2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
20 #include <linux/slab.h>
21 #include <linux/rbtree.h>
24 #include "transaction.h"
25 #include "print-tree.h"
28 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
29 *root
, struct btrfs_path
*path
, int level
);
30 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
31 *root
, struct btrfs_key
*ins_key
,
32 struct btrfs_path
*path
, int data_size
, int extend
);
33 static int push_node_left(struct btrfs_trans_handle
*trans
,
34 struct btrfs_root
*root
, struct extent_buffer
*dst
,
35 struct extent_buffer
*src
, int empty
);
36 static int balance_node_right(struct btrfs_trans_handle
*trans
,
37 struct btrfs_root
*root
,
38 struct extent_buffer
*dst_buf
,
39 struct extent_buffer
*src_buf
);
40 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
42 static int tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
43 struct extent_buffer
*eb
);
45 struct btrfs_path
*btrfs_alloc_path(void)
47 struct btrfs_path
*path
;
48 path
= kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
53 * set all locked nodes in the path to blocking locks. This should
54 * be done before scheduling
56 noinline
void btrfs_set_path_blocking(struct btrfs_path
*p
)
59 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
60 if (!p
->nodes
[i
] || !p
->locks
[i
])
62 btrfs_set_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
63 if (p
->locks
[i
] == BTRFS_READ_LOCK
)
64 p
->locks
[i
] = BTRFS_READ_LOCK_BLOCKING
;
65 else if (p
->locks
[i
] == BTRFS_WRITE_LOCK
)
66 p
->locks
[i
] = BTRFS_WRITE_LOCK_BLOCKING
;
71 * reset all the locked nodes in the patch to spinning locks.
73 * held is used to keep lockdep happy, when lockdep is enabled
74 * we set held to a blocking lock before we go around and
75 * retake all the spinlocks in the path. You can safely use NULL
78 noinline
void btrfs_clear_path_blocking(struct btrfs_path
*p
,
79 struct extent_buffer
*held
, int held_rw
)
83 #ifdef CONFIG_DEBUG_LOCK_ALLOC
84 /* lockdep really cares that we take all of these spinlocks
85 * in the right order. If any of the locks in the path are not
86 * currently blocking, it is going to complain. So, make really
87 * really sure by forcing the path to blocking before we clear
91 btrfs_set_lock_blocking_rw(held
, held_rw
);
92 if (held_rw
== BTRFS_WRITE_LOCK
)
93 held_rw
= BTRFS_WRITE_LOCK_BLOCKING
;
94 else if (held_rw
== BTRFS_READ_LOCK
)
95 held_rw
= BTRFS_READ_LOCK_BLOCKING
;
97 btrfs_set_path_blocking(p
);
100 for (i
= BTRFS_MAX_LEVEL
- 1; i
>= 0; i
--) {
101 if (p
->nodes
[i
] && p
->locks
[i
]) {
102 btrfs_clear_lock_blocking_rw(p
->nodes
[i
], p
->locks
[i
]);
103 if (p
->locks
[i
] == BTRFS_WRITE_LOCK_BLOCKING
)
104 p
->locks
[i
] = BTRFS_WRITE_LOCK
;
105 else if (p
->locks
[i
] == BTRFS_READ_LOCK_BLOCKING
)
106 p
->locks
[i
] = BTRFS_READ_LOCK
;
110 #ifdef CONFIG_DEBUG_LOCK_ALLOC
112 btrfs_clear_lock_blocking_rw(held
, held_rw
);
116 /* this also releases the path */
117 void btrfs_free_path(struct btrfs_path
*p
)
121 btrfs_release_path(p
);
122 kmem_cache_free(btrfs_path_cachep
, p
);
126 * path release drops references on the extent buffers in the path
127 * and it drops any locks held by this path
129 * It is safe to call this on paths that no locks or extent buffers held.
131 noinline
void btrfs_release_path(struct btrfs_path
*p
)
135 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
140 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
143 free_extent_buffer(p
->nodes
[i
]);
149 * safely gets a reference on the root node of a tree. A lock
150 * is not taken, so a concurrent writer may put a different node
151 * at the root of the tree. See btrfs_lock_root_node for the
154 * The extent buffer returned by this has a reference taken, so
155 * it won't disappear. It may stop being the root of the tree
156 * at any time because there are no locks held.
158 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
160 struct extent_buffer
*eb
;
164 eb
= rcu_dereference(root
->node
);
167 * RCU really hurts here, we could free up the root node because
168 * it was cow'ed but we may not get the new root node yet so do
169 * the inc_not_zero dance and if it doesn't work then
170 * synchronize_rcu and try again.
172 if (atomic_inc_not_zero(&eb
->refs
)) {
182 /* loop around taking references on and locking the root node of the
183 * tree until you end up with a lock on the root. A locked buffer
184 * is returned, with a reference held.
186 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
188 struct extent_buffer
*eb
;
191 eb
= btrfs_root_node(root
);
193 if (eb
== root
->node
)
195 btrfs_tree_unlock(eb
);
196 free_extent_buffer(eb
);
201 /* loop around taking references on and locking the root node of the
202 * tree until you end up with a lock on the root. A locked buffer
203 * is returned, with a reference held.
205 static struct extent_buffer
*btrfs_read_lock_root_node(struct btrfs_root
*root
)
207 struct extent_buffer
*eb
;
210 eb
= btrfs_root_node(root
);
211 btrfs_tree_read_lock(eb
);
212 if (eb
== root
->node
)
214 btrfs_tree_read_unlock(eb
);
215 free_extent_buffer(eb
);
220 /* cowonly root (everything not a reference counted cow subvolume), just get
221 * put onto a simple dirty list. transaction.c walks this to make sure they
222 * get properly updated on disk.
224 static void add_root_to_dirty_list(struct btrfs_root
*root
)
226 spin_lock(&root
->fs_info
->trans_lock
);
227 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
228 list_add(&root
->dirty_list
,
229 &root
->fs_info
->dirty_cowonly_roots
);
231 spin_unlock(&root
->fs_info
->trans_lock
);
235 * used by snapshot creation to make a copy of a root for a tree with
236 * a given objectid. The buffer with the new root node is returned in
237 * cow_ret, and this func returns zero on success or a negative error code.
239 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
240 struct btrfs_root
*root
,
241 struct extent_buffer
*buf
,
242 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
244 struct extent_buffer
*cow
;
247 struct btrfs_disk_key disk_key
;
249 WARN_ON(root
->ref_cows
&& trans
->transid
!=
250 root
->fs_info
->running_transaction
->transid
);
251 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
253 level
= btrfs_header_level(buf
);
255 btrfs_item_key(buf
, &disk_key
, 0);
257 btrfs_node_key(buf
, &disk_key
, 0);
259 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, 0,
260 new_root_objectid
, &disk_key
, level
,
265 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
266 btrfs_set_header_bytenr(cow
, cow
->start
);
267 btrfs_set_header_generation(cow
, trans
->transid
);
268 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
269 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
270 BTRFS_HEADER_FLAG_RELOC
);
271 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
272 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
274 btrfs_set_header_owner(cow
, new_root_objectid
);
276 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
279 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
280 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
281 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
283 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
288 btrfs_mark_buffer_dirty(cow
);
297 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
298 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
300 MOD_LOG_ROOT_REPLACE
,
303 struct tree_mod_move
{
308 struct tree_mod_root
{
313 struct tree_mod_elem
{
315 u64 index
; /* shifted logical */
319 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
322 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
325 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
326 struct btrfs_disk_key key
;
329 /* this is used for op == MOD_LOG_MOVE_KEYS */
330 struct tree_mod_move move
;
332 /* this is used for op == MOD_LOG_ROOT_REPLACE */
333 struct tree_mod_root old_root
;
336 static inline void tree_mod_log_read_lock(struct btrfs_fs_info
*fs_info
)
338 read_lock(&fs_info
->tree_mod_log_lock
);
341 static inline void tree_mod_log_read_unlock(struct btrfs_fs_info
*fs_info
)
343 read_unlock(&fs_info
->tree_mod_log_lock
);
346 static inline void tree_mod_log_write_lock(struct btrfs_fs_info
*fs_info
)
348 write_lock(&fs_info
->tree_mod_log_lock
);
351 static inline void tree_mod_log_write_unlock(struct btrfs_fs_info
*fs_info
)
353 write_unlock(&fs_info
->tree_mod_log_lock
);
357 * Increment the upper half of tree_mod_seq, set lower half zero.
359 * Must be called with fs_info->tree_mod_seq_lock held.
361 static inline u64
btrfs_inc_tree_mod_seq_major(struct btrfs_fs_info
*fs_info
)
363 u64 seq
= atomic64_read(&fs_info
->tree_mod_seq
);
364 seq
&= 0xffffffff00000000ull
;
366 atomic64_set(&fs_info
->tree_mod_seq
, seq
);
371 * Increment the lower half of tree_mod_seq.
373 * Must be called with fs_info->tree_mod_seq_lock held. The way major numbers
374 * are generated should not technically require a spin lock here. (Rationale:
375 * incrementing the minor while incrementing the major seq number is between its
376 * atomic64_read and atomic64_set calls doesn't duplicate sequence numbers, it
377 * just returns a unique sequence number as usual.) We have decided to leave
378 * that requirement in here and rethink it once we notice it really imposes a
379 * problem on some workload.
381 static inline u64
btrfs_inc_tree_mod_seq_minor(struct btrfs_fs_info
*fs_info
)
383 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
387 * return the last minor in the previous major tree_mod_seq number
389 u64
btrfs_tree_mod_seq_prev(u64 seq
)
391 return (seq
& 0xffffffff00000000ull
) - 1ull;
395 * This adds a new blocker to the tree mod log's blocker list if the @elem
396 * passed does not already have a sequence number set. So when a caller expects
397 * to record tree modifications, it should ensure to set elem->seq to zero
398 * before calling btrfs_get_tree_mod_seq.
399 * Returns a fresh, unused tree log modification sequence number, even if no new
402 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
403 struct seq_list
*elem
)
407 tree_mod_log_write_lock(fs_info
);
408 spin_lock(&fs_info
->tree_mod_seq_lock
);
410 elem
->seq
= btrfs_inc_tree_mod_seq_major(fs_info
);
411 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
413 seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
414 spin_unlock(&fs_info
->tree_mod_seq_lock
);
415 tree_mod_log_write_unlock(fs_info
);
420 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
421 struct seq_list
*elem
)
423 struct rb_root
*tm_root
;
424 struct rb_node
*node
;
425 struct rb_node
*next
;
426 struct seq_list
*cur_elem
;
427 struct tree_mod_elem
*tm
;
428 u64 min_seq
= (u64
)-1;
429 u64 seq_putting
= elem
->seq
;
434 spin_lock(&fs_info
->tree_mod_seq_lock
);
435 list_del(&elem
->list
);
438 list_for_each_entry(cur_elem
, &fs_info
->tree_mod_seq_list
, list
) {
439 if (cur_elem
->seq
< min_seq
) {
440 if (seq_putting
> cur_elem
->seq
) {
442 * blocker with lower sequence number exists, we
443 * cannot remove anything from the log
445 spin_unlock(&fs_info
->tree_mod_seq_lock
);
448 min_seq
= cur_elem
->seq
;
451 spin_unlock(&fs_info
->tree_mod_seq_lock
);
454 * anything that's lower than the lowest existing (read: blocked)
455 * sequence number can be removed from the tree.
457 tree_mod_log_write_lock(fs_info
);
458 tm_root
= &fs_info
->tree_mod_log
;
459 for (node
= rb_first(tm_root
); node
; node
= next
) {
460 next
= rb_next(node
);
461 tm
= container_of(node
, struct tree_mod_elem
, node
);
462 if (tm
->seq
> min_seq
)
464 rb_erase(node
, tm_root
);
467 tree_mod_log_write_unlock(fs_info
);
471 * key order of the log:
474 * the index is the shifted logical of the *new* root node for root replace
475 * operations, or the shifted logical of the affected block for all other
478 * Note: must be called with write lock (tree_mod_log_write_lock).
481 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
483 struct rb_root
*tm_root
;
484 struct rb_node
**new;
485 struct rb_node
*parent
= NULL
;
486 struct tree_mod_elem
*cur
;
490 spin_lock(&fs_info
->tree_mod_seq_lock
);
491 tm
->seq
= btrfs_inc_tree_mod_seq_minor(fs_info
);
492 spin_unlock(&fs_info
->tree_mod_seq_lock
);
494 tm_root
= &fs_info
->tree_mod_log
;
495 new = &tm_root
->rb_node
;
497 cur
= container_of(*new, struct tree_mod_elem
, node
);
499 if (cur
->index
< tm
->index
)
500 new = &((*new)->rb_left
);
501 else if (cur
->index
> tm
->index
)
502 new = &((*new)->rb_right
);
503 else if (cur
->seq
< tm
->seq
)
504 new = &((*new)->rb_left
);
505 else if (cur
->seq
> tm
->seq
)
506 new = &((*new)->rb_right
);
511 rb_link_node(&tm
->node
, parent
, new);
512 rb_insert_color(&tm
->node
, tm_root
);
517 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
518 * returns zero with the tree_mod_log_lock acquired. The caller must hold
519 * this until all tree mod log insertions are recorded in the rb tree and then
520 * call tree_mod_log_write_unlock() to release.
522 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
523 struct extent_buffer
*eb
) {
525 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
527 if (eb
&& btrfs_header_level(eb
) == 0)
530 tree_mod_log_write_lock(fs_info
);
531 if (list_empty(&(fs_info
)->tree_mod_seq_list
)) {
532 tree_mod_log_write_unlock(fs_info
);
539 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
540 static inline int tree_mod_need_log(const struct btrfs_fs_info
*fs_info
,
541 struct extent_buffer
*eb
)
544 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
546 if (eb
&& btrfs_header_level(eb
) == 0)
552 static struct tree_mod_elem
*
553 alloc_tree_mod_elem(struct extent_buffer
*eb
, int slot
,
554 enum mod_log_op op
, gfp_t flags
)
556 struct tree_mod_elem
*tm
;
558 tm
= kzalloc(sizeof(*tm
), flags
);
562 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
563 if (op
!= MOD_LOG_KEY_ADD
) {
564 btrfs_node_key(eb
, &tm
->key
, slot
);
565 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
569 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
570 RB_CLEAR_NODE(&tm
->node
);
576 tree_mod_log_insert_key(struct btrfs_fs_info
*fs_info
,
577 struct extent_buffer
*eb
, int slot
,
578 enum mod_log_op op
, gfp_t flags
)
580 struct tree_mod_elem
*tm
;
583 if (!tree_mod_need_log(fs_info
, eb
))
586 tm
= alloc_tree_mod_elem(eb
, slot
, op
, flags
);
590 if (tree_mod_dont_log(fs_info
, eb
)) {
595 ret
= __tree_mod_log_insert(fs_info
, tm
);
596 tree_mod_log_write_unlock(fs_info
);
604 tree_mod_log_insert_move(struct btrfs_fs_info
*fs_info
,
605 struct extent_buffer
*eb
, int dst_slot
, int src_slot
,
606 int nr_items
, gfp_t flags
)
608 struct tree_mod_elem
*tm
= NULL
;
609 struct tree_mod_elem
**tm_list
= NULL
;
614 if (!tree_mod_need_log(fs_info
, eb
))
617 tm_list
= kzalloc(nr_items
* sizeof(struct tree_mod_elem
*), flags
);
621 tm
= kzalloc(sizeof(*tm
), flags
);
627 tm
->index
= eb
->start
>> PAGE_CACHE_SHIFT
;
629 tm
->move
.dst_slot
= dst_slot
;
630 tm
->move
.nr_items
= nr_items
;
631 tm
->op
= MOD_LOG_MOVE_KEYS
;
633 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
634 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
+ dst_slot
,
635 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, flags
);
642 if (tree_mod_dont_log(fs_info
, eb
))
647 * When we override something during the move, we log these removals.
648 * This can only happen when we move towards the beginning of the
649 * buffer, i.e. dst_slot < src_slot.
651 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
652 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
657 ret
= __tree_mod_log_insert(fs_info
, tm
);
660 tree_mod_log_write_unlock(fs_info
);
665 for (i
= 0; i
< nr_items
; i
++) {
666 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
667 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
671 tree_mod_log_write_unlock(fs_info
);
679 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
680 struct tree_mod_elem
**tm_list
,
686 for (i
= nritems
- 1; i
>= 0; i
--) {
687 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
689 for (j
= nritems
- 1; j
> i
; j
--)
690 rb_erase(&tm_list
[j
]->node
,
691 &fs_info
->tree_mod_log
);
700 tree_mod_log_insert_root(struct btrfs_fs_info
*fs_info
,
701 struct extent_buffer
*old_root
,
702 struct extent_buffer
*new_root
, gfp_t flags
,
705 struct tree_mod_elem
*tm
= NULL
;
706 struct tree_mod_elem
**tm_list
= NULL
;
711 if (!tree_mod_need_log(fs_info
, NULL
))
714 if (log_removal
&& btrfs_header_level(old_root
) > 0) {
715 nritems
= btrfs_header_nritems(old_root
);
716 tm_list
= kzalloc(nritems
* sizeof(struct tree_mod_elem
*),
722 for (i
= 0; i
< nritems
; i
++) {
723 tm_list
[i
] = alloc_tree_mod_elem(old_root
, i
,
724 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, flags
);
732 tm
= kzalloc(sizeof(*tm
), flags
);
738 tm
->index
= new_root
->start
>> PAGE_CACHE_SHIFT
;
739 tm
->old_root
.logical
= old_root
->start
;
740 tm
->old_root
.level
= btrfs_header_level(old_root
);
741 tm
->generation
= btrfs_header_generation(old_root
);
742 tm
->op
= MOD_LOG_ROOT_REPLACE
;
744 if (tree_mod_dont_log(fs_info
, NULL
))
748 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
750 ret
= __tree_mod_log_insert(fs_info
, tm
);
752 tree_mod_log_write_unlock(fs_info
);
761 for (i
= 0; i
< nritems
; i
++)
770 static struct tree_mod_elem
*
771 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
774 struct rb_root
*tm_root
;
775 struct rb_node
*node
;
776 struct tree_mod_elem
*cur
= NULL
;
777 struct tree_mod_elem
*found
= NULL
;
778 u64 index
= start
>> PAGE_CACHE_SHIFT
;
780 tree_mod_log_read_lock(fs_info
);
781 tm_root
= &fs_info
->tree_mod_log
;
782 node
= tm_root
->rb_node
;
784 cur
= container_of(node
, struct tree_mod_elem
, node
);
785 if (cur
->index
< index
) {
786 node
= node
->rb_left
;
787 } else if (cur
->index
> index
) {
788 node
= node
->rb_right
;
789 } else if (cur
->seq
< min_seq
) {
790 node
= node
->rb_left
;
791 } else if (!smallest
) {
792 /* we want the node with the highest seq */
794 BUG_ON(found
->seq
> cur
->seq
);
796 node
= node
->rb_left
;
797 } else if (cur
->seq
> min_seq
) {
798 /* we want the node with the smallest seq */
800 BUG_ON(found
->seq
< cur
->seq
);
802 node
= node
->rb_right
;
808 tree_mod_log_read_unlock(fs_info
);
814 * this returns the element from the log with the smallest time sequence
815 * value that's in the log (the oldest log item). any element with a time
816 * sequence lower than min_seq will be ignored.
818 static struct tree_mod_elem
*
819 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
822 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
826 * this returns the element from the log with the largest time sequence
827 * value that's in the log (the most recent log item). any element with
828 * a time sequence lower than min_seq will be ignored.
830 static struct tree_mod_elem
*
831 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
833 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
837 tree_mod_log_eb_copy(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
838 struct extent_buffer
*src
, unsigned long dst_offset
,
839 unsigned long src_offset
, int nr_items
)
842 struct tree_mod_elem
**tm_list
= NULL
;
843 struct tree_mod_elem
**tm_list_add
, **tm_list_rem
;
847 if (!tree_mod_need_log(fs_info
, NULL
))
850 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
853 tm_list
= kzalloc(nr_items
* 2 * sizeof(struct tree_mod_elem
*),
858 tm_list_add
= tm_list
;
859 tm_list_rem
= tm_list
+ nr_items
;
860 for (i
= 0; i
< nr_items
; i
++) {
861 tm_list_rem
[i
] = alloc_tree_mod_elem(src
, i
+ src_offset
,
862 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
863 if (!tm_list_rem
[i
]) {
868 tm_list_add
[i
] = alloc_tree_mod_elem(dst
, i
+ dst_offset
,
869 MOD_LOG_KEY_ADD
, GFP_NOFS
);
870 if (!tm_list_add
[i
]) {
876 if (tree_mod_dont_log(fs_info
, NULL
))
880 for (i
= 0; i
< nr_items
; i
++) {
881 ret
= __tree_mod_log_insert(fs_info
, tm_list_rem
[i
]);
884 ret
= __tree_mod_log_insert(fs_info
, tm_list_add
[i
]);
889 tree_mod_log_write_unlock(fs_info
);
895 for (i
= 0; i
< nr_items
* 2; i
++) {
896 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
897 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
901 tree_mod_log_write_unlock(fs_info
);
908 tree_mod_log_eb_move(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*dst
,
909 int dst_offset
, int src_offset
, int nr_items
)
912 ret
= tree_mod_log_insert_move(fs_info
, dst
, dst_offset
, src_offset
,
918 tree_mod_log_set_node_key(struct btrfs_fs_info
*fs_info
,
919 struct extent_buffer
*eb
, int slot
, int atomic
)
923 ret
= tree_mod_log_insert_key(fs_info
, eb
, slot
,
925 atomic
? GFP_ATOMIC
: GFP_NOFS
);
930 tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
)
932 struct tree_mod_elem
**tm_list
= NULL
;
937 if (btrfs_header_level(eb
) == 0)
940 if (!tree_mod_need_log(fs_info
, NULL
))
943 nritems
= btrfs_header_nritems(eb
);
944 tm_list
= kzalloc(nritems
* sizeof(struct tree_mod_elem
*),
949 for (i
= 0; i
< nritems
; i
++) {
950 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
,
951 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
958 if (tree_mod_dont_log(fs_info
, eb
))
961 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
962 tree_mod_log_write_unlock(fs_info
);
970 for (i
= 0; i
< nritems
; i
++)
978 tree_mod_log_set_root_pointer(struct btrfs_root
*root
,
979 struct extent_buffer
*new_root_node
,
983 ret
= tree_mod_log_insert_root(root
->fs_info
, root
->node
,
984 new_root_node
, GFP_NOFS
, log_removal
);
989 * check if the tree block can be shared by multiple trees
991 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
992 struct extent_buffer
*buf
)
995 * Tree blocks not in refernece counted trees and tree roots
996 * are never shared. If a block was allocated after the last
997 * snapshot and the block was not allocated by tree relocation,
998 * we know the block is not shared.
1000 if (root
->ref_cows
&&
1001 buf
!= root
->node
&& buf
!= root
->commit_root
&&
1002 (btrfs_header_generation(buf
) <=
1003 btrfs_root_last_snapshot(&root
->root_item
) ||
1004 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
1006 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
1007 if (root
->ref_cows
&&
1008 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1014 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
1015 struct btrfs_root
*root
,
1016 struct extent_buffer
*buf
,
1017 struct extent_buffer
*cow
,
1027 * Backrefs update rules:
1029 * Always use full backrefs for extent pointers in tree block
1030 * allocated by tree relocation.
1032 * If a shared tree block is no longer referenced by its owner
1033 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
1034 * use full backrefs for extent pointers in tree block.
1036 * If a tree block is been relocating
1037 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
1038 * use full backrefs for extent pointers in tree block.
1039 * The reason for this is some operations (such as drop tree)
1040 * are only allowed for blocks use full backrefs.
1043 if (btrfs_block_can_be_shared(root
, buf
)) {
1044 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
1045 btrfs_header_level(buf
), 1,
1051 btrfs_std_error(root
->fs_info
, ret
);
1056 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1057 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1058 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1063 owner
= btrfs_header_owner(buf
);
1064 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
1065 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
1068 if ((owner
== root
->root_key
.objectid
||
1069 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
1070 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
1071 ret
= btrfs_inc_ref(trans
, root
, buf
, 1, 1);
1072 BUG_ON(ret
); /* -ENOMEM */
1074 if (root
->root_key
.objectid
==
1075 BTRFS_TREE_RELOC_OBJECTID
) {
1076 ret
= btrfs_dec_ref(trans
, root
, buf
, 0, 1);
1077 BUG_ON(ret
); /* -ENOMEM */
1078 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1079 BUG_ON(ret
); /* -ENOMEM */
1081 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
1084 if (root
->root_key
.objectid
==
1085 BTRFS_TREE_RELOC_OBJECTID
)
1086 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1088 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
1089 BUG_ON(ret
); /* -ENOMEM */
1091 if (new_flags
!= 0) {
1092 int level
= btrfs_header_level(buf
);
1094 ret
= btrfs_set_disk_extent_flags(trans
, root
,
1097 new_flags
, level
, 0);
1102 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
1103 if (root
->root_key
.objectid
==
1104 BTRFS_TREE_RELOC_OBJECTID
)
1105 ret
= btrfs_inc_ref(trans
, root
, cow
, 1, 1);
1107 ret
= btrfs_inc_ref(trans
, root
, cow
, 0, 1);
1108 BUG_ON(ret
); /* -ENOMEM */
1109 ret
= btrfs_dec_ref(trans
, root
, buf
, 1, 1);
1110 BUG_ON(ret
); /* -ENOMEM */
1112 clean_tree_block(trans
, root
, buf
);
1119 * does the dirty work in cow of a single block. The parent block (if
1120 * supplied) is updated to point to the new cow copy. The new buffer is marked
1121 * dirty and returned locked. If you modify the block it needs to be marked
1124 * search_start -- an allocation hint for the new block
1126 * empty_size -- a hint that you plan on doing more cow. This is the size in
1127 * bytes the allocator should try to find free next to the block it returns.
1128 * This is just a hint and may be ignored by the allocator.
1130 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1131 struct btrfs_root
*root
,
1132 struct extent_buffer
*buf
,
1133 struct extent_buffer
*parent
, int parent_slot
,
1134 struct extent_buffer
**cow_ret
,
1135 u64 search_start
, u64 empty_size
)
1137 struct btrfs_disk_key disk_key
;
1138 struct extent_buffer
*cow
;
1141 int unlock_orig
= 0;
1144 if (*cow_ret
== buf
)
1147 btrfs_assert_tree_locked(buf
);
1149 WARN_ON(root
->ref_cows
&& trans
->transid
!=
1150 root
->fs_info
->running_transaction
->transid
);
1151 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
1153 level
= btrfs_header_level(buf
);
1156 btrfs_item_key(buf
, &disk_key
, 0);
1158 btrfs_node_key(buf
, &disk_key
, 0);
1160 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
1162 parent_start
= parent
->start
;
1168 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
, parent_start
,
1169 root
->root_key
.objectid
, &disk_key
,
1170 level
, search_start
, empty_size
);
1172 return PTR_ERR(cow
);
1174 /* cow is set to blocking by btrfs_init_new_buffer */
1176 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
1177 btrfs_set_header_bytenr(cow
, cow
->start
);
1178 btrfs_set_header_generation(cow
, trans
->transid
);
1179 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1180 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1181 BTRFS_HEADER_FLAG_RELOC
);
1182 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1183 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1185 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1187 write_extent_buffer(cow
, root
->fs_info
->fsid
, btrfs_header_fsid(),
1190 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1192 btrfs_abort_transaction(trans
, root
, ret
);
1196 if (root
->ref_cows
) {
1197 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1202 if (buf
== root
->node
) {
1203 WARN_ON(parent
&& parent
!= buf
);
1204 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1205 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1206 parent_start
= buf
->start
;
1210 extent_buffer_get(cow
);
1211 tree_mod_log_set_root_pointer(root
, cow
, 1);
1212 rcu_assign_pointer(root
->node
, cow
);
1214 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1216 free_extent_buffer(buf
);
1217 add_root_to_dirty_list(root
);
1219 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1220 parent_start
= parent
->start
;
1224 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1225 tree_mod_log_insert_key(root
->fs_info
, parent
, parent_slot
,
1226 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1227 btrfs_set_node_blockptr(parent
, parent_slot
,
1229 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1231 btrfs_mark_buffer_dirty(parent
);
1233 ret
= tree_mod_log_free_eb(root
->fs_info
, buf
);
1235 btrfs_abort_transaction(trans
, root
, ret
);
1239 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1243 btrfs_tree_unlock(buf
);
1244 free_extent_buffer_stale(buf
);
1245 btrfs_mark_buffer_dirty(cow
);
1251 * returns the logical address of the oldest predecessor of the given root.
1252 * entries older than time_seq are ignored.
1254 static struct tree_mod_elem
*
1255 __tree_mod_log_oldest_root(struct btrfs_fs_info
*fs_info
,
1256 struct extent_buffer
*eb_root
, u64 time_seq
)
1258 struct tree_mod_elem
*tm
;
1259 struct tree_mod_elem
*found
= NULL
;
1260 u64 root_logical
= eb_root
->start
;
1267 * the very last operation that's logged for a root is the replacement
1268 * operation (if it is replaced at all). this has the index of the *new*
1269 * root, making it the very first operation that's logged for this root.
1272 tm
= tree_mod_log_search_oldest(fs_info
, root_logical
,
1277 * if there are no tree operation for the oldest root, we simply
1278 * return it. this should only happen if that (old) root is at
1285 * if there's an operation that's not a root replacement, we
1286 * found the oldest version of our root. normally, we'll find a
1287 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1289 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1293 root_logical
= tm
->old_root
.logical
;
1297 /* if there's no old root to return, return what we found instead */
1305 * tm is a pointer to the first operation to rewind within eb. then, all
1306 * previous operations will be rewinded (until we reach something older than
1310 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1311 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1314 struct rb_node
*next
;
1315 struct tree_mod_elem
*tm
= first_tm
;
1316 unsigned long o_dst
;
1317 unsigned long o_src
;
1318 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1320 n
= btrfs_header_nritems(eb
);
1321 tree_mod_log_read_lock(fs_info
);
1322 while (tm
&& tm
->seq
>= time_seq
) {
1324 * all the operations are recorded with the operator used for
1325 * the modification. as we're going backwards, we do the
1326 * opposite of each operation here.
1329 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1330 BUG_ON(tm
->slot
< n
);
1332 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1333 case MOD_LOG_KEY_REMOVE
:
1334 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1335 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1336 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1340 case MOD_LOG_KEY_REPLACE
:
1341 BUG_ON(tm
->slot
>= n
);
1342 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1343 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1344 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1347 case MOD_LOG_KEY_ADD
:
1348 /* if a move operation is needed it's in the log */
1351 case MOD_LOG_MOVE_KEYS
:
1352 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1353 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1354 memmove_extent_buffer(eb
, o_dst
, o_src
,
1355 tm
->move
.nr_items
* p_size
);
1357 case MOD_LOG_ROOT_REPLACE
:
1359 * this operation is special. for roots, this must be
1360 * handled explicitly before rewinding.
1361 * for non-roots, this operation may exist if the node
1362 * was a root: root A -> child B; then A gets empty and
1363 * B is promoted to the new root. in the mod log, we'll
1364 * have a root-replace operation for B, a tree block
1365 * that is no root. we simply ignore that operation.
1369 next
= rb_next(&tm
->node
);
1372 tm
= container_of(next
, struct tree_mod_elem
, node
);
1373 if (tm
->index
!= first_tm
->index
)
1376 tree_mod_log_read_unlock(fs_info
);
1377 btrfs_set_header_nritems(eb
, n
);
1381 * Called with eb read locked. If the buffer cannot be rewinded, the same buffer
1382 * is returned. If rewind operations happen, a fresh buffer is returned. The
1383 * returned buffer is always read-locked. If the returned buffer is not the
1384 * input buffer, the lock on the input buffer is released and the input buffer
1385 * is freed (its refcount is decremented).
1387 static struct extent_buffer
*
1388 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1389 struct extent_buffer
*eb
, u64 time_seq
)
1391 struct extent_buffer
*eb_rewin
;
1392 struct tree_mod_elem
*tm
;
1397 if (btrfs_header_level(eb
) == 0)
1400 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1404 btrfs_set_path_blocking(path
);
1405 btrfs_set_lock_blocking_rw(eb
, BTRFS_READ_LOCK
);
1407 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1408 BUG_ON(tm
->slot
!= 0);
1409 eb_rewin
= alloc_dummy_extent_buffer(eb
->start
,
1410 fs_info
->tree_root
->nodesize
);
1412 btrfs_tree_read_unlock_blocking(eb
);
1413 free_extent_buffer(eb
);
1416 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1417 btrfs_set_header_backref_rev(eb_rewin
,
1418 btrfs_header_backref_rev(eb
));
1419 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1420 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1422 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1424 btrfs_tree_read_unlock_blocking(eb
);
1425 free_extent_buffer(eb
);
1430 btrfs_clear_path_blocking(path
, NULL
, BTRFS_READ_LOCK
);
1431 btrfs_tree_read_unlock_blocking(eb
);
1432 free_extent_buffer(eb
);
1434 extent_buffer_get(eb_rewin
);
1435 btrfs_tree_read_lock(eb_rewin
);
1436 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1437 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1438 BTRFS_NODEPTRS_PER_BLOCK(fs_info
->tree_root
));
1444 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1445 * value. If there are no changes, the current root->root_node is returned. If
1446 * anything changed in between, there's a fresh buffer allocated on which the
1447 * rewind operations are done. In any case, the returned buffer is read locked.
1448 * Returns NULL on error (with no locks held).
1450 static inline struct extent_buffer
*
1451 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1453 struct tree_mod_elem
*tm
;
1454 struct extent_buffer
*eb
= NULL
;
1455 struct extent_buffer
*eb_root
;
1456 struct extent_buffer
*old
;
1457 struct tree_mod_root
*old_root
= NULL
;
1458 u64 old_generation
= 0;
1462 eb_root
= btrfs_read_lock_root_node(root
);
1463 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1467 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1468 old_root
= &tm
->old_root
;
1469 old_generation
= tm
->generation
;
1470 logical
= old_root
->logical
;
1472 logical
= eb_root
->start
;
1475 tm
= tree_mod_log_search(root
->fs_info
, logical
, time_seq
);
1476 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1477 btrfs_tree_read_unlock(eb_root
);
1478 free_extent_buffer(eb_root
);
1479 blocksize
= btrfs_level_size(root
, old_root
->level
);
1480 old
= read_tree_block(root
, logical
, blocksize
, 0);
1481 if (WARN_ON(!old
|| !extent_buffer_uptodate(old
))) {
1482 free_extent_buffer(old
);
1483 btrfs_warn(root
->fs_info
,
1484 "failed to read tree block %llu from get_old_root", logical
);
1486 eb
= btrfs_clone_extent_buffer(old
);
1487 free_extent_buffer(old
);
1489 } else if (old_root
) {
1490 btrfs_tree_read_unlock(eb_root
);
1491 free_extent_buffer(eb_root
);
1492 eb
= alloc_dummy_extent_buffer(logical
, root
->nodesize
);
1494 btrfs_set_lock_blocking_rw(eb_root
, BTRFS_READ_LOCK
);
1495 eb
= btrfs_clone_extent_buffer(eb_root
);
1496 btrfs_tree_read_unlock_blocking(eb_root
);
1497 free_extent_buffer(eb_root
);
1502 extent_buffer_get(eb
);
1503 btrfs_tree_read_lock(eb
);
1505 btrfs_set_header_bytenr(eb
, eb
->start
);
1506 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1507 btrfs_set_header_owner(eb
, btrfs_header_owner(eb_root
));
1508 btrfs_set_header_level(eb
, old_root
->level
);
1509 btrfs_set_header_generation(eb
, old_generation
);
1512 __tree_mod_log_rewind(root
->fs_info
, eb
, time_seq
, tm
);
1514 WARN_ON(btrfs_header_level(eb
) != 0);
1515 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(root
));
1520 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1522 struct tree_mod_elem
*tm
;
1524 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1526 tm
= __tree_mod_log_oldest_root(root
->fs_info
, eb_root
, time_seq
);
1527 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1528 level
= tm
->old_root
.level
;
1530 level
= btrfs_header_level(eb_root
);
1532 free_extent_buffer(eb_root
);
1537 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1538 struct btrfs_root
*root
,
1539 struct extent_buffer
*buf
)
1541 /* ensure we can see the force_cow */
1545 * We do not need to cow a block if
1546 * 1) this block is not created or changed in this transaction;
1547 * 2) this block does not belong to TREE_RELOC tree;
1548 * 3) the root is not forced COW.
1550 * What is forced COW:
1551 * when we create snapshot during commiting the transaction,
1552 * after we've finished coping src root, we must COW the shared
1553 * block to ensure the metadata consistency.
1555 if (btrfs_header_generation(buf
) == trans
->transid
&&
1556 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1557 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1558 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1565 * cows a single block, see __btrfs_cow_block for the real work.
1566 * This version of it has extra checks so that a block isn't cow'd more than
1567 * once per transaction, as long as it hasn't been written yet
1569 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1570 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1571 struct extent_buffer
*parent
, int parent_slot
,
1572 struct extent_buffer
**cow_ret
)
1577 if (trans
->transaction
!= root
->fs_info
->running_transaction
)
1578 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1580 root
->fs_info
->running_transaction
->transid
);
1582 if (trans
->transid
!= root
->fs_info
->generation
)
1583 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1584 trans
->transid
, root
->fs_info
->generation
);
1586 if (!should_cow_block(trans
, root
, buf
)) {
1591 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
1594 btrfs_set_lock_blocking(parent
);
1595 btrfs_set_lock_blocking(buf
);
1597 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1598 parent_slot
, cow_ret
, search_start
, 0);
1600 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1606 * helper function for defrag to decide if two blocks pointed to by a
1607 * node are actually close by
1609 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1611 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1613 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1619 * compare two keys in a memcmp fashion
1621 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
1623 struct btrfs_key k1
;
1625 btrfs_disk_key_to_cpu(&k1
, disk
);
1627 return btrfs_comp_cpu_keys(&k1
, k2
);
1631 * same as comp_keys only with two btrfs_key's
1633 int btrfs_comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
1635 if (k1
->objectid
> k2
->objectid
)
1637 if (k1
->objectid
< k2
->objectid
)
1639 if (k1
->type
> k2
->type
)
1641 if (k1
->type
< k2
->type
)
1643 if (k1
->offset
> k2
->offset
)
1645 if (k1
->offset
< k2
->offset
)
1651 * this is used by the defrag code to go through all the
1652 * leaves pointed to by a node and reallocate them so that
1653 * disk order is close to key order
1655 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1656 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1657 int start_slot
, u64
*last_ret
,
1658 struct btrfs_key
*progress
)
1660 struct extent_buffer
*cur
;
1663 u64 search_start
= *last_ret
;
1673 int progress_passed
= 0;
1674 struct btrfs_disk_key disk_key
;
1676 parent_level
= btrfs_header_level(parent
);
1678 WARN_ON(trans
->transaction
!= root
->fs_info
->running_transaction
);
1679 WARN_ON(trans
->transid
!= root
->fs_info
->generation
);
1681 parent_nritems
= btrfs_header_nritems(parent
);
1682 blocksize
= btrfs_level_size(root
, parent_level
- 1);
1683 end_slot
= parent_nritems
;
1685 if (parent_nritems
== 1)
1688 btrfs_set_lock_blocking(parent
);
1690 for (i
= start_slot
; i
< end_slot
; i
++) {
1693 btrfs_node_key(parent
, &disk_key
, i
);
1694 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1697 progress_passed
= 1;
1698 blocknr
= btrfs_node_blockptr(parent
, i
);
1699 gen
= btrfs_node_ptr_generation(parent
, i
);
1700 if (last_block
== 0)
1701 last_block
= blocknr
;
1704 other
= btrfs_node_blockptr(parent
, i
- 1);
1705 close
= close_blocks(blocknr
, other
, blocksize
);
1707 if (!close
&& i
< end_slot
- 2) {
1708 other
= btrfs_node_blockptr(parent
, i
+ 1);
1709 close
= close_blocks(blocknr
, other
, blocksize
);
1712 last_block
= blocknr
;
1716 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1718 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1721 if (!cur
|| !uptodate
) {
1723 cur
= read_tree_block(root
, blocknr
,
1725 if (!cur
|| !extent_buffer_uptodate(cur
)) {
1726 free_extent_buffer(cur
);
1729 } else if (!uptodate
) {
1730 err
= btrfs_read_buffer(cur
, gen
);
1732 free_extent_buffer(cur
);
1737 if (search_start
== 0)
1738 search_start
= last_block
;
1740 btrfs_tree_lock(cur
);
1741 btrfs_set_lock_blocking(cur
);
1742 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1745 (end_slot
- i
) * blocksize
));
1747 btrfs_tree_unlock(cur
);
1748 free_extent_buffer(cur
);
1751 search_start
= cur
->start
;
1752 last_block
= cur
->start
;
1753 *last_ret
= search_start
;
1754 btrfs_tree_unlock(cur
);
1755 free_extent_buffer(cur
);
1761 * The leaf data grows from end-to-front in the node.
1762 * this returns the address of the start of the last item,
1763 * which is the stop of the leaf data stack
1765 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
1766 struct extent_buffer
*leaf
)
1768 u32 nr
= btrfs_header_nritems(leaf
);
1770 return BTRFS_LEAF_DATA_SIZE(root
);
1771 return btrfs_item_offset_nr(leaf
, nr
- 1);
1776 * search for key in the extent_buffer. The items start at offset p,
1777 * and they are item_size apart. There are 'max' items in p.
1779 * the slot in the array is returned via slot, and it points to
1780 * the place where you would insert key if it is not found in
1783 * slot may point to max if the key is bigger than all of the keys
1785 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1787 int item_size
, struct btrfs_key
*key
,
1794 struct btrfs_disk_key
*tmp
= NULL
;
1795 struct btrfs_disk_key unaligned
;
1796 unsigned long offset
;
1798 unsigned long map_start
= 0;
1799 unsigned long map_len
= 0;
1802 while (low
< high
) {
1803 mid
= (low
+ high
) / 2;
1804 offset
= p
+ mid
* item_size
;
1806 if (!kaddr
|| offset
< map_start
||
1807 (offset
+ sizeof(struct btrfs_disk_key
)) >
1808 map_start
+ map_len
) {
1810 err
= map_private_extent_buffer(eb
, offset
,
1811 sizeof(struct btrfs_disk_key
),
1812 &kaddr
, &map_start
, &map_len
);
1815 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1818 read_extent_buffer(eb
, &unaligned
,
1819 offset
, sizeof(unaligned
));
1824 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
1827 ret
= comp_keys(tmp
, key
);
1843 * simple bin_search frontend that does the right thing for
1846 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1847 int level
, int *slot
)
1850 return generic_bin_search(eb
,
1851 offsetof(struct btrfs_leaf
, items
),
1852 sizeof(struct btrfs_item
),
1853 key
, btrfs_header_nritems(eb
),
1856 return generic_bin_search(eb
,
1857 offsetof(struct btrfs_node
, ptrs
),
1858 sizeof(struct btrfs_key_ptr
),
1859 key
, btrfs_header_nritems(eb
),
1863 int btrfs_bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
1864 int level
, int *slot
)
1866 return bin_search(eb
, key
, level
, slot
);
1869 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1871 spin_lock(&root
->accounting_lock
);
1872 btrfs_set_root_used(&root
->root_item
,
1873 btrfs_root_used(&root
->root_item
) + size
);
1874 spin_unlock(&root
->accounting_lock
);
1877 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1879 spin_lock(&root
->accounting_lock
);
1880 btrfs_set_root_used(&root
->root_item
,
1881 btrfs_root_used(&root
->root_item
) - size
);
1882 spin_unlock(&root
->accounting_lock
);
1885 /* given a node and slot number, this reads the blocks it points to. The
1886 * extent buffer is returned with a reference taken (but unlocked).
1887 * NULL is returned on error.
1889 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
1890 struct extent_buffer
*parent
, int slot
)
1892 int level
= btrfs_header_level(parent
);
1893 struct extent_buffer
*eb
;
1897 if (slot
>= btrfs_header_nritems(parent
))
1902 eb
= read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
1903 btrfs_level_size(root
, level
- 1),
1904 btrfs_node_ptr_generation(parent
, slot
));
1905 if (eb
&& !extent_buffer_uptodate(eb
)) {
1906 free_extent_buffer(eb
);
1914 * node level balancing, used to make sure nodes are in proper order for
1915 * item deletion. We balance from the top down, so we have to make sure
1916 * that a deletion won't leave an node completely empty later on.
1918 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1919 struct btrfs_root
*root
,
1920 struct btrfs_path
*path
, int level
)
1922 struct extent_buffer
*right
= NULL
;
1923 struct extent_buffer
*mid
;
1924 struct extent_buffer
*left
= NULL
;
1925 struct extent_buffer
*parent
= NULL
;
1929 int orig_slot
= path
->slots
[level
];
1935 mid
= path
->nodes
[level
];
1937 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1938 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1939 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1941 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1943 if (level
< BTRFS_MAX_LEVEL
- 1) {
1944 parent
= path
->nodes
[level
+ 1];
1945 pslot
= path
->slots
[level
+ 1];
1949 * deal with the case where there is only one pointer in the root
1950 * by promoting the node below to a root
1953 struct extent_buffer
*child
;
1955 if (btrfs_header_nritems(mid
) != 1)
1958 /* promote the child to a root */
1959 child
= read_node_slot(root
, mid
, 0);
1962 btrfs_std_error(root
->fs_info
, ret
);
1966 btrfs_tree_lock(child
);
1967 btrfs_set_lock_blocking(child
);
1968 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1970 btrfs_tree_unlock(child
);
1971 free_extent_buffer(child
);
1975 tree_mod_log_set_root_pointer(root
, child
, 1);
1976 rcu_assign_pointer(root
->node
, child
);
1978 add_root_to_dirty_list(root
);
1979 btrfs_tree_unlock(child
);
1981 path
->locks
[level
] = 0;
1982 path
->nodes
[level
] = NULL
;
1983 clean_tree_block(trans
, root
, mid
);
1984 btrfs_tree_unlock(mid
);
1985 /* once for the path */
1986 free_extent_buffer(mid
);
1988 root_sub_used(root
, mid
->len
);
1989 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1990 /* once for the root ptr */
1991 free_extent_buffer_stale(mid
);
1994 if (btrfs_header_nritems(mid
) >
1995 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
1998 left
= read_node_slot(root
, parent
, pslot
- 1);
2000 btrfs_tree_lock(left
);
2001 btrfs_set_lock_blocking(left
);
2002 wret
= btrfs_cow_block(trans
, root
, left
,
2003 parent
, pslot
- 1, &left
);
2009 right
= read_node_slot(root
, parent
, pslot
+ 1);
2011 btrfs_tree_lock(right
);
2012 btrfs_set_lock_blocking(right
);
2013 wret
= btrfs_cow_block(trans
, root
, right
,
2014 parent
, pslot
+ 1, &right
);
2021 /* first, try to make some room in the middle buffer */
2023 orig_slot
+= btrfs_header_nritems(left
);
2024 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2030 * then try to empty the right most buffer into the middle
2033 wret
= push_node_left(trans
, root
, mid
, right
, 1);
2034 if (wret
< 0 && wret
!= -ENOSPC
)
2036 if (btrfs_header_nritems(right
) == 0) {
2037 clean_tree_block(trans
, root
, right
);
2038 btrfs_tree_unlock(right
);
2039 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
2040 root_sub_used(root
, right
->len
);
2041 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
2042 free_extent_buffer_stale(right
);
2045 struct btrfs_disk_key right_key
;
2046 btrfs_node_key(right
, &right_key
, 0);
2047 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2049 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
2050 btrfs_mark_buffer_dirty(parent
);
2053 if (btrfs_header_nritems(mid
) == 1) {
2055 * we're not allowed to leave a node with one item in the
2056 * tree during a delete. A deletion from lower in the tree
2057 * could try to delete the only pointer in this node.
2058 * So, pull some keys from the left.
2059 * There has to be a left pointer at this point because
2060 * otherwise we would have pulled some pointers from the
2065 btrfs_std_error(root
->fs_info
, ret
);
2068 wret
= balance_node_right(trans
, root
, mid
, left
);
2074 wret
= push_node_left(trans
, root
, left
, mid
, 1);
2080 if (btrfs_header_nritems(mid
) == 0) {
2081 clean_tree_block(trans
, root
, mid
);
2082 btrfs_tree_unlock(mid
);
2083 del_ptr(root
, path
, level
+ 1, pslot
);
2084 root_sub_used(root
, mid
->len
);
2085 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
2086 free_extent_buffer_stale(mid
);
2089 /* update the parent key to reflect our changes */
2090 struct btrfs_disk_key mid_key
;
2091 btrfs_node_key(mid
, &mid_key
, 0);
2092 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2094 btrfs_set_node_key(parent
, &mid_key
, pslot
);
2095 btrfs_mark_buffer_dirty(parent
);
2098 /* update the path */
2100 if (btrfs_header_nritems(left
) > orig_slot
) {
2101 extent_buffer_get(left
);
2102 /* left was locked after cow */
2103 path
->nodes
[level
] = left
;
2104 path
->slots
[level
+ 1] -= 1;
2105 path
->slots
[level
] = orig_slot
;
2107 btrfs_tree_unlock(mid
);
2108 free_extent_buffer(mid
);
2111 orig_slot
-= btrfs_header_nritems(left
);
2112 path
->slots
[level
] = orig_slot
;
2115 /* double check we haven't messed things up */
2117 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2121 btrfs_tree_unlock(right
);
2122 free_extent_buffer(right
);
2125 if (path
->nodes
[level
] != left
)
2126 btrfs_tree_unlock(left
);
2127 free_extent_buffer(left
);
2132 /* Node balancing for insertion. Here we only split or push nodes around
2133 * when they are completely full. This is also done top down, so we
2134 * have to be pessimistic.
2136 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2137 struct btrfs_root
*root
,
2138 struct btrfs_path
*path
, int level
)
2140 struct extent_buffer
*right
= NULL
;
2141 struct extent_buffer
*mid
;
2142 struct extent_buffer
*left
= NULL
;
2143 struct extent_buffer
*parent
= NULL
;
2147 int orig_slot
= path
->slots
[level
];
2152 mid
= path
->nodes
[level
];
2153 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2155 if (level
< BTRFS_MAX_LEVEL
- 1) {
2156 parent
= path
->nodes
[level
+ 1];
2157 pslot
= path
->slots
[level
+ 1];
2163 left
= read_node_slot(root
, parent
, pslot
- 1);
2165 /* first, try to make some room in the middle buffer */
2169 btrfs_tree_lock(left
);
2170 btrfs_set_lock_blocking(left
);
2172 left_nr
= btrfs_header_nritems(left
);
2173 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2176 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2181 wret
= push_node_left(trans
, root
,
2188 struct btrfs_disk_key disk_key
;
2189 orig_slot
+= left_nr
;
2190 btrfs_node_key(mid
, &disk_key
, 0);
2191 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2193 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2194 btrfs_mark_buffer_dirty(parent
);
2195 if (btrfs_header_nritems(left
) > orig_slot
) {
2196 path
->nodes
[level
] = left
;
2197 path
->slots
[level
+ 1] -= 1;
2198 path
->slots
[level
] = orig_slot
;
2199 btrfs_tree_unlock(mid
);
2200 free_extent_buffer(mid
);
2203 btrfs_header_nritems(left
);
2204 path
->slots
[level
] = orig_slot
;
2205 btrfs_tree_unlock(left
);
2206 free_extent_buffer(left
);
2210 btrfs_tree_unlock(left
);
2211 free_extent_buffer(left
);
2213 right
= read_node_slot(root
, parent
, pslot
+ 1);
2216 * then try to empty the right most buffer into the middle
2221 btrfs_tree_lock(right
);
2222 btrfs_set_lock_blocking(right
);
2224 right_nr
= btrfs_header_nritems(right
);
2225 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
2228 ret
= btrfs_cow_block(trans
, root
, right
,
2234 wret
= balance_node_right(trans
, root
,
2241 struct btrfs_disk_key disk_key
;
2243 btrfs_node_key(right
, &disk_key
, 0);
2244 tree_mod_log_set_node_key(root
->fs_info
, parent
,
2246 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2247 btrfs_mark_buffer_dirty(parent
);
2249 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2250 path
->nodes
[level
] = right
;
2251 path
->slots
[level
+ 1] += 1;
2252 path
->slots
[level
] = orig_slot
-
2253 btrfs_header_nritems(mid
);
2254 btrfs_tree_unlock(mid
);
2255 free_extent_buffer(mid
);
2257 btrfs_tree_unlock(right
);
2258 free_extent_buffer(right
);
2262 btrfs_tree_unlock(right
);
2263 free_extent_buffer(right
);
2269 * readahead one full node of leaves, finding things that are close
2270 * to the block in 'slot', and triggering ra on them.
2272 static void reada_for_search(struct btrfs_root
*root
,
2273 struct btrfs_path
*path
,
2274 int level
, int slot
, u64 objectid
)
2276 struct extent_buffer
*node
;
2277 struct btrfs_disk_key disk_key
;
2283 int direction
= path
->reada
;
2284 struct extent_buffer
*eb
;
2292 if (!path
->nodes
[level
])
2295 node
= path
->nodes
[level
];
2297 search
= btrfs_node_blockptr(node
, slot
);
2298 blocksize
= btrfs_level_size(root
, level
- 1);
2299 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
2301 free_extent_buffer(eb
);
2307 nritems
= btrfs_header_nritems(node
);
2311 if (direction
< 0) {
2315 } else if (direction
> 0) {
2320 if (path
->reada
< 0 && objectid
) {
2321 btrfs_node_key(node
, &disk_key
, nr
);
2322 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2325 search
= btrfs_node_blockptr(node
, nr
);
2326 if ((search
<= target
&& target
- search
<= 65536) ||
2327 (search
> target
&& search
- target
<= 65536)) {
2328 gen
= btrfs_node_ptr_generation(node
, nr
);
2329 readahead_tree_block(root
, search
, blocksize
, gen
);
2333 if ((nread
> 65536 || nscan
> 32))
2338 static noinline
void reada_for_balance(struct btrfs_root
*root
,
2339 struct btrfs_path
*path
, int level
)
2343 struct extent_buffer
*parent
;
2344 struct extent_buffer
*eb
;
2350 parent
= path
->nodes
[level
+ 1];
2354 nritems
= btrfs_header_nritems(parent
);
2355 slot
= path
->slots
[level
+ 1];
2356 blocksize
= btrfs_level_size(root
, level
);
2359 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2360 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2361 eb
= btrfs_find_tree_block(root
, block1
, blocksize
);
2363 * if we get -eagain from btrfs_buffer_uptodate, we
2364 * don't want to return eagain here. That will loop
2367 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2369 free_extent_buffer(eb
);
2371 if (slot
+ 1 < nritems
) {
2372 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2373 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2374 eb
= btrfs_find_tree_block(root
, block2
, blocksize
);
2375 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2377 free_extent_buffer(eb
);
2381 readahead_tree_block(root
, block1
, blocksize
, 0);
2383 readahead_tree_block(root
, block2
, blocksize
, 0);
2388 * when we walk down the tree, it is usually safe to unlock the higher layers
2389 * in the tree. The exceptions are when our path goes through slot 0, because
2390 * operations on the tree might require changing key pointers higher up in the
2393 * callers might also have set path->keep_locks, which tells this code to keep
2394 * the lock if the path points to the last slot in the block. This is part of
2395 * walking through the tree, and selecting the next slot in the higher block.
2397 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2398 * if lowest_unlock is 1, level 0 won't be unlocked
2400 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2401 int lowest_unlock
, int min_write_lock_level
,
2402 int *write_lock_level
)
2405 int skip_level
= level
;
2407 struct extent_buffer
*t
;
2409 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2410 if (!path
->nodes
[i
])
2412 if (!path
->locks
[i
])
2414 if (!no_skips
&& path
->slots
[i
] == 0) {
2418 if (!no_skips
&& path
->keep_locks
) {
2421 nritems
= btrfs_header_nritems(t
);
2422 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2427 if (skip_level
< i
&& i
>= lowest_unlock
)
2431 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
2432 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2434 if (write_lock_level
&&
2435 i
> min_write_lock_level
&&
2436 i
<= *write_lock_level
) {
2437 *write_lock_level
= i
- 1;
2444 * This releases any locks held in the path starting at level and
2445 * going all the way up to the root.
2447 * btrfs_search_slot will keep the lock held on higher nodes in a few
2448 * corner cases, such as COW of the block at slot zero in the node. This
2449 * ignores those rules, and it should only be called when there are no
2450 * more updates to be done higher up in the tree.
2452 noinline
void btrfs_unlock_up_safe(struct btrfs_path
*path
, int level
)
2456 if (path
->keep_locks
)
2459 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2460 if (!path
->nodes
[i
])
2462 if (!path
->locks
[i
])
2464 btrfs_tree_unlock_rw(path
->nodes
[i
], path
->locks
[i
]);
2470 * helper function for btrfs_search_slot. The goal is to find a block
2471 * in cache without setting the path to blocking. If we find the block
2472 * we return zero and the path is unchanged.
2474 * If we can't find the block, we set the path blocking and do some
2475 * reada. -EAGAIN is returned and the search must be repeated.
2478 read_block_for_search(struct btrfs_trans_handle
*trans
,
2479 struct btrfs_root
*root
, struct btrfs_path
*p
,
2480 struct extent_buffer
**eb_ret
, int level
, int slot
,
2481 struct btrfs_key
*key
, u64 time_seq
)
2486 struct extent_buffer
*b
= *eb_ret
;
2487 struct extent_buffer
*tmp
;
2490 blocknr
= btrfs_node_blockptr(b
, slot
);
2491 gen
= btrfs_node_ptr_generation(b
, slot
);
2492 blocksize
= btrfs_level_size(root
, level
- 1);
2494 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
2496 /* first we do an atomic uptodate check */
2497 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2502 /* the pages were up to date, but we failed
2503 * the generation number check. Do a full
2504 * read for the generation number that is correct.
2505 * We must do this without dropping locks so
2506 * we can trust our generation number
2508 btrfs_set_path_blocking(p
);
2510 /* now we're allowed to do a blocking uptodate check */
2511 ret
= btrfs_read_buffer(tmp
, gen
);
2516 free_extent_buffer(tmp
);
2517 btrfs_release_path(p
);
2522 * reduce lock contention at high levels
2523 * of the btree by dropping locks before
2524 * we read. Don't release the lock on the current
2525 * level because we need to walk this node to figure
2526 * out which blocks to read.
2528 btrfs_unlock_up_safe(p
, level
+ 1);
2529 btrfs_set_path_blocking(p
);
2531 free_extent_buffer(tmp
);
2533 reada_for_search(root
, p
, level
, slot
, key
->objectid
);
2535 btrfs_release_path(p
);
2538 tmp
= read_tree_block(root
, blocknr
, blocksize
, 0);
2541 * If the read above didn't mark this buffer up to date,
2542 * it will never end up being up to date. Set ret to EIO now
2543 * and give up so that our caller doesn't loop forever
2546 if (!btrfs_buffer_uptodate(tmp
, 0, 0))
2548 free_extent_buffer(tmp
);
2554 * helper function for btrfs_search_slot. This does all of the checks
2555 * for node-level blocks and does any balancing required based on
2558 * If no extra work was required, zero is returned. If we had to
2559 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2563 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2564 struct btrfs_root
*root
, struct btrfs_path
*p
,
2565 struct extent_buffer
*b
, int level
, int ins_len
,
2566 int *write_lock_level
)
2569 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2570 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
2573 if (*write_lock_level
< level
+ 1) {
2574 *write_lock_level
= level
+ 1;
2575 btrfs_release_path(p
);
2579 btrfs_set_path_blocking(p
);
2580 reada_for_balance(root
, p
, level
);
2581 sret
= split_node(trans
, root
, p
, level
);
2582 btrfs_clear_path_blocking(p
, NULL
, 0);
2589 b
= p
->nodes
[level
];
2590 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2591 BTRFS_NODEPTRS_PER_BLOCK(root
) / 2) {
2594 if (*write_lock_level
< level
+ 1) {
2595 *write_lock_level
= level
+ 1;
2596 btrfs_release_path(p
);
2600 btrfs_set_path_blocking(p
);
2601 reada_for_balance(root
, p
, level
);
2602 sret
= balance_level(trans
, root
, p
, level
);
2603 btrfs_clear_path_blocking(p
, NULL
, 0);
2609 b
= p
->nodes
[level
];
2611 btrfs_release_path(p
);
2614 BUG_ON(btrfs_header_nritems(b
) == 1);
2624 static void key_search_validate(struct extent_buffer
*b
,
2625 struct btrfs_key
*key
,
2628 #ifdef CONFIG_BTRFS_ASSERT
2629 struct btrfs_disk_key disk_key
;
2631 btrfs_cpu_key_to_disk(&disk_key
, key
);
2634 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2635 offsetof(struct btrfs_leaf
, items
[0].key
),
2638 ASSERT(!memcmp_extent_buffer(b
, &disk_key
,
2639 offsetof(struct btrfs_node
, ptrs
[0].key
),
2644 static int key_search(struct extent_buffer
*b
, struct btrfs_key
*key
,
2645 int level
, int *prev_cmp
, int *slot
)
2647 if (*prev_cmp
!= 0) {
2648 *prev_cmp
= bin_search(b
, key
, level
, slot
);
2652 key_search_validate(b
, key
, level
);
2658 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*found_path
,
2659 u64 iobjectid
, u64 ioff
, u8 key_type
,
2660 struct btrfs_key
*found_key
)
2663 struct btrfs_key key
;
2664 struct extent_buffer
*eb
;
2665 struct btrfs_path
*path
;
2667 key
.type
= key_type
;
2668 key
.objectid
= iobjectid
;
2671 if (found_path
== NULL
) {
2672 path
= btrfs_alloc_path();
2678 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2679 if ((ret
< 0) || (found_key
== NULL
)) {
2680 if (path
!= found_path
)
2681 btrfs_free_path(path
);
2685 eb
= path
->nodes
[0];
2686 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2687 ret
= btrfs_next_leaf(fs_root
, path
);
2690 eb
= path
->nodes
[0];
2693 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2694 if (found_key
->type
!= key
.type
||
2695 found_key
->objectid
!= key
.objectid
)
2702 * look for key in the tree. path is filled in with nodes along the way
2703 * if key is found, we return zero and you can find the item in the leaf
2704 * level of the path (level 0)
2706 * If the key isn't found, the path points to the slot where it should
2707 * be inserted, and 1 is returned. If there are other errors during the
2708 * search a negative error number is returned.
2710 * if ins_len > 0, nodes and leaves will be split as we walk down the
2711 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
2714 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
2715 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
2718 struct extent_buffer
*b
;
2723 int lowest_unlock
= 1;
2725 /* everything at write_lock_level or lower must be write locked */
2726 int write_lock_level
= 0;
2727 u8 lowest_level
= 0;
2728 int min_write_lock_level
;
2731 lowest_level
= p
->lowest_level
;
2732 WARN_ON(lowest_level
&& ins_len
> 0);
2733 WARN_ON(p
->nodes
[0] != NULL
);
2734 BUG_ON(!cow
&& ins_len
);
2739 /* when we are removing items, we might have to go up to level
2740 * two as we update tree pointers Make sure we keep write
2741 * for those levels as well
2743 write_lock_level
= 2;
2744 } else if (ins_len
> 0) {
2746 * for inserting items, make sure we have a write lock on
2747 * level 1 so we can update keys
2749 write_lock_level
= 1;
2753 write_lock_level
= -1;
2755 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2756 write_lock_level
= BTRFS_MAX_LEVEL
;
2758 min_write_lock_level
= write_lock_level
;
2763 * we try very hard to do read locks on the root
2765 root_lock
= BTRFS_READ_LOCK
;
2767 if (p
->search_commit_root
) {
2769 * the commit roots are read only
2770 * so we always do read locks
2772 b
= root
->commit_root
;
2773 extent_buffer_get(b
);
2774 level
= btrfs_header_level(b
);
2775 if (!p
->skip_locking
)
2776 btrfs_tree_read_lock(b
);
2778 if (p
->skip_locking
) {
2779 b
= btrfs_root_node(root
);
2780 level
= btrfs_header_level(b
);
2782 /* we don't know the level of the root node
2783 * until we actually have it read locked
2785 b
= btrfs_read_lock_root_node(root
);
2786 level
= btrfs_header_level(b
);
2787 if (level
<= write_lock_level
) {
2788 /* whoops, must trade for write lock */
2789 btrfs_tree_read_unlock(b
);
2790 free_extent_buffer(b
);
2791 b
= btrfs_lock_root_node(root
);
2792 root_lock
= BTRFS_WRITE_LOCK
;
2794 /* the level might have changed, check again */
2795 level
= btrfs_header_level(b
);
2799 p
->nodes
[level
] = b
;
2800 if (!p
->skip_locking
)
2801 p
->locks
[level
] = root_lock
;
2804 level
= btrfs_header_level(b
);
2807 * setup the path here so we can release it under lock
2808 * contention with the cow code
2812 * if we don't really need to cow this block
2813 * then we don't want to set the path blocking,
2814 * so we test it here
2816 if (!should_cow_block(trans
, root
, b
))
2819 btrfs_set_path_blocking(p
);
2822 * must have write locks on this node and the
2825 if (level
> write_lock_level
||
2826 (level
+ 1 > write_lock_level
&&
2827 level
+ 1 < BTRFS_MAX_LEVEL
&&
2828 p
->nodes
[level
+ 1])) {
2829 write_lock_level
= level
+ 1;
2830 btrfs_release_path(p
);
2834 err
= btrfs_cow_block(trans
, root
, b
,
2835 p
->nodes
[level
+ 1],
2836 p
->slots
[level
+ 1], &b
);
2843 p
->nodes
[level
] = b
;
2844 btrfs_clear_path_blocking(p
, NULL
, 0);
2847 * we have a lock on b and as long as we aren't changing
2848 * the tree, there is no way to for the items in b to change.
2849 * It is safe to drop the lock on our parent before we
2850 * go through the expensive btree search on b.
2852 * If we're inserting or deleting (ins_len != 0), then we might
2853 * be changing slot zero, which may require changing the parent.
2854 * So, we can't drop the lock until after we know which slot
2855 * we're operating on.
2857 if (!ins_len
&& !p
->keep_locks
) {
2860 if (u
< BTRFS_MAX_LEVEL
&& p
->locks
[u
]) {
2861 btrfs_tree_unlock_rw(p
->nodes
[u
], p
->locks
[u
]);
2866 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
2870 if (ret
&& slot
> 0) {
2874 p
->slots
[level
] = slot
;
2875 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
,
2876 ins_len
, &write_lock_level
);
2883 b
= p
->nodes
[level
];
2884 slot
= p
->slots
[level
];
2887 * slot 0 is special, if we change the key
2888 * we have to update the parent pointer
2889 * which means we must have a write lock
2892 if (slot
== 0 && ins_len
&&
2893 write_lock_level
< level
+ 1) {
2894 write_lock_level
= level
+ 1;
2895 btrfs_release_path(p
);
2899 unlock_up(p
, level
, lowest_unlock
,
2900 min_write_lock_level
, &write_lock_level
);
2902 if (level
== lowest_level
) {
2908 err
= read_block_for_search(trans
, root
, p
,
2909 &b
, level
, slot
, key
, 0);
2917 if (!p
->skip_locking
) {
2918 level
= btrfs_header_level(b
);
2919 if (level
<= write_lock_level
) {
2920 err
= btrfs_try_tree_write_lock(b
);
2922 btrfs_set_path_blocking(p
);
2924 btrfs_clear_path_blocking(p
, b
,
2927 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2929 err
= btrfs_try_tree_read_lock(b
);
2931 btrfs_set_path_blocking(p
);
2932 btrfs_tree_read_lock(b
);
2933 btrfs_clear_path_blocking(p
, b
,
2936 p
->locks
[level
] = BTRFS_READ_LOCK
;
2938 p
->nodes
[level
] = b
;
2941 p
->slots
[level
] = slot
;
2943 btrfs_leaf_free_space(root
, b
) < ins_len
) {
2944 if (write_lock_level
< 1) {
2945 write_lock_level
= 1;
2946 btrfs_release_path(p
);
2950 btrfs_set_path_blocking(p
);
2951 err
= split_leaf(trans
, root
, key
,
2952 p
, ins_len
, ret
== 0);
2953 btrfs_clear_path_blocking(p
, NULL
, 0);
2961 if (!p
->search_for_split
)
2962 unlock_up(p
, level
, lowest_unlock
,
2963 min_write_lock_level
, &write_lock_level
);
2970 * we don't really know what they plan on doing with the path
2971 * from here on, so for now just mark it as blocking
2973 if (!p
->leave_spinning
)
2974 btrfs_set_path_blocking(p
);
2976 btrfs_release_path(p
);
2981 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2982 * current state of the tree together with the operations recorded in the tree
2983 * modification log to search for the key in a previous version of this tree, as
2984 * denoted by the time_seq parameter.
2986 * Naturally, there is no support for insert, delete or cow operations.
2988 * The resulting path and return value will be set up as if we called
2989 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2991 int btrfs_search_old_slot(struct btrfs_root
*root
, struct btrfs_key
*key
,
2992 struct btrfs_path
*p
, u64 time_seq
)
2994 struct extent_buffer
*b
;
2999 int lowest_unlock
= 1;
3000 u8 lowest_level
= 0;
3003 lowest_level
= p
->lowest_level
;
3004 WARN_ON(p
->nodes
[0] != NULL
);
3006 if (p
->search_commit_root
) {
3008 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3012 b
= get_old_root(root
, time_seq
);
3013 level
= btrfs_header_level(b
);
3014 p
->locks
[level
] = BTRFS_READ_LOCK
;
3017 level
= btrfs_header_level(b
);
3018 p
->nodes
[level
] = b
;
3019 btrfs_clear_path_blocking(p
, NULL
, 0);
3022 * we have a lock on b and as long as we aren't changing
3023 * the tree, there is no way to for the items in b to change.
3024 * It is safe to drop the lock on our parent before we
3025 * go through the expensive btree search on b.
3027 btrfs_unlock_up_safe(p
, level
+ 1);
3030 * Since we can unwind eb's we want to do a real search every
3034 ret
= key_search(b
, key
, level
, &prev_cmp
, &slot
);
3038 if (ret
&& slot
> 0) {
3042 p
->slots
[level
] = slot
;
3043 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3045 if (level
== lowest_level
) {
3051 err
= read_block_for_search(NULL
, root
, p
, &b
, level
,
3052 slot
, key
, time_seq
);
3060 level
= btrfs_header_level(b
);
3061 err
= btrfs_try_tree_read_lock(b
);
3063 btrfs_set_path_blocking(p
);
3064 btrfs_tree_read_lock(b
);
3065 btrfs_clear_path_blocking(p
, b
,
3068 b
= tree_mod_log_rewind(root
->fs_info
, p
, b
, time_seq
);
3073 p
->locks
[level
] = BTRFS_READ_LOCK
;
3074 p
->nodes
[level
] = b
;
3076 p
->slots
[level
] = slot
;
3077 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
3083 if (!p
->leave_spinning
)
3084 btrfs_set_path_blocking(p
);
3086 btrfs_release_path(p
);
3092 * helper to use instead of search slot if no exact match is needed but
3093 * instead the next or previous item should be returned.
3094 * When find_higher is true, the next higher item is returned, the next lower
3096 * When return_any and find_higher are both true, and no higher item is found,
3097 * return the next lower instead.
3098 * When return_any is true and find_higher is false, and no lower item is found,
3099 * return the next higher instead.
3100 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3103 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3104 struct btrfs_key
*key
, struct btrfs_path
*p
,
3105 int find_higher
, int return_any
)
3108 struct extent_buffer
*leaf
;
3111 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3115 * a return value of 1 means the path is at the position where the
3116 * item should be inserted. Normally this is the next bigger item,
3117 * but in case the previous item is the last in a leaf, path points
3118 * to the first free slot in the previous leaf, i.e. at an invalid
3124 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3125 ret
= btrfs_next_leaf(root
, p
);
3131 * no higher item found, return the next
3136 btrfs_release_path(p
);
3140 if (p
->slots
[0] == 0) {
3141 ret
= btrfs_prev_leaf(root
, p
);
3145 p
->slots
[0] = btrfs_header_nritems(leaf
) - 1;
3151 * no lower item found, return the next
3156 btrfs_release_path(p
);
3166 * adjust the pointers going up the tree, starting at level
3167 * making sure the right key of each node is points to 'key'.
3168 * This is used after shifting pointers to the left, so it stops
3169 * fixing up pointers when a given leaf/node is not in slot 0 of the
3173 static void fixup_low_keys(struct btrfs_root
*root
, struct btrfs_path
*path
,
3174 struct btrfs_disk_key
*key
, int level
)
3177 struct extent_buffer
*t
;
3179 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3180 int tslot
= path
->slots
[i
];
3181 if (!path
->nodes
[i
])
3184 tree_mod_log_set_node_key(root
->fs_info
, t
, tslot
, 1);
3185 btrfs_set_node_key(t
, key
, tslot
);
3186 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3195 * This function isn't completely safe. It's the caller's responsibility
3196 * that the new key won't break the order
3198 void btrfs_set_item_key_safe(struct btrfs_root
*root
, struct btrfs_path
*path
,
3199 struct btrfs_key
*new_key
)
3201 struct btrfs_disk_key disk_key
;
3202 struct extent_buffer
*eb
;
3205 eb
= path
->nodes
[0];
3206 slot
= path
->slots
[0];
3208 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3209 BUG_ON(comp_keys(&disk_key
, new_key
) >= 0);
3211 if (slot
< btrfs_header_nritems(eb
) - 1) {
3212 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3213 BUG_ON(comp_keys(&disk_key
, new_key
) <= 0);
3216 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3217 btrfs_set_item_key(eb
, &disk_key
, slot
);
3218 btrfs_mark_buffer_dirty(eb
);
3220 fixup_low_keys(root
, path
, &disk_key
, 1);
3224 * try to push data from one node into the next node left in the
3227 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3228 * error, and > 0 if there was no room in the left hand block.
3230 static int push_node_left(struct btrfs_trans_handle
*trans
,
3231 struct btrfs_root
*root
, struct extent_buffer
*dst
,
3232 struct extent_buffer
*src
, int empty
)
3239 src_nritems
= btrfs_header_nritems(src
);
3240 dst_nritems
= btrfs_header_nritems(dst
);
3241 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3242 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3243 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3245 if (!empty
&& src_nritems
<= 8)
3248 if (push_items
<= 0)
3252 push_items
= min(src_nritems
, push_items
);
3253 if (push_items
< src_nritems
) {
3254 /* leave at least 8 pointers in the node if
3255 * we aren't going to empty it
3257 if (src_nritems
- push_items
< 8) {
3258 if (push_items
<= 8)
3264 push_items
= min(src_nritems
- 8, push_items
);
3266 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, dst_nritems
, 0,
3269 btrfs_abort_transaction(trans
, root
, ret
);
3272 copy_extent_buffer(dst
, src
,
3273 btrfs_node_key_ptr_offset(dst_nritems
),
3274 btrfs_node_key_ptr_offset(0),
3275 push_items
* sizeof(struct btrfs_key_ptr
));
3277 if (push_items
< src_nritems
) {
3279 * don't call tree_mod_log_eb_move here, key removal was already
3280 * fully logged by tree_mod_log_eb_copy above.
3282 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3283 btrfs_node_key_ptr_offset(push_items
),
3284 (src_nritems
- push_items
) *
3285 sizeof(struct btrfs_key_ptr
));
3287 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3288 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3289 btrfs_mark_buffer_dirty(src
);
3290 btrfs_mark_buffer_dirty(dst
);
3296 * try to push data from one node into the next node right in the
3299 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3300 * error, and > 0 if there was no room in the right hand block.
3302 * this will only push up to 1/2 the contents of the left node over
3304 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3305 struct btrfs_root
*root
,
3306 struct extent_buffer
*dst
,
3307 struct extent_buffer
*src
)
3315 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3316 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3318 src_nritems
= btrfs_header_nritems(src
);
3319 dst_nritems
= btrfs_header_nritems(dst
);
3320 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
3321 if (push_items
<= 0)
3324 if (src_nritems
< 4)
3327 max_push
= src_nritems
/ 2 + 1;
3328 /* don't try to empty the node */
3329 if (max_push
>= src_nritems
)
3332 if (max_push
< push_items
)
3333 push_items
= max_push
;
3335 tree_mod_log_eb_move(root
->fs_info
, dst
, push_items
, 0, dst_nritems
);
3336 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3337 btrfs_node_key_ptr_offset(0),
3339 sizeof(struct btrfs_key_ptr
));
3341 ret
= tree_mod_log_eb_copy(root
->fs_info
, dst
, src
, 0,
3342 src_nritems
- push_items
, push_items
);
3344 btrfs_abort_transaction(trans
, root
, ret
);
3347 copy_extent_buffer(dst
, src
,
3348 btrfs_node_key_ptr_offset(0),
3349 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3350 push_items
* sizeof(struct btrfs_key_ptr
));
3352 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3353 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3355 btrfs_mark_buffer_dirty(src
);
3356 btrfs_mark_buffer_dirty(dst
);
3362 * helper function to insert a new root level in the tree.
3363 * A new node is allocated, and a single item is inserted to
3364 * point to the existing root
3366 * returns zero on success or < 0 on failure.
3368 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3369 struct btrfs_root
*root
,
3370 struct btrfs_path
*path
, int level
)
3373 struct extent_buffer
*lower
;
3374 struct extent_buffer
*c
;
3375 struct extent_buffer
*old
;
3376 struct btrfs_disk_key lower_key
;
3378 BUG_ON(path
->nodes
[level
]);
3379 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3381 lower
= path
->nodes
[level
-1];
3383 btrfs_item_key(lower
, &lower_key
, 0);
3385 btrfs_node_key(lower
, &lower_key
, 0);
3387 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3388 root
->root_key
.objectid
, &lower_key
,
3389 level
, root
->node
->start
, 0);
3393 root_add_used(root
, root
->nodesize
);
3395 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
3396 btrfs_set_header_nritems(c
, 1);
3397 btrfs_set_header_level(c
, level
);
3398 btrfs_set_header_bytenr(c
, c
->start
);
3399 btrfs_set_header_generation(c
, trans
->transid
);
3400 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
3401 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
3403 write_extent_buffer(c
, root
->fs_info
->fsid
, btrfs_header_fsid(),
3406 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
3407 btrfs_header_chunk_tree_uuid(c
), BTRFS_UUID_SIZE
);
3409 btrfs_set_node_key(c
, &lower_key
, 0);
3410 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3411 lower_gen
= btrfs_header_generation(lower
);
3412 WARN_ON(lower_gen
!= trans
->transid
);
3414 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3416 btrfs_mark_buffer_dirty(c
);
3419 tree_mod_log_set_root_pointer(root
, c
, 0);
3420 rcu_assign_pointer(root
->node
, c
);
3422 /* the super has an extra ref to root->node */
3423 free_extent_buffer(old
);
3425 add_root_to_dirty_list(root
);
3426 extent_buffer_get(c
);
3427 path
->nodes
[level
] = c
;
3428 path
->locks
[level
] = BTRFS_WRITE_LOCK
;
3429 path
->slots
[level
] = 0;
3434 * worker function to insert a single pointer in a node.
3435 * the node should have enough room for the pointer already
3437 * slot and level indicate where you want the key to go, and
3438 * blocknr is the block the key points to.
3440 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3441 struct btrfs_root
*root
, struct btrfs_path
*path
,
3442 struct btrfs_disk_key
*key
, u64 bytenr
,
3443 int slot
, int level
)
3445 struct extent_buffer
*lower
;
3449 BUG_ON(!path
->nodes
[level
]);
3450 btrfs_assert_tree_locked(path
->nodes
[level
]);
3451 lower
= path
->nodes
[level
];
3452 nritems
= btrfs_header_nritems(lower
);
3453 BUG_ON(slot
> nritems
);
3454 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
));
3455 if (slot
!= nritems
) {
3457 tree_mod_log_eb_move(root
->fs_info
, lower
, slot
+ 1,
3458 slot
, nritems
- slot
);
3459 memmove_extent_buffer(lower
,
3460 btrfs_node_key_ptr_offset(slot
+ 1),
3461 btrfs_node_key_ptr_offset(slot
),
3462 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3465 ret
= tree_mod_log_insert_key(root
->fs_info
, lower
, slot
,
3466 MOD_LOG_KEY_ADD
, GFP_NOFS
);
3469 btrfs_set_node_key(lower
, key
, slot
);
3470 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3471 WARN_ON(trans
->transid
== 0);
3472 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3473 btrfs_set_header_nritems(lower
, nritems
+ 1);
3474 btrfs_mark_buffer_dirty(lower
);
3478 * split the node at the specified level in path in two.
3479 * The path is corrected to point to the appropriate node after the split
3481 * Before splitting this tries to make some room in the node by pushing
3482 * left and right, if either one works, it returns right away.
3484 * returns 0 on success and < 0 on failure
3486 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3487 struct btrfs_root
*root
,
3488 struct btrfs_path
*path
, int level
)
3490 struct extent_buffer
*c
;
3491 struct extent_buffer
*split
;
3492 struct btrfs_disk_key disk_key
;
3497 c
= path
->nodes
[level
];
3498 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3499 if (c
== root
->node
) {
3501 * trying to split the root, lets make a new one
3503 * tree mod log: We don't log_removal old root in
3504 * insert_new_root, because that root buffer will be kept as a
3505 * normal node. We are going to log removal of half of the
3506 * elements below with tree_mod_log_eb_copy. We're holding a
3507 * tree lock on the buffer, which is why we cannot race with
3508 * other tree_mod_log users.
3510 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3514 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3515 c
= path
->nodes
[level
];
3516 if (!ret
&& btrfs_header_nritems(c
) <
3517 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
3523 c_nritems
= btrfs_header_nritems(c
);
3524 mid
= (c_nritems
+ 1) / 2;
3525 btrfs_node_key(c
, &disk_key
, mid
);
3527 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
3528 root
->root_key
.objectid
,
3529 &disk_key
, level
, c
->start
, 0);
3531 return PTR_ERR(split
);
3533 root_add_used(root
, root
->nodesize
);
3535 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
3536 btrfs_set_header_level(split
, btrfs_header_level(c
));
3537 btrfs_set_header_bytenr(split
, split
->start
);
3538 btrfs_set_header_generation(split
, trans
->transid
);
3539 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
3540 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
3541 write_extent_buffer(split
, root
->fs_info
->fsid
,
3542 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
3543 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
3544 btrfs_header_chunk_tree_uuid(split
),
3547 ret
= tree_mod_log_eb_copy(root
->fs_info
, split
, c
, 0,
3548 mid
, c_nritems
- mid
);
3550 btrfs_abort_transaction(trans
, root
, ret
);
3553 copy_extent_buffer(split
, c
,
3554 btrfs_node_key_ptr_offset(0),
3555 btrfs_node_key_ptr_offset(mid
),
3556 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3557 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3558 btrfs_set_header_nritems(c
, mid
);
3561 btrfs_mark_buffer_dirty(c
);
3562 btrfs_mark_buffer_dirty(split
);
3564 insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
3565 path
->slots
[level
+ 1] + 1, level
+ 1);
3567 if (path
->slots
[level
] >= mid
) {
3568 path
->slots
[level
] -= mid
;
3569 btrfs_tree_unlock(c
);
3570 free_extent_buffer(c
);
3571 path
->nodes
[level
] = split
;
3572 path
->slots
[level
+ 1] += 1;
3574 btrfs_tree_unlock(split
);
3575 free_extent_buffer(split
);
3581 * how many bytes are required to store the items in a leaf. start
3582 * and nr indicate which items in the leaf to check. This totals up the
3583 * space used both by the item structs and the item data
3585 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3587 struct btrfs_item
*start_item
;
3588 struct btrfs_item
*end_item
;
3589 struct btrfs_map_token token
;
3591 int nritems
= btrfs_header_nritems(l
);
3592 int end
= min(nritems
, start
+ nr
) - 1;
3596 btrfs_init_map_token(&token
);
3597 start_item
= btrfs_item_nr(start
);
3598 end_item
= btrfs_item_nr(end
);
3599 data_len
= btrfs_token_item_offset(l
, start_item
, &token
) +
3600 btrfs_token_item_size(l
, start_item
, &token
);
3601 data_len
= data_len
- btrfs_token_item_offset(l
, end_item
, &token
);
3602 data_len
+= sizeof(struct btrfs_item
) * nr
;
3603 WARN_ON(data_len
< 0);
3608 * The space between the end of the leaf items and
3609 * the start of the leaf data. IOW, how much room
3610 * the leaf has left for both items and data
3612 noinline
int btrfs_leaf_free_space(struct btrfs_root
*root
,
3613 struct extent_buffer
*leaf
)
3615 int nritems
= btrfs_header_nritems(leaf
);
3617 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
3619 btrfs_crit(root
->fs_info
,
3620 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3621 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
3622 leaf_space_used(leaf
, 0, nritems
), nritems
);
3628 * min slot controls the lowest index we're willing to push to the
3629 * right. We'll push up to and including min_slot, but no lower
3631 static noinline
int __push_leaf_right(struct btrfs_trans_handle
*trans
,
3632 struct btrfs_root
*root
,
3633 struct btrfs_path
*path
,
3634 int data_size
, int empty
,
3635 struct extent_buffer
*right
,
3636 int free_space
, u32 left_nritems
,
3639 struct extent_buffer
*left
= path
->nodes
[0];
3640 struct extent_buffer
*upper
= path
->nodes
[1];
3641 struct btrfs_map_token token
;
3642 struct btrfs_disk_key disk_key
;
3647 struct btrfs_item
*item
;
3653 btrfs_init_map_token(&token
);
3658 nr
= max_t(u32
, 1, min_slot
);
3660 if (path
->slots
[0] >= left_nritems
)
3661 push_space
+= data_size
;
3663 slot
= path
->slots
[1];
3664 i
= left_nritems
- 1;
3666 item
= btrfs_item_nr(i
);
3668 if (!empty
&& push_items
> 0) {
3669 if (path
->slots
[0] > i
)
3671 if (path
->slots
[0] == i
) {
3672 int space
= btrfs_leaf_free_space(root
, left
);
3673 if (space
+ push_space
* 2 > free_space
)
3678 if (path
->slots
[0] == i
)
3679 push_space
+= data_size
;
3681 this_item_size
= btrfs_item_size(left
, item
);
3682 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3686 push_space
+= this_item_size
+ sizeof(*item
);
3692 if (push_items
== 0)
3695 WARN_ON(!empty
&& push_items
== left_nritems
);
3697 /* push left to right */
3698 right_nritems
= btrfs_header_nritems(right
);
3700 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3701 push_space
-= leaf_data_end(root
, left
);
3703 /* make room in the right data area */
3704 data_end
= leaf_data_end(root
, right
);
3705 memmove_extent_buffer(right
,
3706 btrfs_leaf_data(right
) + data_end
- push_space
,
3707 btrfs_leaf_data(right
) + data_end
,
3708 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
3710 /* copy from the left data area */
3711 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
3712 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3713 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
3716 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3717 btrfs_item_nr_offset(0),
3718 right_nritems
* sizeof(struct btrfs_item
));
3720 /* copy the items from left to right */
3721 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3722 btrfs_item_nr_offset(left_nritems
- push_items
),
3723 push_items
* sizeof(struct btrfs_item
));
3725 /* update the item pointers */
3726 right_nritems
+= push_items
;
3727 btrfs_set_header_nritems(right
, right_nritems
);
3728 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3729 for (i
= 0; i
< right_nritems
; i
++) {
3730 item
= btrfs_item_nr(i
);
3731 push_space
-= btrfs_token_item_size(right
, item
, &token
);
3732 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3735 left_nritems
-= push_items
;
3736 btrfs_set_header_nritems(left
, left_nritems
);
3739 btrfs_mark_buffer_dirty(left
);
3741 clean_tree_block(trans
, root
, left
);
3743 btrfs_mark_buffer_dirty(right
);
3745 btrfs_item_key(right
, &disk_key
, 0);
3746 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3747 btrfs_mark_buffer_dirty(upper
);
3749 /* then fixup the leaf pointer in the path */
3750 if (path
->slots
[0] >= left_nritems
) {
3751 path
->slots
[0] -= left_nritems
;
3752 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3753 clean_tree_block(trans
, root
, path
->nodes
[0]);
3754 btrfs_tree_unlock(path
->nodes
[0]);
3755 free_extent_buffer(path
->nodes
[0]);
3756 path
->nodes
[0] = right
;
3757 path
->slots
[1] += 1;
3759 btrfs_tree_unlock(right
);
3760 free_extent_buffer(right
);
3765 btrfs_tree_unlock(right
);
3766 free_extent_buffer(right
);
3771 * push some data in the path leaf to the right, trying to free up at
3772 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3774 * returns 1 if the push failed because the other node didn't have enough
3775 * room, 0 if everything worked out and < 0 if there were major errors.
3777 * this will push starting from min_slot to the end of the leaf. It won't
3778 * push any slot lower than min_slot
3780 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3781 *root
, struct btrfs_path
*path
,
3782 int min_data_size
, int data_size
,
3783 int empty
, u32 min_slot
)
3785 struct extent_buffer
*left
= path
->nodes
[0];
3786 struct extent_buffer
*right
;
3787 struct extent_buffer
*upper
;
3793 if (!path
->nodes
[1])
3796 slot
= path
->slots
[1];
3797 upper
= path
->nodes
[1];
3798 if (slot
>= btrfs_header_nritems(upper
) - 1)
3801 btrfs_assert_tree_locked(path
->nodes
[1]);
3803 right
= read_node_slot(root
, upper
, slot
+ 1);
3807 btrfs_tree_lock(right
);
3808 btrfs_set_lock_blocking(right
);
3810 free_space
= btrfs_leaf_free_space(root
, right
);
3811 if (free_space
< data_size
)
3814 /* cow and double check */
3815 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3820 free_space
= btrfs_leaf_free_space(root
, right
);
3821 if (free_space
< data_size
)
3824 left_nritems
= btrfs_header_nritems(left
);
3825 if (left_nritems
== 0)
3828 if (path
->slots
[0] == left_nritems
&& !empty
) {
3829 /* Key greater than all keys in the leaf, right neighbor has
3830 * enough room for it and we're not emptying our leaf to delete
3831 * it, therefore use right neighbor to insert the new item and
3832 * no need to touch/dirty our left leaft. */
3833 btrfs_tree_unlock(left
);
3834 free_extent_buffer(left
);
3835 path
->nodes
[0] = right
;
3841 return __push_leaf_right(trans
, root
, path
, min_data_size
, empty
,
3842 right
, free_space
, left_nritems
, min_slot
);
3844 btrfs_tree_unlock(right
);
3845 free_extent_buffer(right
);
3850 * push some data in the path leaf to the left, trying to free up at
3851 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3853 * max_slot can put a limit on how far into the leaf we'll push items. The
3854 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3857 static noinline
int __push_leaf_left(struct btrfs_trans_handle
*trans
,
3858 struct btrfs_root
*root
,
3859 struct btrfs_path
*path
, int data_size
,
3860 int empty
, struct extent_buffer
*left
,
3861 int free_space
, u32 right_nritems
,
3864 struct btrfs_disk_key disk_key
;
3865 struct extent_buffer
*right
= path
->nodes
[0];
3869 struct btrfs_item
*item
;
3870 u32 old_left_nritems
;
3874 u32 old_left_item_size
;
3875 struct btrfs_map_token token
;
3877 btrfs_init_map_token(&token
);
3880 nr
= min(right_nritems
, max_slot
);
3882 nr
= min(right_nritems
- 1, max_slot
);
3884 for (i
= 0; i
< nr
; i
++) {
3885 item
= btrfs_item_nr(i
);
3887 if (!empty
&& push_items
> 0) {
3888 if (path
->slots
[0] < i
)
3890 if (path
->slots
[0] == i
) {
3891 int space
= btrfs_leaf_free_space(root
, right
);
3892 if (space
+ push_space
* 2 > free_space
)
3897 if (path
->slots
[0] == i
)
3898 push_space
+= data_size
;
3900 this_item_size
= btrfs_item_size(right
, item
);
3901 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3905 push_space
+= this_item_size
+ sizeof(*item
);
3908 if (push_items
== 0) {
3912 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3914 /* push data from right to left */
3915 copy_extent_buffer(left
, right
,
3916 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3917 btrfs_item_nr_offset(0),
3918 push_items
* sizeof(struct btrfs_item
));
3920 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
3921 btrfs_item_offset_nr(right
, push_items
- 1);
3923 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
3924 leaf_data_end(root
, left
) - push_space
,
3925 btrfs_leaf_data(right
) +
3926 btrfs_item_offset_nr(right
, push_items
- 1),
3928 old_left_nritems
= btrfs_header_nritems(left
);
3929 BUG_ON(old_left_nritems
<= 0);
3931 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3932 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3935 item
= btrfs_item_nr(i
);
3937 ioff
= btrfs_token_item_offset(left
, item
, &token
);
3938 btrfs_set_token_item_offset(left
, item
,
3939 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
),
3942 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3944 /* fixup right node */
3945 if (push_items
> right_nritems
)
3946 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3949 if (push_items
< right_nritems
) {
3950 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3951 leaf_data_end(root
, right
);
3952 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
3953 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
3954 btrfs_leaf_data(right
) +
3955 leaf_data_end(root
, right
), push_space
);
3957 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3958 btrfs_item_nr_offset(push_items
),
3959 (btrfs_header_nritems(right
) - push_items
) *
3960 sizeof(struct btrfs_item
));
3962 right_nritems
-= push_items
;
3963 btrfs_set_header_nritems(right
, right_nritems
);
3964 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
3965 for (i
= 0; i
< right_nritems
; i
++) {
3966 item
= btrfs_item_nr(i
);
3968 push_space
= push_space
- btrfs_token_item_size(right
,
3970 btrfs_set_token_item_offset(right
, item
, push_space
, &token
);
3973 btrfs_mark_buffer_dirty(left
);
3975 btrfs_mark_buffer_dirty(right
);
3977 clean_tree_block(trans
, root
, right
);
3979 btrfs_item_key(right
, &disk_key
, 0);
3980 fixup_low_keys(root
, path
, &disk_key
, 1);
3982 /* then fixup the leaf pointer in the path */
3983 if (path
->slots
[0] < push_items
) {
3984 path
->slots
[0] += old_left_nritems
;
3985 btrfs_tree_unlock(path
->nodes
[0]);
3986 free_extent_buffer(path
->nodes
[0]);
3987 path
->nodes
[0] = left
;
3988 path
->slots
[1] -= 1;
3990 btrfs_tree_unlock(left
);
3991 free_extent_buffer(left
);
3992 path
->slots
[0] -= push_items
;
3994 BUG_ON(path
->slots
[0] < 0);
3997 btrfs_tree_unlock(left
);
3998 free_extent_buffer(left
);
4003 * push some data in the path leaf to the left, trying to free up at
4004 * least data_size bytes. returns zero if the push worked, nonzero otherwise
4006 * max_slot can put a limit on how far into the leaf we'll push items. The
4007 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
4010 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
4011 *root
, struct btrfs_path
*path
, int min_data_size
,
4012 int data_size
, int empty
, u32 max_slot
)
4014 struct extent_buffer
*right
= path
->nodes
[0];
4015 struct extent_buffer
*left
;
4021 slot
= path
->slots
[1];
4024 if (!path
->nodes
[1])
4027 right_nritems
= btrfs_header_nritems(right
);
4028 if (right_nritems
== 0)
4031 btrfs_assert_tree_locked(path
->nodes
[1]);
4033 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
4037 btrfs_tree_lock(left
);
4038 btrfs_set_lock_blocking(left
);
4040 free_space
= btrfs_leaf_free_space(root
, left
);
4041 if (free_space
< data_size
) {
4046 /* cow and double check */
4047 ret
= btrfs_cow_block(trans
, root
, left
,
4048 path
->nodes
[1], slot
- 1, &left
);
4050 /* we hit -ENOSPC, but it isn't fatal here */
4056 free_space
= btrfs_leaf_free_space(root
, left
);
4057 if (free_space
< data_size
) {
4062 return __push_leaf_left(trans
, root
, path
, min_data_size
,
4063 empty
, left
, free_space
, right_nritems
,
4066 btrfs_tree_unlock(left
);
4067 free_extent_buffer(left
);
4072 * split the path's leaf in two, making sure there is at least data_size
4073 * available for the resulting leaf level of the path.
4075 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4076 struct btrfs_root
*root
,
4077 struct btrfs_path
*path
,
4078 struct extent_buffer
*l
,
4079 struct extent_buffer
*right
,
4080 int slot
, int mid
, int nritems
)
4085 struct btrfs_disk_key disk_key
;
4086 struct btrfs_map_token token
;
4088 btrfs_init_map_token(&token
);
4090 nritems
= nritems
- mid
;
4091 btrfs_set_header_nritems(right
, nritems
);
4092 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
4094 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4095 btrfs_item_nr_offset(mid
),
4096 nritems
* sizeof(struct btrfs_item
));
4098 copy_extent_buffer(right
, l
,
4099 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
4100 data_copy_size
, btrfs_leaf_data(l
) +
4101 leaf_data_end(root
, l
), data_copy_size
);
4103 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
4104 btrfs_item_end_nr(l
, mid
);
4106 for (i
= 0; i
< nritems
; i
++) {
4107 struct btrfs_item
*item
= btrfs_item_nr(i
);
4110 ioff
= btrfs_token_item_offset(right
, item
, &token
);
4111 btrfs_set_token_item_offset(right
, item
,
4112 ioff
+ rt_data_off
, &token
);
4115 btrfs_set_header_nritems(l
, mid
);
4116 btrfs_item_key(right
, &disk_key
, 0);
4117 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4118 path
->slots
[1] + 1, 1);
4120 btrfs_mark_buffer_dirty(right
);
4121 btrfs_mark_buffer_dirty(l
);
4122 BUG_ON(path
->slots
[0] != slot
);
4125 btrfs_tree_unlock(path
->nodes
[0]);
4126 free_extent_buffer(path
->nodes
[0]);
4127 path
->nodes
[0] = right
;
4128 path
->slots
[0] -= mid
;
4129 path
->slots
[1] += 1;
4131 btrfs_tree_unlock(right
);
4132 free_extent_buffer(right
);
4135 BUG_ON(path
->slots
[0] < 0);
4139 * double splits happen when we need to insert a big item in the middle
4140 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4141 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4144 * We avoid this by trying to push the items on either side of our target
4145 * into the adjacent leaves. If all goes well we can avoid the double split
4148 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4149 struct btrfs_root
*root
,
4150 struct btrfs_path
*path
,
4157 int space_needed
= data_size
;
4159 slot
= path
->slots
[0];
4160 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4161 space_needed
-= btrfs_leaf_free_space(root
, path
->nodes
[0]);
4164 * try to push all the items after our slot into the
4167 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4174 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4176 * our goal is to get our slot at the start or end of a leaf. If
4177 * we've done so we're done
4179 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4182 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4185 /* try to push all the items before our slot into the next leaf */
4186 slot
= path
->slots
[0];
4187 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4200 * split the path's leaf in two, making sure there is at least data_size
4201 * available for the resulting leaf level of the path.
4203 * returns 0 if all went well and < 0 on failure.
4205 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4206 struct btrfs_root
*root
,
4207 struct btrfs_key
*ins_key
,
4208 struct btrfs_path
*path
, int data_size
,
4211 struct btrfs_disk_key disk_key
;
4212 struct extent_buffer
*l
;
4216 struct extent_buffer
*right
;
4220 int num_doubles
= 0;
4221 int tried_avoid_double
= 0;
4224 slot
= path
->slots
[0];
4225 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4226 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(root
))
4229 /* first try to make some room by pushing left and right */
4230 if (data_size
&& path
->nodes
[1]) {
4231 int space_needed
= data_size
;
4233 if (slot
< btrfs_header_nritems(l
))
4234 space_needed
-= btrfs_leaf_free_space(root
, l
);
4236 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4237 space_needed
, 0, 0);
4241 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4242 space_needed
, 0, (u32
)-1);
4248 /* did the pushes work? */
4249 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
4253 if (!path
->nodes
[1]) {
4254 ret
= insert_new_root(trans
, root
, path
, 1);
4261 slot
= path
->slots
[0];
4262 nritems
= btrfs_header_nritems(l
);
4263 mid
= (nritems
+ 1) / 2;
4267 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4268 BTRFS_LEAF_DATA_SIZE(root
)) {
4269 if (slot
>= nritems
) {
4273 if (mid
!= nritems
&&
4274 leaf_space_used(l
, mid
, nritems
- mid
) +
4275 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4276 if (data_size
&& !tried_avoid_double
)
4277 goto push_for_double
;
4283 if (leaf_space_used(l
, 0, mid
) + data_size
>
4284 BTRFS_LEAF_DATA_SIZE(root
)) {
4285 if (!extend
&& data_size
&& slot
== 0) {
4287 } else if ((extend
|| !data_size
) && slot
== 0) {
4291 if (mid
!= nritems
&&
4292 leaf_space_used(l
, mid
, nritems
- mid
) +
4293 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
4294 if (data_size
&& !tried_avoid_double
)
4295 goto push_for_double
;
4303 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4305 btrfs_item_key(l
, &disk_key
, mid
);
4307 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
, 0,
4308 root
->root_key
.objectid
,
4309 &disk_key
, 0, l
->start
, 0);
4311 return PTR_ERR(right
);
4313 root_add_used(root
, root
->leafsize
);
4315 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
4316 btrfs_set_header_bytenr(right
, right
->start
);
4317 btrfs_set_header_generation(right
, trans
->transid
);
4318 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
4319 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
4320 btrfs_set_header_level(right
, 0);
4321 write_extent_buffer(right
, root
->fs_info
->fsid
,
4322 btrfs_header_fsid(), BTRFS_FSID_SIZE
);
4324 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
4325 btrfs_header_chunk_tree_uuid(right
),
4330 btrfs_set_header_nritems(right
, 0);
4331 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4332 path
->slots
[1] + 1, 1);
4333 btrfs_tree_unlock(path
->nodes
[0]);
4334 free_extent_buffer(path
->nodes
[0]);
4335 path
->nodes
[0] = right
;
4337 path
->slots
[1] += 1;
4339 btrfs_set_header_nritems(right
, 0);
4340 insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
4342 btrfs_tree_unlock(path
->nodes
[0]);
4343 free_extent_buffer(path
->nodes
[0]);
4344 path
->nodes
[0] = right
;
4346 if (path
->slots
[1] == 0)
4347 fixup_low_keys(root
, path
, &disk_key
, 1);
4349 btrfs_mark_buffer_dirty(right
);
4353 copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
4356 BUG_ON(num_doubles
!= 0);
4364 push_for_double_split(trans
, root
, path
, data_size
);
4365 tried_avoid_double
= 1;
4366 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= data_size
)
4371 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4372 struct btrfs_root
*root
,
4373 struct btrfs_path
*path
, int ins_len
)
4375 struct btrfs_key key
;
4376 struct extent_buffer
*leaf
;
4377 struct btrfs_file_extent_item
*fi
;
4382 leaf
= path
->nodes
[0];
4383 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4385 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4386 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4388 if (btrfs_leaf_free_space(root
, leaf
) >= ins_len
)
4391 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4392 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4393 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4394 struct btrfs_file_extent_item
);
4395 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4397 btrfs_release_path(path
);
4399 path
->keep_locks
= 1;
4400 path
->search_for_split
= 1;
4401 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4402 path
->search_for_split
= 0;
4407 leaf
= path
->nodes
[0];
4408 /* if our item isn't there or got smaller, return now */
4409 if (ret
> 0 || item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4412 /* the leaf has changed, it now has room. return now */
4413 if (btrfs_leaf_free_space(root
, path
->nodes
[0]) >= ins_len
)
4416 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4417 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4418 struct btrfs_file_extent_item
);
4419 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4423 btrfs_set_path_blocking(path
);
4424 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4428 path
->keep_locks
= 0;
4429 btrfs_unlock_up_safe(path
, 1);
4432 path
->keep_locks
= 0;
4436 static noinline
int split_item(struct btrfs_trans_handle
*trans
,
4437 struct btrfs_root
*root
,
4438 struct btrfs_path
*path
,
4439 struct btrfs_key
*new_key
,
4440 unsigned long split_offset
)
4442 struct extent_buffer
*leaf
;
4443 struct btrfs_item
*item
;
4444 struct btrfs_item
*new_item
;
4450 struct btrfs_disk_key disk_key
;
4452 leaf
= path
->nodes
[0];
4453 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
4455 btrfs_set_path_blocking(path
);
4457 item
= btrfs_item_nr(path
->slots
[0]);
4458 orig_offset
= btrfs_item_offset(leaf
, item
);
4459 item_size
= btrfs_item_size(leaf
, item
);
4461 buf
= kmalloc(item_size
, GFP_NOFS
);
4465 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4466 path
->slots
[0]), item_size
);
4468 slot
= path
->slots
[0] + 1;
4469 nritems
= btrfs_header_nritems(leaf
);
4470 if (slot
!= nritems
) {
4471 /* shift the items */
4472 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4473 btrfs_item_nr_offset(slot
),
4474 (nritems
- slot
) * sizeof(struct btrfs_item
));
4477 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4478 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4480 new_item
= btrfs_item_nr(slot
);
4482 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4483 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4485 btrfs_set_item_offset(leaf
, item
,
4486 orig_offset
+ item_size
- split_offset
);
4487 btrfs_set_item_size(leaf
, item
, split_offset
);
4489 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4491 /* write the data for the start of the original item */
4492 write_extent_buffer(leaf
, buf
,
4493 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4496 /* write the data for the new item */
4497 write_extent_buffer(leaf
, buf
+ split_offset
,
4498 btrfs_item_ptr_offset(leaf
, slot
),
4499 item_size
- split_offset
);
4500 btrfs_mark_buffer_dirty(leaf
);
4502 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
4508 * This function splits a single item into two items,
4509 * giving 'new_key' to the new item and splitting the
4510 * old one at split_offset (from the start of the item).
4512 * The path may be released by this operation. After
4513 * the split, the path is pointing to the old item. The
4514 * new item is going to be in the same node as the old one.
4516 * Note, the item being split must be smaller enough to live alone on
4517 * a tree block with room for one extra struct btrfs_item
4519 * This allows us to split the item in place, keeping a lock on the
4520 * leaf the entire time.
4522 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4523 struct btrfs_root
*root
,
4524 struct btrfs_path
*path
,
4525 struct btrfs_key
*new_key
,
4526 unsigned long split_offset
)
4529 ret
= setup_leaf_for_split(trans
, root
, path
,
4530 sizeof(struct btrfs_item
));
4534 ret
= split_item(trans
, root
, path
, new_key
, split_offset
);
4539 * This function duplicate a item, giving 'new_key' to the new item.
4540 * It guarantees both items live in the same tree leaf and the new item
4541 * is contiguous with the original item.
4543 * This allows us to split file extent in place, keeping a lock on the
4544 * leaf the entire time.
4546 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4547 struct btrfs_root
*root
,
4548 struct btrfs_path
*path
,
4549 struct btrfs_key
*new_key
)
4551 struct extent_buffer
*leaf
;
4555 leaf
= path
->nodes
[0];
4556 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4557 ret
= setup_leaf_for_split(trans
, root
, path
,
4558 item_size
+ sizeof(struct btrfs_item
));
4563 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4564 item_size
, item_size
+
4565 sizeof(struct btrfs_item
), 1);
4566 leaf
= path
->nodes
[0];
4567 memcpy_extent_buffer(leaf
,
4568 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4569 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4575 * make the item pointed to by the path smaller. new_size indicates
4576 * how small to make it, and from_end tells us if we just chop bytes
4577 * off the end of the item or if we shift the item to chop bytes off
4580 void btrfs_truncate_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4581 u32 new_size
, int from_end
)
4584 struct extent_buffer
*leaf
;
4585 struct btrfs_item
*item
;
4587 unsigned int data_end
;
4588 unsigned int old_data_start
;
4589 unsigned int old_size
;
4590 unsigned int size_diff
;
4592 struct btrfs_map_token token
;
4594 btrfs_init_map_token(&token
);
4596 leaf
= path
->nodes
[0];
4597 slot
= path
->slots
[0];
4599 old_size
= btrfs_item_size_nr(leaf
, slot
);
4600 if (old_size
== new_size
)
4603 nritems
= btrfs_header_nritems(leaf
);
4604 data_end
= leaf_data_end(root
, leaf
);
4606 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4608 size_diff
= old_size
- new_size
;
4611 BUG_ON(slot
>= nritems
);
4614 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4616 /* first correct the data pointers */
4617 for (i
= slot
; i
< nritems
; i
++) {
4619 item
= btrfs_item_nr(i
);
4621 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4622 btrfs_set_token_item_offset(leaf
, item
,
4623 ioff
+ size_diff
, &token
);
4626 /* shift the data */
4628 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4629 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4630 data_end
, old_data_start
+ new_size
- data_end
);
4632 struct btrfs_disk_key disk_key
;
4635 btrfs_item_key(leaf
, &disk_key
, slot
);
4637 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4639 struct btrfs_file_extent_item
*fi
;
4641 fi
= btrfs_item_ptr(leaf
, slot
,
4642 struct btrfs_file_extent_item
);
4643 fi
= (struct btrfs_file_extent_item
*)(
4644 (unsigned long)fi
- size_diff
);
4646 if (btrfs_file_extent_type(leaf
, fi
) ==
4647 BTRFS_FILE_EXTENT_INLINE
) {
4648 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4649 memmove_extent_buffer(leaf
, ptr
,
4651 offsetof(struct btrfs_file_extent_item
,
4656 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4657 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
4658 data_end
, old_data_start
- data_end
);
4660 offset
= btrfs_disk_key_offset(&disk_key
);
4661 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4662 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4664 fixup_low_keys(root
, path
, &disk_key
, 1);
4667 item
= btrfs_item_nr(slot
);
4668 btrfs_set_item_size(leaf
, item
, new_size
);
4669 btrfs_mark_buffer_dirty(leaf
);
4671 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4672 btrfs_print_leaf(root
, leaf
);
4678 * make the item pointed to by the path bigger, data_size is the added size.
4680 void btrfs_extend_item(struct btrfs_root
*root
, struct btrfs_path
*path
,
4684 struct extent_buffer
*leaf
;
4685 struct btrfs_item
*item
;
4687 unsigned int data_end
;
4688 unsigned int old_data
;
4689 unsigned int old_size
;
4691 struct btrfs_map_token token
;
4693 btrfs_init_map_token(&token
);
4695 leaf
= path
->nodes
[0];
4697 nritems
= btrfs_header_nritems(leaf
);
4698 data_end
= leaf_data_end(root
, leaf
);
4700 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
4701 btrfs_print_leaf(root
, leaf
);
4704 slot
= path
->slots
[0];
4705 old_data
= btrfs_item_end_nr(leaf
, slot
);
4708 if (slot
>= nritems
) {
4709 btrfs_print_leaf(root
, leaf
);
4710 btrfs_crit(root
->fs_info
, "slot %d too large, nritems %d",
4716 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4718 /* first correct the data pointers */
4719 for (i
= slot
; i
< nritems
; i
++) {
4721 item
= btrfs_item_nr(i
);
4723 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4724 btrfs_set_token_item_offset(leaf
, item
,
4725 ioff
- data_size
, &token
);
4728 /* shift the data */
4729 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4730 data_end
- data_size
, btrfs_leaf_data(leaf
) +
4731 data_end
, old_data
- data_end
);
4733 data_end
= old_data
;
4734 old_size
= btrfs_item_size_nr(leaf
, slot
);
4735 item
= btrfs_item_nr(slot
);
4736 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4737 btrfs_mark_buffer_dirty(leaf
);
4739 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4740 btrfs_print_leaf(root
, leaf
);
4746 * this is a helper for btrfs_insert_empty_items, the main goal here is
4747 * to save stack depth by doing the bulk of the work in a function
4748 * that doesn't call btrfs_search_slot
4750 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4751 struct btrfs_key
*cpu_key
, u32
*data_size
,
4752 u32 total_data
, u32 total_size
, int nr
)
4754 struct btrfs_item
*item
;
4757 unsigned int data_end
;
4758 struct btrfs_disk_key disk_key
;
4759 struct extent_buffer
*leaf
;
4761 struct btrfs_map_token token
;
4763 btrfs_init_map_token(&token
);
4765 leaf
= path
->nodes
[0];
4766 slot
= path
->slots
[0];
4768 nritems
= btrfs_header_nritems(leaf
);
4769 data_end
= leaf_data_end(root
, leaf
);
4771 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
4772 btrfs_print_leaf(root
, leaf
);
4773 btrfs_crit(root
->fs_info
, "not enough freespace need %u have %d",
4774 total_size
, btrfs_leaf_free_space(root
, leaf
));
4778 if (slot
!= nritems
) {
4779 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4781 if (old_data
< data_end
) {
4782 btrfs_print_leaf(root
, leaf
);
4783 btrfs_crit(root
->fs_info
, "slot %d old_data %d data_end %d",
4784 slot
, old_data
, data_end
);
4788 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4790 /* first correct the data pointers */
4791 for (i
= slot
; i
< nritems
; i
++) {
4794 item
= btrfs_item_nr( i
);
4795 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
4796 btrfs_set_token_item_offset(leaf
, item
,
4797 ioff
- total_data
, &token
);
4799 /* shift the items */
4800 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4801 btrfs_item_nr_offset(slot
),
4802 (nritems
- slot
) * sizeof(struct btrfs_item
));
4804 /* shift the data */
4805 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
4806 data_end
- total_data
, btrfs_leaf_data(leaf
) +
4807 data_end
, old_data
- data_end
);
4808 data_end
= old_data
;
4811 /* setup the item for the new data */
4812 for (i
= 0; i
< nr
; i
++) {
4813 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4814 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4815 item
= btrfs_item_nr(slot
+ i
);
4816 btrfs_set_token_item_offset(leaf
, item
,
4817 data_end
- data_size
[i
], &token
);
4818 data_end
-= data_size
[i
];
4819 btrfs_set_token_item_size(leaf
, item
, data_size
[i
], &token
);
4822 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4825 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4826 fixup_low_keys(root
, path
, &disk_key
, 1);
4828 btrfs_unlock_up_safe(path
, 1);
4829 btrfs_mark_buffer_dirty(leaf
);
4831 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
4832 btrfs_print_leaf(root
, leaf
);
4838 * Given a key and some data, insert items into the tree.
4839 * This does all the path init required, making room in the tree if needed.
4841 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4842 struct btrfs_root
*root
,
4843 struct btrfs_path
*path
,
4844 struct btrfs_key
*cpu_key
, u32
*data_size
,
4853 for (i
= 0; i
< nr
; i
++)
4854 total_data
+= data_size
[i
];
4856 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4857 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4863 slot
= path
->slots
[0];
4866 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4867 total_data
, total_size
, nr
);
4872 * Given a key and some data, insert an item into the tree.
4873 * This does all the path init required, making room in the tree if needed.
4875 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
4876 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
4880 struct btrfs_path
*path
;
4881 struct extent_buffer
*leaf
;
4884 path
= btrfs_alloc_path();
4887 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4889 leaf
= path
->nodes
[0];
4890 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4891 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4892 btrfs_mark_buffer_dirty(leaf
);
4894 btrfs_free_path(path
);
4899 * delete the pointer from a given node.
4901 * the tree should have been previously balanced so the deletion does not
4904 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4905 int level
, int slot
)
4907 struct extent_buffer
*parent
= path
->nodes
[level
];
4911 nritems
= btrfs_header_nritems(parent
);
4912 if (slot
!= nritems
- 1) {
4914 tree_mod_log_eb_move(root
->fs_info
, parent
, slot
,
4915 slot
+ 1, nritems
- slot
- 1);
4916 memmove_extent_buffer(parent
,
4917 btrfs_node_key_ptr_offset(slot
),
4918 btrfs_node_key_ptr_offset(slot
+ 1),
4919 sizeof(struct btrfs_key_ptr
) *
4920 (nritems
- slot
- 1));
4922 ret
= tree_mod_log_insert_key(root
->fs_info
, parent
, slot
,
4923 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
4928 btrfs_set_header_nritems(parent
, nritems
);
4929 if (nritems
== 0 && parent
== root
->node
) {
4930 BUG_ON(btrfs_header_level(root
->node
) != 1);
4931 /* just turn the root into a leaf and break */
4932 btrfs_set_header_level(root
->node
, 0);
4933 } else if (slot
== 0) {
4934 struct btrfs_disk_key disk_key
;
4936 btrfs_node_key(parent
, &disk_key
, 0);
4937 fixup_low_keys(root
, path
, &disk_key
, level
+ 1);
4939 btrfs_mark_buffer_dirty(parent
);
4943 * a helper function to delete the leaf pointed to by path->slots[1] and
4946 * This deletes the pointer in path->nodes[1] and frees the leaf
4947 * block extent. zero is returned if it all worked out, < 0 otherwise.
4949 * The path must have already been setup for deleting the leaf, including
4950 * all the proper balancing. path->nodes[1] must be locked.
4952 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4953 struct btrfs_root
*root
,
4954 struct btrfs_path
*path
,
4955 struct extent_buffer
*leaf
)
4957 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4958 del_ptr(root
, path
, 1, path
->slots
[1]);
4961 * btrfs_free_extent is expensive, we want to make sure we
4962 * aren't holding any locks when we call it
4964 btrfs_unlock_up_safe(path
, 0);
4966 root_sub_used(root
, leaf
->len
);
4968 extent_buffer_get(leaf
);
4969 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4970 free_extent_buffer_stale(leaf
);
4973 * delete the item at the leaf level in path. If that empties
4974 * the leaf, remove it from the tree
4976 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4977 struct btrfs_path
*path
, int slot
, int nr
)
4979 struct extent_buffer
*leaf
;
4980 struct btrfs_item
*item
;
4987 struct btrfs_map_token token
;
4989 btrfs_init_map_token(&token
);
4991 leaf
= path
->nodes
[0];
4992 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4994 for (i
= 0; i
< nr
; i
++)
4995 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4997 nritems
= btrfs_header_nritems(leaf
);
4999 if (slot
+ nr
!= nritems
) {
5000 int data_end
= leaf_data_end(root
, leaf
);
5002 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
5004 btrfs_leaf_data(leaf
) + data_end
,
5005 last_off
- data_end
);
5007 for (i
= slot
+ nr
; i
< nritems
; i
++) {
5010 item
= btrfs_item_nr(i
);
5011 ioff
= btrfs_token_item_offset(leaf
, item
, &token
);
5012 btrfs_set_token_item_offset(leaf
, item
,
5013 ioff
+ dsize
, &token
);
5016 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
5017 btrfs_item_nr_offset(slot
+ nr
),
5018 sizeof(struct btrfs_item
) *
5019 (nritems
- slot
- nr
));
5021 btrfs_set_header_nritems(leaf
, nritems
- nr
);
5024 /* delete the leaf if we've emptied it */
5026 if (leaf
== root
->node
) {
5027 btrfs_set_header_level(leaf
, 0);
5029 btrfs_set_path_blocking(path
);
5030 clean_tree_block(trans
, root
, leaf
);
5031 btrfs_del_leaf(trans
, root
, path
, leaf
);
5034 int used
= leaf_space_used(leaf
, 0, nritems
);
5036 struct btrfs_disk_key disk_key
;
5038 btrfs_item_key(leaf
, &disk_key
, 0);
5039 fixup_low_keys(root
, path
, &disk_key
, 1);
5042 /* delete the leaf if it is mostly empty */
5043 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
5044 /* push_leaf_left fixes the path.
5045 * make sure the path still points to our leaf
5046 * for possible call to del_ptr below
5048 slot
= path
->slots
[1];
5049 extent_buffer_get(leaf
);
5051 btrfs_set_path_blocking(path
);
5052 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
5054 if (wret
< 0 && wret
!= -ENOSPC
)
5057 if (path
->nodes
[0] == leaf
&&
5058 btrfs_header_nritems(leaf
)) {
5059 wret
= push_leaf_right(trans
, root
, path
, 1,
5061 if (wret
< 0 && wret
!= -ENOSPC
)
5065 if (btrfs_header_nritems(leaf
) == 0) {
5066 path
->slots
[1] = slot
;
5067 btrfs_del_leaf(trans
, root
, path
, leaf
);
5068 free_extent_buffer(leaf
);
5071 /* if we're still in the path, make sure
5072 * we're dirty. Otherwise, one of the
5073 * push_leaf functions must have already
5074 * dirtied this buffer
5076 if (path
->nodes
[0] == leaf
)
5077 btrfs_mark_buffer_dirty(leaf
);
5078 free_extent_buffer(leaf
);
5081 btrfs_mark_buffer_dirty(leaf
);
5088 * search the tree again to find a leaf with lesser keys
5089 * returns 0 if it found something or 1 if there are no lesser leaves.
5090 * returns < 0 on io errors.
5092 * This may release the path, and so you may lose any locks held at the
5095 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5097 struct btrfs_key key
;
5098 struct btrfs_disk_key found_key
;
5101 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5103 if (key
.offset
> 0) {
5105 } else if (key
.type
> 0) {
5107 key
.offset
= (u64
)-1;
5108 } else if (key
.objectid
> 0) {
5111 key
.offset
= (u64
)-1;
5116 btrfs_release_path(path
);
5117 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5120 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5121 ret
= comp_keys(&found_key
, &key
);
5128 * A helper function to walk down the tree starting at min_key, and looking
5129 * for nodes or leaves that are have a minimum transaction id.
5130 * This is used by the btree defrag code, and tree logging
5132 * This does not cow, but it does stuff the starting key it finds back
5133 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5134 * key and get a writable path.
5136 * This does lock as it descends, and path->keep_locks should be set
5137 * to 1 by the caller.
5139 * This honors path->lowest_level to prevent descent past a given level
5142 * min_trans indicates the oldest transaction that you are interested
5143 * in walking through. Any nodes or leaves older than min_trans are
5144 * skipped over (without reading them).
5146 * returns zero if something useful was found, < 0 on error and 1 if there
5147 * was nothing in the tree that matched the search criteria.
5149 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5150 struct btrfs_path
*path
,
5153 struct extent_buffer
*cur
;
5154 struct btrfs_key found_key
;
5161 WARN_ON(!path
->keep_locks
);
5163 cur
= btrfs_read_lock_root_node(root
);
5164 level
= btrfs_header_level(cur
);
5165 WARN_ON(path
->nodes
[level
]);
5166 path
->nodes
[level
] = cur
;
5167 path
->locks
[level
] = BTRFS_READ_LOCK
;
5169 if (btrfs_header_generation(cur
) < min_trans
) {
5174 nritems
= btrfs_header_nritems(cur
);
5175 level
= btrfs_header_level(cur
);
5176 sret
= bin_search(cur
, min_key
, level
, &slot
);
5178 /* at the lowest level, we're done, setup the path and exit */
5179 if (level
== path
->lowest_level
) {
5180 if (slot
>= nritems
)
5183 path
->slots
[level
] = slot
;
5184 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5187 if (sret
&& slot
> 0)
5190 * check this node pointer against the min_trans parameters.
5191 * If it is too old, old, skip to the next one.
5193 while (slot
< nritems
) {
5196 gen
= btrfs_node_ptr_generation(cur
, slot
);
5197 if (gen
< min_trans
) {
5205 * we didn't find a candidate key in this node, walk forward
5206 * and find another one
5208 if (slot
>= nritems
) {
5209 path
->slots
[level
] = slot
;
5210 btrfs_set_path_blocking(path
);
5211 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5214 btrfs_release_path(path
);
5220 /* save our key for returning back */
5221 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5222 path
->slots
[level
] = slot
;
5223 if (level
== path
->lowest_level
) {
5225 unlock_up(path
, level
, 1, 0, NULL
);
5228 btrfs_set_path_blocking(path
);
5229 cur
= read_node_slot(root
, cur
, slot
);
5230 BUG_ON(!cur
); /* -ENOMEM */
5232 btrfs_tree_read_lock(cur
);
5234 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5235 path
->nodes
[level
- 1] = cur
;
5236 unlock_up(path
, level
, 1, 0, NULL
);
5237 btrfs_clear_path_blocking(path
, NULL
, 0);
5241 memcpy(min_key
, &found_key
, sizeof(found_key
));
5242 btrfs_set_path_blocking(path
);
5246 static void tree_move_down(struct btrfs_root
*root
,
5247 struct btrfs_path
*path
,
5248 int *level
, int root_level
)
5250 BUG_ON(*level
== 0);
5251 path
->nodes
[*level
- 1] = read_node_slot(root
, path
->nodes
[*level
],
5252 path
->slots
[*level
]);
5253 path
->slots
[*level
- 1] = 0;
5257 static int tree_move_next_or_upnext(struct btrfs_root
*root
,
5258 struct btrfs_path
*path
,
5259 int *level
, int root_level
)
5263 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5265 path
->slots
[*level
]++;
5267 while (path
->slots
[*level
] >= nritems
) {
5268 if (*level
== root_level
)
5272 path
->slots
[*level
] = 0;
5273 free_extent_buffer(path
->nodes
[*level
]);
5274 path
->nodes
[*level
] = NULL
;
5276 path
->slots
[*level
]++;
5278 nritems
= btrfs_header_nritems(path
->nodes
[*level
]);
5285 * Returns 1 if it had to move up and next. 0 is returned if it moved only next
5288 static int tree_advance(struct btrfs_root
*root
,
5289 struct btrfs_path
*path
,
5290 int *level
, int root_level
,
5292 struct btrfs_key
*key
)
5296 if (*level
== 0 || !allow_down
) {
5297 ret
= tree_move_next_or_upnext(root
, path
, level
, root_level
);
5299 tree_move_down(root
, path
, level
, root_level
);
5304 btrfs_item_key_to_cpu(path
->nodes
[*level
], key
,
5305 path
->slots
[*level
]);
5307 btrfs_node_key_to_cpu(path
->nodes
[*level
], key
,
5308 path
->slots
[*level
]);
5313 static int tree_compare_item(struct btrfs_root
*left_root
,
5314 struct btrfs_path
*left_path
,
5315 struct btrfs_path
*right_path
,
5320 unsigned long off1
, off2
;
5322 len1
= btrfs_item_size_nr(left_path
->nodes
[0], left_path
->slots
[0]);
5323 len2
= btrfs_item_size_nr(right_path
->nodes
[0], right_path
->slots
[0]);
5327 off1
= btrfs_item_ptr_offset(left_path
->nodes
[0], left_path
->slots
[0]);
5328 off2
= btrfs_item_ptr_offset(right_path
->nodes
[0],
5329 right_path
->slots
[0]);
5331 read_extent_buffer(left_path
->nodes
[0], tmp_buf
, off1
, len1
);
5333 cmp
= memcmp_extent_buffer(right_path
->nodes
[0], tmp_buf
, off2
, len1
);
5340 #define ADVANCE_ONLY_NEXT -1
5343 * This function compares two trees and calls the provided callback for
5344 * every changed/new/deleted item it finds.
5345 * If shared tree blocks are encountered, whole subtrees are skipped, making
5346 * the compare pretty fast on snapshotted subvolumes.
5348 * This currently works on commit roots only. As commit roots are read only,
5349 * we don't do any locking. The commit roots are protected with transactions.
5350 * Transactions are ended and rejoined when a commit is tried in between.
5352 * This function checks for modifications done to the trees while comparing.
5353 * If it detects a change, it aborts immediately.
5355 int btrfs_compare_trees(struct btrfs_root
*left_root
,
5356 struct btrfs_root
*right_root
,
5357 btrfs_changed_cb_t changed_cb
, void *ctx
)
5361 struct btrfs_trans_handle
*trans
= NULL
;
5362 struct btrfs_path
*left_path
= NULL
;
5363 struct btrfs_path
*right_path
= NULL
;
5364 struct btrfs_key left_key
;
5365 struct btrfs_key right_key
;
5366 char *tmp_buf
= NULL
;
5367 int left_root_level
;
5368 int right_root_level
;
5371 int left_end_reached
;
5372 int right_end_reached
;
5377 u64 left_start_ctransid
;
5378 u64 right_start_ctransid
;
5381 left_path
= btrfs_alloc_path();
5386 right_path
= btrfs_alloc_path();
5392 tmp_buf
= kmalloc(left_root
->leafsize
, GFP_NOFS
);
5398 left_path
->search_commit_root
= 1;
5399 left_path
->skip_locking
= 1;
5400 right_path
->search_commit_root
= 1;
5401 right_path
->skip_locking
= 1;
5403 spin_lock(&left_root
->root_item_lock
);
5404 left_start_ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5405 spin_unlock(&left_root
->root_item_lock
);
5407 spin_lock(&right_root
->root_item_lock
);
5408 right_start_ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5409 spin_unlock(&right_root
->root_item_lock
);
5411 trans
= btrfs_join_transaction(left_root
);
5412 if (IS_ERR(trans
)) {
5413 ret
= PTR_ERR(trans
);
5419 * Strategy: Go to the first items of both trees. Then do
5421 * If both trees are at level 0
5422 * Compare keys of current items
5423 * If left < right treat left item as new, advance left tree
5425 * If left > right treat right item as deleted, advance right tree
5427 * If left == right do deep compare of items, treat as changed if
5428 * needed, advance both trees and repeat
5429 * If both trees are at the same level but not at level 0
5430 * Compare keys of current nodes/leafs
5431 * If left < right advance left tree and repeat
5432 * If left > right advance right tree and repeat
5433 * If left == right compare blockptrs of the next nodes/leafs
5434 * If they match advance both trees but stay at the same level
5436 * If they don't match advance both trees while allowing to go
5438 * If tree levels are different
5439 * Advance the tree that needs it and repeat
5441 * Advancing a tree means:
5442 * If we are at level 0, try to go to the next slot. If that's not
5443 * possible, go one level up and repeat. Stop when we found a level
5444 * where we could go to the next slot. We may at this point be on a
5447 * If we are not at level 0 and not on shared tree blocks, go one
5450 * If we are not at level 0 and on shared tree blocks, go one slot to
5451 * the right if possible or go up and right.
5454 left_level
= btrfs_header_level(left_root
->commit_root
);
5455 left_root_level
= left_level
;
5456 left_path
->nodes
[left_level
] = left_root
->commit_root
;
5457 extent_buffer_get(left_path
->nodes
[left_level
]);
5459 right_level
= btrfs_header_level(right_root
->commit_root
);
5460 right_root_level
= right_level
;
5461 right_path
->nodes
[right_level
] = right_root
->commit_root
;
5462 extent_buffer_get(right_path
->nodes
[right_level
]);
5464 if (left_level
== 0)
5465 btrfs_item_key_to_cpu(left_path
->nodes
[left_level
],
5466 &left_key
, left_path
->slots
[left_level
]);
5468 btrfs_node_key_to_cpu(left_path
->nodes
[left_level
],
5469 &left_key
, left_path
->slots
[left_level
]);
5470 if (right_level
== 0)
5471 btrfs_item_key_to_cpu(right_path
->nodes
[right_level
],
5472 &right_key
, right_path
->slots
[right_level
]);
5474 btrfs_node_key_to_cpu(right_path
->nodes
[right_level
],
5475 &right_key
, right_path
->slots
[right_level
]);
5477 left_end_reached
= right_end_reached
= 0;
5478 advance_left
= advance_right
= 0;
5482 * We need to make sure the transaction does not get committed
5483 * while we do anything on commit roots. This means, we need to
5484 * join and leave transactions for every item that we process.
5486 if (trans
&& btrfs_should_end_transaction(trans
, left_root
)) {
5487 btrfs_release_path(left_path
);
5488 btrfs_release_path(right_path
);
5490 ret
= btrfs_end_transaction(trans
, left_root
);
5495 /* now rejoin the transaction */
5497 trans
= btrfs_join_transaction(left_root
);
5498 if (IS_ERR(trans
)) {
5499 ret
= PTR_ERR(trans
);
5504 spin_lock(&left_root
->root_item_lock
);
5505 ctransid
= btrfs_root_ctransid(&left_root
->root_item
);
5506 spin_unlock(&left_root
->root_item_lock
);
5507 if (ctransid
!= left_start_ctransid
)
5508 left_start_ctransid
= 0;
5510 spin_lock(&right_root
->root_item_lock
);
5511 ctransid
= btrfs_root_ctransid(&right_root
->root_item
);
5512 spin_unlock(&right_root
->root_item_lock
);
5513 if (ctransid
!= right_start_ctransid
)
5514 right_start_ctransid
= 0;
5516 if (!left_start_ctransid
|| !right_start_ctransid
) {
5517 WARN(1, KERN_WARNING
5518 "BTRFS: btrfs_compare_tree detected "
5519 "a change in one of the trees while "
5520 "iterating. This is probably a "
5527 * the commit root may have changed, so start again
5530 left_path
->lowest_level
= left_level
;
5531 right_path
->lowest_level
= right_level
;
5532 ret
= btrfs_search_slot(NULL
, left_root
,
5533 &left_key
, left_path
, 0, 0);
5536 ret
= btrfs_search_slot(NULL
, right_root
,
5537 &right_key
, right_path
, 0, 0);
5542 if (advance_left
&& !left_end_reached
) {
5543 ret
= tree_advance(left_root
, left_path
, &left_level
,
5545 advance_left
!= ADVANCE_ONLY_NEXT
,
5548 left_end_reached
= ADVANCE
;
5551 if (advance_right
&& !right_end_reached
) {
5552 ret
= tree_advance(right_root
, right_path
, &right_level
,
5554 advance_right
!= ADVANCE_ONLY_NEXT
,
5557 right_end_reached
= ADVANCE
;
5561 if (left_end_reached
&& right_end_reached
) {
5564 } else if (left_end_reached
) {
5565 if (right_level
== 0) {
5566 ret
= changed_cb(left_root
, right_root
,
5567 left_path
, right_path
,
5569 BTRFS_COMPARE_TREE_DELETED
,
5574 advance_right
= ADVANCE
;
5576 } else if (right_end_reached
) {
5577 if (left_level
== 0) {
5578 ret
= changed_cb(left_root
, right_root
,
5579 left_path
, right_path
,
5581 BTRFS_COMPARE_TREE_NEW
,
5586 advance_left
= ADVANCE
;
5590 if (left_level
== 0 && right_level
== 0) {
5591 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5593 ret
= changed_cb(left_root
, right_root
,
5594 left_path
, right_path
,
5596 BTRFS_COMPARE_TREE_NEW
,
5600 advance_left
= ADVANCE
;
5601 } else if (cmp
> 0) {
5602 ret
= changed_cb(left_root
, right_root
,
5603 left_path
, right_path
,
5605 BTRFS_COMPARE_TREE_DELETED
,
5609 advance_right
= ADVANCE
;
5611 enum btrfs_compare_tree_result cmp
;
5613 WARN_ON(!extent_buffer_uptodate(left_path
->nodes
[0]));
5614 ret
= tree_compare_item(left_root
, left_path
,
5615 right_path
, tmp_buf
);
5617 cmp
= BTRFS_COMPARE_TREE_CHANGED
;
5619 cmp
= BTRFS_COMPARE_TREE_SAME
;
5620 ret
= changed_cb(left_root
, right_root
,
5621 left_path
, right_path
,
5622 &left_key
, cmp
, ctx
);
5625 advance_left
= ADVANCE
;
5626 advance_right
= ADVANCE
;
5628 } else if (left_level
== right_level
) {
5629 cmp
= btrfs_comp_cpu_keys(&left_key
, &right_key
);
5631 advance_left
= ADVANCE
;
5632 } else if (cmp
> 0) {
5633 advance_right
= ADVANCE
;
5635 left_blockptr
= btrfs_node_blockptr(
5636 left_path
->nodes
[left_level
],
5637 left_path
->slots
[left_level
]);
5638 right_blockptr
= btrfs_node_blockptr(
5639 right_path
->nodes
[right_level
],
5640 right_path
->slots
[right_level
]);
5641 if (left_blockptr
== right_blockptr
) {
5643 * As we're on a shared block, don't
5644 * allow to go deeper.
5646 advance_left
= ADVANCE_ONLY_NEXT
;
5647 advance_right
= ADVANCE_ONLY_NEXT
;
5649 advance_left
= ADVANCE
;
5650 advance_right
= ADVANCE
;
5653 } else if (left_level
< right_level
) {
5654 advance_right
= ADVANCE
;
5656 advance_left
= ADVANCE
;
5661 btrfs_free_path(left_path
);
5662 btrfs_free_path(right_path
);
5667 ret
= btrfs_end_transaction(trans
, left_root
);
5669 btrfs_end_transaction(trans
, left_root
);
5676 * this is similar to btrfs_next_leaf, but does not try to preserve
5677 * and fixup the path. It looks for and returns the next key in the
5678 * tree based on the current path and the min_trans parameters.
5680 * 0 is returned if another key is found, < 0 if there are any errors
5681 * and 1 is returned if there are no higher keys in the tree
5683 * path->keep_locks should be set to 1 on the search made before
5684 * calling this function.
5686 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5687 struct btrfs_key
*key
, int level
, u64 min_trans
)
5690 struct extent_buffer
*c
;
5692 WARN_ON(!path
->keep_locks
);
5693 while (level
< BTRFS_MAX_LEVEL
) {
5694 if (!path
->nodes
[level
])
5697 slot
= path
->slots
[level
] + 1;
5698 c
= path
->nodes
[level
];
5700 if (slot
>= btrfs_header_nritems(c
)) {
5703 struct btrfs_key cur_key
;
5704 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5705 !path
->nodes
[level
+ 1])
5708 if (path
->locks
[level
+ 1]) {
5713 slot
= btrfs_header_nritems(c
) - 1;
5715 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5717 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5719 orig_lowest
= path
->lowest_level
;
5720 btrfs_release_path(path
);
5721 path
->lowest_level
= level
;
5722 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5724 path
->lowest_level
= orig_lowest
;
5728 c
= path
->nodes
[level
];
5729 slot
= path
->slots
[level
];
5736 btrfs_item_key_to_cpu(c
, key
, slot
);
5738 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5740 if (gen
< min_trans
) {
5744 btrfs_node_key_to_cpu(c
, key
, slot
);
5752 * search the tree again to find a leaf with greater keys
5753 * returns 0 if it found something or 1 if there are no greater leaves.
5754 * returns < 0 on io errors.
5756 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5758 return btrfs_next_old_leaf(root
, path
, 0);
5761 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5766 struct extent_buffer
*c
;
5767 struct extent_buffer
*next
;
5768 struct btrfs_key key
;
5771 int old_spinning
= path
->leave_spinning
;
5772 int next_rw_lock
= 0;
5774 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5778 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5783 btrfs_release_path(path
);
5785 path
->keep_locks
= 1;
5786 path
->leave_spinning
= 1;
5789 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5791 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5792 path
->keep_locks
= 0;
5797 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5799 * by releasing the path above we dropped all our locks. A balance
5800 * could have added more items next to the key that used to be
5801 * at the very end of the block. So, check again here and
5802 * advance the path if there are now more items available.
5804 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5811 while (level
< BTRFS_MAX_LEVEL
) {
5812 if (!path
->nodes
[level
]) {
5817 slot
= path
->slots
[level
] + 1;
5818 c
= path
->nodes
[level
];
5819 if (slot
>= btrfs_header_nritems(c
)) {
5821 if (level
== BTRFS_MAX_LEVEL
) {
5829 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5830 free_extent_buffer(next
);
5834 next_rw_lock
= path
->locks
[level
];
5835 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5841 btrfs_release_path(path
);
5845 if (!path
->skip_locking
) {
5846 ret
= btrfs_try_tree_read_lock(next
);
5847 if (!ret
&& time_seq
) {
5849 * If we don't get the lock, we may be racing
5850 * with push_leaf_left, holding that lock while
5851 * itself waiting for the leaf we've currently
5852 * locked. To solve this situation, we give up
5853 * on our lock and cycle.
5855 free_extent_buffer(next
);
5856 btrfs_release_path(path
);
5861 btrfs_set_path_blocking(path
);
5862 btrfs_tree_read_lock(next
);
5863 btrfs_clear_path_blocking(path
, next
,
5866 next_rw_lock
= BTRFS_READ_LOCK
;
5870 path
->slots
[level
] = slot
;
5873 c
= path
->nodes
[level
];
5874 if (path
->locks
[level
])
5875 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5877 free_extent_buffer(c
);
5878 path
->nodes
[level
] = next
;
5879 path
->slots
[level
] = 0;
5880 if (!path
->skip_locking
)
5881 path
->locks
[level
] = next_rw_lock
;
5885 ret
= read_block_for_search(NULL
, root
, path
, &next
, level
,
5891 btrfs_release_path(path
);
5895 if (!path
->skip_locking
) {
5896 ret
= btrfs_try_tree_read_lock(next
);
5898 btrfs_set_path_blocking(path
);
5899 btrfs_tree_read_lock(next
);
5900 btrfs_clear_path_blocking(path
, next
,
5903 next_rw_lock
= BTRFS_READ_LOCK
;
5908 unlock_up(path
, 0, 1, 0, NULL
);
5909 path
->leave_spinning
= old_spinning
;
5911 btrfs_set_path_blocking(path
);
5917 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5918 * searching until it gets past min_objectid or finds an item of 'type'
5920 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5922 int btrfs_previous_item(struct btrfs_root
*root
,
5923 struct btrfs_path
*path
, u64 min_objectid
,
5926 struct btrfs_key found_key
;
5927 struct extent_buffer
*leaf
;
5932 if (path
->slots
[0] == 0) {
5933 btrfs_set_path_blocking(path
);
5934 ret
= btrfs_prev_leaf(root
, path
);
5940 leaf
= path
->nodes
[0];
5941 nritems
= btrfs_header_nritems(leaf
);
5944 if (path
->slots
[0] == nritems
)
5947 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5948 if (found_key
.objectid
< min_objectid
)
5950 if (found_key
.type
== type
)
5952 if (found_key
.objectid
== min_objectid
&&
5953 found_key
.type
< type
)
5960 * search in extent tree to find a previous Metadata/Data extent item with
5963 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5965 int btrfs_previous_extent_item(struct btrfs_root
*root
,
5966 struct btrfs_path
*path
, u64 min_objectid
)
5968 struct btrfs_key found_key
;
5969 struct extent_buffer
*leaf
;
5974 if (path
->slots
[0] == 0) {
5975 btrfs_set_path_blocking(path
);
5976 ret
= btrfs_prev_leaf(root
, path
);
5982 leaf
= path
->nodes
[0];
5983 nritems
= btrfs_header_nritems(leaf
);
5986 if (path
->slots
[0] == nritems
)
5989 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5990 if (found_key
.objectid
< min_objectid
)
5992 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
||
5993 found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
5995 if (found_key
.objectid
== min_objectid
&&
5996 found_key
.type
< BTRFS_EXTENT_ITEM_KEY
)