old_root
->logical
;
1264 logical
= root
->node
->start
;
1267 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1268 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1269 btrfs_tree_read_unlock(root
->node
);
1270 free_extent_buffer(root
->node
);
1271 blocksize
= btrfs_level_size(root
, old_root
->level
);
1272 old
= read_tree_block(root
, logical
, blocksize
, 0);
1274 pr_warn("btrfs: failed to read tree block %llu from get_old_root\n",
1278 eb
= btrfs_clone_extent_buffer(old
);
1279 free_extent_buffer(old
);
1281 } else if (old_root
) {
1282 btrfs_tree_read_unlock(root
->node
);
1283 free_extent_buffer(root
->node
);
1284 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1286 eb
= btrfs_clone_extent_buffer(root
->node
);
1287 btrfs_tree_read_unlock(root
->node
);
1288 free_extent_buffer(root
->node
);
1293 extent_buffer_get(eb
);
1294 btrfs_tree_read_lock(eb
);
1296 btrfs_set_header_bytenr(eb
, eb
->start
);
1297 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1298 btrfs_set_header_owner(eb
, root
->root_key
.objectid
);
1299 btrfs_set_header_level(eb
, old_root
->level
);
1300 btrfs_set_header_generation(eb
, old_generation
);
1303 __tree_mod_log_rewind(eb
, time_seq
, tm
);
1305 WARN_ON(btrfs_header_level(eb
) != 0);
1306 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1311 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1313 struct tree_mod_elem
*tm
;
1316 tm
= __tree_mod_log_oldest_root(root
->fs_info
, root
, time_seq
);
1317 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1318 level
= tm
->old_root
.level
;
1321 level
= btrfs_header_level(root
->node
);
1328 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1329 struct btrfs_root
*root
,
1330 struct extent_buffer
*buf
)
1332 /* ensure we can see the force_cow */
1336 * We do not need to cow a block if
1337 * 1) this block is not created or changed in this transaction;
1338 * 2) this block does not belong to TREE_RELOC tree;
1339 * 3) the root is not forced COW.
1341 * What is forced COW:
1342 * when we create snapshot during commiting the transaction,
1343 * after we've finished coping src root, we must COW the shared
1344 * block to ensure the metadata consistency.
1346 if (btrfs_header_generation(buf
) == trans
->transid
&&
1347 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1348 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1349 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1356 * cows a single block, see __btrfs_cow_block for the real work.
1357 * This version of it has extra checks so that a block isn't cow'd more than
1358 * once per transaction, as long as it hasn't been written yet
1360 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1361 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1362 struct extent_buffer
*parent
, int parent_slot
,
1363 struct extent_buffer
**cow_ret
)
1368 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1369 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1370 (unsigned long long)trans
->transid
,
1371 (unsigned long long)
1372 root
->fs_info
->running_transaction
->transid
);
1374 if (trans
->transid
!= root
->fs_info
->generation
)
1375 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1376 (unsigned long long)trans
->transid
,
1377 (unsigned long long)root
->fs_info
->generation
);
1379 if (!should_cow_block(trans
, root
, buf
)) {
1384 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1387 btrfs_set_lock_blocking(parent
);
1388 btrfs_set_lock_blocking(buf
);
1390 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1391 parent_slot
, cow_ret
, search_start
, 0);
1393 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1399 * helper function for defrag to decide if two blocks pointed to by a
1400 * node are actually close by
1402 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1404 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1406 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1412 * compare two keys in a memcmp fashion
1414 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1416 struct btrfs_key k1
;
1418 btrfs_disk_key_to_cpu(&k1
, disk
);
1420 return btrfs_comp_cpu_keys(&k1
, k2
);
1424 * same as comp_keys only with two btrfs_key's
1426 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1428 if (k1
->objectid
> k2
->objectid
)
1430 if (k1
->objectid
< k2
->objectid
)
1432 if (k1
->type
> k2
->type
)
1434 if (k1
->type
< k2
->type
)
1436 if (k1
->offset
> k2
->offset
)
1438 if (k1
->offset
< k2
->offset
)
1444 * this is used by the defrag code to go through all the
1445 * leaves pointed to by a node and reallocate them so that
1446 * disk order is close to key order
1448 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1449 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1450 int start_slot
, u64
*last_ret
,
1451 struct btrfs_key
*progress
)
1453 struct extent_buffer
*cur
;
1456 u64 search_start
= *last_ret
;
1466 int progress_passed
= 0;
1467 struct btrfs_disk_key disk_key
;
1469 parent_level
= btrfs_header_level(parent
);
1471 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1472 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1474 parent_nritems
= btrfs_header_nritems(parent
);
1475 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1476 end_slot
= parent_nritems
;
1478 if (parent_nritems
== 1)
1481 btrfs_set_lock_blocking(parent
);
1483 for (i
= start_slot
; i
< end_slot
; i
++) {
1486 btrfs_node_key(parent
, &disk_key
, i
);
1487 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1490 progress_passed
= 1;
1491 blocknr
= btrfs_node_blockptr(parent
, i
);
1492 gen
= btrfs_node_ptr_generation(parent
, i
);
1493 if (last_block
== 0)
1494 last_block
= blocknr
;
1497 other
= btrfs_node_blockptr(parent
, i
- 1);
1498 close
= close_blocks(blocknr
, other
, blocksize
);
1500 if (!close
&& i
< end_slot
- 2) {
1501 other
= btrfs_node_blockptr(parent
, i
+ 1);
1502 close
= close_blocks(blocknr
, other
, blocksize
);
1505 last_block
= blocknr
;
1509 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1511 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1514 if (!cur
|| !uptodate
) {
1516 cur
= read_tree_block(root
, blocknr
,
1520 } else if (!uptodate
) {
1521 err
= btrfs_read_buffer(cur
, gen
);
1523 free_extent_buffer(cur
);
1528 if (search_start
== 0)
1529 search_start
= last_block
;
1531 btrfs_tree_lock(cur
);
1532 btrfs_set_lock_blocking(cur
);
1533 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1536 (end_slot
- i
) * blocksize
));
1538 btrfs_tree_unlock(cur
);
1539 free_extent_buffer(cur
);
1542 search_start
= cur
->start
;
1543 last_block
= cur
->start
;
1544 *last_ret
= search_start
;
1545 btrfs_tree_unlock(cur
);
1546 free_extent_buffer(cur
);
1552 * The leaf data grows from end-to-front in the node.
1553 * this returns the address of the start of the last item,
1554 * which is the stop of the leaf data stack
1556 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1557 struct extent_buffer
*leaf
)
1559 u32 nr
= btrfs_header_nritems(leaf
);
1561 return BTRFS_LEAF_DATA_SIZE(root
);
1562 return btrfs_item_offset_nr(leaf
, nr
- 1);
1567 * search for key in the extent_buffer. The items start at offset p,
1568 * and they are item_size apart. There are 'max' items in p.
1570 * the slot in the array is returned via slot, and it points to
1571 * the place where you would insert key if it is not found in
1574 * slot may point to max if the key is bigger than all of the keys
1576 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1578 int item_size
, struct btrfs_key
*key
,
1585 struct btrfs_disk_key
*tmp
= NULL
;
1586 struct btrfs_disk_key unaligned
;
1587 unsigned long offset
;
1589 unsigned long map_start
= 0;
1590 unsigned long map_len
= 0;
1593 while (low
< high
) {
1594 mid
= (low
+ high
) / 2;
1595 offset
= p
+ mid
* item_size
;
1597 if (!kaddr
|| offset
< map_start
||
1598 (offset
+ sizeof(struct btrfs_disk_key
)) >
1599 map_start
+ map_len
) {
1601 err
= map_private_extent_buffer(eb
, offset
,
1602 sizeof(struct btrfs_disk_key
),
1603 &kaddr
, &map_start
, &map_len
);
1606 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1609 read_extent_buffer(eb
, &unaligned
,
1610 offset
, sizeof(unaligned
));
1615 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1618 ret
= comp_keys(tmp
, key
);
1634 * simple bin_search frontend that does the right thing for
1637 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1638 int level
, int *slot
)
1641 return generic_bin_search(eb
,
1642 offsetof(struct btrfs_leaf
, items
),
1643 sizeof(struct btrfs_item
),
1644 key
, btrfs_header_nritems(eb
),
1647 return generic_bin_search(eb
,
1648 offsetof(struct btrfs_node
, ptrs
),
1649 sizeof(struct btrfs_key_ptr
),
1650 key
, btrfs_header_nritems(eb
),
1654 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1655 int level
, int *slot
)
1657 return bin_search(eb
, key
, level
, slot
);
1660 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1662 spin_lock(&root
->accounting_lock
);
1663 btrfs_set_root_used(&root
->root_item
,
1664 btrfs_root_used(&root
->root_item
) + size
);
1665 spin_unlock(&root
->accounting_lock
);
1668 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1670 spin_lock(&root
->accounting_lock
);
1671 btrfs_set_root_used(&root
->root_item
,
1672 btrfs_root_used(&root
->root_item
) - size
);
1673 spin_unlock(&root
->accounting_lock
);
1676 /* given a node and slot number, this reads the blocks it points to. The
1677 * extent buffer is returned with a reference taken (but unlocked).
1678 * NULL is returned on error.
1680 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1681 struct extent_buffer
*parent
, int slot
)
1683 int level
= btrfs_header_level(parent
);
1686 if (slot
>= btrfs_header_nritems(parent
))
1691 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1692 btrfs_level_size(root
, level
- 1),
1693 btrfs_node_ptr_generation(parent
, slot
));
1697 * node level balancing, used to make sure nodes are in proper order for
1698 * item deletion. We balance from the top down, so we have to make sure
1699 * that a deletion won't leave an node completely empty later on.
1701 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1702 struct btrfs_root
*root
,
1703 struct btrfs_path
*path
, int level
)
1705 struct extent_buffer
*right
= NULL
;
1706 struct extent_buffer
*mid
;
1707 struct extent_buffer
*left
= NULL
;
1708 struct extent_buffer
*parent
= NULL
;
1712 int orig_slot
= path
->slots
[level
];
1718 mid
= path
->nodes
[level
];
1720 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1721 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1722 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1724 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1726 if (level
< BTRFS_MAX_LEVEL
- 1) {
1727 parent
= path
->nodes
[level
+ 1];
1728 pslot
= path
->slots
[level
+ 1];
1732 * deal with the case where there is only one pointer in the root
1733 * by promoting the node below to a root
1736 struct extent_buffer
*child
;
1738 if (btrfs_header_nritems(mid
) != 1)
1741 /* promote the child to a root */
1742 child
= read_node_slot(root
, mid
, 0);
1745 btrfs_std_error(root
->fs_info
, ret
);
1749 btrfs_tree_lock(child
);
1750 btrfs_set_lock_blocking(child
);
1751 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1753 btrfs_tree_unlock(child
);
1754 free_extent_buffer(child
);
1758 tree_mod_log_set_root_pointer(root
, child
);
1759 rcu_assign_pointer(root
->node
, child
);
1761 add_root_to_dirty_list(root
);
1762 btrfs_tree_unlock(child
);
1764 path
->locks
[level
] = 0;
1765 path
->nodes
[level
] = NULL
;
1766 clean_tree_block(trans
, root
, mid
);
1767 btrfs_tree_unlock(mid
);
1768 /* once for the path */
1769 free_extent_buffer(mid
);
1771 root_sub_used(root
, mid
->len
);
1772 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1773 /* once for the root ptr */
1774 free_extent_buffer_stale(mid
);
1777 if (btrfs_header_nritems(mid
) >
1778 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1781 left
= read_node_slot(root
, parent
, pslot
- 1);
1783 btrfs_tree_lock(left
);
1784 btrfs_set_lock_blocking(left
);
1785 wret
= btrfs_cow_block(trans
, root
, left
,
1786 parent
, pslot
- 1, &left
);
1792 right
= read_node_slot(root
, parent
, pslot
+ 1);
1794 btrfs_tree_lock(right
);
1795 btrfs_set_lock_blocking(right
);
1796 wret
= btrfs_cow_block(trans
, root
, right
,
1797 parent
, pslot
+ 1, &right
);
1804 /* first, try to make some room in the middle buffer */
1806 orig_slot
+= btrfs_header_nritems(left
);
1807 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1813 * then try to empty the right most buffer into the middle
1816 wret
= push_node_left(trans
, root
, mid
, right
, 1);
1817 if (wret
< 0 && wret
!= -ENOSPC
)
1819 if (btrfs_header_nritems(right
) == 0) {
1820 clean_tree_block(trans
, root
, right
);
1821 btrfs_tree_unlock(right
);
1822 del_ptr(trans
, root
, path
, level
+ 1, pslot
+ 1);
1823 root_sub_used(root
, right
->len
);
1824 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
1825 free_extent_buffer_stale(right
);
1828 struct btrfs_disk_key right_key
;
1829 btrfs_node_key(right
, &right_key
, 0);
1830 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1832 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1833 btrfs_mark_buffer_dirty(parent
);
1836 if (btrfs_header_nritems(mid
) == 1) {
1838 * we're not allowed to leave a node with one item in the
1839 * tree during a delete. A deletion from lower in the tree
1840 * could try to delete the only pointer in this node.
1841 * So, pull some keys from the left.
1842 * There has to be a left pointer at this point because
1843 * otherwise we would have pulled some pointers from the
1848 btrfs_std_error(root
->fs_info
, ret
);
1851 wret
= balance_node_right(trans
, root
, mid
, left
);
1857 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1863 if (btrfs_header_nritems(mid
) == 0) {
1864 clean_tree_block(trans
, root
, mid
);
1865 btrfs_tree_unlock(mid
);
1866 del_ptr(trans
, root
, path
, level
+ 1, pslot
);
1867 root_sub_used(root
, mid
->len
);
1868 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1869 free_extent_buffer_stale(mid
);
1872 /* update the parent key to reflect our changes */
1873 struct btrfs_disk_key mid_key
;
1874 btrfs_node_key(mid
, &mid_key
, 0);
1875 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1877 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1878 btrfs_mark_buffer_dirty(parent
);
1881 /* update the path */
1883 if (btrfs_header_nritems(left
) > orig_slot
) {
1884 extent_buffer_get(left
);
1885 /* left was locked after cow */
1886 path
->nodes
[level
] = left
;
1887 path
->slots
[level
+ 1] -= 1;
1888 path
->slots
[level
] = orig_slot
;
1890 btrfs_tree_unlock(mid
);
1891 free_extent_buffer(mid
);
1894 orig_slot
-= btrfs_header_nritems(left
);
1895 path
->slots
[level
] = orig_slot
;
1898 /* double check we haven't messed things up */
1900 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
1904 btrfs_tree_unlock(right
);
1905 free_extent_buffer(right
);
1908 if (path
->nodes
[level
] != left
)
1909 btrfs_tree_unlock(left
);
1910 free_extent_buffer(left
);
1915 /* Node balancing for insertion. Here we only split or push nodes around
1916 * when they are completely full. This is also done top down, so we
1917 * have to be pessimistic.
1919 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
1920 struct btrfs_root
*root
,
1921 struct btrfs_path
*path
, int level
)
1923 struct extent_buffer
*right
= NULL
;
1924 struct extent_buffer
*mid
;
1925 struct extent_buffer
*left
= NULL
;
1926 struct extent_buffer
*parent
= NULL
;
1930 int orig_slot
= path
->slots
[level
];
1935 mid
= path
->nodes
[level
];
1936 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1938 if (level
< BTRFS_MAX_LEVEL
- 1) {
1939 parent
= path
->nodes
[level
+ 1];
1940 pslot
= path
->slots
[level
+ 1];
1946 left
= read_node_slot(root
, parent
, pslot
- 1);
1948 /* first, try to make some room in the middle buffer */
1952 btrfs_tree_lock(left
);
1953 btrfs_set_lock_blocking(left
);
1955 left_nr
= btrfs_header_nritems(left
);
1956 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1959 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1964 wret
= push_node_left(trans
, root
,
1971 struct btrfs_disk_key disk_key
;
1972 orig_slot
+= left_nr
;
1973 btrfs_node_key(mid
, &disk_key
, 0);
1974 tree_mod_log_set_node_key(root
->fs_info
, parent
,
1976 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1977 btrfs_mark_buffer_dirty(parent
);
1978 if (btrfs_header_nritems(left
) > orig_slot
) {
1979 path
->nodes
[level
] = left
;
1980 path
->slots
[level
+ 1] -= 1;
1981 path
->slots
[level
] = orig_slot
;
1982 btrfs_tree_unlock(mid
);
1983 free_extent_buffer(mid
);
1986 btrfs_header_nritems(left
);
1987 path
->slots
[level
] = orig_slot
;
1988 btrfs_tree_unlock(left
);
1989 free_extent_buffer(left
);
1993 btrfs_tree_unlock(left
);
1994 free_extent_buffer(left
);
1996 right
= read_node_slot(root
, parent
, pslot
+ 1);
1999 * then try to empty the right most buffer into the middle
2004 btrfs_tree_lock(right
);
2005 btrfs_set_lock_blocking(right
);
2007 right_nr
= btrfs_header_nritems(right
);
2008 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2011 ret
= btrfs_cow_block(trans
, root
, right
,
2017 wret
= balance_node_right(trans
, root
,
2024 struct btrfs_disk_key disk_key
;
2026 btrfs_node_key(right
, &disk_key
, 0);
2027 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2029 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2030 btrfs_mark_buffer_dirty(parent
);
2032 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2033 path
->nodes
[level
] = right
;
2034 path
->slots
[level
+ 1] += 1;
2035 path
->slots
[level
] = orig_slot
-
2036 btrfs_header_nritems(mid
);
2037 btrfs_tree_unlock(mid
);
2038 free_extent_buffer(mid
);
2040 btrfs_tree_unlock(right
);
2041 free_extent_buffer(right
);
2045 btrfs_tree_unlock(right
);
2046 free_extent_buffer(right
);
2052 * readahead one full node of leaves, finding things that are close
2053 * to the block in 'slot', and triggering ra on them.
2055 static void reada_for_search(struct btrfs_root
*root
,
2056 struct btrfs_path
*path
,
2057 int level
, int slot
, u64 objectid
)
2059 struct extent_buffer
*node
;
2060 struct btrfs_disk_key disk_key
;
2066 int direction
= path
->reada
;
2067 struct extent_buffer
*eb
;
2075 if (!path
->nodes
[level
])
2078 node
= path
->nodes
[level
];
2080 search
= btrfs_node_blockptr(node
, slot
);
2081 blocksize
= btrfs_level_size(root
, level
- 1);
2082 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
2084 free_extent_buffer(eb
);
2090 nritems
= btrfs_header_nritems(node
);
2094 if (direction
< 0) {
2098 } else if (direction
> 0) {
2103 if (path
->reada
< 0 && objectid
) {
2104 btrfs_node_key(node
, &disk_key
, nr
);
2105 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2108 search
= btrfs_node_blockptr(node
, nr
);
2109 if ((search
<= target
&& target
- search
<= 65536) ||
2110 (search
> target
&& search
- target
<= 65536)) {
2111 gen
= btrfs_node_ptr_generation(node
, nr
);
2112 readahead_tree_block(root
, search
, blocksize
, gen
);
2116 if ((nread
> 65536 || nscan
> 32))
2122 * returns -EAGAIN if it had to drop the path, or zero if everything was in
2125 static noinline
int reada_for_balance(struct btrfs_root
*root
,
2126 struct btrfs_path
*path
, int level
)
2130 struct extent_buffer
*parent
;
2131 struct extent_buffer
*eb
;
2138 parent
= path
->nodes
[level
+ 1];
2142 nritems
= btrfs_header_nritems(parent
);
2143 slot
= path
->slots
[level
+ 1];
2144 blocksize
= btrfs_level_size(root
, level
);
2147 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2148 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2149 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2151 * if we get -eagain from btrfs_buffer_uptodate, we
2152 * don't want to return eagain here. That will loop
2155 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2157 free_extent_buffer(eb
);
2159 if (slot
+ 1 < nritems
) {
2160 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2161 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2162 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2163 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2165 free_extent_buffer(eb
);
2167 if (block1
|| block2
) {
2170 /* release the whole path */
2171 btrfs_release_path(path
);
2173 /* read the blocks */
2175 readahead_tree_block(root
, block1
, blocksize
, 0);
2177 readahead_tree_block(root
, block2
, blocksize
, 0);
2180 eb
= read_tree_block(root
, block1
, blocksize
, 0);
2181 free_extent_buffer(eb
);
2184 eb
= read_tree_block(root
, block2
, blocksize
, 0);
2185 free_extent_buffer(eb
);
2193 * when we walk down the tree, it is usually safe to unlock the higher layers
2194 * in the tree. The exceptions are when our path goes through slot 0, because
2195 * operations on the tree might require changing key pointers higher up in the
2198 * callers might also have set path->keep_locks, which tells this code to keep
2199 * the lock if the path points to the last slot in the block. This is part of
2200 * walking through the tree, and selecting the next slot in the higher block.
2202 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2203 * if lowest_unlock is 1, level 0 won't be unlocked
2205 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2206 int lowest_unlock
, int min_write_lock_level
,
2207 int *write_lock_level
)
2210 int skip_level
= level
;
2212 struct extent_buffer
*t
;
2214 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2215 if (!path
->nodes
[i
])
2217 if (!path
->locks
[i
])
2219 if (!no_skips
&& path
->slots
[i
] == 0) {
2223 if (!no_skips
&& path
->keep_locks
) {
2226 nritems
= btrfs_header_nritems(t
);
2227 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2232 if (skip_level
< i
&& i
>= lowest_unlock
)
2236 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2237 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2239 if (write_lock_level
&&
2240 i
> min_write_lock_level
&&
2241 i
<= *write_lock_level
) {
2242 *write_lock_level
= i
- 1;
2249 * This releases any locks held in the path starting at level and
2250 * going all the way up to the root.
2252 * btrfs_search_slot will keep the lock held on higher nodes in a few
2253 * corner cases, such as COW of the block at slot zero in the node. This
2254 * ignores those rules, and it should only be called when there are no
2255 * more updates to be done higher up in the tree.
2257 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2261 if (path
->keep_locks
)
2264 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2265 if (!path
->nodes
[i
])
2267 if (!path
->locks
[i
])
2269 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2275 * helper function for btrfs_search_slot. The goal is to find a block
2276 * in cache without setting the path to blocking. If we find the block
2277 * we return zero and the path is unchanged.
2279 * If we can't find the block, we set the path blocking and do some
2280 * reada. -EAGAIN is returned and the search must be repeated.
2283 read_block_for_search(struct btrfs_trans_handle
*trans
,
2284 struct btrfs_root
*root
, struct btrfs_path
*p
,
2285 struct extent_buffer
**eb_ret
, int level
, int slot
,
2286 struct btrfs_key
*key
, u64 time_seq
)
2291 struct extent_buffer
*b
= *eb_ret
;
2292 struct extent_buffer
*tmp
;
2295 blocknr
= btrfs_node_blockptr(b
, slot
);
2296 gen
= btrfs_node_ptr_generation(b
, slot
);
2297 blocksize
= btrfs_level_size(root
, level
- 1);
2299 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2301 /* first we do an atomic uptodate check */
2302 if (btrfs_buffer_uptodate(tmp
, 0, 1) > 0) {
2303 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2305 * we found an up to date block without
2312 /* the pages were up to date, but we failed
2313 * the generation number check. Do a full
2314 * read for the generation number that is correct.
2315 * We must do this without dropping locks so
2316 * we can trust our generation number
2318 free_extent_buffer(tmp
);
2319 btrfs_set_path_blocking(p
);
2321 /* now we're allowed to do a blocking uptodate check */
2322 tmp
= read_tree_block(root
, blocknr
, blocksize
, gen
);
2323 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
, 0) > 0) {
2327 free_extent_buffer(tmp
);
2328 btrfs_release_path(p
);
2334 * reduce lock contention at high levels
2335 * of the btree by dropping locks before
2336 * we read. Don't release the lock on the current
2337 * level because we need to walk this node to figure
2338 * out which blocks to read.
2340 btrfs_unlock_up_safe(p
, level
+ 1);
2341 btrfs_set_path_blocking(p
);
2343 free_extent_buffer(tmp
);
2345 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2347 btrfs_release_path(p
);
2350 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2353 * If the read above didn't mark this buffer up to date,
2354 * it will never end up being up to date. Set ret to EIO now
2355 * and give up so that our caller doesn't loop forever
2358 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2360 free_extent_buffer(tmp
);
2366 * helper function for btrfs_search_slot. This does all of the checks
2367 * for node-level blocks and does any balancing required based on
2370 * If no extra work was required, zero is returned. If we had to
2371 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2375 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2376 struct btrfs_root
*root
, struct btrfs_path
*p
,
2377 struct extent_buffer
*b
, int level
, int ins_len
,
2378 int *write_lock_level
)
2381 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2382 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2385 if (*write_lock_level
< level
+ 1) {
2386 *write_lock_level
= level
+ 1;
2387 btrfs_release_path(p
);
2391 sret
= reada_for_balance(root
, p
, level
);
2395 btrfs_set_path_blocking(p
);
2396 sret
= split_node(trans
, root
, p
, level
);
2397 btrfs_clear_path_blocking(p
, NULL
, 0);
2404 b
= p
->nodes
[level
];
2405 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2406 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2409 if (*write_lock_level
< level
+ 1) {
2410 *write_lock_level
= level
+ 1;
2411 btrfs_release_path(p
);
2415 sret
= reada_for_balance(root
, p
, level
);
2419 btrfs_set_path_blocking(p
);
2420 sret
= balance_level(trans
, root
, p
, level
);
2421 btrfs_clear_path_blocking(p
, NULL
, 0);
2427 b
= p
->nodes
[level
];
2429 btrfs_release_path(p
);
2432 BUG_ON(btrfs_header_nritems(b
) == 1);
2443 * look for key in the tree. path is filled in with nodes along the way
2444 * if key is found, we return zero and you can find the item in the leaf
2445 * level of the path (level 0)
2447 * If the key isn't found, the path points to the slot where it should
2448 * be inserted, and 1 is returned. If there are other errors during the
2449 * search a negative error number is returned.
2451 * if ins_len > 0, nodes and leaves will be split as we walk down the
2452 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2455 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2456 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2459 struct extent_buffer
*b
;
2464 int lowest_unlock
= 1;
2466 /* everything at write_lock_level or lower must be write locked */
2467 int write_lock_level
= 0;
2468 u8 lowest_level
= 0;
2469 int min_write_lock_level
;
2471 lowest_level
= p
->lowest_level
;
2472 WARN_ON(lowest_level
&& ins_len
> 0);
2473 WARN_ON(p
->nodes
[0] != NULL
);
2478 /* when we are removing items, we might have to go up to level
2479 * two as we update tree pointers Make sure we keep write
2480 * for those levels as well
2482 write_lock_level
= 2;
2483 } else if (ins_len
> 0) {
2485 * for inserting items, make sure we have a write lock on
2486 * level 1 so we can update keys
2488 write_lock_level
= 1;
2492 write_lock_level
= -1;
2494 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2495 write_lock_level
= BTRFS_MAX_LEVEL
;
2497 min_write_lock_level
= write_lock_level
;
2501 * we try very hard to do read locks on the root
2503 root_lock
= BTRFS_READ_LOCK
;
2505 if (p
->search_commit_root
) {
2507 * the commit roots are read only
2508 * so we always do read locks
2510 b
= root
->commit_root
;
2511 extent_buffer_get(b
);
2512 level
= btrfs_header_level(b
);
2513 if (!p
->skip_locking
)
2514 btrfs_tree_read_lock(b
);
2516 if (p
->skip_locking
) {
2517 b
= btrfs_root_node(root
);
2518 level
= btrfs_header_level(b
);
2520 /* we don't know the level of the root node
2521 * until we actually have it read locked
2523 b
= btrfs_read_lock_root_node(root
);
2524 level
= btrfs_header_level(b
);
2525 if (level
<= write_lock_level
) {
2526 /* whoops, must trade for write lock */
2527 btrfs_tree_read_unlock(b
);
2528 free_extent_buffer(b
);
2529 b
= btrfs_lock_root_node(root
);
2530 root_lock
= BTRFS_WRITE_LOCK
;
2532 /* the level might have changed, check again */
2533 level
= btrfs_header_level(b
);
2537 p
->nodes
[level
] = b
;
2538 if (!p
->skip_locking
)
2539 p
->locks
[level
] = root_lock
;
2542 level
= btrfs_header_level(b
);
2545 * setup the path here so we can release it under lock
2546 * contention with the cow code
2550 * if we don't really need to cow this block
2551 * then we don't want to set the path blocking,
2552 * so we test it here
2554 if (!should_cow_block(trans
, root
, b
))
2557 btrfs_set_path_blocking(p
);
2560 * must have write locks on this node and the
2563 if (level
> write_lock_level
||
2564 (level
+ 1 > write_lock_level
&&
2565 level
+ 1 < BTRFS_MAX_LEVEL
&&
2566 p
->nodes
[level
+ 1])) {
2567 write_lock_level
= level
+ 1;
2568 btrfs_release_path(p
);
2572 err
= btrfs_cow_block(trans
, root
, b
,
2573 p
->nodes
[level
+ 1],
2574 p
->slots
[level
+ 1], &b
);
2581 BUG_ON(!cow
&& ins_len
);
2583 p
->nodes
[level
] = b
;
2584 btrfs_clear_path_blocking(p
, NULL
, 0);
2587 * we have a lock on b and as long as we aren't changing
2588 * the tree, there is no way to for the items in b to change.
2589 * It is safe to drop the lock on our parent before we
2590 * go through the expensive btree search on b.
2592 * If cow is true, then we might be changing slot zero,
2593 * which may require changing the parent. So, we can't
2594 * drop the lock until after we know which slot we're
2598 btrfs_unlock_up_safe(p
, level
+ 1);
2600 ret
= bin_search(b
, key
, level
, &slot
);
2604 if (ret
&& slot
> 0) {
2608 p
->slots
[level
] = slot
;
2609 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2610 ins_len
, &write_lock_level
);
2617 b
= p
->nodes
[level
];
2618 slot
= p
->slots
[level
];
2621 * slot 0 is special, if we change the key
2622 * we have to update the parent pointer
2623 * which means we must have a write lock
2626 if (slot
== 0 && cow
&&
2627 write_lock_level
< level
+ 1) {
2628 write_lock_level
= level
+ 1;
2629 btrfs_release_path(p
);
2633 unlock_up(p
, level
, lowest_unlock
,
2634 min_write_lock_level
, &write_lock_level
);
2636 if (level
== lowest_level
) {
2642 err
= read_block_for_search(trans
, root
, p
,
2643 &b
, level
, slot
, key
, 0);
2651 if (!p
->skip_locking
) {
2652 level
= btrfs_header_level(b
);
2653 if (level
<= write_lock_level
) {
2654 err
= btrfs_try_tree_write_lock(b
);
2656 btrfs_set_path_blocking(p
);
2658 btrfs_clear_path_blocking(p
, b
,
2661 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2663 err
= btrfs_try_tree_read_lock(b
);
2665 btrfs_set_path_blocking(p
);
2666 btrfs_tree_read_lock(b
);
2667 btrfs_clear_path_blocking(p
, b
,
2670 p
->locks
[level
] = BTRFS_READ_LOCK
;
2672 p
->nodes
[level
] = b
;
2675 p
->slots
[level
] = slot
;
2677 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2678 if (write_lock_level
< 1) {
2679 write_lock_level
= 1;
2680 btrfs_release_path(p
);
2684 btrfs_set_path_blocking(p
);
2685 err
= split_leaf(trans
, root
, key
,
2686 p
, ins_len
, ret
== 0);
2687 btrfs_clear_path_blocking(p
, NULL
, 0);
2695 if (!p
->search_for_split
)
2696 unlock_up(p
, level
, lowest_unlock
,
2697 min_write_lock_level
, &write_lock_level
);
2704 * we don't really know what they plan on doing with the path
2705 * from here on, so for now just mark it as blocking
2707 if (!p
->leave_spinning
)
2708 btrfs_set_path_blocking(p
);
2710 btrfs_release_path(p
);
2715 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2716 * current state of the tree together with the operations recorded in the tree
2717 * modification log to search for the key in a previous version of this tree, as
2718 * denoted by the time_seq parameter.
2720 * Naturally, there is no support for insert, delete or cow operations.
2722 * The resulting path and return value will be set up as if we called
2723 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2725 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2726 struct btrfs_path
*p
, u64 time_seq
)
2728 struct extent_buffer
*b
;
2733 int lowest_unlock
= 1;
2734 u8 lowest_level
= 0;
2736 lowest_level
= p
->lowest_level
;
2737 WARN_ON(p
->nodes
[0] != NULL
);
2739 if (p
->search_commit_root
) {
2741 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2745 b
= get_old_root(root
, time_seq
);
2746 level
= btrfs_header_level(b
);
2747 p
->locks
[level
] = BTRFS_READ_LOCK
;
2750 level
= btrfs_header_level(b
);
2751 p
->nodes
[level
] = b
;
2752 btrfs_clear_path_blocking(p
, NULL
, 0);
2755 * we have a lock on b and as long as we aren't changing
2756 * the tree, there is no way to for the items in b to change.
2757 * It is safe to drop the lock on our parent before we
2758 * go through the expensive btree search on b.
2760 btrfs_unlock_up_safe(p
, level
+ 1);
2762 ret
= bin_search(b
, key
, level
, &slot
);
2766 if (ret
&& slot
> 0) {
2770 p
->slots
[level
] = slot
;
2771 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2773 if (level
== lowest_level
) {
2779 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
2780 slot
, key
, time_seq
);
2788 level
= btrfs_header_level(b
);
2789 err
= btrfs_try_tree_read_lock(b
);
2791 btrfs_set_path_blocking(p
);
2792 btrfs_tree_read_lock(b
);
2793 btrfs_clear_path_blocking(p
, b
,
2796 p
->locks
[level
] = BTRFS_READ_LOCK
;
2797 p
->nodes
[level
] = b
;
2798 b
= tree_mod_log_rewind(root
->fs_info
, b
, time_seq
);
2799 if (b
!= p
->nodes
[level
]) {
2800 btrfs_tree_unlock_rw(p
->nodes
[level
],
2802 p
->locks
[level
] = 0;
2803 p
->nodes
[level
] = b
;
2806 p
->slots
[level
] = slot
;
2807 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2813 if (!p
->leave_spinning
)
2814 btrfs_set_path_blocking(p
);
2816 btrfs_release_path(p
);
2822 * helper to use instead of search slot if no exact match is needed but
2823 * instead the next or previous item should be returned.
2824 * When find_higher is true, the next higher item is returned, the next lower
2826 * When return_any and find_higher are both true, and no higher item is found,
2827 * return the next lower instead.
2828 * When return_any is true and find_higher is false, and no lower item is found,
2829 * return the next higher instead.
2830 * It returns 0 if any item is found, 1 if none is found (tree empty), and
2833 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
2834 struct btrfs_key
*key
, struct btrfs_path
*p
,
2835 int find_higher
, int return_any
)
2838 struct extent_buffer
*leaf
;
2841 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2845 * a return value of 1 means the path is at the position where the
2846 * item should be inserted. Normally this is the next bigger item,
2847 * but in case the previous item is the last in a leaf, path points
2848 * to the first free slot in the previous leaf, i.e. at an invalid
2854 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
2855 ret
= btrfs_next_leaf(root
, p
);
2861 * no higher item found, return the next
2866 btrfs_release_path(p
);
2870 if (p
->slots
[0] == 0) {
2871 ret
= btrfs_prev_leaf(root
, p
);
2875 p
->slots
[0] = btrfs_header_nritems(leaf
) - 1;
2881 * no lower item found, return the next
2886 btrfs_release_path(p
);
2896 * adjust the pointers going up the tree, starting at level
2897 * making sure the right key of each node is points to 'key'.
2898 * This is used after shifting pointers to the left, so it stops
2899 * fixing up pointers when a given leaf/node is not in slot 0 of the
2903 static void fixup_low_keys(struct btrfs_trans_handle
*trans
,
2904 struct btrfs_root
*root
, struct btrfs_path
*path
,
2905 struct btrfs_disk_key
*key
, int level
)
2908 struct extent_buffer
*t
;
2910 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2911 int tslot
= path
->slots
[i
];
2912 if (!path
->nodes
[i
])
2915 tree_mod_log_set_node_key(root
->fs_info
, t
, tslot
, 1);
2916 btrfs_set_node_key(t
, key
, tslot
);
2917 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
2926 * This function isn't completely safe. It's the caller's responsibility
2927 * that the new key won't break the order
2929 void btrfs_set_item_key_safe(struct btrfs_trans_handle
*trans
,
2930 struct btrfs_root
*root
, struct btrfs_path
*path
,
2931 struct btrfs_key
*new_key
)
2933 struct btrfs_disk_key disk_key
;
2934 struct extent_buffer
*eb
;
2937 eb
= path
->nodes
[0];
2938 slot
= path
->slots
[0];
2940 btrfs_item_key(eb
, &disk_key
, slot
- 1);
2941 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
2943 if (slot
< btrfs_header_nritems(eb
) - 1) {
2944 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
2945 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
2948 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2949 btrfs_set_item_key(eb
, &disk_key
, slot
);
2950 btrfs_mark_buffer_dirty(eb
);
2952 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2956 * try to push data from one node into the next node left in the
2959 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
2960 * error, and > 0 if there was no room in the left hand block.
2962 static int push_node_left(struct btrfs_trans_handle
*trans
,
2963 struct btrfs_root
*root
, struct extent_buffer
*dst
,
2964 struct extent_buffer
*src
, int empty
)
2971 src_nritems
= btrfs_header_nritems(src
);
2972 dst_nritems
= btrfs_header_nritems(dst
);
2973 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
2974 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
2975 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
2977 if (!empty
&& src_nritems
<= 8)
2980 if (push_items
<= 0)
2984 push_items
= min(src_nritems
, push_items
);
2985 if (push_items
< src_nritems
) {
2986 /* leave at least 8 pointers in the node if
2987 * we aren't going to empty it
2989 if (src_nritems
- push_items
< 8) {
2990 if (push_items
<= 8)
2996 push_items
= min(src_nritems
- 8, push_items
);
2998 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3000 copy_extent_buffer(dst
, src
,
3001 btrfs_node_key_ptr_offset(dst_nritems
),
3002 btrfs_node_key_ptr_offset(0),
3003 push_items
* sizeof(struct btrfs_key_ptr
));
3005 if (push_items
< src_nritems
) {
3007 * don't call tree_mod_log_eb_move here, key removal was already
3008 * fully logged by tree_mod_log_eb_copy above.
3010 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3011 btrfs_node_key_ptr_offset(push_items
),
3012 (src_nritems
- push_items
) *
3013 sizeof(struct btrfs_key_ptr
));
3015 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3016 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3017 btrfs_mark_buffer_dirty(src
);
3018 btrfs_mark_buffer_dirty(dst
);
3024 * try to push data from one node into the next node right in the
3027 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3028 * error, and > 0 if there was no room in the right hand block.
3030 * this will only push up to 1/2 the contents of the left node over
3032 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3033 struct btrfs_root
*root
,
3034 struct extent_buffer
*dst
,
3035 struct extent_buffer
*src
)
3043 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3044 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3046 src_nritems
= btrfs_header_nritems(src
);
3047 dst_nritems
= btrfs_header_nritems(dst
);
3048 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3049 if (push_items
<= 0)
3052 if (src_nritems
< 4)
3055 max_push
= src_nritems
/ 2 + 1;
3056 /* don't try to empty the node */
3057 if (max_push
>= src_nritems
)
3060 if (max_push
< push_items
)
3061 push_items
= max_push
;
3063 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3064 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3065 btrfs_node_key_ptr_offset(0),
3067 sizeof(struct btrfs_key_ptr
));
3069 tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3070 src_nritems
- push_items
, push_items
, 1);
3071 copy_extent_buffer(dst
, src
,
3072 btrfs_node_key_ptr_offset(0),
3073 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3074 push_items
* sizeof(struct btrfs_key_ptr
));
3076 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3077 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3079 btrfs_mark_buffer_dirty(src
);
3080 btrfs_mark_buffer_dirty(dst
);
3086 * helper function to insert a new root level in the tree.
3087 * A new node is allocated, and a single item is inserted to
3088 * point to the existing root
3090 * returns zero on success or < 0 on failure.
3092 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3093 struct btrfs_root
*root
,
3094 struct btrfs_path
*path
, int level
)
3097 struct extent_buffer
*lower
;
3098 struct extent_buffer
*c
;
3099 struct extent_buffer
*old
;
3100 struct btrfs_disk_key lower_key
;
3102 BUG_ON(path
->nodes
[level
]);
3103 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3105 lower
= path
->nodes
[level
-1];
3107 btrfs_item_key(lower
, &lower_key
, 0);
3109 btrfs_node_key(lower
, &lower_key
, 0);
3111 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3112 root
->root_key
.objectid
, &lower_key
,
3113 level
, root
->node
->start
, 0);
3117 root_add_used(root
, root
->nodesize
);
3119 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3120 btrfs_set_header_nritems(c
, 1);
3121 btrfs_set_header_level(c
, level
);
3122 btrfs_set_header_bytenr(c
, c
->start
);
3123 btrfs_set_header_generation(c
, trans
->transid
);
3124 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3125 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3127 write_extent_buffer(c
, root
->fs_info
->fsid
,
3128 (unsigned long)btrfs_header_fsid(c
),
3131 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3132 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
3135 btrfs_set_node_key(c
, &lower_key
, 0);
3136 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3137 lower_gen
= btrfs_header_generation(lower
);
3138 WARN_ON(lower_gen
!= trans
->transid
);
3140 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3142 btrfs_mark_buffer_dirty(c
);
3145 tree_mod_log_set_root_pointer(root
, c
);
3146 rcu_assign_pointer(root
->node
, c
);
3148 /* the super has an extra ref to root->node */
3149 free_extent_buffer(old
);
3151 add_root_to_dirty_list(root
);
3152 extent_buffer_get(c
);
3153 path
->nodes
[level
] = c
;
3154 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
3155 path
->slots
[level
] = 0;
3160 * worker function to insert a single pointer in a node.
3161 * the node should have enough room for the pointer already
3163 * slot and level indicate where you want the key to go, and
3164 * blocknr is the block the key points to.
3166 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3167 struct btrfs_root
*root
, struct btrfs_path
*path
,
3168 struct btrfs_disk_key
*key
, u64 bytenr
,
3169 int slot
, int level
)
3171 struct extent_buffer
*lower
;
3175 BUG_ON(!path
->nodes
[level
]);
3176 btrfs_assert_tree_locked(path
->nodes
[level
]);
3177 lower
= path
->nodes
[level
];
3178 nritems
= btrfs_header_nritems(lower
);
3179 BUG_ON(slot
> nritems
);
3180 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3181 if (slot
!= nritems
) {
3183 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3184 slot
, nritems
- slot
);
3185 memmove_extent_buffer(lower
,
3186 btrfs_node_key_ptr_offset(slot
+ 1),
3187 btrfs_node_key_ptr_offset(slot
),
3188 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3191 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3195 btrfs_set_node_key(lower
, key
, slot
);
3196 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3197 WARN_ON(trans
->transid
== 0);
3198 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3199 btrfs_set_header_nritems(lower
, nritems
+ 1);
3200 btrfs_mark_buffer_dirty(lower
);
3204 * split the node at the specified level in path in two.
3205 * The path is corrected to point to the appropriate node after the split
3207 * Before splitting this tries to make some room in the node by pushing
3208 * left and right, if either one works, it returns right away.
3210 * returns 0 on success and < 0 on failure
3212 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3213 struct btrfs_root
*root
,
3214 struct btrfs_path
*path
, int level
)
3216 struct extent_buffer
*c
;
3217 struct extent_buffer
*split
;
3218 struct btrfs_disk_key disk_key
;
3222 int tree_mod_log_removal
= 1;
3224 c
= path
->nodes
[level
];
3225 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3226 if (c
== root
->node
) {
3227 /* trying to split the root, lets make a new one */
3228 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3230 * removal of root nodes has been logged by
3231 * tree_mod_log_set_root_pointer due to locking
3233 tree_mod_log_removal
= 0;
3237 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3238 c
= path
->nodes
[level
];
3239 if (!ret
&& btrfs_header_nritems(c
) <
3240 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3246 c_nritems
= btrfs_header_nritems(c
);
3247 mid
= (c_nritems
+ 1) / 2;
3248 btrfs_node_key(c
, &disk_key
, mid
);
3250 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3251 root
->root_key
.objectid
,
3252 &disk_key
, level
, c
->start
, 0);
3254 return PTR_ERR(split
);
3256 root_add_used(root
, root
->nodesize
);
3258 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3259 btrfs_set_header_level(split
, btrfs_header_level(c
));
3260 btrfs_set_header_bytenr(split
, split
->start
);
3261 btrfs_set_header_generation(split
, trans
->transid
);
3262 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3263 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3264 write_extent_buffer(split
, root
->fs_info
->fsid
,
3265 (unsigned long)btrfs_header_fsid(split
),
3267 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3268 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
3271 tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0, mid
, c_nritems
- mid
,
3272 tree_mod_log_removal
);
3273 copy_extent_buffer(split
, c
,
3274 btrfs_node_key_ptr_offset(0),
3275 btrfs_node_key_ptr_offset(mid
),
3276 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3277 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3278 btrfs_set_header_nritems(c
, mid
);
3281 btrfs_mark_buffer_dirty(c
);
3282 btrfs_mark_buffer_dirty(split
);
3284 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3285 path
->slots
[level
+ 1] + 1, level
+ 1);
3287 if (path
->slots
[level
] >= mid
) {
3288 path
->slots
[level
] -= mid
;
3289 btrfs_tree_unlock(c
);
3290 free_extent_buffer(c
);
3291 path
->nodes
[level
] = split
;
3292 path
->slots
[level
+ 1] += 1;
3294 btrfs_tree_unlock(split
);
3295 free_extent_buffer(split
);
3301 * how many bytes are required to store the items in a leaf. start
3302 * and nr indicate which items in the leaf to check. This totals up the
3303 * space used both by the item structs and the item data
3305 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3307 struct btrfs_item
*start_item
;
3308 struct btrfs_item
*end_item
;
3309 struct btrfs_map_token token
;
3311 int nritems
= btrfs_header_nritems(l
);
3312 int end
= min(nritems
, start
+ nr
) - 1;
3316 btrfs_init_map_token(&token
);
3317 start_item
= btrfs_item_nr(l
, start
);
3318 end_item
= btrfs_item_nr(l
, end
);
3319 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3320 btrfs_token_item_size(l
, start_item
, &token
);
3321 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3322 data_len
+= sizeof(struct btrfs_item
) * nr
;
3323 WARN_ON(data_len
< 0);
3328 * The space between the end of the leaf items and
3329 * the start of the leaf data. IOW, how much room
3330 * the leaf has left for both items and data
3332 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3333 struct extent_buffer
*leaf
)
3335 int nritems
= btrfs_header_nritems(leaf
);
3337 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3339 printk(KERN_CRIT
"leaf free space ret %d, leaf data size %lu, "
3340 "used %d nritems %d\n",
3341 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3342 leaf_space_used(leaf
, 0, nritems
), nritems
);
3348 * min slot controls the lowest index we're willing to push to the
3349 * right. We'll push up to and including min_slot, but no lower
3351 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3352 struct btrfs_root
*root
,
3353 struct btrfs_path
*path
,
3354 int data_size
, int empty
,
3355 struct extent_buffer
*right
,
3356 int free_space
, u32 left_nritems
,
3359 struct extent_buffer
*left
= path
->nodes
[0];
3360 struct extent_buffer
*upper
= path
->nodes
[1];
3361 struct btrfs_map_token token
;
3362 struct btrfs_disk_key disk_key
;
3367 struct btrfs_item
*item
;
3373 btrfs_init_map_token(&token
);
3378 nr
= max_t(u32
, 1, min_slot
);
3380 if (path
->slots
[0] >= left_nritems
)
3381 push_space
+= data_size
;
3383 slot
= path
->slots
[1];
3384 i
= left_nritems
- 1;
3386 item
= btrfs_item_nr(left
, i
);
3388 if (!empty
&& push_items
> 0) {
3389 if (path
->slots
[0] > i
)
3391 if (path
->slots
[0] == i
) {
3392 int space
= btrfs_leaf_free_space(root
, left
);
3393 if (space
+ push_space
* 2 > free_space
)
3398 if (path
->slots
[0] == i
)
3399 push_space
+= data_size
;
3401 this_item_size
= btrfs_item_size(left
, item
);
3402 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3406 push_space
+= this_item_size
+ sizeof(*item
);
3412 if (push_items
== 0)
3415 WARN_ON(!empty
&& push_items
== left_nritems
);
3417 /* push left to right */
3418 right_nritems
= btrfs_header_nritems(right
);
3420 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3421 push_space
-= leaf_data_end(root
, left
);
3423 /* make room in the right data area */
3424 data_end
= leaf_data_end(root
, right
);
3425 memmove_extent_buffer(right
,
3426 btrfs_leaf_data(right
) + data_end
- push_space
,
3427 btrfs_leaf_data(right
) + data_end
,
3428 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3430 /* copy from the left data area */
3431 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3432 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3433 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3436 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3437 btrfs_item_nr_offset(0),
3438 right_nritems
* sizeof(struct btrfs_item
));
3440 /* copy the items from left to right */
3441 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3442 btrfs_item_nr_offset(left_nritems
- push_items
),
3443 push_items
* sizeof(struct btrfs_item
));
3445 /* update the item pointers */
3446 right_nritems
+= push_items
;
3447 btrfs_set_header_nritems(right
, right_nritems
);
3448 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3449 for (i
= 0; i
< right_nritems
; i
++) {
3450 item
= btrfs_item_nr(right
, i
);
3451 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3452 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3455 left_nritems
-= push_items
;
3456 btrfs_set_header_nritems(left
, left_nritems
);
3459 btrfs_mark_buffer_dirty(left
);
3461 clean_tree_block(trans
, root
, left
);
3463 btrfs_mark_buffer_dirty(right
);
3465 btrfs_item_key(right
, &disk_key
, 0);
3466 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3467 btrfs_mark_buffer_dirty(upper
);
3469 /* then fixup the leaf pointer in the path */
3470 if (path
->slots
[0] >= left_nritems
) {
3471 path
->slots
[0] -= left_nritems
;
3472 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3473 clean_tree_block(trans
, root
, path
->nodes
[0]);
3474 btrfs_tree_unlock(path
->nodes
[0]);
3475 free_extent_buffer(path
->nodes
[0]);
3476 path
->nodes
[0] = right
;
3477 path
->slots
[1] += 1;
3479 btrfs_tree_unlock(right
);
3480 free_extent_buffer(right
);
3485 btrfs_tree_unlock(right
);
3486 free_extent_buffer(right
);
3491 * push some data in the path leaf to the right, trying to free up at
3492 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3494 * returns 1 if the push failed because the other node didn't have enough
3495 * room, 0 if everything worked out and < 0 if there were major errors.
3497 * this will push starting from min_slot to the end of the leaf. It won't
3498 * push any slot lower than min_slot
3500 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3501 *root
, struct btrfs_path
*path
,
3502 int min_data_size
, int data_size
,
3503 int empty
, u32 min_slot
)
3505 struct extent_buffer
*left
= path
->nodes
[0];
3506 struct extent_buffer
*right
;
3507 struct extent_buffer
*upper
;
3513 if (!path
->nodes
[1])
3516 slot
= path
->slots
[1];
3517 upper
= path
->nodes
[1];
3518 if (slot
>= btrfs_header_nritems(upper
) - 1)
3521 btrfs_assert_tree_locked(path
->nodes
[1]);
3523 right
= read_node_slot(root
, upper
, slot
+ 1);
3527 btrfs_tree_lock(right
);
3528 btrfs_set_lock_blocking(right
);
3530 free_space
= btrfs_leaf_free_space(root
, right
);
3531 if (free_space
< data_size
)
3534 /* cow and double check */
3535 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3540 free_space
= btrfs_leaf_free_space(root
, right
);
3541 if (free_space
< data_size
)
3544 left_nritems
= btrfs_header_nritems(left
);
3545 if (left_nritems
== 0)
3548 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3549 right
, free_space
, left_nritems
, min_slot
);
3551 btrfs_tree_unlock(right
);
3552 free_extent_buffer(right
);
3557 * push some data in the path leaf to the left, trying to free up at
3558 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3560 * max_slot can put a limit on how far into the leaf we'll push items. The
3561 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3564 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3565 struct btrfs_root
*root
,
3566 struct btrfs_path
*path
, int data_size
,
3567 int empty
, struct extent_buffer
*left
,
3568 int free_space
, u32 right_nritems
,
3571 struct btrfs_disk_key disk_key
;
3572 struct extent_buffer
*right
= path
->nodes
[0];
3576 struct btrfs_item
*item
;
3577 u32 old_left_nritems
;
3581 u32 old_left_item_size
;
3582 struct btrfs_map_token token
;
3584 btrfs_init_map_token(&token
);
3587 nr
= min(right_nritems
, max_slot
);
3589 nr
= min(right_nritems
- 1, max_slot
);
3591 for (i
= 0; i
< nr
; i
++) {
3592 item
= btrfs_item_nr(right
, i
);
3594 if (!empty
&& push_items
> 0) {
3595 if (path
->slots
[0] < i
)
3597 if (path
->slots
[0] == i
) {
3598 int space
= btrfs_leaf_free_space(root
, right
);
3599 if (space
+ push_space
* 2 > free_space
)
3604 if (path
->slots
[0] == i
)
3605 push_space
+= data_size
;
3607 this_item_size
= btrfs_item_size(right
, item
);
3608 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3612 push_space
+= this_item_size
+ sizeof(*item
);
3615 if (push_items
== 0) {
3619 if (!empty
&& push_items
== btrfs_header_nritems(right
))
3622 /* push data from right to left */
3623 copy_extent_buffer(left
, right
,
3624 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3625 btrfs_item_nr_offset(0),
3626 push_items
* sizeof(struct btrfs_item
));
3628 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3629 btrfs_item_offset_nr(right
, push_items
- 1);
3631 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3632 leaf_data_end(root
, left
) - push_space
,
3633 btrfs_leaf_data(right
) +
3634 btrfs_item_offset_nr(right
, push_items
- 1),
3636 old_left_nritems
= btrfs_header_nritems(left
);
3637 BUG_ON(old_left_nritems
<= 0);
3639 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3640 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3643 item
= btrfs_item_nr(left
, i
);
3645 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3646 btrfs_set_token_item_offset(left
, item
,
3647 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3650 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3652 /* fixup right node */
3653 if (push_items
> right_nritems
)
3654 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3657 if (push_items
< right_nritems
) {
3658 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3659 leaf_data_end(root
, right
);
3660 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3661 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3662 btrfs_leaf_data(right
) +
3663 leaf_data_end(root
, right
), push_space
);
3665 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3666 btrfs_item_nr_offset(push_items
),
3667 (btrfs_header_nritems(right
) - push_items
) *
3668 sizeof(struct btrfs_item
));
3670 right_nritems
-= push_items
;
3671 btrfs_set_header_nritems(right
, right_nritems
);
3672 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3673 for (i
= 0; i
< right_nritems
; i
++) {
3674 item
= btrfs_item_nr(right
, i
);
3676 push_space
= push_space
- btrfs_token_item_size(right
,
3678 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3681 btrfs_mark_buffer_dirty(left
);
3683 btrfs_mark_buffer_dirty(right
);
3685 clean_tree_block(trans
, root
, right
);
3687 btrfs_item_key(right
, &disk_key
, 0);
3688 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3690 /* then fixup the leaf pointer in the path */
3691 if (path
->slots
[0] < push_items
) {
3692 path
->slots
[0] += old_left_nritems
;
3693 btrfs_tree_unlock(path
->nodes
[0]);
3694 free_extent_buffer(path
->nodes
[0]);
3695 path
->nodes
[0] = left
;
3696 path
->slots
[1] -= 1;
3698 btrfs_tree_unlock(left
);
3699 free_extent_buffer(left
);
3700 path
->slots
[0] -= push_items
;
3702 BUG_ON(path
->slots
[0] < 0);
3705 btrfs_tree_unlock(left
);
3706 free_extent_buffer(left
);
3711 * push some data in the path leaf to the left, trying to free up at
3712 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3714 * max_slot can put a limit on how far into the leaf we'll push items. The
3715 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3718 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3719 *root
, struct btrfs_path
*path
, int min_data_size
,
3720 int data_size
, int empty
, u32 max_slot
)
3722 struct extent_buffer
*right
= path
->nodes
[0];
3723 struct extent_buffer
*left
;
3729 slot
= path
->slots
[1];
3732 if (!path
->nodes
[1])
3735 right_nritems
= btrfs_header_nritems(right
);
3736 if (right_nritems
== 0)
3739 btrfs_assert_tree_locked(path
->nodes
[1]);
3741 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
3745 btrfs_tree_lock(left
);
3746 btrfs_set_lock_blocking(left
);
3748 free_space
= btrfs_leaf_free_space(root
, left
);
3749 if (free_space
< data_size
) {
3754 /* cow and double check */
3755 ret
= btrfs_cow_block(trans
, root
, left
,
3756 path
->nodes
[1], slot
- 1, &left
);
3758 /* we hit -ENOSPC, but it isn't fatal here */
3764 free_space
= btrfs_leaf_free_space(root
, left
);
3765 if (free_space
< data_size
) {
3770 return __push_leaf_left(trans
, root
, path
, min_data_size
,
3771 empty
, left
, free_space
, right_nritems
,
3774 btrfs_tree_unlock(left
);
3775 free_extent_buffer(left
);
3780 * split the path's leaf in two, making sure there is at least data_size
3781 * available for the resulting leaf level of the path.
3783 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
3784 struct btrfs_root
*root
,
3785 struct btrfs_path
*path
,
3786 struct extent_buffer
*l
,
3787 struct extent_buffer
*right
,
3788 int slot
, int mid
, int nritems
)
3793 struct btrfs_disk_key disk_key
;
3794 struct btrfs_map_token token
;
3796 btrfs_init_map_token(&token
);
3798 nritems
= nritems
- mid
;
3799 btrfs_set_header_nritems(right
, nritems
);
3800 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
3802 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
3803 btrfs_item_nr_offset(mid
),
3804 nritems
* sizeof(struct btrfs_item
));
3806 copy_extent_buffer(right
, l
,
3807 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
3808 data_copy_size
, btrfs_leaf_data(l
) +
3809 leaf_data_end(root
, l
), data_copy_size
);
3811 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
3812 btrfs_item_end_nr(l
, mid
);
3814 for (i
= 0; i
< nritems
; i
++) {
3815 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
3818 ioff
= btrfs_token_item_offset(right
, item
, &token
);
3819 btrfs_set_token_item_offset(right
, item
,
3820 ioff
+ rt_data_off
, &token
);
3823 btrfs_set_header_nritems(l
, mid
);
3824 btrfs_item_key(right
, &disk_key
, 0);
3825 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
3826 path
->slots
[1] + 1, 1);
3828 btrfs_mark_buffer_dirty(right
);
3829 btrfs_mark_buffer_dirty(l
);
3830 BUG_ON(path
->slots
[0] != slot
);
3833 btrfs_tree_unlock(path
->nodes
[0]);
3834 free_extent_buffer(path
->nodes
[0]);
3835 path
->nodes
[0] = right
;
3836 path
->slots
[0] -= mid
;
3837 path
->slots
[1] += 1;
3839 btrfs_tree_unlock(right
);
3840 free_extent_buffer(right
);
3843 BUG_ON(path
->slots
[0] < 0);
3847 * double splits happen when we need to insert a big item in the middle
3848 * of a leaf. A double split can leave us with 3 mostly empty leaves:
3849 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
3852 * We avoid this by trying to push the items on either side of our target
3853 * into the adjacent leaves. If all goes well we can avoid the double split
3856 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
3857 struct btrfs_root
*root
,
3858 struct btrfs_path
*path
,
3866 slot
= path
->slots
[0];
3869 * try to push all the items after our slot into the
3872 ret
= push_leaf_right(trans
, root
, path
, 1, data_size
, 0, slot
);
3879 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3881 * our goal is to get our slot at the start or end of a leaf. If
3882 * we've done so we're done
3884 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
3887 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
3890 /* try to push all the items before our slot into the next leaf */
3891 slot
= path
->slots
[0];
3892 ret
= push_leaf_left(trans
, root
, path
, 1, data_size
, 0, slot
);
3905 * split the path's leaf in two, making sure there is at least data_size
3906 * available for the resulting leaf level of the path.
3908 * returns 0 if all went well and < 0 on failure.
3910 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
3911 struct btrfs_root
*root
,
3912 struct btrfs_key
*ins_key
,
3913 struct btrfs_path
*path
, int data_size
,
3916 struct btrfs_disk_key disk_key
;
3917 struct extent_buffer
*l
;
3921 struct extent_buffer
*right
;
3925 int num_doubles
= 0;
3926 int tried_avoid_double
= 0;
3929 slot
= path
->slots
[0];
3930 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
3931 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
3934 /* first try to make some room by pushing left and right */
3936 wret
= push_leaf_right(trans
, root
, path
, data_size
,
3941 wret
= push_leaf_left(trans
, root
, path
, data_size
,
3942 data_size
, 0, (u32
)-1);
3948 /* did the pushes work? */
3949 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
3953 if (!path
->nodes
[1]) {
3954 ret
= insert_new_root(trans
, root
, path
, 1);
3961 slot
= path
->slots
[0];
3962 nritems
= btrfs_header_nritems(l
);
3963 mid
= (nritems
+ 1) / 2;
3967 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
3968 BTRFS_LEAF_DATA_SIZE(root
)) {
3969 if (slot
>= nritems
) {
3973 if (mid
!= nritems
&&
3974 leaf_space_used(l
, mid
, nritems
- mid
) +
3975 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3976 if (data_size
&& !tried_avoid_double
)
3977 goto push_for_double
;
3983 if (leaf_space_used(l
, 0, mid
) + data_size
>
3984 BTRFS_LEAF_DATA_SIZE(root
)) {
3985 if (!extend
&& data_size
&& slot
== 0) {
3987 } else if ((extend
|| !data_size
) && slot
== 0) {
3991 if (mid
!= nritems
&&
3992 leaf_space_used(l
, mid
, nritems
- mid
) +
3993 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
3994 if (data_size
&& !tried_avoid_double
)
3995 goto push_for_double
;
4003 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4005 btrfs_item_key(l
, &disk_key
, mid
);
4007 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
4008 root
->root_key
.objectid
,
4009 &disk_key
, 0, l
->start
, 0);
4011 return PTR_ERR(right
);
4013 root_add_used(root
, root
->leafsize
);
4015 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4016 btrfs_set_header_bytenr(right
, right
->start
);
4017 btrfs_set_header_generation(right
, trans
->transid
);
4018 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4019 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4020 btrfs_set_header_level(right
, 0);
4021 write_extent_buffer(right
, root
->fs_info
->fsid
,
4022 (unsigned long)btrfs_header_fsid(right
),
4025 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
4026 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
4031 btrfs_set_header_nritems(right
, 0);
4032 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4033 path
->slots
[1] + 1, 1);
4034 btrfs_tree_unlock(path
->nodes
[0]);
4035 free_extent_buffer(path
->nodes
[0]);
4036 path
->nodes
[0] = right
;
4038 path
->slots
[1] += 1;
4040 btrfs_set_header_nritems(right
, 0);
4041 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4043 btrfs_tree_unlock(path
->nodes
[0]);
4044 free_extent_buffer(path
->nodes
[0]);
4045 path
->nodes
[0] = right
;
4047 if (path
->slots
[1] == 0)
4048 fixup_low_keys(trans
, root
, path
,
4051 btrfs_mark_buffer_dirty(right
);
4055 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4058 BUG_ON(num_doubles
!= 0);
4066 push_for_double_split(trans
, root
, path
, data_size
);
4067 tried_avoid_double
= 1;
4068 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4073 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4074 struct btrfs_root
*root
,
4075 struct btrfs_path
*path
, int ins_len
)
4077 struct btrfs_key key
;
4078 struct extent_buffer
*leaf
;
4079 struct btrfs_file_extent_item
*fi
;
4084 leaf
= path
->nodes
[0];
4085 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4087 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4088 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4090 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4093 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4094 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4095 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4096 struct btrfs_file_extent_item
);
4097 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4099 btrfs_release_path(path
);
4101 path
->keep_locks
= 1;
4102 path
->search_for_split
= 1;
4103 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4104 path
->search_for_split
= 0;
4109 leaf
= path
->nodes
[0];
4110 /* if our item isn't there or got smaller, return now */
4111 if (ret
> 0 || item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4114 /* the leaf has changed, it now has room. return now */
4115 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4118 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4119 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4120 struct btrfs_file_extent_item
);
4121 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4125 btrfs_set_path_blocking(path
);
4126 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4130 path
->keep_locks
= 0;
4131 btrfs_unlock_up_safe(path
, 1);
4134 path
->keep_locks
= 0;
4138 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4139 struct btrfs_root
*root
,
4140 struct btrfs_path
*path
,
4141 struct btrfs_key
*new_key
,
4142 unsigned long split_offset
)
4144 struct extent_buffer
*leaf
;
4145 struct btrfs_item
*item
;
4146 struct btrfs_item
*new_item
;
4152 struct btrfs_disk_key disk_key
;
4154 leaf
= path
->nodes
[0];
4155 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4157 btrfs_set_path_blocking(path
);
4159 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
4160 orig_offset
= btrfs_item_offset(leaf
, item
);
4161 item_size
= btrfs_item_size(leaf
, item
);
4163 buf
= kmalloc(item_size
, GFP_NOFS
);
4167 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4168 path
->slots
[0]), item_size
);
4170 slot
= path
->slots
[0] + 1;
4171 nritems
= btrfs_header_nritems(leaf
);
4172 if (slot
!= nritems
) {
4173 /* shift the items */
4174 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4175 btrfs_item_nr_offset(slot
),
4176 (nritems
- slot
) * sizeof(struct btrfs_item
));
4179 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4180 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4182 new_item
= btrfs_item_nr(leaf
, slot
);
4184 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4185 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4187 btrfs_set_item_offset(leaf
, item
,
4188 orig_offset
+ item_size
- split_offset
);
4189 btrfs_set_item_size(leaf
, item
, split_offset
);
4191 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4193 /* write the data for the start of the original item */
4194 write_extent_buffer(leaf
, buf
,
4195 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4198 /* write the data for the new item */
4199 write_extent_buffer(leaf
, buf
+ split_offset
,
4200 btrfs_item_ptr_offset(leaf
, slot
),
4201 item_size
- split_offset
);
4202 btrfs_mark_buffer_dirty(leaf
);
4204 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4210 * This function splits a single item into two items,
4211 * giving 'new_key' to the new item and splitting the
4212 * old one at split_offset (from the start of the item).
4214 * The path may be released by this operation. After
4215 * the split, the path is pointing to the old item. The
4216 * new item is going to be in the same node as the old one.
4218 * Note, the item being split must be smaller enough to live alone on
4219 * a tree block with room for one extra struct btrfs_item
4221 * This allows us to split the item in place, keeping a lock on the
4222 * leaf the entire time.
4224 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4225 struct btrfs_root
*root
,
4226 struct btrfs_path
*path
,
4227 struct btrfs_key
*new_key
,
4228 unsigned long split_offset
)
4231 ret
= setup_leaf_for_split(trans
, root
, path
,
4232 sizeof(struct btrfs_item
));
4236 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4241 * This function duplicate a item, giving 'new_key' to the new item.
4242 * It guarantees both items live in the same tree leaf and the new item
4243 * is contiguous with the original item.
4245 * This allows us to split file extent in place, keeping a lock on the
4246 * leaf the entire time.
4248 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4249 struct btrfs_root
*root
,
4250 struct btrfs_path
*path
,
4251 struct btrfs_key
*new_key
)
4253 struct extent_buffer
*leaf
;
4257 leaf
= path
->nodes
[0];
4258 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4259 ret
= setup_leaf_for_split(trans
, root
, path
,
4260 item_size
+ sizeof(struct btrfs_item
));
4265 setup_items_for_insert(trans
, root
, path
, new_key
, &item_size
,
4266 item_size
, item_size
+
4267 sizeof(struct btrfs_item
), 1);
4268 leaf
= path
->nodes
[0];
4269 memcpy_extent_buffer(leaf
,
4270 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4271 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4277 * make the item pointed to by the path smaller. new_size indicates
4278 * how small to make it, and from_end tells us if we just chop bytes
4279 * off the end of the item or if we shift the item to chop bytes off
4282 void btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
4283 struct btrfs_root
*root
,
4284 struct btrfs_path
*path
,
4285 u32 new_size
, int from_end
)
4288 struct extent_buffer
*leaf
;
4289 struct btrfs_item
*item
;
4291 unsigned int data_end
;
4292 unsigned int old_data_start
;
4293 unsigned int old_size
;
4294 unsigned int size_diff
;
4296 struct btrfs_map_token token
;
4298 btrfs_init_map_token(&token
);
4300 leaf
= path
->nodes
[0];
4301 slot
= path
->slots
[0];
4303 old_size
= btrfs_item_size_nr(leaf
, slot
);
4304 if (old_size
== new_size
)
4307 nritems
= btrfs_header_nritems(leaf
);
4308 data_end
= leaf_data_end(root
, leaf
);
4310 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4312 size_diff
= old_size
- new_size
;
4315 BUG_ON(slot
>= nritems
);
4318 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4320 /* first correct the data pointers */
4321 for (i
= slot
; i
< nritems
; i
++) {
4323 item
= btrfs_item_nr(leaf
, i
);
4325 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4326 btrfs_set_token_item_offset(leaf
, item
,
4327 ioff
+ size_diff
, &token
);
4330 /* shift the data */
4332 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4333 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4334 data_end
, old_data_start
+ new_size
- data_end
);
4336 struct btrfs_disk_key disk_key
;
4339 btrfs_item_key(leaf
, &disk_key
, slot
);
4341 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4343 struct btrfs_file_extent_item
*fi
;
4345 fi
= btrfs_item_ptr(leaf
, slot
,
4346 struct btrfs_file_extent_item
);
4347 fi
= (struct btrfs_file_extent_item
*)(
4348 (unsigned long)fi
- size_diff
);
4350 if (btrfs_file_extent_type(leaf
, fi
) ==
4351 BTRFS_FILE_EXTENT_INLINE
) {
4352 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4353 memmove_extent_buffer(leaf
, ptr
,
4355 offsetof(struct btrfs_file_extent_item
,
4360 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4361 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4362 data_end
, old_data_start
- data_end
);
4364 offset
= btrfs_disk_key_offset(&disk_key
);
4365 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4366 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4368 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4371 item
= btrfs_item_nr(leaf
, slot
);
4372 btrfs_set_item_size(leaf
, item
, new_size
);
4373 btrfs_mark_buffer_dirty(leaf
);
4375 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4376 btrfs_print_leaf(root
, leaf
);
4382 * make the item pointed to by the path bigger, data_size is the new size.
4384 void btrfs_extend_item(struct btrfs_trans_handle
*trans
,
4385 struct btrfs_root
*root
, struct btrfs_path
*path
,
4389 struct extent_buffer
*leaf
;
4390 struct btrfs_item
*item
;
4392 unsigned int data_end
;
4393 unsigned int old_data
;
4394 unsigned int old_size
;
4396 struct btrfs_map_token token
;
4398 btrfs_init_map_token(&token
);
4400 leaf
= path
->nodes
[0];
4402 nritems
= btrfs_header_nritems(leaf
);
4403 data_end
= leaf_data_end(root
, leaf
);
4405 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4406 btrfs_print_leaf(root
, leaf
);
4409 slot
= path
->slots
[0];
4410 old_data
= btrfs_item_end_nr(leaf
, slot
);
4413 if (slot
>= nritems
) {
4414 btrfs_print_leaf(root
, leaf
);
4415 printk(KERN_CRIT
"slot %d too large, nritems %d\n",
4421 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4423 /* first correct the data pointers */
4424 for (i
= slot
; i
< nritems
; i
++) {
4426 item
= btrfs_item_nr(leaf
, i
);
4428 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4429 btrfs_set_token_item_offset(leaf
, item
,
4430 ioff
- data_size
, &token
);
4433 /* shift the data */
4434 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4435 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4436 data_end
, old_data
- data_end
);
4438 data_end
= old_data
;
4439 old_size
= btrfs_item_size_nr(leaf
, slot
);
4440 item
= btrfs_item_nr(leaf
, slot
);
4441 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4442 btrfs_mark_buffer_dirty(leaf
);
4444 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4445 btrfs_print_leaf(root
, leaf
);
4451 * this is a helper for btrfs_insert_empty_items, the main goal here is
4452 * to save stack depth by doing the bulk of the work in a function
4453 * that doesn't call btrfs_search_slot
4455 void setup_items_for_insert(struct btrfs_trans_handle
*trans
,
4456 struct btrfs_root
*root
, struct btrfs_path
*path
,
4457 struct btrfs_key
*cpu_key
, u32
*data_size
,
4458 u32 total_data
, u32 total_size
, int nr
)
4460 struct btrfs_item
*item
;
4463 unsigned int data_end
;
4464 struct btrfs_disk_key disk_key
;
4465 struct extent_buffer
*leaf
;
4467 struct btrfs_map_token token
;
4469 btrfs_init_map_token(&token
);
4471 leaf
= path
->nodes
[0];
4472 slot
= path
->slots
[0];
4474 nritems
= btrfs_header_nritems(leaf
);
4475 data_end
= leaf_data_end(root
, leaf
);
4477 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4478 btrfs_print_leaf(root
, leaf
);
4479 printk(KERN_CRIT
"not enough freespace need %u have %d\n",
4480 total_size
, btrfs_leaf_free_space(root
, leaf
));
4484 if (slot
!= nritems
) {
4485 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4487 if (old_data
< data_end
) {
4488 btrfs_print_leaf(root
, leaf
);
4489 printk(KERN_CRIT
"slot %d old_data %d data_end %d\n",
4490 slot
, old_data
, data_end
);
4494 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4496 /* first correct the data pointers */
4497 for (i
= slot
; i
< nritems
; i
++) {
4500 item
= btrfs_item_nr(leaf
, i
);
4501 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4502 btrfs_set_token_item_offset(leaf
, item
,
4503 ioff
- total_data
, &token
);
4505 /* shift the items */
4506 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4507 btrfs_item_nr_offset(slot
),
4508 (nritems
- slot
) * sizeof(struct btrfs_item
));
4510 /* shift the data */
4511 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4512 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4513 data_end
, old_data
- data_end
);
4514 data_end
= old_data
;
4517 /* setup the item for the new data */
4518 for (i
= 0; i
< nr
; i
++) {
4519 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4520 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4521 item
= btrfs_item_nr(leaf
, slot
+ i
);
4522 btrfs_set_token_item_offset(leaf
, item
,
4523 data_end
- data_size
[i
], &token
);
4524 data_end
-= data_size
[i
];
4525 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4528 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4531 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4532 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4534 btrfs_unlock_up_safe(path
, 1);
4535 btrfs_mark_buffer_dirty(leaf
);
4537 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4538 btrfs_print_leaf(root
, leaf
);
4544 * Given a key and some data, insert items into the tree.
4545 * This does all the path init required, making room in the tree if needed.
4547 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4548 struct btrfs_root
*root
,
4549 struct btrfs_path
*path
,
4550 struct btrfs_key
*cpu_key
, u32
*data_size
,
4559 for (i
= 0; i
< nr
; i
++)
4560 total_data
+= data_size
[i
];
4562 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4563 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4569 slot
= path
->slots
[0];
4572 setup_items_for_insert(trans
, root
, path
, cpu_key
, data_size
,
4573 total_data
, total_size
, nr
);
4578 * Given a key and some data, insert an item into the tree.
4579 * This does all the path init required, making room in the tree if needed.
4581 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4582 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4586 struct btrfs_path
*path
;
4587 struct extent_buffer
*leaf
;
4590 path
= btrfs_alloc_path();
4593 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4595 leaf
= path
->nodes
[0];
4596 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4597 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4598 btrfs_mark_buffer_dirty(leaf
);
4600 btrfs_free_path(path
);
4605 * delete the pointer from a given node.
4607 * the tree should have been previously balanced so the deletion does not
4610 static void del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4611 struct btrfs_path
*path
, int level
, int slot
)
4613 struct extent_buffer
*parent
= path
->nodes
[level
];
4617 nritems
= btrfs_header_nritems(parent
);
4618 if (slot
!= nritems
- 1) {
4620 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4621 slot
+ 1, nritems
- slot
- 1);
4622 memmove_extent_buffer(parent
,
4623 btrfs_node_key_ptr_offset(slot
),
4624 btrfs_node_key_ptr_offset(slot
+ 1),
4625 sizeof(struct btrfs_key_ptr
) *
4626 (nritems
- slot
- 1));
4628 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4629 MOD_LOG_KEY_REMOVE
);
4634 btrfs_set_header_nritems(parent
, nritems
);
4635 if (nritems
== 0 && parent
== root
->node
) {
4636 BUG_ON(btrfs_header_level(root
->node
) != 1);
4637 /* just turn the root into a leaf and break */
4638 btrfs_set_header_level(root
->node
, 0);
4639 } else if (slot
== 0) {
4640 struct btrfs_disk_key disk_key
;
4642 btrfs_node_key(parent
, &disk_key
, 0);
4643 fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
4645 btrfs_mark_buffer_dirty(parent
);
4649 * a helper function to delete the leaf pointed to by path->slots[1] and
4652 * This deletes the pointer in path->nodes[1] and frees the leaf
4653 * block extent. zero is returned if it all worked out, < 0 otherwise.
4655 * The path must have already been setup for deleting the leaf, including
4656 * all the proper balancing. path->nodes[1] must be locked.
4658 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4659 struct btrfs_root
*root
,
4660 struct btrfs_path
*path
,
4661 struct extent_buffer
*leaf
)
4663 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4664 del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
4667 * btrfs_free_extent is expensive, we want to make sure we
4668 * aren't holding any locks when we call it
4670 btrfs_unlock_up_safe(path
, 0);
4672 root_sub_used(root
, leaf
->len
);
4674 extent_buffer_get(leaf
);
4675 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4676 free_extent_buffer_stale(leaf
);
4679 * delete the item at the leaf level in path. If that empties
4680 * the leaf, remove it from the tree
4682 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4683 struct btrfs_path
*path
, int slot
, int nr
)
4685 struct extent_buffer
*leaf
;
4686 struct btrfs_item
*item
;
4693 struct btrfs_map_token token
;
4695 btrfs_init_map_token(&token
);
4697 leaf
= path
->nodes
[0];
4698 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4700 for (i
= 0; i
< nr
; i
++)
4701 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4703 nritems
= btrfs_header_nritems(leaf
);
4705 if (slot
+ nr
!= nritems
) {
4706 int data_end
= leaf_data_end(root
, leaf
);
4708 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4710 btrfs_leaf_data(leaf
) + data_end
,
4711 last_off
- data_end
);
4713 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4716 item
= btrfs_item_nr(leaf
, i
);
4717 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4718 btrfs_set_token_item_offset(leaf
, item
,
4719 ioff
+ dsize
, &token
);
4722 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4723 btrfs_item_nr_offset(slot
+ nr
),
4724 sizeof(struct btrfs_item
) *
4725 (nritems
- slot
- nr
));
4727 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4730 /* delete the leaf if we've emptied it */
4732 if (leaf
== root
->node
) {
4733 btrfs_set_header_level(leaf
, 0);
4735 btrfs_set_path_blocking(path
);
4736 clean_tree_block(trans
, root
, leaf
);
4737 btrfs_del_leaf(trans
, root
, path
, leaf
);
4740 int used
= leaf_space_used(leaf
, 0, nritems
);
4742 struct btrfs_disk_key disk_key
;
4744 btrfs_item_key(leaf
, &disk_key
, 0);
4745 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
4748 /* delete the leaf if it is mostly empty */
4749 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
4750 /* push_leaf_left fixes the path.
4751 * make sure the path still points to our leaf
4752 * for possible call to del_ptr below
4754 slot
= path
->slots
[1];
4755 extent_buffer_get(leaf
);
4757 btrfs_set_path_blocking(path
);
4758 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
4760 if (wret
< 0 && wret
!= -ENOSPC
)
4763 if (path
->nodes
[0] == leaf
&&
4764 btrfs_header_nritems(leaf
)) {
4765 wret
= push_leaf_right(trans
, root
, path
, 1,
4767 if (wret
< 0 && wret
!= -ENOSPC
)
4771 if (btrfs_header_nritems(leaf
) == 0) {
4772 path
->slots
[1] = slot
;
4773 btrfs_del_leaf(trans
, root
, path
, leaf
);
4774 free_extent_buffer(leaf
);
4777 /* if we're still in the path, make sure
4778 * we're dirty. Otherwise, one of the
4779 * push_leaf functions must have already
4780 * dirtied this buffer
4782 if (path
->nodes
[0] == leaf
)
4783 btrfs_mark_buffer_dirty(leaf
);
4784 free_extent_buffer(leaf
);
4787 btrfs_mark_buffer_dirty(leaf
);
4794 * search the tree again to find a leaf with lesser keys
4795 * returns 0 if it found something or 1 if there are no lesser leaves.
4796 * returns < 0 on io errors.
4798 * This may release the path, and so you may lose any locks held at the
4801 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
4803 struct btrfs_key key
;
4804 struct btrfs_disk_key found_key
;
4807 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
4811 else if (key
.type
> 0)
4813 else if (key
.objectid
> 0)
4818 btrfs_release_path(path
);
4819 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
4822 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
4823 ret
= comp_keys(&found_key
, &key
);
4830 * A helper function to walk down the tree starting at min_key, and looking
4831 * for nodes or leaves that are have a minimum transaction id.
4832 * This is used by the btree defrag code, and tree logging
4834 * This does not cow, but it does stuff the starting key it finds back
4835 * into min_key, so you can call btrfs_search_slot with cow=1 on the
4836 * key and get a writable path.
4838 * This does lock as it descends, and path->keep_locks should be set
4839 * to 1 by the caller.
4841 * This honors path->lowest_level to prevent descent past a given level
4844 * min_trans indicates the oldest transaction that you are interested
4845 * in walking through. Any nodes or leaves older than min_trans are
4846 * skipped over (without reading them).
4848 * returns zero if something useful was found, < 0 on error and 1 if there
4849 * was nothing in the tree that matched the search criteria.
4851 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
4852 struct btrfs_key
*max_key
,
4853 struct btrfs_path
*path
,
4856 struct extent_buffer
*cur
;
4857 struct btrfs_key found_key
;
4864 WARN_ON(!path
->keep_locks
);
4866 cur
= btrfs_read_lock_root_node(root
);
4867 level
= btrfs_header_level(cur
);
4868 WARN_ON(path
->nodes
[level
]);
4869 path
->nodes
[level
] = cur
;
4870 path
->locks
[level
] = BTRFS_READ_LOCK
;
4872 if (btrfs_header_generation(cur
) < min_trans
) {
4877 nritems
= btrfs_header_nritems(cur
);
4878 level
= btrfs_header_level(cur
);
4879 sret
= bin_search(cur
, min_key
, level
, &slot
);
4881 /* at the lowest level, we're done, setup the path and exit */
4882 if (level
== path
->lowest_level
) {
4883 if (slot
>= nritems
)
4886 path
->slots
[level
] = slot
;
4887 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
4890 if (sret
&& slot
> 0)
4893 * check this node pointer against the min_trans parameters.
4894 * If it is too old, old, skip to the next one.
4896 while (slot
< nritems
) {
4900 blockptr
= btrfs_node_blockptr(cur
, slot
);
4901 gen
= btrfs_node_ptr_generation(cur
, slot
);
4902 if (gen
< min_trans
) {
4910 * we didn't find a candidate key in this node, walk forward
4911 * and find another one
4913 if (slot
>= nritems
) {
4914 path
->slots
[level
] = slot
;
4915 btrfs_set_path_blocking(path
);
4916 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
4919 btrfs_release_path(path
);
4925 /* save our key for returning back */
4926 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
4927 path
->slots
[level
] = slot
;
4928 if (level
== path
->lowest_level
) {
4930 unlock_up(path
, level
, 1, 0, NULL
);
4933 btrfs_set_path_blocking(path
);
4934 cur
= read_node_slot(root
, cur
, slot
);
4935 BUG_ON(!cur
); /* -ENOMEM */
4937 btrfs_tree_read_lock(cur
);
4939 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
4940 path
->nodes
[level
- 1] = cur
;
4941 unlock_up(path
, level
, 1, 0, NULL
);
4942 btrfs_clear_path_blocking(path
, NULL
, 0);
4946 memcpy(min_key
, &found_key
, sizeof(found_key
));
4947 btrfs_set_path_blocking(path
);
4951 static void tree_move_down(struct btrfs_root
*root
,
4952 struct btrfs_path
*path
,
4953 int *level
, int root_level
)
4955 BUG_ON(*level
== 0);
4956 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
4957 path
->slots
[*level
]);
4958 path
->slots
[*level
- 1] = 0;
4962 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
4963 struct btrfs_path
*path
,
4964 int *level
, int root_level
)
4968 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
4970 path
->slots
[*level
]++;
4972 while (path
->slots
[*level
] >= nritems
) {
4973 if (*level
== root_level
)
4977 path
->slots
[*level
] = 0;
4978 free_extent_buffer(path
->nodes
[*level
]);
4979 path
->nodes
[*level
] = NULL
;
4981 path
->slots
[*level
]++;
4983 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
4990 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
4993 static int tree_advance(struct btrfs_root
*root
,
4994 struct btrfs_path
*path
,
4995 int *level
, int root_level
,
4997 struct btrfs_key
*key
)
5001 if (*level
== 0 || !allow_down
) {
5002 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5004 tree_move_down(root
, path
, level
, root_level
);
5009 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5010 path
->slots
[*level
]);
5012 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5013 path
->slots
[*level
]);
5018 static int tree_compare_item(struct btrfs_root
*left_root
,
5019 struct btrfs_path
*left_path
,
5020 struct btrfs_path
*right_path
,
5025 unsigned long off1
, off2
;
5027 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5028 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5032 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5033 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5034 right_path
->slots
[0]);
5036 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5038 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5045 #define ADVANCE_ONLY_NEXT -1
5048 * This function compares two trees and calls the provided callback for
5049 * every changed/new/deleted item it finds.
5050 * If shared tree blocks are encountered, whole subtrees are skipped, making
5051 * the compare pretty fast on snapshotted subvolumes.
5053 * This currently works on commit roots only. As commit roots are read only,
5054 * we don't do any locking. The commit roots are protected with transactions.
5055 * Transactions are ended and rejoined when a commit is tried in between.
5057 * This function checks for modifications done to the trees while comparing.
5058 * If it detects a change, it aborts immediately.
5060 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5061 struct btrfs_root
*right_root
,
5062 btrfs_changed_cb_t changed_cb
, void *ctx
)
5066 struct btrfs_trans_handle
*trans
= NULL
;
5067 struct btrfs_path
*left_path
= NULL
;
5068 struct btrfs_path
*right_path
= NULL
;
5069 struct btrfs_key left_key
;
5070 struct btrfs_key right_key
;
5071 char *tmp_buf
= NULL
;
5072 int left_root_level
;
5073 int right_root_level
;
5076 int left_end_reached
;
5077 int right_end_reached
;
5082 u64 left_start_ctransid
;
5083 u64 right_start_ctransid
;
5086 left_path
= btrfs_alloc_path();
5091 right_path
= btrfs_alloc_path();
5097 tmp_buf
= kmalloc(left_root
->leafsize
, GFP_NOFS
);
5103 left_path
->search_commit_root
= 1;
5104 left_path
->skip_locking
= 1;
5105 right_path
->search_commit_root
= 1;
5106 right_path
->skip_locking
= 1;
5108 spin_lock(&left_root
->root_item_lock
);
5109 left_start_ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5110 spin_unlock(&left_root
->root_item_lock
);
5112 spin_lock(&right_root
->root_item_lock
);
5113 right_start_ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5114 spin_unlock(&right_root
->root_item_lock
);
5116 trans
= btrfs_join_transaction(left_root
);
5117 if (IS_ERR(trans
)) {
5118 ret
= PTR_ERR(trans
);
5124 * Strategy: Go to the first items of both trees. Then do
5126 * If both trees are at level 0
5127 * Compare keys of current items
5128 * If left < right treat left item as new, advance left tree
5130 * If left > right treat right item as deleted, advance right tree
5132 * If left == right do deep compare of items, treat as changed if
5133 * needed, advance both trees and repeat
5134 * If both trees are at the same level but not at level 0
5135 * Compare keys of current nodes/leafs
5136 * If left < right advance left tree and repeat
5137 * If left > right advance right tree and repeat
5138 * If left == right compare blockptrs of the next nodes/leafs
5139 * If they match advance both trees but stay at the same level
5141 * If they don't match advance both trees while allowing to go
5143 * If tree levels are different
5144 * Advance the tree that needs it and repeat
5146 * Advancing a tree means:
5147 * If we are at level 0, try to go to the next slot. If that's not
5148 * possible, go one level up and repeat. Stop when we found a level
5149 * where we could go to the next slot. We may at this point be on a
5152 * If we are not at level 0 and not on shared tree blocks, go one
5155 * If we are not at level 0 and on shared tree blocks, go one slot to
5156 * the right if possible or go up and right.
5159 left_level
= btrfs_header_level(left_root
->commit_root
);
5160 left_root_level
= left_level
;
5161 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5162 extent_buffer_get(left_path
->nodes
[left_level
]);
5164 right_level
= btrfs_header_level(right_root
->commit_root
);
5165 right_root_level
= right_level
;
5166 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5167 extent_buffer_get(right_path
->nodes
[right_level
]);
5169 if (left_level
== 0)
5170 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5171 &left_key
, left_path
->slots
[left_level
]);
5173 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5174 &left_key
, left_path
->slots
[left_level
]);
5175 if (right_level
== 0)
5176 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5177 &right_key
, right_path
->slots
[right_level
]);
5179 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5180 &right_key
, right_path
->slots
[right_level
]);
5182 left_end_reached
= right_end_reached
= 0;
5183 advance_left
= advance_right
= 0;
5187 * We need to make sure the transaction does not get committed
5188 * while we do anything on commit roots. This means, we need to
5189 * join and leave transactions for every item that we process.
5191 if (trans
&& btrfs_should_end_transaction(trans
, left_root
)) {
5192 btrfs_release_path(left_path
);
5193 btrfs_release_path(right_path
);
5195 ret
= btrfs_end_transaction(trans
, left_root
);
5200 /* now rejoin the transaction */
5202 trans
= btrfs_join_transaction(left_root
);
5203 if (IS_ERR(trans
)) {
5204 ret
= PTR_ERR(trans
);
5209 spin_lock(&left_root
->root_item_lock
);
5210 ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5211 spin_unlock(&left_root
->root_item_lock
);
5212 if (ctransid
!= left_start_ctransid
)
5213 left_start_ctransid
= 0;
5215 spin_lock(&right_root
->root_item_lock
);
5216 ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5217 spin_unlock(&right_root
->root_item_lock
);
5218 if (ctransid
!= right_start_ctransid
)
5219 right_start_ctransid
= 0;
5221 if (!left_start_ctransid
|| !right_start_ctransid
) {
5222 WARN(1, KERN_WARNING
5223 "btrfs: btrfs_compare_tree detected "
5224 "a change in one of the trees while "
5225 "iterating. This is probably a "
5232 * the commit root may have changed, so start again
5235 left_path
->lowest_level
= left_level
;
5236 right_path
->lowest_level
= right_level
;
5237 ret
= btrfs_search_slot(NULL
, left_root
,
5238 &left_key
, left_path
, 0, 0);
5241 ret
= btrfs_search_slot(NULL
, right_root
,
5242 &right_key
, right_path
, 0, 0);
5247 if (advance_left
&& !left_end_reached
) {
5248 ret
= tree_advance(left_root
, left_path
, &left_level
,
5250 advance_left
!= ADVANCE_ONLY_NEXT
,
5253 left_end_reached
= ADVANCE
;
5256 if (advance_right
&& !right_end_reached
) {
5257 ret
= tree_advance(right_root
, right_path
, &right_level
,
5259 advance_right
!= ADVANCE_ONLY_NEXT
,
5262 right_end_reached
= ADVANCE
;
5266 if (left_end_reached
&& right_end_reached
) {
5269 } else if (left_end_reached
) {
5270 if (right_level
== 0) {
5271 ret
= changed_cb(left_root
, right_root
,
5272 left_path
, right_path
,
5274 BTRFS_COMPARE_TREE_DELETED
,
5279 advance_right
= ADVANCE
;
5281 } else if (right_end_reached
) {
5282 if (left_level
== 0) {
5283 ret
= changed_cb(left_root
, right_root
,
5284 left_path
, right_path
,
5286 BTRFS_COMPARE_TREE_NEW
,
5291 advance_left
= ADVANCE
;
5295 if (left_level
== 0 && right_level
== 0) {
5296 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5298 ret
= changed_cb(left_root
, right_root
,
5299 left_path
, right_path
,
5301 BTRFS_COMPARE_TREE_NEW
,
5305 advance_left
= ADVANCE
;
5306 } else if (cmp
> 0) {
5307 ret
= changed_cb(left_root
, right_root
,
5308 left_path
, right_path
,
5310 BTRFS_COMPARE_TREE_DELETED
,
5314 advance_right
= ADVANCE
;
5316 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5317 ret
= tree_compare_item(left_root
, left_path
,
5318 right_path
, tmp_buf
);
5320 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5321 ret
= changed_cb(left_root
, right_root
,
5322 left_path
, right_path
,
5324 BTRFS_COMPARE_TREE_CHANGED
,
5329 advance_left
= ADVANCE
;
5330 advance_right
= ADVANCE
;
5332 } else if (left_level
== right_level
) {
5333 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5335 advance_left
= ADVANCE
;
5336 } else if (cmp
> 0) {
5337 advance_right
= ADVANCE
;
5339 left_blockptr
= btrfs_node_blockptr(
5340 left_path
->nodes
[left_level
],
5341 left_path
->slots
[left_level
]);
5342 right_blockptr
= btrfs_node_blockptr(
5343 right_path
->nodes
[right_level
],
5344 right_path
->slots
[right_level
]);
5345 if (left_blockptr
== right_blockptr
) {
5347 * As we're on a shared block, don't
5348 * allow to go deeper.
5350 advance_left
= ADVANCE_ONLY_NEXT
;
5351 advance_right
= ADVANCE_ONLY_NEXT
;
5353 advance_left
= ADVANCE
;
5354 advance_right
= ADVANCE
;
5357 } else if (left_level
< right_level
) {
5358 advance_right
= ADVANCE
;
5360 advance_left
= ADVANCE
;
5365 btrfs_free_path(left_path
);
5366 btrfs_free_path(right_path
);
5371 ret
= btrfs_end_transaction(trans
, left_root
);
5373 btrfs_end_transaction(trans
, left_root
);
5380 * this is similar to btrfs_next_leaf, but does not try to preserve
5381 * and fixup the path. It looks for and returns the next key in the
5382 * tree based on the current path and the min_trans parameters.
5384 * 0 is returned if another key is found, < 0 if there are any errors
5385 * and 1 is returned if there are no higher keys in the tree
5387 * path->keep_locks should be set to 1 on the search made before
5388 * calling this function.
5390 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5391 struct btrfs_key
*key
, int level
, u64 min_trans
)
5394 struct extent_buffer
*c
;
5396 WARN_ON(!path
->keep_locks
);
5397 while (level
< BTRFS_MAX_LEVEL
) {
5398 if (!path
->nodes
[level
])
5401 slot
= path
->slots
[level
] + 1;
5402 c
= path
->nodes
[level
];
5404 if (slot
>= btrfs_header_nritems(c
)) {
5407 struct btrfs_key cur_key
;
5408 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5409 !path
->nodes
[level
+ 1])
5412 if (path
->locks
[level
+ 1]) {
5417 slot
= btrfs_header_nritems(c
) - 1;
5419 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5421 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5423 orig_lowest
= path
->lowest_level
;
5424 btrfs_release_path(path
);
5425 path
->lowest_level
= level
;
5426 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5428 path
->lowest_level
= orig_lowest
;
5432 c
= path
->nodes
[level
];
5433 slot
= path
->slots
[level
];
5440 btrfs_item_key_to_cpu(c
, key
, slot
);
5442 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5444 if (gen
< min_trans
) {
5448 btrfs_node_key_to_cpu(c
, key
, slot
);
5456 * search the tree again to find a leaf with greater keys
5457 * returns 0 if it found something or 1 if there are no greater leaves.
5458 * returns < 0 on io errors.
5460 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5462 return btrfs_next_old_leaf(root
, path
, 0);
5465 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5470 struct extent_buffer
*c
;
5471 struct extent_buffer
*next
;
5472 struct btrfs_key key
;
5475 int old_spinning
= path
->leave_spinning
;
5476 int next_rw_lock
= 0;
5478 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5482 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5487 btrfs_release_path(path
);
5489 path
->keep_locks
= 1;
5490 path
->leave_spinning
= 1;
5493 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5495 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5496 path
->keep_locks
= 0;
5501 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5503 * by releasing the path above we dropped all our locks. A balance
5504 * could have added more items next to the key that used to be
5505 * at the very end of the block. So, check again here and
5506 * advance the path if there are now more items available.
5508 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5515 while (level
< BTRFS_MAX_LEVEL
) {
5516 if (!path
->nodes
[level
]) {
5521 slot
= path
->slots
[level
] + 1;
5522 c
= path
->nodes
[level
];
5523 if (slot
>= btrfs_header_nritems(c
)) {
5525 if (level
== BTRFS_MAX_LEVEL
) {
5533 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5534 free_extent_buffer(next
);
5538 next_rw_lock
= path
->locks
[level
];
5539 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5545 btrfs_release_path(path
);
5549 if (!path
->skip_locking
) {
5550 ret
= btrfs_try_tree_read_lock(next
);
5551 if (!ret
&& time_seq
) {
5553 * If we don't get the lock, we may be racing
5554 * with push_leaf_left, holding that lock while
5555 * itself waiting for the leaf we've currently
5556 * locked. To solve this situation, we give up
5557 * on our lock and cycle.
5559 free_extent_buffer(next
);
5560 btrfs_release_path(path
);
5565 btrfs_set_path_blocking(path
);
5566 btrfs_tree_read_lock(next
);
5567 btrfs_clear_path_blocking(path
, next
,
5570 next_rw_lock
= BTRFS_READ_LOCK
;
5574 path
->slots
[level
] = slot
;
5577 c
= path
->nodes
[level
];
5578 if (path
->locks
[level
])
5579 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5581 free_extent_buffer(c
);
5582 path
->nodes
[level
] = next
;
5583 path
->slots
[level
] = 0;
5584 if (!path
->skip_locking
)
5585 path
->locks
[level
] = next_rw_lock
;
5589 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5595 btrfs_release_path(path
);
5599 if (!path
->skip_locking
) {
5600 ret
= btrfs_try_tree_read_lock(next
);
5602 btrfs_set_path_blocking(path
);
5603 btrfs_tree_read_lock(next
);
5604 btrfs_clear_path_blocking(path
, next
,
5607 next_rw_lock
= BTRFS_READ_LOCK
;
5612 unlock_up(path
, 0, 1, 0, NULL
);
5613 path
->leave_spinning
= old_spinning
;
5615 btrfs_set_path_blocking(path
);
5621 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5622 * searching until it gets past min_objectid or finds an item of 'type'
5624 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5626 int btrfs_previous_item(struct btrfs_root
*root
,
5627 struct btrfs_path
*path
, u64 min_objectid
,
5630 struct btrfs_key found_key
;
5631 struct extent_buffer
*leaf
;
5636 if (path
->slots
[0] == 0) {
5637 btrfs_set_path_blocking(path
);
5638 ret
= btrfs_prev_leaf(root
, path
);
5644 leaf
= path
->nodes
[0];
5645 nritems
= btrfs_header_nritems(leaf
);
5648 if (path
->slots
[0] == nritems
)
5651 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5652 if (found_key
.objectid
< min_objectid
)
5654 if (found_key
.type
== type
)
5656 if (found_key
.objectid
== min_objectid
&&
5657 found_key
.type
< type
)