1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (C) 2007,2008 Oracle. All rights reserved.
6 #include <linux/sched.h>
7 #include <linux/slab.h>
8 #include <linux/rbtree.h>
12 #include "transaction.h"
13 #include "print-tree.h"
18 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
19 *root
, struct btrfs_path
*path
, int level
);
20 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
21 const struct btrfs_key
*ins_key
, struct btrfs_path
*path
,
22 int data_size
, int extend
);
23 static int push_node_left(struct btrfs_trans_handle
*trans
,
24 struct extent_buffer
*dst
,
25 struct extent_buffer
*src
, int empty
);
26 static int balance_node_right(struct btrfs_trans_handle
*trans
,
27 struct extent_buffer
*dst_buf
,
28 struct extent_buffer
*src_buf
);
29 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
32 static const struct btrfs_csums
{
35 const char driver
[12];
37 [BTRFS_CSUM_TYPE_CRC32
] = { .size
= 4, .name
= "crc32c" },
38 [BTRFS_CSUM_TYPE_XXHASH
] = { .size
= 8, .name
= "xxhash64" },
39 [BTRFS_CSUM_TYPE_SHA256
] = { .size
= 32, .name
= "sha256" },
40 [BTRFS_CSUM_TYPE_BLAKE2
] = { .size
= 32, .name
= "blake2b",
41 .driver
= "blake2b-256" },
44 int btrfs_super_csum_size(const struct btrfs_super_block
*s
)
46 u16 t
= btrfs_super_csum_type(s
);
48 * csum type is validated at mount time
50 return btrfs_csums
[t
].size
;
53 const char *btrfs_super_csum_name(u16 csum_type
)
55 /* csum type is validated at mount time */
56 return btrfs_csums
[csum_type
].name
;
60 * Return driver name if defined, otherwise the name that's also a valid driver
63 const char *btrfs_super_csum_driver(u16 csum_type
)
65 /* csum type is validated at mount time */
66 return btrfs_csums
[csum_type
].driver
[0] ?
67 btrfs_csums
[csum_type
].driver
:
68 btrfs_csums
[csum_type
].name
;
71 size_t __attribute_const__
btrfs_get_num_csums(void)
73 return ARRAY_SIZE(btrfs_csums
);
76 struct btrfs_path
*btrfs_alloc_path(void)
78 return kmem_cache_zalloc(btrfs_path_cachep
, GFP_NOFS
);
81 /* this also releases the path */
82 void btrfs_free_path(struct btrfs_path
*p
)
86 btrfs_release_path(p
);
87 kmem_cache_free(btrfs_path_cachep
, p
);
91 * path release drops references on the extent buffers in the path
92 * and it drops any locks held by this path
94 * It is safe to call this on paths that no locks or extent buffers held.
96 noinline
void btrfs_release_path(struct btrfs_path
*p
)
100 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
105 btrfs_tree_unlock_rw(p
->nodes
[i
], p
->locks
[i
]);
108 free_extent_buffer(p
->nodes
[i
]);
114 * safely gets a reference on the root node of a tree. A lock
115 * is not taken, so a concurrent writer may put a different node
116 * at the root of the tree. See btrfs_lock_root_node for the
119 * The extent buffer returned by this has a reference taken, so
120 * it won't disappear. It may stop being the root of the tree
121 * at any time because there are no locks held.
123 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
125 struct extent_buffer
*eb
;
129 eb
= rcu_dereference(root
->node
);
132 * RCU really hurts here, we could free up the root node because
133 * it was COWed but we may not get the new root node yet so do
134 * the inc_not_zero dance and if it doesn't work then
135 * synchronize_rcu and try again.
137 if (atomic_inc_not_zero(&eb
->refs
)) {
148 * Cowonly root (not-shareable trees, everything not subvolume or reloc roots),
149 * just get put onto a simple dirty list. Transaction walks this list to make
150 * sure they get properly updated on disk.
152 static void add_root_to_dirty_list(struct btrfs_root
*root
)
154 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
156 if (test_bit(BTRFS_ROOT_DIRTY
, &root
->state
) ||
157 !test_bit(BTRFS_ROOT_TRACK_DIRTY
, &root
->state
))
160 spin_lock(&fs_info
->trans_lock
);
161 if (!test_and_set_bit(BTRFS_ROOT_DIRTY
, &root
->state
)) {
162 /* Want the extent tree to be the last on the list */
163 if (root
->root_key
.objectid
== BTRFS_EXTENT_TREE_OBJECTID
)
164 list_move_tail(&root
->dirty_list
,
165 &fs_info
->dirty_cowonly_roots
);
167 list_move(&root
->dirty_list
,
168 &fs_info
->dirty_cowonly_roots
);
170 spin_unlock(&fs_info
->trans_lock
);
174 * used by snapshot creation to make a copy of a root for a tree with
175 * a given objectid. The buffer with the new root node is returned in
176 * cow_ret, and this func returns zero on success or a negative error code.
178 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
179 struct btrfs_root
*root
,
180 struct extent_buffer
*buf
,
181 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
183 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
184 struct extent_buffer
*cow
;
187 struct btrfs_disk_key disk_key
;
189 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
190 trans
->transid
!= fs_info
->running_transaction
->transid
);
191 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
192 trans
->transid
!= root
->last_trans
);
194 level
= btrfs_header_level(buf
);
196 btrfs_item_key(buf
, &disk_key
, 0);
198 btrfs_node_key(buf
, &disk_key
, 0);
200 cow
= btrfs_alloc_tree_block(trans
, root
, 0, new_root_objectid
,
201 &disk_key
, level
, buf
->start
, 0);
205 copy_extent_buffer_full(cow
, buf
);
206 btrfs_set_header_bytenr(cow
, cow
->start
);
207 btrfs_set_header_generation(cow
, trans
->transid
);
208 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
209 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
210 BTRFS_HEADER_FLAG_RELOC
);
211 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
212 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
214 btrfs_set_header_owner(cow
, new_root_objectid
);
216 write_extent_buffer_fsid(cow
, fs_info
->fs_devices
->metadata_uuid
);
218 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
219 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
220 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
222 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
227 btrfs_mark_buffer_dirty(cow
);
236 MOD_LOG_KEY_REMOVE_WHILE_FREEING
,
237 MOD_LOG_KEY_REMOVE_WHILE_MOVING
,
239 MOD_LOG_ROOT_REPLACE
,
242 struct tree_mod_root
{
247 struct tree_mod_elem
{
253 /* this is used for MOD_LOG_KEY_* and MOD_LOG_MOVE_KEYS operations */
256 /* this is used for MOD_LOG_KEY* and MOD_LOG_ROOT_REPLACE */
259 /* those are used for op == MOD_LOG_KEY_{REPLACE,REMOVE} */
260 struct btrfs_disk_key key
;
263 /* this is used for op == MOD_LOG_MOVE_KEYS */
269 /* this is used for op == MOD_LOG_ROOT_REPLACE */
270 struct tree_mod_root old_root
;
274 * Pull a new tree mod seq number for our operation.
276 static inline u64
btrfs_inc_tree_mod_seq(struct btrfs_fs_info
*fs_info
)
278 return atomic64_inc_return(&fs_info
->tree_mod_seq
);
282 * This adds a new blocker to the tree mod log's blocker list if the @elem
283 * passed does not already have a sequence number set. So when a caller expects
284 * to record tree modifications, it should ensure to set elem->seq to zero
285 * before calling btrfs_get_tree_mod_seq.
286 * Returns a fresh, unused tree log modification sequence number, even if no new
289 u64
btrfs_get_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
290 struct seq_list
*elem
)
292 write_lock(&fs_info
->tree_mod_log_lock
);
294 elem
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
295 list_add_tail(&elem
->list
, &fs_info
->tree_mod_seq_list
);
297 write_unlock(&fs_info
->tree_mod_log_lock
);
302 void btrfs_put_tree_mod_seq(struct btrfs_fs_info
*fs_info
,
303 struct seq_list
*elem
)
305 struct rb_root
*tm_root
;
306 struct rb_node
*node
;
307 struct rb_node
*next
;
308 struct tree_mod_elem
*tm
;
309 u64 min_seq
= (u64
)-1;
310 u64 seq_putting
= elem
->seq
;
315 write_lock(&fs_info
->tree_mod_log_lock
);
316 list_del(&elem
->list
);
319 if (!list_empty(&fs_info
->tree_mod_seq_list
)) {
320 struct seq_list
*first
;
322 first
= list_first_entry(&fs_info
->tree_mod_seq_list
,
323 struct seq_list
, list
);
324 if (seq_putting
> first
->seq
) {
326 * Blocker with lower sequence number exists, we
327 * cannot remove anything from the log.
329 write_unlock(&fs_info
->tree_mod_log_lock
);
332 min_seq
= first
->seq
;
336 * anything that's lower than the lowest existing (read: blocked)
337 * sequence number can be removed from the tree.
339 tm_root
= &fs_info
->tree_mod_log
;
340 for (node
= rb_first(tm_root
); node
; node
= next
) {
341 next
= rb_next(node
);
342 tm
= rb_entry(node
, struct tree_mod_elem
, node
);
343 if (tm
->seq
>= min_seq
)
345 rb_erase(node
, tm_root
);
348 write_unlock(&fs_info
->tree_mod_log_lock
);
352 * key order of the log:
353 * node/leaf start address -> sequence
355 * The 'start address' is the logical address of the *new* root node
356 * for root replace operations, or the logical address of the affected
357 * block for all other operations.
360 __tree_mod_log_insert(struct btrfs_fs_info
*fs_info
, struct tree_mod_elem
*tm
)
362 struct rb_root
*tm_root
;
363 struct rb_node
**new;
364 struct rb_node
*parent
= NULL
;
365 struct tree_mod_elem
*cur
;
367 lockdep_assert_held_write(&fs_info
->tree_mod_log_lock
);
369 tm
->seq
= btrfs_inc_tree_mod_seq(fs_info
);
371 tm_root
= &fs_info
->tree_mod_log
;
372 new = &tm_root
->rb_node
;
374 cur
= rb_entry(*new, struct tree_mod_elem
, node
);
376 if (cur
->logical
< tm
->logical
)
377 new = &((*new)->rb_left
);
378 else if (cur
->logical
> tm
->logical
)
379 new = &((*new)->rb_right
);
380 else if (cur
->seq
< tm
->seq
)
381 new = &((*new)->rb_left
);
382 else if (cur
->seq
> tm
->seq
)
383 new = &((*new)->rb_right
);
388 rb_link_node(&tm
->node
, parent
, new);
389 rb_insert_color(&tm
->node
, tm_root
);
394 * Determines if logging can be omitted. Returns 1 if it can. Otherwise, it
395 * returns zero with the tree_mod_log_lock acquired. The caller must hold
396 * this until all tree mod log insertions are recorded in the rb tree and then
397 * write unlock fs_info::tree_mod_log_lock.
399 static inline int tree_mod_dont_log(struct btrfs_fs_info
*fs_info
,
400 struct extent_buffer
*eb
) {
402 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
404 if (eb
&& btrfs_header_level(eb
) == 0)
407 write_lock(&fs_info
->tree_mod_log_lock
);
408 if (list_empty(&(fs_info
)->tree_mod_seq_list
)) {
409 write_unlock(&fs_info
->tree_mod_log_lock
);
416 /* Similar to tree_mod_dont_log, but doesn't acquire any locks. */
417 static inline int tree_mod_need_log(const struct btrfs_fs_info
*fs_info
,
418 struct extent_buffer
*eb
)
421 if (list_empty(&(fs_info
)->tree_mod_seq_list
))
423 if (eb
&& btrfs_header_level(eb
) == 0)
429 static struct tree_mod_elem
*
430 alloc_tree_mod_elem(struct extent_buffer
*eb
, int slot
,
431 enum mod_log_op op
, gfp_t flags
)
433 struct tree_mod_elem
*tm
;
435 tm
= kzalloc(sizeof(*tm
), flags
);
439 tm
->logical
= eb
->start
;
440 if (op
!= MOD_LOG_KEY_ADD
) {
441 btrfs_node_key(eb
, &tm
->key
, slot
);
442 tm
->blockptr
= btrfs_node_blockptr(eb
, slot
);
446 tm
->generation
= btrfs_node_ptr_generation(eb
, slot
);
447 RB_CLEAR_NODE(&tm
->node
);
452 static noinline
int tree_mod_log_insert_key(struct extent_buffer
*eb
, int slot
,
453 enum mod_log_op op
, gfp_t flags
)
455 struct tree_mod_elem
*tm
;
458 if (!tree_mod_need_log(eb
->fs_info
, eb
))
461 tm
= alloc_tree_mod_elem(eb
, slot
, op
, flags
);
465 if (tree_mod_dont_log(eb
->fs_info
, eb
)) {
470 ret
= __tree_mod_log_insert(eb
->fs_info
, tm
);
471 write_unlock(&eb
->fs_info
->tree_mod_log_lock
);
478 static noinline
int tree_mod_log_insert_move(struct extent_buffer
*eb
,
479 int dst_slot
, int src_slot
, int nr_items
)
481 struct tree_mod_elem
*tm
= NULL
;
482 struct tree_mod_elem
**tm_list
= NULL
;
487 if (!tree_mod_need_log(eb
->fs_info
, eb
))
490 tm_list
= kcalloc(nr_items
, sizeof(struct tree_mod_elem
*), GFP_NOFS
);
494 tm
= kzalloc(sizeof(*tm
), GFP_NOFS
);
500 tm
->logical
= eb
->start
;
502 tm
->move
.dst_slot
= dst_slot
;
503 tm
->move
.nr_items
= nr_items
;
504 tm
->op
= MOD_LOG_MOVE_KEYS
;
506 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
507 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
+ dst_slot
,
508 MOD_LOG_KEY_REMOVE_WHILE_MOVING
, GFP_NOFS
);
515 if (tree_mod_dont_log(eb
->fs_info
, eb
))
520 * When we override something during the move, we log these removals.
521 * This can only happen when we move towards the beginning of the
522 * buffer, i.e. dst_slot < src_slot.
524 for (i
= 0; i
+ dst_slot
< src_slot
&& i
< nr_items
; i
++) {
525 ret
= __tree_mod_log_insert(eb
->fs_info
, tm_list
[i
]);
530 ret
= __tree_mod_log_insert(eb
->fs_info
, tm
);
533 write_unlock(&eb
->fs_info
->tree_mod_log_lock
);
538 for (i
= 0; i
< nr_items
; i
++) {
539 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
540 rb_erase(&tm_list
[i
]->node
, &eb
->fs_info
->tree_mod_log
);
544 write_unlock(&eb
->fs_info
->tree_mod_log_lock
);
552 __tree_mod_log_free_eb(struct btrfs_fs_info
*fs_info
,
553 struct tree_mod_elem
**tm_list
,
559 for (i
= nritems
- 1; i
>= 0; i
--) {
560 ret
= __tree_mod_log_insert(fs_info
, tm_list
[i
]);
562 for (j
= nritems
- 1; j
> i
; j
--)
563 rb_erase(&tm_list
[j
]->node
,
564 &fs_info
->tree_mod_log
);
572 static noinline
int tree_mod_log_insert_root(struct extent_buffer
*old_root
,
573 struct extent_buffer
*new_root
, int log_removal
)
575 struct btrfs_fs_info
*fs_info
= old_root
->fs_info
;
576 struct tree_mod_elem
*tm
= NULL
;
577 struct tree_mod_elem
**tm_list
= NULL
;
582 if (!tree_mod_need_log(fs_info
, NULL
))
585 if (log_removal
&& btrfs_header_level(old_root
) > 0) {
586 nritems
= btrfs_header_nritems(old_root
);
587 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*),
593 for (i
= 0; i
< nritems
; i
++) {
594 tm_list
[i
] = alloc_tree_mod_elem(old_root
, i
,
595 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
603 tm
= kzalloc(sizeof(*tm
), GFP_NOFS
);
609 tm
->logical
= new_root
->start
;
610 tm
->old_root
.logical
= old_root
->start
;
611 tm
->old_root
.level
= btrfs_header_level(old_root
);
612 tm
->generation
= btrfs_header_generation(old_root
);
613 tm
->op
= MOD_LOG_ROOT_REPLACE
;
615 if (tree_mod_dont_log(fs_info
, NULL
))
619 ret
= __tree_mod_log_free_eb(fs_info
, tm_list
, nritems
);
621 ret
= __tree_mod_log_insert(fs_info
, tm
);
623 write_unlock(&fs_info
->tree_mod_log_lock
);
632 for (i
= 0; i
< nritems
; i
++)
641 static struct tree_mod_elem
*
642 __tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
,
645 struct rb_root
*tm_root
;
646 struct rb_node
*node
;
647 struct tree_mod_elem
*cur
= NULL
;
648 struct tree_mod_elem
*found
= NULL
;
650 read_lock(&fs_info
->tree_mod_log_lock
);
651 tm_root
= &fs_info
->tree_mod_log
;
652 node
= tm_root
->rb_node
;
654 cur
= rb_entry(node
, struct tree_mod_elem
, node
);
655 if (cur
->logical
< start
) {
656 node
= node
->rb_left
;
657 } else if (cur
->logical
> start
) {
658 node
= node
->rb_right
;
659 } else if (cur
->seq
< min_seq
) {
660 node
= node
->rb_left
;
661 } else if (!smallest
) {
662 /* we want the node with the highest seq */
664 BUG_ON(found
->seq
> cur
->seq
);
666 node
= node
->rb_left
;
667 } else if (cur
->seq
> min_seq
) {
668 /* we want the node with the smallest seq */
670 BUG_ON(found
->seq
< cur
->seq
);
672 node
= node
->rb_right
;
678 read_unlock(&fs_info
->tree_mod_log_lock
);
684 * this returns the element from the log with the smallest time sequence
685 * value that's in the log (the oldest log item). any element with a time
686 * sequence lower than min_seq will be ignored.
688 static struct tree_mod_elem
*
689 tree_mod_log_search_oldest(struct btrfs_fs_info
*fs_info
, u64 start
,
692 return __tree_mod_log_search(fs_info
, start
, min_seq
, 1);
696 * this returns the element from the log with the largest time sequence
697 * value that's in the log (the most recent log item). any element with
698 * a time sequence lower than min_seq will be ignored.
700 static struct tree_mod_elem
*
701 tree_mod_log_search(struct btrfs_fs_info
*fs_info
, u64 start
, u64 min_seq
)
703 return __tree_mod_log_search(fs_info
, start
, min_seq
, 0);
706 static noinline
int tree_mod_log_eb_copy(struct extent_buffer
*dst
,
707 struct extent_buffer
*src
, unsigned long dst_offset
,
708 unsigned long src_offset
, int nr_items
)
710 struct btrfs_fs_info
*fs_info
= dst
->fs_info
;
712 struct tree_mod_elem
**tm_list
= NULL
;
713 struct tree_mod_elem
**tm_list_add
, **tm_list_rem
;
717 if (!tree_mod_need_log(fs_info
, NULL
))
720 if (btrfs_header_level(dst
) == 0 && btrfs_header_level(src
) == 0)
723 tm_list
= kcalloc(nr_items
* 2, sizeof(struct tree_mod_elem
*),
728 tm_list_add
= tm_list
;
729 tm_list_rem
= tm_list
+ nr_items
;
730 for (i
= 0; i
< nr_items
; i
++) {
731 tm_list_rem
[i
] = alloc_tree_mod_elem(src
, i
+ src_offset
,
732 MOD_LOG_KEY_REMOVE
, GFP_NOFS
);
733 if (!tm_list_rem
[i
]) {
738 tm_list_add
[i
] = alloc_tree_mod_elem(dst
, i
+ dst_offset
,
739 MOD_LOG_KEY_ADD
, GFP_NOFS
);
740 if (!tm_list_add
[i
]) {
746 if (tree_mod_dont_log(fs_info
, NULL
))
750 for (i
= 0; i
< nr_items
; i
++) {
751 ret
= __tree_mod_log_insert(fs_info
, tm_list_rem
[i
]);
754 ret
= __tree_mod_log_insert(fs_info
, tm_list_add
[i
]);
759 write_unlock(&fs_info
->tree_mod_log_lock
);
765 for (i
= 0; i
< nr_items
* 2; i
++) {
766 if (tm_list
[i
] && !RB_EMPTY_NODE(&tm_list
[i
]->node
))
767 rb_erase(&tm_list
[i
]->node
, &fs_info
->tree_mod_log
);
771 write_unlock(&fs_info
->tree_mod_log_lock
);
777 static noinline
int tree_mod_log_free_eb(struct extent_buffer
*eb
)
779 struct tree_mod_elem
**tm_list
= NULL
;
784 if (btrfs_header_level(eb
) == 0)
787 if (!tree_mod_need_log(eb
->fs_info
, NULL
))
790 nritems
= btrfs_header_nritems(eb
);
791 tm_list
= kcalloc(nritems
, sizeof(struct tree_mod_elem
*), GFP_NOFS
);
795 for (i
= 0; i
< nritems
; i
++) {
796 tm_list
[i
] = alloc_tree_mod_elem(eb
, i
,
797 MOD_LOG_KEY_REMOVE_WHILE_FREEING
, GFP_NOFS
);
804 if (tree_mod_dont_log(eb
->fs_info
, eb
))
807 ret
= __tree_mod_log_free_eb(eb
->fs_info
, tm_list
, nritems
);
808 write_unlock(&eb
->fs_info
->tree_mod_log_lock
);
816 for (i
= 0; i
< nritems
; i
++)
824 * check if the tree block can be shared by multiple trees
826 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
827 struct extent_buffer
*buf
)
830 * Tree blocks not in shareable trees and tree roots are never shared.
831 * If a block was allocated after the last snapshot and the block was
832 * not allocated by tree relocation, we know the block is not shared.
834 if (test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
835 buf
!= root
->node
&& buf
!= root
->commit_root
&&
836 (btrfs_header_generation(buf
) <=
837 btrfs_root_last_snapshot(&root
->root_item
) ||
838 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
844 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
845 struct btrfs_root
*root
,
846 struct extent_buffer
*buf
,
847 struct extent_buffer
*cow
,
850 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
858 * Backrefs update rules:
860 * Always use full backrefs for extent pointers in tree block
861 * allocated by tree relocation.
863 * If a shared tree block is no longer referenced by its owner
864 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
865 * use full backrefs for extent pointers in tree block.
867 * If a tree block is been relocating
868 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
869 * use full backrefs for extent pointers in tree block.
870 * The reason for this is some operations (such as drop tree)
871 * are only allowed for blocks use full backrefs.
874 if (btrfs_block_can_be_shared(root
, buf
)) {
875 ret
= btrfs_lookup_extent_info(trans
, fs_info
, buf
->start
,
876 btrfs_header_level(buf
), 1,
882 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
887 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
888 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
889 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
894 owner
= btrfs_header_owner(buf
);
895 BUG_ON(owner
== BTRFS_TREE_RELOC_OBJECTID
&&
896 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
));
899 if ((owner
== root
->root_key
.objectid
||
900 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
901 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
902 ret
= btrfs_inc_ref(trans
, root
, buf
, 1);
906 if (root
->root_key
.objectid
==
907 BTRFS_TREE_RELOC_OBJECTID
) {
908 ret
= btrfs_dec_ref(trans
, root
, buf
, 0);
911 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
915 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
918 if (root
->root_key
.objectid
==
919 BTRFS_TREE_RELOC_OBJECTID
)
920 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
922 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
926 if (new_flags
!= 0) {
927 int level
= btrfs_header_level(buf
);
929 ret
= btrfs_set_disk_extent_flags(trans
, buf
,
930 new_flags
, level
, 0);
935 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
936 if (root
->root_key
.objectid
==
937 BTRFS_TREE_RELOC_OBJECTID
)
938 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
940 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
943 ret
= btrfs_dec_ref(trans
, root
, buf
, 1);
947 btrfs_clean_tree_block(buf
);
953 static struct extent_buffer
*alloc_tree_block_no_bg_flush(
954 struct btrfs_trans_handle
*trans
,
955 struct btrfs_root
*root
,
957 const struct btrfs_disk_key
*disk_key
,
962 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
963 struct extent_buffer
*ret
;
966 * If we are COWing a node/leaf from the extent, chunk, device or free
967 * space trees, make sure that we do not finish block group creation of
968 * pending block groups. We do this to avoid a deadlock.
969 * COWing can result in allocation of a new chunk, and flushing pending
970 * block groups (btrfs_create_pending_block_groups()) can be triggered
971 * when finishing allocation of a new chunk. Creation of a pending block
972 * group modifies the extent, chunk, device and free space trees,
973 * therefore we could deadlock with ourselves since we are holding a
974 * lock on an extent buffer that btrfs_create_pending_block_groups() may
976 * For similar reasons, we also need to delay flushing pending block
977 * groups when splitting a leaf or node, from one of those trees, since
978 * we are holding a write lock on it and its parent or when inserting a
979 * new root node for one of those trees.
981 if (root
== fs_info
->extent_root
||
982 root
== fs_info
->chunk_root
||
983 root
== fs_info
->dev_root
||
984 root
== fs_info
->free_space_root
)
985 trans
->can_flush_pending_bgs
= false;
987 ret
= btrfs_alloc_tree_block(trans
, root
, parent_start
,
988 root
->root_key
.objectid
, disk_key
, level
,
990 trans
->can_flush_pending_bgs
= true;
996 * does the dirty work in cow of a single block. The parent block (if
997 * supplied) is updated to point to the new cow copy. The new buffer is marked
998 * dirty and returned locked. If you modify the block it needs to be marked
1001 * search_start -- an allocation hint for the new block
1003 * empty_size -- a hint that you plan on doing more cow. This is the size in
1004 * bytes the allocator should try to find free next to the block it returns.
1005 * This is just a hint and may be ignored by the allocator.
1007 static noinline
int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1008 struct btrfs_root
*root
,
1009 struct extent_buffer
*buf
,
1010 struct extent_buffer
*parent
, int parent_slot
,
1011 struct extent_buffer
**cow_ret
,
1012 u64 search_start
, u64 empty_size
)
1014 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1015 struct btrfs_disk_key disk_key
;
1016 struct extent_buffer
*cow
;
1019 int unlock_orig
= 0;
1020 u64 parent_start
= 0;
1022 if (*cow_ret
== buf
)
1025 btrfs_assert_tree_locked(buf
);
1027 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
1028 trans
->transid
!= fs_info
->running_transaction
->transid
);
1029 WARN_ON(test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
) &&
1030 trans
->transid
!= root
->last_trans
);
1032 level
= btrfs_header_level(buf
);
1035 btrfs_item_key(buf
, &disk_key
, 0);
1037 btrfs_node_key(buf
, &disk_key
, 0);
1039 if ((root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) && parent
)
1040 parent_start
= parent
->start
;
1042 cow
= alloc_tree_block_no_bg_flush(trans
, root
, parent_start
, &disk_key
,
1043 level
, search_start
, empty_size
);
1045 return PTR_ERR(cow
);
1047 /* cow is set to blocking by btrfs_init_new_buffer */
1049 copy_extent_buffer_full(cow
, buf
);
1050 btrfs_set_header_bytenr(cow
, cow
->start
);
1051 btrfs_set_header_generation(cow
, trans
->transid
);
1052 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
1053 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
1054 BTRFS_HEADER_FLAG_RELOC
);
1055 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
1056 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
1058 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
1060 write_extent_buffer_fsid(cow
, fs_info
->fs_devices
->metadata_uuid
);
1062 ret
= update_ref_for_cow(trans
, root
, buf
, cow
, &last_ref
);
1064 btrfs_abort_transaction(trans
, ret
);
1068 if (test_bit(BTRFS_ROOT_SHAREABLE
, &root
->state
)) {
1069 ret
= btrfs_reloc_cow_block(trans
, root
, buf
, cow
);
1071 btrfs_abort_transaction(trans
, ret
);
1076 if (buf
== root
->node
) {
1077 WARN_ON(parent
&& parent
!= buf
);
1078 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
1079 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
1080 parent_start
= buf
->start
;
1082 atomic_inc(&cow
->refs
);
1083 ret
= tree_mod_log_insert_root(root
->node
, cow
, 1);
1085 rcu_assign_pointer(root
->node
, cow
);
1087 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1089 free_extent_buffer(buf
);
1090 add_root_to_dirty_list(root
);
1092 WARN_ON(trans
->transid
!= btrfs_header_generation(parent
));
1093 tree_mod_log_insert_key(parent
, parent_slot
,
1094 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1095 btrfs_set_node_blockptr(parent
, parent_slot
,
1097 btrfs_set_node_ptr_generation(parent
, parent_slot
,
1099 btrfs_mark_buffer_dirty(parent
);
1101 ret
= tree_mod_log_free_eb(buf
);
1103 btrfs_abort_transaction(trans
, ret
);
1107 btrfs_free_tree_block(trans
, root
, buf
, parent_start
,
1111 btrfs_tree_unlock(buf
);
1112 free_extent_buffer_stale(buf
);
1113 btrfs_mark_buffer_dirty(cow
);
1119 * returns the logical address of the oldest predecessor of the given root.
1120 * entries older than time_seq are ignored.
1122 static struct tree_mod_elem
*__tree_mod_log_oldest_root(
1123 struct extent_buffer
*eb_root
, u64 time_seq
)
1125 struct tree_mod_elem
*tm
;
1126 struct tree_mod_elem
*found
= NULL
;
1127 u64 root_logical
= eb_root
->start
;
1134 * the very last operation that's logged for a root is the
1135 * replacement operation (if it is replaced at all). this has
1136 * the logical address of the *new* root, making it the very
1137 * first operation that's logged for this root.
1140 tm
= tree_mod_log_search_oldest(eb_root
->fs_info
, root_logical
,
1145 * if there are no tree operation for the oldest root, we simply
1146 * return it. this should only happen if that (old) root is at
1153 * if there's an operation that's not a root replacement, we
1154 * found the oldest version of our root. normally, we'll find a
1155 * MOD_LOG_KEY_REMOVE_WHILE_FREEING operation here.
1157 if (tm
->op
!= MOD_LOG_ROOT_REPLACE
)
1161 root_logical
= tm
->old_root
.logical
;
1165 /* if there's no old root to return, return what we found instead */
1173 * tm is a pointer to the first operation to rewind within eb. then, all
1174 * previous operations will be rewound (until we reach something older than
1178 __tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct extent_buffer
*eb
,
1179 u64 time_seq
, struct tree_mod_elem
*first_tm
)
1182 struct rb_node
*next
;
1183 struct tree_mod_elem
*tm
= first_tm
;
1184 unsigned long o_dst
;
1185 unsigned long o_src
;
1186 unsigned long p_size
= sizeof(struct btrfs_key_ptr
);
1188 n
= btrfs_header_nritems(eb
);
1189 read_lock(&fs_info
->tree_mod_log_lock
);
1190 while (tm
&& tm
->seq
>= time_seq
) {
1192 * all the operations are recorded with the operator used for
1193 * the modification. as we're going backwards, we do the
1194 * opposite of each operation here.
1197 case MOD_LOG_KEY_REMOVE_WHILE_FREEING
:
1198 BUG_ON(tm
->slot
< n
);
1200 case MOD_LOG_KEY_REMOVE_WHILE_MOVING
:
1201 case MOD_LOG_KEY_REMOVE
:
1202 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1203 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1204 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1208 case MOD_LOG_KEY_REPLACE
:
1209 BUG_ON(tm
->slot
>= n
);
1210 btrfs_set_node_key(eb
, &tm
->key
, tm
->slot
);
1211 btrfs_set_node_blockptr(eb
, tm
->slot
, tm
->blockptr
);
1212 btrfs_set_node_ptr_generation(eb
, tm
->slot
,
1215 case MOD_LOG_KEY_ADD
:
1216 /* if a move operation is needed it's in the log */
1219 case MOD_LOG_MOVE_KEYS
:
1220 o_dst
= btrfs_node_key_ptr_offset(tm
->slot
);
1221 o_src
= btrfs_node_key_ptr_offset(tm
->move
.dst_slot
);
1222 memmove_extent_buffer(eb
, o_dst
, o_src
,
1223 tm
->move
.nr_items
* p_size
);
1225 case MOD_LOG_ROOT_REPLACE
:
1227 * this operation is special. for roots, this must be
1228 * handled explicitly before rewinding.
1229 * for non-roots, this operation may exist if the node
1230 * was a root: root A -> child B; then A gets empty and
1231 * B is promoted to the new root. in the mod log, we'll
1232 * have a root-replace operation for B, a tree block
1233 * that is no root. we simply ignore that operation.
1237 next
= rb_next(&tm
->node
);
1240 tm
= rb_entry(next
, struct tree_mod_elem
, node
);
1241 if (tm
->logical
!= first_tm
->logical
)
1244 read_unlock(&fs_info
->tree_mod_log_lock
);
1245 btrfs_set_header_nritems(eb
, n
);
1249 * Called with eb read locked. If the buffer cannot be rewound, the same buffer
1250 * is returned. If rewind operations happen, a fresh buffer is returned. The
1251 * returned buffer is always read-locked. If the returned buffer is not the
1252 * input buffer, the lock on the input buffer is released and the input buffer
1253 * is freed (its refcount is decremented).
1255 static struct extent_buffer
*
1256 tree_mod_log_rewind(struct btrfs_fs_info
*fs_info
, struct btrfs_path
*path
,
1257 struct extent_buffer
*eb
, u64 time_seq
)
1259 struct extent_buffer
*eb_rewin
;
1260 struct tree_mod_elem
*tm
;
1265 if (btrfs_header_level(eb
) == 0)
1268 tm
= tree_mod_log_search(fs_info
, eb
->start
, time_seq
);
1272 btrfs_set_path_blocking(path
);
1273 btrfs_set_lock_blocking_read(eb
);
1275 if (tm
->op
== MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1276 BUG_ON(tm
->slot
!= 0);
1277 eb_rewin
= alloc_dummy_extent_buffer(fs_info
, eb
->start
);
1279 btrfs_tree_read_unlock_blocking(eb
);
1280 free_extent_buffer(eb
);
1283 btrfs_set_header_bytenr(eb_rewin
, eb
->start
);
1284 btrfs_set_header_backref_rev(eb_rewin
,
1285 btrfs_header_backref_rev(eb
));
1286 btrfs_set_header_owner(eb_rewin
, btrfs_header_owner(eb
));
1287 btrfs_set_header_level(eb_rewin
, btrfs_header_level(eb
));
1289 eb_rewin
= btrfs_clone_extent_buffer(eb
);
1291 btrfs_tree_read_unlock_blocking(eb
);
1292 free_extent_buffer(eb
);
1297 btrfs_tree_read_unlock_blocking(eb
);
1298 free_extent_buffer(eb
);
1300 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb_rewin
),
1301 eb_rewin
, btrfs_header_level(eb_rewin
));
1302 btrfs_tree_read_lock(eb_rewin
);
1303 __tree_mod_log_rewind(fs_info
, eb_rewin
, time_seq
, tm
);
1304 WARN_ON(btrfs_header_nritems(eb_rewin
) >
1305 BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
1311 * get_old_root() rewinds the state of @root's root node to the given @time_seq
1312 * value. If there are no changes, the current root->root_node is returned. If
1313 * anything changed in between, there's a fresh buffer allocated on which the
1314 * rewind operations are done. In any case, the returned buffer is read locked.
1315 * Returns NULL on error (with no locks held).
1317 static inline struct extent_buffer
*
1318 get_old_root(struct btrfs_root
*root
, u64 time_seq
)
1320 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1321 struct tree_mod_elem
*tm
;
1322 struct extent_buffer
*eb
= NULL
;
1323 struct extent_buffer
*eb_root
;
1324 u64 eb_root_owner
= 0;
1325 struct extent_buffer
*old
;
1326 struct tree_mod_root
*old_root
= NULL
;
1327 u64 old_generation
= 0;
1331 eb_root
= btrfs_read_lock_root_node(root
);
1332 tm
= __tree_mod_log_oldest_root(eb_root
, time_seq
);
1336 if (tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1337 old_root
= &tm
->old_root
;
1338 old_generation
= tm
->generation
;
1339 logical
= old_root
->logical
;
1340 level
= old_root
->level
;
1342 logical
= eb_root
->start
;
1343 level
= btrfs_header_level(eb_root
);
1346 tm
= tree_mod_log_search(fs_info
, logical
, time_seq
);
1347 if (old_root
&& tm
&& tm
->op
!= MOD_LOG_KEY_REMOVE_WHILE_FREEING
) {
1348 btrfs_tree_read_unlock(eb_root
);
1349 free_extent_buffer(eb_root
);
1350 old
= read_tree_block(fs_info
, logical
, 0, level
, NULL
);
1351 if (WARN_ON(IS_ERR(old
) || !extent_buffer_uptodate(old
))) {
1353 free_extent_buffer(old
);
1355 "failed to read tree block %llu from get_old_root",
1358 eb
= btrfs_clone_extent_buffer(old
);
1359 free_extent_buffer(old
);
1361 } else if (old_root
) {
1362 eb_root_owner
= btrfs_header_owner(eb_root
);
1363 btrfs_tree_read_unlock(eb_root
);
1364 free_extent_buffer(eb_root
);
1365 eb
= alloc_dummy_extent_buffer(fs_info
, logical
);
1367 btrfs_set_lock_blocking_read(eb_root
);
1368 eb
= btrfs_clone_extent_buffer(eb_root
);
1369 btrfs_tree_read_unlock_blocking(eb_root
);
1370 free_extent_buffer(eb_root
);
1376 btrfs_set_header_bytenr(eb
, eb
->start
);
1377 btrfs_set_header_backref_rev(eb
, BTRFS_MIXED_BACKREF_REV
);
1378 btrfs_set_header_owner(eb
, eb_root_owner
);
1379 btrfs_set_header_level(eb
, old_root
->level
);
1380 btrfs_set_header_generation(eb
, old_generation
);
1382 btrfs_set_buffer_lockdep_class(btrfs_header_owner(eb
), eb
,
1383 btrfs_header_level(eb
));
1384 btrfs_tree_read_lock(eb
);
1386 __tree_mod_log_rewind(fs_info
, eb
, time_seq
, tm
);
1388 WARN_ON(btrfs_header_level(eb
) != 0);
1389 WARN_ON(btrfs_header_nritems(eb
) > BTRFS_NODEPTRS_PER_BLOCK(fs_info
));
1394 int btrfs_old_root_level(struct btrfs_root
*root
, u64 time_seq
)
1396 struct tree_mod_elem
*tm
;
1398 struct extent_buffer
*eb_root
= btrfs_root_node(root
);
1400 tm
= __tree_mod_log_oldest_root(eb_root
, time_seq
);
1401 if (tm
&& tm
->op
== MOD_LOG_ROOT_REPLACE
) {
1402 level
= tm
->old_root
.level
;
1404 level
= btrfs_header_level(eb_root
);
1406 free_extent_buffer(eb_root
);
1411 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
1412 struct btrfs_root
*root
,
1413 struct extent_buffer
*buf
)
1415 if (btrfs_is_testing(root
->fs_info
))
1418 /* Ensure we can see the FORCE_COW bit */
1419 smp_mb__before_atomic();
1422 * We do not need to cow a block if
1423 * 1) this block is not created or changed in this transaction;
1424 * 2) this block does not belong to TREE_RELOC tree;
1425 * 3) the root is not forced COW.
1427 * What is forced COW:
1428 * when we create snapshot during committing the transaction,
1429 * after we've finished copying src root, we must COW the shared
1430 * block to ensure the metadata consistency.
1432 if (btrfs_header_generation(buf
) == trans
->transid
&&
1433 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
1434 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
1435 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)) &&
1436 !test_bit(BTRFS_ROOT_FORCE_COW
, &root
->state
))
1442 * cows a single block, see __btrfs_cow_block for the real work.
1443 * This version of it has extra checks so that a block isn't COWed more than
1444 * once per transaction, as long as it hasn't been written yet
1446 noinline
int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
1447 struct btrfs_root
*root
, struct extent_buffer
*buf
,
1448 struct extent_buffer
*parent
, int parent_slot
,
1449 struct extent_buffer
**cow_ret
)
1451 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1455 if (test_bit(BTRFS_ROOT_DELETING
, &root
->state
))
1457 "COW'ing blocks on a fs root that's being dropped");
1459 if (trans
->transaction
!= fs_info
->running_transaction
)
1460 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1462 fs_info
->running_transaction
->transid
);
1464 if (trans
->transid
!= fs_info
->generation
)
1465 WARN(1, KERN_CRIT
"trans %llu running %llu\n",
1466 trans
->transid
, fs_info
->generation
);
1468 if (!should_cow_block(trans
, root
, buf
)) {
1469 trans
->dirty
= true;
1474 search_start
= buf
->start
& ~((u64
)SZ_1G
- 1);
1477 btrfs_set_lock_blocking_write(parent
);
1478 btrfs_set_lock_blocking_write(buf
);
1481 * Before CoWing this block for later modification, check if it's
1482 * the subtree root and do the delayed subtree trace if needed.
1484 * Also We don't care about the error, as it's handled internally.
1486 btrfs_qgroup_trace_subtree_after_cow(trans
, root
, buf
);
1487 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
1488 parent_slot
, cow_ret
, search_start
, 0);
1490 trace_btrfs_cow_block(root
, buf
, *cow_ret
);
1496 * helper function for defrag to decide if two blocks pointed to by a
1497 * node are actually close by
1499 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
1501 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
1503 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
1508 #ifdef __LITTLE_ENDIAN
1511 * Compare two keys, on little-endian the disk order is same as CPU order and
1512 * we can avoid the conversion.
1514 static int comp_keys(const struct btrfs_disk_key
*disk_key
,
1515 const struct btrfs_key
*k2
)
1517 const struct btrfs_key
*k1
= (const struct btrfs_key
*)disk_key
;
1519 return btrfs_comp_cpu_keys(k1
, k2
);
1525 * compare two keys in a memcmp fashion
1527 static int comp_keys(const struct btrfs_disk_key
*disk
,
1528 const struct btrfs_key
*k2
)
1530 struct btrfs_key k1
;
1532 btrfs_disk_key_to_cpu(&k1
, disk
);
1534 return btrfs_comp_cpu_keys(&k1
, k2
);
1539 * same as comp_keys only with two btrfs_key's
1541 int __pure
btrfs_comp_cpu_keys(const struct btrfs_key
*k1
, const struct btrfs_key
*k2
)
1543 if (k1
->objectid
> k2
->objectid
)
1545 if (k1
->objectid
< k2
->objectid
)
1547 if (k1
->type
> k2
->type
)
1549 if (k1
->type
< k2
->type
)
1551 if (k1
->offset
> k2
->offset
)
1553 if (k1
->offset
< k2
->offset
)
1559 * this is used by the defrag code to go through all the
1560 * leaves pointed to by a node and reallocate them so that
1561 * disk order is close to key order
1563 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
1564 struct btrfs_root
*root
, struct extent_buffer
*parent
,
1565 int start_slot
, u64
*last_ret
,
1566 struct btrfs_key
*progress
)
1568 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1569 struct extent_buffer
*cur
;
1572 u64 search_start
= *last_ret
;
1582 int progress_passed
= 0;
1583 struct btrfs_disk_key disk_key
;
1585 parent_level
= btrfs_header_level(parent
);
1587 WARN_ON(trans
->transaction
!= fs_info
->running_transaction
);
1588 WARN_ON(trans
->transid
!= fs_info
->generation
);
1590 parent_nritems
= btrfs_header_nritems(parent
);
1591 blocksize
= fs_info
->nodesize
;
1592 end_slot
= parent_nritems
- 1;
1594 if (parent_nritems
<= 1)
1597 btrfs_set_lock_blocking_write(parent
);
1599 for (i
= start_slot
; i
<= end_slot
; i
++) {
1600 struct btrfs_key first_key
;
1603 btrfs_node_key(parent
, &disk_key
, i
);
1604 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
1607 progress_passed
= 1;
1608 blocknr
= btrfs_node_blockptr(parent
, i
);
1609 gen
= btrfs_node_ptr_generation(parent
, i
);
1610 btrfs_node_key_to_cpu(parent
, &first_key
, i
);
1611 if (last_block
== 0)
1612 last_block
= blocknr
;
1615 other
= btrfs_node_blockptr(parent
, i
- 1);
1616 close
= close_blocks(blocknr
, other
, blocksize
);
1618 if (!close
&& i
< end_slot
) {
1619 other
= btrfs_node_blockptr(parent
, i
+ 1);
1620 close
= close_blocks(blocknr
, other
, blocksize
);
1623 last_block
= blocknr
;
1627 cur
= find_extent_buffer(fs_info
, blocknr
);
1629 uptodate
= btrfs_buffer_uptodate(cur
, gen
, 0);
1632 if (!cur
|| !uptodate
) {
1634 cur
= read_tree_block(fs_info
, blocknr
, gen
,
1638 return PTR_ERR(cur
);
1639 } else if (!extent_buffer_uptodate(cur
)) {
1640 free_extent_buffer(cur
);
1643 } else if (!uptodate
) {
1644 err
= btrfs_read_buffer(cur
, gen
,
1645 parent_level
- 1,&first_key
);
1647 free_extent_buffer(cur
);
1652 if (search_start
== 0)
1653 search_start
= last_block
;
1655 btrfs_tree_lock(cur
);
1656 btrfs_set_lock_blocking_write(cur
);
1657 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
1660 (end_slot
- i
) * blocksize
));
1662 btrfs_tree_unlock(cur
);
1663 free_extent_buffer(cur
);
1666 search_start
= cur
->start
;
1667 last_block
= cur
->start
;
1668 *last_ret
= search_start
;
1669 btrfs_tree_unlock(cur
);
1670 free_extent_buffer(cur
);
1676 * search for key in the extent_buffer. The items start at offset p,
1677 * and they are item_size apart. There are 'max' items in p.
1679 * the slot in the array is returned via slot, and it points to
1680 * the place where you would insert key if it is not found in
1683 * slot may point to max if the key is bigger than all of the keys
1685 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
1686 unsigned long p
, int item_size
,
1687 const struct btrfs_key
*key
,
1693 const int key_size
= sizeof(struct btrfs_disk_key
);
1696 btrfs_err(eb
->fs_info
,
1697 "%s: low (%d) > high (%d) eb %llu owner %llu level %d",
1698 __func__
, low
, high
, eb
->start
,
1699 btrfs_header_owner(eb
), btrfs_header_level(eb
));
1703 while (low
< high
) {
1705 unsigned long offset
;
1706 struct btrfs_disk_key
*tmp
;
1707 struct btrfs_disk_key unaligned
;
1710 mid
= (low
+ high
) / 2;
1711 offset
= p
+ mid
* item_size
;
1712 oip
= offset_in_page(offset
);
1714 if (oip
+ key_size
<= PAGE_SIZE
) {
1715 const unsigned long idx
= offset
>> PAGE_SHIFT
;
1716 char *kaddr
= page_address(eb
->pages
[idx
]);
1718 tmp
= (struct btrfs_disk_key
*)(kaddr
+ oip
);
1720 read_extent_buffer(eb
, &unaligned
, offset
, key_size
);
1724 ret
= comp_keys(tmp
, key
);
1740 * simple bin_search frontend that does the right thing for
1743 int btrfs_bin_search(struct extent_buffer
*eb
, const struct btrfs_key
*key
,
1746 if (btrfs_header_level(eb
) == 0)
1747 return generic_bin_search(eb
,
1748 offsetof(struct btrfs_leaf
, items
),
1749 sizeof(struct btrfs_item
),
1750 key
, btrfs_header_nritems(eb
),
1753 return generic_bin_search(eb
,
1754 offsetof(struct btrfs_node
, ptrs
),
1755 sizeof(struct btrfs_key_ptr
),
1756 key
, btrfs_header_nritems(eb
),
1760 static void root_add_used(struct btrfs_root
*root
, u32 size
)
1762 spin_lock(&root
->accounting_lock
);
1763 btrfs_set_root_used(&root
->root_item
,
1764 btrfs_root_used(&root
->root_item
) + size
);
1765 spin_unlock(&root
->accounting_lock
);
1768 static void root_sub_used(struct btrfs_root
*root
, u32 size
)
1770 spin_lock(&root
->accounting_lock
);
1771 btrfs_set_root_used(&root
->root_item
,
1772 btrfs_root_used(&root
->root_item
) - size
);
1773 spin_unlock(&root
->accounting_lock
);
1776 /* given a node and slot number, this reads the blocks it points to. The
1777 * extent buffer is returned with a reference taken (but unlocked).
1779 struct extent_buffer
*btrfs_read_node_slot(struct extent_buffer
*parent
,
1782 int level
= btrfs_header_level(parent
);
1783 struct extent_buffer
*eb
;
1784 struct btrfs_key first_key
;
1786 if (slot
< 0 || slot
>= btrfs_header_nritems(parent
))
1787 return ERR_PTR(-ENOENT
);
1791 btrfs_node_key_to_cpu(parent
, &first_key
, slot
);
1792 eb
= read_tree_block(parent
->fs_info
, btrfs_node_blockptr(parent
, slot
),
1793 btrfs_node_ptr_generation(parent
, slot
),
1794 level
- 1, &first_key
);
1795 if (!IS_ERR(eb
) && !extent_buffer_uptodate(eb
)) {
1796 free_extent_buffer(eb
);
1804 * node level balancing, used to make sure nodes are in proper order for
1805 * item deletion. We balance from the top down, so we have to make sure
1806 * that a deletion won't leave an node completely empty later on.
1808 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
1809 struct btrfs_root
*root
,
1810 struct btrfs_path
*path
, int level
)
1812 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
1813 struct extent_buffer
*right
= NULL
;
1814 struct extent_buffer
*mid
;
1815 struct extent_buffer
*left
= NULL
;
1816 struct extent_buffer
*parent
= NULL
;
1820 int orig_slot
= path
->slots
[level
];
1825 mid
= path
->nodes
[level
];
1827 WARN_ON(path
->locks
[level
] != BTRFS_WRITE_LOCK
&&
1828 path
->locks
[level
] != BTRFS_WRITE_LOCK_BLOCKING
);
1829 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1831 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1833 if (level
< BTRFS_MAX_LEVEL
- 1) {
1834 parent
= path
->nodes
[level
+ 1];
1835 pslot
= path
->slots
[level
+ 1];
1839 * deal with the case where there is only one pointer in the root
1840 * by promoting the node below to a root
1843 struct extent_buffer
*child
;
1845 if (btrfs_header_nritems(mid
) != 1)
1848 /* promote the child to a root */
1849 child
= btrfs_read_node_slot(mid
, 0);
1850 if (IS_ERR(child
)) {
1851 ret
= PTR_ERR(child
);
1852 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
1856 btrfs_tree_lock(child
);
1857 btrfs_set_lock_blocking_write(child
);
1858 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
1860 btrfs_tree_unlock(child
);
1861 free_extent_buffer(child
);
1865 ret
= tree_mod_log_insert_root(root
->node
, child
, 1);
1867 rcu_assign_pointer(root
->node
, child
);
1869 add_root_to_dirty_list(root
);
1870 btrfs_tree_unlock(child
);
1872 path
->locks
[level
] = 0;
1873 path
->nodes
[level
] = NULL
;
1874 btrfs_clean_tree_block(mid
);
1875 btrfs_tree_unlock(mid
);
1876 /* once for the path */
1877 free_extent_buffer(mid
);
1879 root_sub_used(root
, mid
->len
);
1880 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1881 /* once for the root ptr */
1882 free_extent_buffer_stale(mid
);
1885 if (btrfs_header_nritems(mid
) >
1886 BTRFS_NODEPTRS_PER_BLOCK(fs_info
) / 4)
1889 left
= btrfs_read_node_slot(parent
, pslot
- 1);
1894 btrfs_tree_lock(left
);
1895 btrfs_set_lock_blocking_write(left
);
1896 wret
= btrfs_cow_block(trans
, root
, left
,
1897 parent
, pslot
- 1, &left
);
1904 right
= btrfs_read_node_slot(parent
, pslot
+ 1);
1909 btrfs_tree_lock(right
);
1910 btrfs_set_lock_blocking_write(right
);
1911 wret
= btrfs_cow_block(trans
, root
, right
,
1912 parent
, pslot
+ 1, &right
);
1919 /* first, try to make some room in the middle buffer */
1921 orig_slot
+= btrfs_header_nritems(left
);
1922 wret
= push_node_left(trans
, left
, mid
, 1);
1928 * then try to empty the right most buffer into the middle
1931 wret
= push_node_left(trans
, mid
, right
, 1);
1932 if (wret
< 0 && wret
!= -ENOSPC
)
1934 if (btrfs_header_nritems(right
) == 0) {
1935 btrfs_clean_tree_block(right
);
1936 btrfs_tree_unlock(right
);
1937 del_ptr(root
, path
, level
+ 1, pslot
+ 1);
1938 root_sub_used(root
, right
->len
);
1939 btrfs_free_tree_block(trans
, root
, right
, 0, 1);
1940 free_extent_buffer_stale(right
);
1943 struct btrfs_disk_key right_key
;
1944 btrfs_node_key(right
, &right_key
, 0);
1945 ret
= tree_mod_log_insert_key(parent
, pslot
+ 1,
1946 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1948 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1949 btrfs_mark_buffer_dirty(parent
);
1952 if (btrfs_header_nritems(mid
) == 1) {
1954 * we're not allowed to leave a node with one item in the
1955 * tree during a delete. A deletion from lower in the tree
1956 * could try to delete the only pointer in this node.
1957 * So, pull some keys from the left.
1958 * There has to be a left pointer at this point because
1959 * otherwise we would have pulled some pointers from the
1964 btrfs_handle_fs_error(fs_info
, ret
, NULL
);
1967 wret
= balance_node_right(trans
, mid
, left
);
1973 wret
= push_node_left(trans
, left
, mid
, 1);
1979 if (btrfs_header_nritems(mid
) == 0) {
1980 btrfs_clean_tree_block(mid
);
1981 btrfs_tree_unlock(mid
);
1982 del_ptr(root
, path
, level
+ 1, pslot
);
1983 root_sub_used(root
, mid
->len
);
1984 btrfs_free_tree_block(trans
, root
, mid
, 0, 1);
1985 free_extent_buffer_stale(mid
);
1988 /* update the parent key to reflect our changes */
1989 struct btrfs_disk_key mid_key
;
1990 btrfs_node_key(mid
, &mid_key
, 0);
1991 ret
= tree_mod_log_insert_key(parent
, pslot
,
1992 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
1994 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1995 btrfs_mark_buffer_dirty(parent
);
1998 /* update the path */
2000 if (btrfs_header_nritems(left
) > orig_slot
) {
2001 atomic_inc(&left
->refs
);
2002 /* left was locked after cow */
2003 path
->nodes
[level
] = left
;
2004 path
->slots
[level
+ 1] -= 1;
2005 path
->slots
[level
] = orig_slot
;
2007 btrfs_tree_unlock(mid
);
2008 free_extent_buffer(mid
);
2011 orig_slot
-= btrfs_header_nritems(left
);
2012 path
->slots
[level
] = orig_slot
;
2015 /* double check we haven't messed things up */
2017 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
2021 btrfs_tree_unlock(right
);
2022 free_extent_buffer(right
);
2025 if (path
->nodes
[level
] != left
)
2026 btrfs_tree_unlock(left
);
2027 free_extent_buffer(left
);
2032 /* Node balancing for insertion. Here we only split or push nodes around
2033 * when they are completely full. This is also done top down, so we
2034 * have to be pessimistic.
2036 static noinline
int push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
2037 struct btrfs_root
*root
,
2038 struct btrfs_path
*path
, int level
)
2040 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2041 struct extent_buffer
*right
= NULL
;
2042 struct extent_buffer
*mid
;
2043 struct extent_buffer
*left
= NULL
;
2044 struct extent_buffer
*parent
= NULL
;
2048 int orig_slot
= path
->slots
[level
];
2053 mid
= path
->nodes
[level
];
2054 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
2056 if (level
< BTRFS_MAX_LEVEL
- 1) {
2057 parent
= path
->nodes
[level
+ 1];
2058 pslot
= path
->slots
[level
+ 1];
2064 left
= btrfs_read_node_slot(parent
, pslot
- 1);
2068 /* first, try to make some room in the middle buffer */
2072 btrfs_tree_lock(left
);
2073 btrfs_set_lock_blocking_write(left
);
2075 left_nr
= btrfs_header_nritems(left
);
2076 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - 1) {
2079 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
2084 wret
= push_node_left(trans
, left
, mid
, 0);
2090 struct btrfs_disk_key disk_key
;
2091 orig_slot
+= left_nr
;
2092 btrfs_node_key(mid
, &disk_key
, 0);
2093 ret
= tree_mod_log_insert_key(parent
, pslot
,
2094 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
2096 btrfs_set_node_key(parent
, &disk_key
, pslot
);
2097 btrfs_mark_buffer_dirty(parent
);
2098 if (btrfs_header_nritems(left
) > orig_slot
) {
2099 path
->nodes
[level
] = left
;
2100 path
->slots
[level
+ 1] -= 1;
2101 path
->slots
[level
] = orig_slot
;
2102 btrfs_tree_unlock(mid
);
2103 free_extent_buffer(mid
);
2106 btrfs_header_nritems(left
);
2107 path
->slots
[level
] = orig_slot
;
2108 btrfs_tree_unlock(left
);
2109 free_extent_buffer(left
);
2113 btrfs_tree_unlock(left
);
2114 free_extent_buffer(left
);
2116 right
= btrfs_read_node_slot(parent
, pslot
+ 1);
2121 * then try to empty the right most buffer into the middle
2126 btrfs_tree_lock(right
);
2127 btrfs_set_lock_blocking_write(right
);
2129 right_nr
= btrfs_header_nritems(right
);
2130 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - 1) {
2133 ret
= btrfs_cow_block(trans
, root
, right
,
2139 wret
= balance_node_right(trans
, right
, mid
);
2145 struct btrfs_disk_key disk_key
;
2147 btrfs_node_key(right
, &disk_key
, 0);
2148 ret
= tree_mod_log_insert_key(parent
, pslot
+ 1,
2149 MOD_LOG_KEY_REPLACE
, GFP_NOFS
);
2151 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
2152 btrfs_mark_buffer_dirty(parent
);
2154 if (btrfs_header_nritems(mid
) <= orig_slot
) {
2155 path
->nodes
[level
] = right
;
2156 path
->slots
[level
+ 1] += 1;
2157 path
->slots
[level
] = orig_slot
-
2158 btrfs_header_nritems(mid
);
2159 btrfs_tree_unlock(mid
);
2160 free_extent_buffer(mid
);
2162 btrfs_tree_unlock(right
);
2163 free_extent_buffer(right
);
2167 btrfs_tree_unlock(right
);
2168 free_extent_buffer(right
);
2174 * readahead one full node of leaves, finding things that are close
2175 * to the block in 'slot', and triggering ra on them.
2177 static void reada_for_search(struct btrfs_fs_info
*fs_info
,
2178 struct btrfs_path
*path
,
2179 int level
, int slot
, u64 objectid
)
2181 struct extent_buffer
*node
;
2182 struct btrfs_disk_key disk_key
;
2187 struct extent_buffer
*eb
;
2195 if (!path
->nodes
[level
])
2198 node
= path
->nodes
[level
];
2200 search
= btrfs_node_blockptr(node
, slot
);
2201 blocksize
= fs_info
->nodesize
;
2202 eb
= find_extent_buffer(fs_info
, search
);
2204 free_extent_buffer(eb
);
2210 nritems
= btrfs_header_nritems(node
);
2214 if (path
->reada
== READA_BACK
) {
2218 } else if (path
->reada
== READA_FORWARD
) {
2223 if (path
->reada
== READA_BACK
&& objectid
) {
2224 btrfs_node_key(node
, &disk_key
, nr
);
2225 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
2228 search
= btrfs_node_blockptr(node
, nr
);
2229 if ((search
<= target
&& target
- search
<= 65536) ||
2230 (search
> target
&& search
- target
<= 65536)) {
2231 readahead_tree_block(fs_info
, search
);
2235 if ((nread
> 65536 || nscan
> 32))
2240 static noinline
void reada_for_balance(struct btrfs_fs_info
*fs_info
,
2241 struct btrfs_path
*path
, int level
)
2245 struct extent_buffer
*parent
;
2246 struct extent_buffer
*eb
;
2251 parent
= path
->nodes
[level
+ 1];
2255 nritems
= btrfs_header_nritems(parent
);
2256 slot
= path
->slots
[level
+ 1];
2259 block1
= btrfs_node_blockptr(parent
, slot
- 1);
2260 gen
= btrfs_node_ptr_generation(parent
, slot
- 1);
2261 eb
= find_extent_buffer(fs_info
, block1
);
2263 * if we get -eagain from btrfs_buffer_uptodate, we
2264 * don't want to return eagain here. That will loop
2267 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2269 free_extent_buffer(eb
);
2271 if (slot
+ 1 < nritems
) {
2272 block2
= btrfs_node_blockptr(parent
, slot
+ 1);
2273 gen
= btrfs_node_ptr_generation(parent
, slot
+ 1);
2274 eb
= find_extent_buffer(fs_info
, block2
);
2275 if (eb
&& btrfs_buffer_uptodate(eb
, gen
, 1) != 0)
2277 free_extent_buffer(eb
);
2281 readahead_tree_block(fs_info
, block1
);
2283 readahead_tree_block(fs_info
, block2
);
2288 * when we walk down the tree, it is usually safe to unlock the higher layers
2289 * in the tree. The exceptions are when our path goes through slot 0, because
2290 * operations on the tree might require changing key pointers higher up in the
2293 * callers might also have set path->keep_locks, which tells this code to keep
2294 * the lock if the path points to the last slot in the block. This is part of
2295 * walking through the tree, and selecting the next slot in the higher block.
2297 * lowest_unlock sets the lowest level in the tree we're allowed to unlock. so
2298 * if lowest_unlock is 1, level 0 won't be unlocked
2300 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
2301 int lowest_unlock
, int min_write_lock_level
,
2302 int *write_lock_level
)
2305 int skip_level
= level
;
2307 struct extent_buffer
*t
;
2309 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
2310 if (!path
->nodes
[i
])
2312 if (!path
->locks
[i
])
2314 if (!no_skips
&& path
->slots
[i
] == 0) {
2318 if (!no_skips
&& path
->keep_locks
) {
2321 nritems
= btrfs_header_nritems(t
);
2322 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
2327 if (skip_level
< i
&& i
>= lowest_unlock
)
2331 if (i
>= lowest_unlock
&& i
> skip_level
) {
2332 btrfs_tree_unlock_rw(t
, path
->locks
[i
]);
2334 if (write_lock_level
&&
2335 i
> min_write_lock_level
&&
2336 i
<= *write_lock_level
) {
2337 *write_lock_level
= i
- 1;
2344 * helper function for btrfs_search_slot. The goal is to find a block
2345 * in cache without setting the path to blocking. If we find the block
2346 * we return zero and the path is unchanged.
2348 * If we can't find the block, we set the path blocking and do some
2349 * reada. -EAGAIN is returned and the search must be repeated.
2352 read_block_for_search(struct btrfs_root
*root
, struct btrfs_path
*p
,
2353 struct extent_buffer
**eb_ret
, int level
, int slot
,
2354 const struct btrfs_key
*key
)
2356 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2359 struct extent_buffer
*tmp
;
2360 struct btrfs_key first_key
;
2364 blocknr
= btrfs_node_blockptr(*eb_ret
, slot
);
2365 gen
= btrfs_node_ptr_generation(*eb_ret
, slot
);
2366 parent_level
= btrfs_header_level(*eb_ret
);
2367 btrfs_node_key_to_cpu(*eb_ret
, &first_key
, slot
);
2369 tmp
= find_extent_buffer(fs_info
, blocknr
);
2371 /* first we do an atomic uptodate check */
2372 if (btrfs_buffer_uptodate(tmp
, gen
, 1) > 0) {
2374 * Do extra check for first_key, eb can be stale due to
2375 * being cached, read from scrub, or have multiple
2376 * parents (shared tree blocks).
2378 if (btrfs_verify_level_key(tmp
,
2379 parent_level
- 1, &first_key
, gen
)) {
2380 free_extent_buffer(tmp
);
2387 /* the pages were up to date, but we failed
2388 * the generation number check. Do a full
2389 * read for the generation number that is correct.
2390 * We must do this without dropping locks so
2391 * we can trust our generation number
2393 btrfs_set_path_blocking(p
);
2395 /* now we're allowed to do a blocking uptodate check */
2396 ret
= btrfs_read_buffer(tmp
, gen
, parent_level
- 1, &first_key
);
2401 free_extent_buffer(tmp
);
2402 btrfs_release_path(p
);
2407 * reduce lock contention at high levels
2408 * of the btree by dropping locks before
2409 * we read. Don't release the lock on the current
2410 * level because we need to walk this node to figure
2411 * out which blocks to read.
2413 btrfs_unlock_up_safe(p
, level
+ 1);
2414 btrfs_set_path_blocking(p
);
2416 if (p
->reada
!= READA_NONE
)
2417 reada_for_search(fs_info
, p
, level
, slot
, key
->objectid
);
2420 tmp
= read_tree_block(fs_info
, blocknr
, gen
, parent_level
- 1,
2424 * If the read above didn't mark this buffer up to date,
2425 * it will never end up being up to date. Set ret to EIO now
2426 * and give up so that our caller doesn't loop forever
2429 if (!extent_buffer_uptodate(tmp
))
2431 free_extent_buffer(tmp
);
2436 btrfs_release_path(p
);
2441 * helper function for btrfs_search_slot. This does all of the checks
2442 * for node-level blocks and does any balancing required based on
2445 * If no extra work was required, zero is returned. If we had to
2446 * drop the path, -EAGAIN is returned and btrfs_search_slot must
2450 setup_nodes_for_search(struct btrfs_trans_handle
*trans
,
2451 struct btrfs_root
*root
, struct btrfs_path
*p
,
2452 struct extent_buffer
*b
, int level
, int ins_len
,
2453 int *write_lock_level
)
2455 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2458 if ((p
->search_for_split
|| ins_len
> 0) && btrfs_header_nritems(b
) >=
2459 BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - 3) {
2462 if (*write_lock_level
< level
+ 1) {
2463 *write_lock_level
= level
+ 1;
2464 btrfs_release_path(p
);
2468 btrfs_set_path_blocking(p
);
2469 reada_for_balance(fs_info
, p
, level
);
2470 sret
= split_node(trans
, root
, p
, level
);
2477 b
= p
->nodes
[level
];
2478 } else if (ins_len
< 0 && btrfs_header_nritems(b
) <
2479 BTRFS_NODEPTRS_PER_BLOCK(fs_info
) / 2) {
2482 if (*write_lock_level
< level
+ 1) {
2483 *write_lock_level
= level
+ 1;
2484 btrfs_release_path(p
);
2488 btrfs_set_path_blocking(p
);
2489 reada_for_balance(fs_info
, p
, level
);
2490 sret
= balance_level(trans
, root
, p
, level
);
2496 b
= p
->nodes
[level
];
2498 btrfs_release_path(p
);
2501 BUG_ON(btrfs_header_nritems(b
) == 1);
2511 int btrfs_find_item(struct btrfs_root
*fs_root
, struct btrfs_path
*path
,
2512 u64 iobjectid
, u64 ioff
, u8 key_type
,
2513 struct btrfs_key
*found_key
)
2516 struct btrfs_key key
;
2517 struct extent_buffer
*eb
;
2522 key
.type
= key_type
;
2523 key
.objectid
= iobjectid
;
2526 ret
= btrfs_search_slot(NULL
, fs_root
, &key
, path
, 0, 0);
2530 eb
= path
->nodes
[0];
2531 if (ret
&& path
->slots
[0] >= btrfs_header_nritems(eb
)) {
2532 ret
= btrfs_next_leaf(fs_root
, path
);
2535 eb
= path
->nodes
[0];
2538 btrfs_item_key_to_cpu(eb
, found_key
, path
->slots
[0]);
2539 if (found_key
->type
!= key
.type
||
2540 found_key
->objectid
!= key
.objectid
)
2546 static struct extent_buffer
*btrfs_search_slot_get_root(struct btrfs_root
*root
,
2547 struct btrfs_path
*p
,
2548 int write_lock_level
)
2550 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2551 struct extent_buffer
*b
;
2555 /* We try very hard to do read locks on the root */
2556 root_lock
= BTRFS_READ_LOCK
;
2558 if (p
->search_commit_root
) {
2560 * The commit roots are read only so we always do read locks,
2561 * and we always must hold the commit_root_sem when doing
2562 * searches on them, the only exception is send where we don't
2563 * want to block transaction commits for a long time, so
2564 * we need to clone the commit root in order to avoid races
2565 * with transaction commits that create a snapshot of one of
2566 * the roots used by a send operation.
2568 if (p
->need_commit_sem
) {
2569 down_read(&fs_info
->commit_root_sem
);
2570 b
= btrfs_clone_extent_buffer(root
->commit_root
);
2571 up_read(&fs_info
->commit_root_sem
);
2573 return ERR_PTR(-ENOMEM
);
2576 b
= root
->commit_root
;
2577 atomic_inc(&b
->refs
);
2579 level
= btrfs_header_level(b
);
2581 * Ensure that all callers have set skip_locking when
2582 * p->search_commit_root = 1.
2584 ASSERT(p
->skip_locking
== 1);
2589 if (p
->skip_locking
) {
2590 b
= btrfs_root_node(root
);
2591 level
= btrfs_header_level(b
);
2596 * If the level is set to maximum, we can skip trying to get the read
2599 if (write_lock_level
< BTRFS_MAX_LEVEL
) {
2601 * We don't know the level of the root node until we actually
2602 * have it read locked
2604 b
= btrfs_read_lock_root_node(root
);
2605 level
= btrfs_header_level(b
);
2606 if (level
> write_lock_level
)
2609 /* Whoops, must trade for write lock */
2610 btrfs_tree_read_unlock(b
);
2611 free_extent_buffer(b
);
2614 b
= btrfs_lock_root_node(root
);
2615 root_lock
= BTRFS_WRITE_LOCK
;
2617 /* The level might have changed, check again */
2618 level
= btrfs_header_level(b
);
2621 p
->nodes
[level
] = b
;
2622 if (!p
->skip_locking
)
2623 p
->locks
[level
] = root_lock
;
2625 * Callers are responsible for dropping b's references.
2632 * btrfs_search_slot - look for a key in a tree and perform necessary
2633 * modifications to preserve tree invariants.
2635 * @trans: Handle of transaction, used when modifying the tree
2636 * @p: Holds all btree nodes along the search path
2637 * @root: The root node of the tree
2638 * @key: The key we are looking for
2639 * @ins_len: Indicates purpose of search, for inserts it is 1, for
2640 * deletions it's -1. 0 for plain searches
2641 * @cow: boolean should CoW operations be performed. Must always be 1
2642 * when modifying the tree.
2644 * If @ins_len > 0, nodes and leaves will be split as we walk down the tree.
2645 * If @ins_len < 0, nodes will be merged as we walk down the tree (if possible)
2647 * If @key is found, 0 is returned and you can find the item in the leaf level
2648 * of the path (level 0)
2650 * If @key isn't found, 1 is returned and the leaf level of the path (level 0)
2651 * points to the slot where it should be inserted
2653 * If an error is encountered while searching the tree a negative error number
2656 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2657 const struct btrfs_key
*key
, struct btrfs_path
*p
,
2658 int ins_len
, int cow
)
2660 struct extent_buffer
*b
;
2665 int lowest_unlock
= 1;
2666 /* everything at write_lock_level or lower must be write locked */
2667 int write_lock_level
= 0;
2668 u8 lowest_level
= 0;
2669 int min_write_lock_level
;
2672 lowest_level
= p
->lowest_level
;
2673 WARN_ON(lowest_level
&& ins_len
> 0);
2674 WARN_ON(p
->nodes
[0] != NULL
);
2675 BUG_ON(!cow
&& ins_len
);
2680 /* when we are removing items, we might have to go up to level
2681 * two as we update tree pointers Make sure we keep write
2682 * for those levels as well
2684 write_lock_level
= 2;
2685 } else if (ins_len
> 0) {
2687 * for inserting items, make sure we have a write lock on
2688 * level 1 so we can update keys
2690 write_lock_level
= 1;
2694 write_lock_level
= -1;
2696 if (cow
&& (p
->keep_locks
|| p
->lowest_level
))
2697 write_lock_level
= BTRFS_MAX_LEVEL
;
2699 min_write_lock_level
= write_lock_level
;
2703 b
= btrfs_search_slot_get_root(root
, p
, write_lock_level
);
2712 level
= btrfs_header_level(b
);
2715 bool last_level
= (level
== (BTRFS_MAX_LEVEL
- 1));
2718 * if we don't really need to cow this block
2719 * then we don't want to set the path blocking,
2720 * so we test it here
2722 if (!should_cow_block(trans
, root
, b
)) {
2723 trans
->dirty
= true;
2728 * must have write locks on this node and the
2731 if (level
> write_lock_level
||
2732 (level
+ 1 > write_lock_level
&&
2733 level
+ 1 < BTRFS_MAX_LEVEL
&&
2734 p
->nodes
[level
+ 1])) {
2735 write_lock_level
= level
+ 1;
2736 btrfs_release_path(p
);
2740 btrfs_set_path_blocking(p
);
2742 err
= btrfs_cow_block(trans
, root
, b
, NULL
, 0,
2745 err
= btrfs_cow_block(trans
, root
, b
,
2746 p
->nodes
[level
+ 1],
2747 p
->slots
[level
+ 1], &b
);
2754 p
->nodes
[level
] = b
;
2756 * Leave path with blocking locks to avoid massive
2757 * lock context switch, this is made on purpose.
2761 * we have a lock on b and as long as we aren't changing
2762 * the tree, there is no way to for the items in b to change.
2763 * It is safe to drop the lock on our parent before we
2764 * go through the expensive btree search on b.
2766 * If we're inserting or deleting (ins_len != 0), then we might
2767 * be changing slot zero, which may require changing the parent.
2768 * So, we can't drop the lock until after we know which slot
2769 * we're operating on.
2771 if (!ins_len
&& !p
->keep_locks
) {
2774 if (u
< BTRFS_MAX_LEVEL
&& p
->locks
[u
]) {
2775 btrfs_tree_unlock_rw(p
->nodes
[u
], p
->locks
[u
]);
2781 * If btrfs_bin_search returns an exact match (prev_cmp == 0)
2782 * we can safely assume the target key will always be in slot 0
2783 * on lower levels due to the invariants BTRFS' btree provides,
2784 * namely that a btrfs_key_ptr entry always points to the
2785 * lowest key in the child node, thus we can skip searching
2788 if (prev_cmp
== 0) {
2792 ret
= btrfs_bin_search(b
, key
, &slot
);
2799 p
->slots
[level
] = slot
;
2801 btrfs_leaf_free_space(b
) < ins_len
) {
2802 if (write_lock_level
< 1) {
2803 write_lock_level
= 1;
2804 btrfs_release_path(p
);
2808 btrfs_set_path_blocking(p
);
2809 err
= split_leaf(trans
, root
, key
,
2810 p
, ins_len
, ret
== 0);
2818 if (!p
->search_for_split
)
2819 unlock_up(p
, level
, lowest_unlock
,
2820 min_write_lock_level
, NULL
);
2823 if (ret
&& slot
> 0) {
2827 p
->slots
[level
] = slot
;
2828 err
= setup_nodes_for_search(trans
, root
, p
, b
, level
, ins_len
,
2836 b
= p
->nodes
[level
];
2837 slot
= p
->slots
[level
];
2840 * Slot 0 is special, if we change the key we have to update
2841 * the parent pointer which means we must have a write lock on
2844 if (slot
== 0 && ins_len
&& write_lock_level
< level
+ 1) {
2845 write_lock_level
= level
+ 1;
2846 btrfs_release_path(p
);
2850 unlock_up(p
, level
, lowest_unlock
, min_write_lock_level
,
2853 if (level
== lowest_level
) {
2859 err
= read_block_for_search(root
, p
, &b
, level
, slot
, key
);
2867 if (!p
->skip_locking
) {
2868 level
= btrfs_header_level(b
);
2869 if (level
<= write_lock_level
) {
2870 if (!btrfs_try_tree_write_lock(b
)) {
2871 btrfs_set_path_blocking(p
);
2874 p
->locks
[level
] = BTRFS_WRITE_LOCK
;
2876 if (!btrfs_tree_read_lock_atomic(b
)) {
2877 btrfs_set_path_blocking(p
);
2878 btrfs_tree_read_lock(b
);
2880 p
->locks
[level
] = BTRFS_READ_LOCK
;
2882 p
->nodes
[level
] = b
;
2888 * we don't really know what they plan on doing with the path
2889 * from here on, so for now just mark it as blocking
2891 if (!p
->leave_spinning
)
2892 btrfs_set_path_blocking(p
);
2893 if (ret
< 0 && !p
->skip_release_on_error
)
2894 btrfs_release_path(p
);
2899 * Like btrfs_search_slot, this looks for a key in the given tree. It uses the
2900 * current state of the tree together with the operations recorded in the tree
2901 * modification log to search for the key in a previous version of this tree, as
2902 * denoted by the time_seq parameter.
2904 * Naturally, there is no support for insert, delete or cow operations.
2906 * The resulting path and return value will be set up as if we called
2907 * btrfs_search_slot at that point in time with ins_len and cow both set to 0.
2909 int btrfs_search_old_slot(struct btrfs_root
*root
, const struct btrfs_key
*key
,
2910 struct btrfs_path
*p
, u64 time_seq
)
2912 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
2913 struct extent_buffer
*b
;
2918 int lowest_unlock
= 1;
2919 u8 lowest_level
= 0;
2921 lowest_level
= p
->lowest_level
;
2922 WARN_ON(p
->nodes
[0] != NULL
);
2924 if (p
->search_commit_root
) {
2926 return btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
2930 b
= get_old_root(root
, time_seq
);
2935 level
= btrfs_header_level(b
);
2936 p
->locks
[level
] = BTRFS_READ_LOCK
;
2941 level
= btrfs_header_level(b
);
2942 p
->nodes
[level
] = b
;
2945 * we have a lock on b and as long as we aren't changing
2946 * the tree, there is no way to for the items in b to change.
2947 * It is safe to drop the lock on our parent before we
2948 * go through the expensive btree search on b.
2950 btrfs_unlock_up_safe(p
, level
+ 1);
2952 ret
= btrfs_bin_search(b
, key
, &slot
);
2957 p
->slots
[level
] = slot
;
2958 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2962 if (ret
&& slot
> 0) {
2966 p
->slots
[level
] = slot
;
2967 unlock_up(p
, level
, lowest_unlock
, 0, NULL
);
2969 if (level
== lowest_level
) {
2975 err
= read_block_for_search(root
, p
, &b
, level
, slot
, key
);
2983 level
= btrfs_header_level(b
);
2984 if (!btrfs_tree_read_lock_atomic(b
)) {
2985 btrfs_set_path_blocking(p
);
2986 btrfs_tree_read_lock(b
);
2988 b
= tree_mod_log_rewind(fs_info
, p
, b
, time_seq
);
2993 p
->locks
[level
] = BTRFS_READ_LOCK
;
2994 p
->nodes
[level
] = b
;
2998 if (!p
->leave_spinning
)
2999 btrfs_set_path_blocking(p
);
3001 btrfs_release_path(p
);
3007 * helper to use instead of search slot if no exact match is needed but
3008 * instead the next or previous item should be returned.
3009 * When find_higher is true, the next higher item is returned, the next lower
3011 * When return_any and find_higher are both true, and no higher item is found,
3012 * return the next lower instead.
3013 * When return_any is true and find_higher is false, and no lower item is found,
3014 * return the next higher instead.
3015 * It returns 0 if any item is found, 1 if none is found (tree empty), and
3018 int btrfs_search_slot_for_read(struct btrfs_root
*root
,
3019 const struct btrfs_key
*key
,
3020 struct btrfs_path
*p
, int find_higher
,
3024 struct extent_buffer
*leaf
;
3027 ret
= btrfs_search_slot(NULL
, root
, key
, p
, 0, 0);
3031 * a return value of 1 means the path is at the position where the
3032 * item should be inserted. Normally this is the next bigger item,
3033 * but in case the previous item is the last in a leaf, path points
3034 * to the first free slot in the previous leaf, i.e. at an invalid
3040 if (p
->slots
[0] >= btrfs_header_nritems(leaf
)) {
3041 ret
= btrfs_next_leaf(root
, p
);
3047 * no higher item found, return the next
3052 btrfs_release_path(p
);
3056 if (p
->slots
[0] == 0) {
3057 ret
= btrfs_prev_leaf(root
, p
);
3062 if (p
->slots
[0] == btrfs_header_nritems(leaf
))
3069 * no lower item found, return the next
3074 btrfs_release_path(p
);
3084 * adjust the pointers going up the tree, starting at level
3085 * making sure the right key of each node is points to 'key'.
3086 * This is used after shifting pointers to the left, so it stops
3087 * fixing up pointers when a given leaf/node is not in slot 0 of the
3091 static void fixup_low_keys(struct btrfs_path
*path
,
3092 struct btrfs_disk_key
*key
, int level
)
3095 struct extent_buffer
*t
;
3098 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
3099 int tslot
= path
->slots
[i
];
3101 if (!path
->nodes
[i
])
3104 ret
= tree_mod_log_insert_key(t
, tslot
, MOD_LOG_KEY_REPLACE
,
3107 btrfs_set_node_key(t
, key
, tslot
);
3108 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
3117 * This function isn't completely safe. It's the caller's responsibility
3118 * that the new key won't break the order
3120 void btrfs_set_item_key_safe(struct btrfs_fs_info
*fs_info
,
3121 struct btrfs_path
*path
,
3122 const struct btrfs_key
*new_key
)
3124 struct btrfs_disk_key disk_key
;
3125 struct extent_buffer
*eb
;
3128 eb
= path
->nodes
[0];
3129 slot
= path
->slots
[0];
3131 btrfs_item_key(eb
, &disk_key
, slot
- 1);
3132 if (unlikely(comp_keys(&disk_key
, new_key
) >= 0)) {
3134 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3135 slot
, btrfs_disk_key_objectid(&disk_key
),
3136 btrfs_disk_key_type(&disk_key
),
3137 btrfs_disk_key_offset(&disk_key
),
3138 new_key
->objectid
, new_key
->type
,
3140 btrfs_print_leaf(eb
);
3144 if (slot
< btrfs_header_nritems(eb
) - 1) {
3145 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
3146 if (unlikely(comp_keys(&disk_key
, new_key
) <= 0)) {
3148 "slot %u key (%llu %u %llu) new key (%llu %u %llu)",
3149 slot
, btrfs_disk_key_objectid(&disk_key
),
3150 btrfs_disk_key_type(&disk_key
),
3151 btrfs_disk_key_offset(&disk_key
),
3152 new_key
->objectid
, new_key
->type
,
3154 btrfs_print_leaf(eb
);
3159 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
3160 btrfs_set_item_key(eb
, &disk_key
, slot
);
3161 btrfs_mark_buffer_dirty(eb
);
3163 fixup_low_keys(path
, &disk_key
, 1);
3167 * try to push data from one node into the next node left in the
3170 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
3171 * error, and > 0 if there was no room in the left hand block.
3173 static int push_node_left(struct btrfs_trans_handle
*trans
,
3174 struct extent_buffer
*dst
,
3175 struct extent_buffer
*src
, int empty
)
3177 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3183 src_nritems
= btrfs_header_nritems(src
);
3184 dst_nritems
= btrfs_header_nritems(dst
);
3185 push_items
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - dst_nritems
;
3186 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3187 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3189 if (!empty
&& src_nritems
<= 8)
3192 if (push_items
<= 0)
3196 push_items
= min(src_nritems
, push_items
);
3197 if (push_items
< src_nritems
) {
3198 /* leave at least 8 pointers in the node if
3199 * we aren't going to empty it
3201 if (src_nritems
- push_items
< 8) {
3202 if (push_items
<= 8)
3208 push_items
= min(src_nritems
- 8, push_items
);
3210 ret
= tree_mod_log_eb_copy(dst
, src
, dst_nritems
, 0, push_items
);
3212 btrfs_abort_transaction(trans
, ret
);
3215 copy_extent_buffer(dst
, src
,
3216 btrfs_node_key_ptr_offset(dst_nritems
),
3217 btrfs_node_key_ptr_offset(0),
3218 push_items
* sizeof(struct btrfs_key_ptr
));
3220 if (push_items
< src_nritems
) {
3222 * Don't call tree_mod_log_insert_move here, key removal was
3223 * already fully logged by tree_mod_log_eb_copy above.
3225 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
3226 btrfs_node_key_ptr_offset(push_items
),
3227 (src_nritems
- push_items
) *
3228 sizeof(struct btrfs_key_ptr
));
3230 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3231 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3232 btrfs_mark_buffer_dirty(src
);
3233 btrfs_mark_buffer_dirty(dst
);
3239 * try to push data from one node into the next node right in the
3242 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
3243 * error, and > 0 if there was no room in the right hand block.
3245 * this will only push up to 1/2 the contents of the left node over
3247 static int balance_node_right(struct btrfs_trans_handle
*trans
,
3248 struct extent_buffer
*dst
,
3249 struct extent_buffer
*src
)
3251 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
3258 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
3259 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
3261 src_nritems
= btrfs_header_nritems(src
);
3262 dst_nritems
= btrfs_header_nritems(dst
);
3263 push_items
= BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - dst_nritems
;
3264 if (push_items
<= 0)
3267 if (src_nritems
< 4)
3270 max_push
= src_nritems
/ 2 + 1;
3271 /* don't try to empty the node */
3272 if (max_push
>= src_nritems
)
3275 if (max_push
< push_items
)
3276 push_items
= max_push
;
3278 ret
= tree_mod_log_insert_move(dst
, push_items
, 0, dst_nritems
);
3280 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
3281 btrfs_node_key_ptr_offset(0),
3283 sizeof(struct btrfs_key_ptr
));
3285 ret
= tree_mod_log_eb_copy(dst
, src
, 0, src_nritems
- push_items
,
3288 btrfs_abort_transaction(trans
, ret
);
3291 copy_extent_buffer(dst
, src
,
3292 btrfs_node_key_ptr_offset(0),
3293 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
3294 push_items
* sizeof(struct btrfs_key_ptr
));
3296 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
3297 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
3299 btrfs_mark_buffer_dirty(src
);
3300 btrfs_mark_buffer_dirty(dst
);
3306 * helper function to insert a new root level in the tree.
3307 * A new node is allocated, and a single item is inserted to
3308 * point to the existing root
3310 * returns zero on success or < 0 on failure.
3312 static noinline
int insert_new_root(struct btrfs_trans_handle
*trans
,
3313 struct btrfs_root
*root
,
3314 struct btrfs_path
*path
, int level
)
3316 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3318 struct extent_buffer
*lower
;
3319 struct extent_buffer
*c
;
3320 struct extent_buffer
*old
;
3321 struct btrfs_disk_key lower_key
;
3324 BUG_ON(path
->nodes
[level
]);
3325 BUG_ON(path
->nodes
[level
-1] != root
->node
);
3327 lower
= path
->nodes
[level
-1];
3329 btrfs_item_key(lower
, &lower_key
, 0);
3331 btrfs_node_key(lower
, &lower_key
, 0);
3333 c
= alloc_tree_block_no_bg_flush(trans
, root
, 0, &lower_key
, level
,
3334 root
->node
->start
, 0);
3338 root_add_used(root
, fs_info
->nodesize
);
3340 btrfs_set_header_nritems(c
, 1);
3341 btrfs_set_node_key(c
, &lower_key
, 0);
3342 btrfs_set_node_blockptr(c
, 0, lower
->start
);
3343 lower_gen
= btrfs_header_generation(lower
);
3344 WARN_ON(lower_gen
!= trans
->transid
);
3346 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
3348 btrfs_mark_buffer_dirty(c
);
3351 ret
= tree_mod_log_insert_root(root
->node
, c
, 0);
3353 rcu_assign_pointer(root
->node
, c
);
3355 /* the super has an extra ref to root->node */
3356 free_extent_buffer(old
);
3358 add_root_to_dirty_list(root
);
3359 atomic_inc(&c
->refs
);
3360 path
->nodes
[level
] = c
;
3361 path
->locks
[level
] = BTRFS_WRITE_LOCK_BLOCKING
;
3362 path
->slots
[level
] = 0;
3367 * worker function to insert a single pointer in a node.
3368 * the node should have enough room for the pointer already
3370 * slot and level indicate where you want the key to go, and
3371 * blocknr is the block the key points to.
3373 static void insert_ptr(struct btrfs_trans_handle
*trans
,
3374 struct btrfs_path
*path
,
3375 struct btrfs_disk_key
*key
, u64 bytenr
,
3376 int slot
, int level
)
3378 struct extent_buffer
*lower
;
3382 BUG_ON(!path
->nodes
[level
]);
3383 btrfs_assert_tree_locked(path
->nodes
[level
]);
3384 lower
= path
->nodes
[level
];
3385 nritems
= btrfs_header_nritems(lower
);
3386 BUG_ON(slot
> nritems
);
3387 BUG_ON(nritems
== BTRFS_NODEPTRS_PER_BLOCK(trans
->fs_info
));
3388 if (slot
!= nritems
) {
3390 ret
= tree_mod_log_insert_move(lower
, slot
+ 1, slot
,
3394 memmove_extent_buffer(lower
,
3395 btrfs_node_key_ptr_offset(slot
+ 1),
3396 btrfs_node_key_ptr_offset(slot
),
3397 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
3400 ret
= tree_mod_log_insert_key(lower
, slot
, MOD_LOG_KEY_ADD
,
3404 btrfs_set_node_key(lower
, key
, slot
);
3405 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
3406 WARN_ON(trans
->transid
== 0);
3407 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
3408 btrfs_set_header_nritems(lower
, nritems
+ 1);
3409 btrfs_mark_buffer_dirty(lower
);
3413 * split the node at the specified level in path in two.
3414 * The path is corrected to point to the appropriate node after the split
3416 * Before splitting this tries to make some room in the node by pushing
3417 * left and right, if either one works, it returns right away.
3419 * returns 0 on success and < 0 on failure
3421 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
3422 struct btrfs_root
*root
,
3423 struct btrfs_path
*path
, int level
)
3425 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
3426 struct extent_buffer
*c
;
3427 struct extent_buffer
*split
;
3428 struct btrfs_disk_key disk_key
;
3433 c
= path
->nodes
[level
];
3434 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
3435 if (c
== root
->node
) {
3437 * trying to split the root, lets make a new one
3439 * tree mod log: We don't log_removal old root in
3440 * insert_new_root, because that root buffer will be kept as a
3441 * normal node. We are going to log removal of half of the
3442 * elements below with tree_mod_log_eb_copy. We're holding a
3443 * tree lock on the buffer, which is why we cannot race with
3444 * other tree_mod_log users.
3446 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
3450 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
3451 c
= path
->nodes
[level
];
3452 if (!ret
&& btrfs_header_nritems(c
) <
3453 BTRFS_NODEPTRS_PER_BLOCK(fs_info
) - 3)
3459 c_nritems
= btrfs_header_nritems(c
);
3460 mid
= (c_nritems
+ 1) / 2;
3461 btrfs_node_key(c
, &disk_key
, mid
);
3463 split
= alloc_tree_block_no_bg_flush(trans
, root
, 0, &disk_key
, level
,
3466 return PTR_ERR(split
);
3468 root_add_used(root
, fs_info
->nodesize
);
3469 ASSERT(btrfs_header_level(c
) == level
);
3471 ret
= tree_mod_log_eb_copy(split
, c
, 0, mid
, c_nritems
- mid
);
3473 btrfs_abort_transaction(trans
, ret
);
3476 copy_extent_buffer(split
, c
,
3477 btrfs_node_key_ptr_offset(0),
3478 btrfs_node_key_ptr_offset(mid
),
3479 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
3480 btrfs_set_header_nritems(split
, c_nritems
- mid
);
3481 btrfs_set_header_nritems(c
, mid
);
3484 btrfs_mark_buffer_dirty(c
);
3485 btrfs_mark_buffer_dirty(split
);
3487 insert_ptr(trans
, path
, &disk_key
, split
->start
,
3488 path
->slots
[level
+ 1] + 1, level
+ 1);
3490 if (path
->slots
[level
] >= mid
) {
3491 path
->slots
[level
] -= mid
;
3492 btrfs_tree_unlock(c
);
3493 free_extent_buffer(c
);
3494 path
->nodes
[level
] = split
;
3495 path
->slots
[level
+ 1] += 1;
3497 btrfs_tree_unlock(split
);
3498 free_extent_buffer(split
);
3504 * how many bytes are required to store the items in a leaf. start
3505 * and nr indicate which items in the leaf to check. This totals up the
3506 * space used both by the item structs and the item data
3508 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
3510 struct btrfs_item
*start_item
;
3511 struct btrfs_item
*end_item
;
3513 int nritems
= btrfs_header_nritems(l
);
3514 int end
= min(nritems
, start
+ nr
) - 1;
3518 start_item
= btrfs_item_nr(start
);
3519 end_item
= btrfs_item_nr(end
);
3520 data_len
= btrfs_item_offset(l
, start_item
) +
3521 btrfs_item_size(l
, start_item
);
3522 data_len
= data_len
- btrfs_item_offset(l
, end_item
);
3523 data_len
+= sizeof(struct btrfs_item
) * nr
;
3524 WARN_ON(data_len
< 0);
3529 * The space between the end of the leaf items and
3530 * the start of the leaf data. IOW, how much room
3531 * the leaf has left for both items and data
3533 noinline
int btrfs_leaf_free_space(struct extent_buffer
*leaf
)
3535 struct btrfs_fs_info
*fs_info
= leaf
->fs_info
;
3536 int nritems
= btrfs_header_nritems(leaf
);
3539 ret
= BTRFS_LEAF_DATA_SIZE(fs_info
) - leaf_space_used(leaf
, 0, nritems
);
3542 "leaf free space ret %d, leaf data size %lu, used %d nritems %d",
3544 (unsigned long) BTRFS_LEAF_DATA_SIZE(fs_info
),
3545 leaf_space_used(leaf
, 0, nritems
), nritems
);
3551 * min slot controls the lowest index we're willing to push to the
3552 * right. We'll push up to and including min_slot, but no lower
3554 static noinline
int __push_leaf_right(struct btrfs_path
*path
,
3555 int data_size
, int empty
,
3556 struct extent_buffer
*right
,
3557 int free_space
, u32 left_nritems
,
3560 struct btrfs_fs_info
*fs_info
= right
->fs_info
;
3561 struct extent_buffer
*left
= path
->nodes
[0];
3562 struct extent_buffer
*upper
= path
->nodes
[1];
3563 struct btrfs_map_token token
;
3564 struct btrfs_disk_key disk_key
;
3569 struct btrfs_item
*item
;
3578 nr
= max_t(u32
, 1, min_slot
);
3580 if (path
->slots
[0] >= left_nritems
)
3581 push_space
+= data_size
;
3583 slot
= path
->slots
[1];
3584 i
= left_nritems
- 1;
3586 item
= btrfs_item_nr(i
);
3588 if (!empty
&& push_items
> 0) {
3589 if (path
->slots
[0] > i
)
3591 if (path
->slots
[0] == i
) {
3592 int space
= btrfs_leaf_free_space(left
);
3594 if (space
+ push_space
* 2 > free_space
)
3599 if (path
->slots
[0] == i
)
3600 push_space
+= data_size
;
3602 this_item_size
= btrfs_item_size(left
, item
);
3603 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3607 push_space
+= this_item_size
+ sizeof(*item
);
3613 if (push_items
== 0)
3616 WARN_ON(!empty
&& push_items
== left_nritems
);
3618 /* push left to right */
3619 right_nritems
= btrfs_header_nritems(right
);
3621 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
3622 push_space
-= leaf_data_end(left
);
3624 /* make room in the right data area */
3625 data_end
= leaf_data_end(right
);
3626 memmove_extent_buffer(right
,
3627 BTRFS_LEAF_DATA_OFFSET
+ data_end
- push_space
,
3628 BTRFS_LEAF_DATA_OFFSET
+ data_end
,
3629 BTRFS_LEAF_DATA_SIZE(fs_info
) - data_end
);
3631 /* copy from the left data area */
3632 copy_extent_buffer(right
, left
, BTRFS_LEAF_DATA_OFFSET
+
3633 BTRFS_LEAF_DATA_SIZE(fs_info
) - push_space
,
3634 BTRFS_LEAF_DATA_OFFSET
+ leaf_data_end(left
),
3637 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
3638 btrfs_item_nr_offset(0),
3639 right_nritems
* sizeof(struct btrfs_item
));
3641 /* copy the items from left to right */
3642 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
3643 btrfs_item_nr_offset(left_nritems
- push_items
),
3644 push_items
* sizeof(struct btrfs_item
));
3646 /* update the item pointers */
3647 btrfs_init_map_token(&token
, right
);
3648 right_nritems
+= push_items
;
3649 btrfs_set_header_nritems(right
, right_nritems
);
3650 push_space
= BTRFS_LEAF_DATA_SIZE(fs_info
);
3651 for (i
= 0; i
< right_nritems
; i
++) {
3652 item
= btrfs_item_nr(i
);
3653 push_space
-= btrfs_token_item_size(&token
, item
);
3654 btrfs_set_token_item_offset(&token
, item
, push_space
);
3657 left_nritems
-= push_items
;
3658 btrfs_set_header_nritems(left
, left_nritems
);
3661 btrfs_mark_buffer_dirty(left
);
3663 btrfs_clean_tree_block(left
);
3665 btrfs_mark_buffer_dirty(right
);
3667 btrfs_item_key(right
, &disk_key
, 0);
3668 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
3669 btrfs_mark_buffer_dirty(upper
);
3671 /* then fixup the leaf pointer in the path */
3672 if (path
->slots
[0] >= left_nritems
) {
3673 path
->slots
[0] -= left_nritems
;
3674 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
3675 btrfs_clean_tree_block(path
->nodes
[0]);
3676 btrfs_tree_unlock(path
->nodes
[0]);
3677 free_extent_buffer(path
->nodes
[0]);
3678 path
->nodes
[0] = right
;
3679 path
->slots
[1] += 1;
3681 btrfs_tree_unlock(right
);
3682 free_extent_buffer(right
);
3687 btrfs_tree_unlock(right
);
3688 free_extent_buffer(right
);
3693 * push some data in the path leaf to the right, trying to free up at
3694 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3696 * returns 1 if the push failed because the other node didn't have enough
3697 * room, 0 if everything worked out and < 0 if there were major errors.
3699 * this will push starting from min_slot to the end of the leaf. It won't
3700 * push any slot lower than min_slot
3702 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
3703 *root
, struct btrfs_path
*path
,
3704 int min_data_size
, int data_size
,
3705 int empty
, u32 min_slot
)
3707 struct extent_buffer
*left
= path
->nodes
[0];
3708 struct extent_buffer
*right
;
3709 struct extent_buffer
*upper
;
3715 if (!path
->nodes
[1])
3718 slot
= path
->slots
[1];
3719 upper
= path
->nodes
[1];
3720 if (slot
>= btrfs_header_nritems(upper
) - 1)
3723 btrfs_assert_tree_locked(path
->nodes
[1]);
3725 right
= btrfs_read_node_slot(upper
, slot
+ 1);
3727 * slot + 1 is not valid or we fail to read the right node,
3728 * no big deal, just return.
3733 btrfs_tree_lock(right
);
3734 btrfs_set_lock_blocking_write(right
);
3736 free_space
= btrfs_leaf_free_space(right
);
3737 if (free_space
< data_size
)
3740 /* cow and double check */
3741 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
3746 free_space
= btrfs_leaf_free_space(right
);
3747 if (free_space
< data_size
)
3750 left_nritems
= btrfs_header_nritems(left
);
3751 if (left_nritems
== 0)
3754 if (path
->slots
[0] == left_nritems
&& !empty
) {
3755 /* Key greater than all keys in the leaf, right neighbor has
3756 * enough room for it and we're not emptying our leaf to delete
3757 * it, therefore use right neighbor to insert the new item and
3758 * no need to touch/dirty our left leaf. */
3759 btrfs_tree_unlock(left
);
3760 free_extent_buffer(left
);
3761 path
->nodes
[0] = right
;
3767 return __push_leaf_right(path
, min_data_size
, empty
,
3768 right
, free_space
, left_nritems
, min_slot
);
3770 btrfs_tree_unlock(right
);
3771 free_extent_buffer(right
);
3776 * push some data in the path leaf to the left, trying to free up at
3777 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3779 * max_slot can put a limit on how far into the leaf we'll push items. The
3780 * item at 'max_slot' won't be touched. Use (u32)-1 to make us do all the
3783 static noinline
int __push_leaf_left(struct btrfs_path
*path
, int data_size
,
3784 int empty
, struct extent_buffer
*left
,
3785 int free_space
, u32 right_nritems
,
3788 struct btrfs_fs_info
*fs_info
= left
->fs_info
;
3789 struct btrfs_disk_key disk_key
;
3790 struct extent_buffer
*right
= path
->nodes
[0];
3794 struct btrfs_item
*item
;
3795 u32 old_left_nritems
;
3799 u32 old_left_item_size
;
3800 struct btrfs_map_token token
;
3803 nr
= min(right_nritems
, max_slot
);
3805 nr
= min(right_nritems
- 1, max_slot
);
3807 for (i
= 0; i
< nr
; i
++) {
3808 item
= btrfs_item_nr(i
);
3810 if (!empty
&& push_items
> 0) {
3811 if (path
->slots
[0] < i
)
3813 if (path
->slots
[0] == i
) {
3814 int space
= btrfs_leaf_free_space(right
);
3816 if (space
+ push_space
* 2 > free_space
)
3821 if (path
->slots
[0] == i
)
3822 push_space
+= data_size
;
3824 this_item_size
= btrfs_item_size(right
, item
);
3825 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
3829 push_space
+= this_item_size
+ sizeof(*item
);
3832 if (push_items
== 0) {
3836 WARN_ON(!empty
&& push_items
== btrfs_header_nritems(right
));
3838 /* push data from right to left */
3839 copy_extent_buffer(left
, right
,
3840 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
3841 btrfs_item_nr_offset(0),
3842 push_items
* sizeof(struct btrfs_item
));
3844 push_space
= BTRFS_LEAF_DATA_SIZE(fs_info
) -
3845 btrfs_item_offset_nr(right
, push_items
- 1);
3847 copy_extent_buffer(left
, right
, BTRFS_LEAF_DATA_OFFSET
+
3848 leaf_data_end(left
) - push_space
,
3849 BTRFS_LEAF_DATA_OFFSET
+
3850 btrfs_item_offset_nr(right
, push_items
- 1),
3852 old_left_nritems
= btrfs_header_nritems(left
);
3853 BUG_ON(old_left_nritems
<= 0);
3855 btrfs_init_map_token(&token
, left
);
3856 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
3857 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
3860 item
= btrfs_item_nr(i
);
3862 ioff
= btrfs_token_item_offset(&token
, item
);
3863 btrfs_set_token_item_offset(&token
, item
,
3864 ioff
- (BTRFS_LEAF_DATA_SIZE(fs_info
) - old_left_item_size
));
3866 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
3868 /* fixup right node */
3869 if (push_items
> right_nritems
)
3870 WARN(1, KERN_CRIT
"push items %d nr %u\n", push_items
,
3873 if (push_items
< right_nritems
) {
3874 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
3875 leaf_data_end(right
);
3876 memmove_extent_buffer(right
, BTRFS_LEAF_DATA_OFFSET
+
3877 BTRFS_LEAF_DATA_SIZE(fs_info
) - push_space
,
3878 BTRFS_LEAF_DATA_OFFSET
+
3879 leaf_data_end(right
), push_space
);
3881 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
3882 btrfs_item_nr_offset(push_items
),
3883 (btrfs_header_nritems(right
) - push_items
) *
3884 sizeof(struct btrfs_item
));
3887 btrfs_init_map_token(&token
, right
);
3888 right_nritems
-= push_items
;
3889 btrfs_set_header_nritems(right
, right_nritems
);
3890 push_space
= BTRFS_LEAF_DATA_SIZE(fs_info
);
3891 for (i
= 0; i
< right_nritems
; i
++) {
3892 item
= btrfs_item_nr(i
);
3894 push_space
= push_space
- btrfs_token_item_size(&token
, item
);
3895 btrfs_set_token_item_offset(&token
, item
, push_space
);
3898 btrfs_mark_buffer_dirty(left
);
3900 btrfs_mark_buffer_dirty(right
);
3902 btrfs_clean_tree_block(right
);
3904 btrfs_item_key(right
, &disk_key
, 0);
3905 fixup_low_keys(path
, &disk_key
, 1);
3907 /* then fixup the leaf pointer in the path */
3908 if (path
->slots
[0] < push_items
) {
3909 path
->slots
[0] += old_left_nritems
;
3910 btrfs_tree_unlock(path
->nodes
[0]);
3911 free_extent_buffer(path
->nodes
[0]);
3912 path
->nodes
[0] = left
;
3913 path
->slots
[1] -= 1;
3915 btrfs_tree_unlock(left
);
3916 free_extent_buffer(left
);
3917 path
->slots
[0] -= push_items
;
3919 BUG_ON(path
->slots
[0] < 0);
3922 btrfs_tree_unlock(left
);
3923 free_extent_buffer(left
);
3928 * push some data in the path leaf to the left, trying to free up at
3929 * least data_size bytes. returns zero if the push worked, nonzero otherwise
3931 * max_slot can put a limit on how far into the leaf we'll push items. The
3932 * item at 'max_slot' won't be touched. Use (u32)-1 to make us push all the
3935 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
3936 *root
, struct btrfs_path
*path
, int min_data_size
,
3937 int data_size
, int empty
, u32 max_slot
)
3939 struct extent_buffer
*right
= path
->nodes
[0];
3940 struct extent_buffer
*left
;
3946 slot
= path
->slots
[1];
3949 if (!path
->nodes
[1])
3952 right_nritems
= btrfs_header_nritems(right
);
3953 if (right_nritems
== 0)
3956 btrfs_assert_tree_locked(path
->nodes
[1]);
3958 left
= btrfs_read_node_slot(path
->nodes
[1], slot
- 1);
3960 * slot - 1 is not valid or we fail to read the left node,
3961 * no big deal, just return.
3966 btrfs_tree_lock(left
);
3967 btrfs_set_lock_blocking_write(left
);
3969 free_space
= btrfs_leaf_free_space(left
);
3970 if (free_space
< data_size
) {
3975 /* cow and double check */
3976 ret
= btrfs_cow_block(trans
, root
, left
,
3977 path
->nodes
[1], slot
- 1, &left
);
3979 /* we hit -ENOSPC, but it isn't fatal here */
3985 free_space
= btrfs_leaf_free_space(left
);
3986 if (free_space
< data_size
) {
3991 return __push_leaf_left(path
, min_data_size
,
3992 empty
, left
, free_space
, right_nritems
,
3995 btrfs_tree_unlock(left
);
3996 free_extent_buffer(left
);
4001 * split the path's leaf in two, making sure there is at least data_size
4002 * available for the resulting leaf level of the path.
4004 static noinline
void copy_for_split(struct btrfs_trans_handle
*trans
,
4005 struct btrfs_path
*path
,
4006 struct extent_buffer
*l
,
4007 struct extent_buffer
*right
,
4008 int slot
, int mid
, int nritems
)
4010 struct btrfs_fs_info
*fs_info
= trans
->fs_info
;
4014 struct btrfs_disk_key disk_key
;
4015 struct btrfs_map_token token
;
4017 nritems
= nritems
- mid
;
4018 btrfs_set_header_nritems(right
, nritems
);
4019 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(l
);
4021 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
4022 btrfs_item_nr_offset(mid
),
4023 nritems
* sizeof(struct btrfs_item
));
4025 copy_extent_buffer(right
, l
,
4026 BTRFS_LEAF_DATA_OFFSET
+ BTRFS_LEAF_DATA_SIZE(fs_info
) -
4027 data_copy_size
, BTRFS_LEAF_DATA_OFFSET
+
4028 leaf_data_end(l
), data_copy_size
);
4030 rt_data_off
= BTRFS_LEAF_DATA_SIZE(fs_info
) - btrfs_item_end_nr(l
, mid
);
4032 btrfs_init_map_token(&token
, right
);
4033 for (i
= 0; i
< nritems
; i
++) {
4034 struct btrfs_item
*item
= btrfs_item_nr(i
);
4037 ioff
= btrfs_token_item_offset(&token
, item
);
4038 btrfs_set_token_item_offset(&token
, item
, ioff
+ rt_data_off
);
4041 btrfs_set_header_nritems(l
, mid
);
4042 btrfs_item_key(right
, &disk_key
, 0);
4043 insert_ptr(trans
, path
, &disk_key
, right
->start
, path
->slots
[1] + 1, 1);
4045 btrfs_mark_buffer_dirty(right
);
4046 btrfs_mark_buffer_dirty(l
);
4047 BUG_ON(path
->slots
[0] != slot
);
4050 btrfs_tree_unlock(path
->nodes
[0]);
4051 free_extent_buffer(path
->nodes
[0]);
4052 path
->nodes
[0] = right
;
4053 path
->slots
[0] -= mid
;
4054 path
->slots
[1] += 1;
4056 btrfs_tree_unlock(right
);
4057 free_extent_buffer(right
);
4060 BUG_ON(path
->slots
[0] < 0);
4064 * double splits happen when we need to insert a big item in the middle
4065 * of a leaf. A double split can leave us with 3 mostly empty leaves:
4066 * leaf: [ slots 0 - N] [ our target ] [ N + 1 - total in leaf ]
4069 * We avoid this by trying to push the items on either side of our target
4070 * into the adjacent leaves. If all goes well we can avoid the double split
4073 static noinline
int push_for_double_split(struct btrfs_trans_handle
*trans
,
4074 struct btrfs_root
*root
,
4075 struct btrfs_path
*path
,
4082 int space_needed
= data_size
;
4084 slot
= path
->slots
[0];
4085 if (slot
< btrfs_header_nritems(path
->nodes
[0]))
4086 space_needed
-= btrfs_leaf_free_space(path
->nodes
[0]);
4089 * try to push all the items after our slot into the
4092 ret
= push_leaf_right(trans
, root
, path
, 1, space_needed
, 0, slot
);
4099 nritems
= btrfs_header_nritems(path
->nodes
[0]);
4101 * our goal is to get our slot at the start or end of a leaf. If
4102 * we've done so we're done
4104 if (path
->slots
[0] == 0 || path
->slots
[0] == nritems
)
4107 if (btrfs_leaf_free_space(path
->nodes
[0]) >= data_size
)
4110 /* try to push all the items before our slot into the next leaf */
4111 slot
= path
->slots
[0];
4112 space_needed
= data_size
;
4114 space_needed
-= btrfs_leaf_free_space(path
->nodes
[0]);
4115 ret
= push_leaf_left(trans
, root
, path
, 1, space_needed
, 0, slot
);
4128 * split the path's leaf in two, making sure there is at least data_size
4129 * available for the resulting leaf level of the path.
4131 * returns 0 if all went well and < 0 on failure.
4133 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
4134 struct btrfs_root
*root
,
4135 const struct btrfs_key
*ins_key
,
4136 struct btrfs_path
*path
, int data_size
,
4139 struct btrfs_disk_key disk_key
;
4140 struct extent_buffer
*l
;
4144 struct extent_buffer
*right
;
4145 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4149 int num_doubles
= 0;
4150 int tried_avoid_double
= 0;
4153 slot
= path
->slots
[0];
4154 if (extend
&& data_size
+ btrfs_item_size_nr(l
, slot
) +
4155 sizeof(struct btrfs_item
) > BTRFS_LEAF_DATA_SIZE(fs_info
))
4158 /* first try to make some room by pushing left and right */
4159 if (data_size
&& path
->nodes
[1]) {
4160 int space_needed
= data_size
;
4162 if (slot
< btrfs_header_nritems(l
))
4163 space_needed
-= btrfs_leaf_free_space(l
);
4165 wret
= push_leaf_right(trans
, root
, path
, space_needed
,
4166 space_needed
, 0, 0);
4170 space_needed
= data_size
;
4172 space_needed
-= btrfs_leaf_free_space(l
);
4173 wret
= push_leaf_left(trans
, root
, path
, space_needed
,
4174 space_needed
, 0, (u32
)-1);
4180 /* did the pushes work? */
4181 if (btrfs_leaf_free_space(l
) >= data_size
)
4185 if (!path
->nodes
[1]) {
4186 ret
= insert_new_root(trans
, root
, path
, 1);
4193 slot
= path
->slots
[0];
4194 nritems
= btrfs_header_nritems(l
);
4195 mid
= (nritems
+ 1) / 2;
4199 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
4200 BTRFS_LEAF_DATA_SIZE(fs_info
)) {
4201 if (slot
>= nritems
) {
4205 if (mid
!= nritems
&&
4206 leaf_space_used(l
, mid
, nritems
- mid
) +
4207 data_size
> BTRFS_LEAF_DATA_SIZE(fs_info
)) {
4208 if (data_size
&& !tried_avoid_double
)
4209 goto push_for_double
;
4215 if (leaf_space_used(l
, 0, mid
) + data_size
>
4216 BTRFS_LEAF_DATA_SIZE(fs_info
)) {
4217 if (!extend
&& data_size
&& slot
== 0) {
4219 } else if ((extend
|| !data_size
) && slot
== 0) {
4223 if (mid
!= nritems
&&
4224 leaf_space_used(l
, mid
, nritems
- mid
) +
4225 data_size
> BTRFS_LEAF_DATA_SIZE(fs_info
)) {
4226 if (data_size
&& !tried_avoid_double
)
4227 goto push_for_double
;
4235 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
4237 btrfs_item_key(l
, &disk_key
, mid
);
4239 right
= alloc_tree_block_no_bg_flush(trans
, root
, 0, &disk_key
, 0,
4242 return PTR_ERR(right
);
4244 root_add_used(root
, fs_info
->nodesize
);
4248 btrfs_set_header_nritems(right
, 0);
4249 insert_ptr(trans
, path
, &disk_key
,
4250 right
->start
, path
->slots
[1] + 1, 1);
4251 btrfs_tree_unlock(path
->nodes
[0]);
4252 free_extent_buffer(path
->nodes
[0]);
4253 path
->nodes
[0] = right
;
4255 path
->slots
[1] += 1;
4257 btrfs_set_header_nritems(right
, 0);
4258 insert_ptr(trans
, path
, &disk_key
,
4259 right
->start
, path
->slots
[1], 1);
4260 btrfs_tree_unlock(path
->nodes
[0]);
4261 free_extent_buffer(path
->nodes
[0]);
4262 path
->nodes
[0] = right
;
4264 if (path
->slots
[1] == 0)
4265 fixup_low_keys(path
, &disk_key
, 1);
4268 * We create a new leaf 'right' for the required ins_len and
4269 * we'll do btrfs_mark_buffer_dirty() on this leaf after copying
4270 * the content of ins_len to 'right'.
4275 copy_for_split(trans
, path
, l
, right
, slot
, mid
, nritems
);
4278 BUG_ON(num_doubles
!= 0);
4286 push_for_double_split(trans
, root
, path
, data_size
);
4287 tried_avoid_double
= 1;
4288 if (btrfs_leaf_free_space(path
->nodes
[0]) >= data_size
)
4293 static noinline
int setup_leaf_for_split(struct btrfs_trans_handle
*trans
,
4294 struct btrfs_root
*root
,
4295 struct btrfs_path
*path
, int ins_len
)
4297 struct btrfs_key key
;
4298 struct extent_buffer
*leaf
;
4299 struct btrfs_file_extent_item
*fi
;
4304 leaf
= path
->nodes
[0];
4305 btrfs_item_key_to_cpu(leaf
, &key
, path
->slots
[0]);
4307 BUG_ON(key
.type
!= BTRFS_EXTENT_DATA_KEY
&&
4308 key
.type
!= BTRFS_EXTENT_CSUM_KEY
);
4310 if (btrfs_leaf_free_space(leaf
) >= ins_len
)
4313 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4314 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4315 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4316 struct btrfs_file_extent_item
);
4317 extent_len
= btrfs_file_extent_num_bytes(leaf
, fi
);
4319 btrfs_release_path(path
);
4321 path
->keep_locks
= 1;
4322 path
->search_for_split
= 1;
4323 ret
= btrfs_search_slot(trans
, root
, &key
, path
, 0, 1);
4324 path
->search_for_split
= 0;
4331 leaf
= path
->nodes
[0];
4332 /* if our item isn't there, return now */
4333 if (item_size
!= btrfs_item_size_nr(leaf
, path
->slots
[0]))
4336 /* the leaf has changed, it now has room. return now */
4337 if (btrfs_leaf_free_space(path
->nodes
[0]) >= ins_len
)
4340 if (key
.type
== BTRFS_EXTENT_DATA_KEY
) {
4341 fi
= btrfs_item_ptr(leaf
, path
->slots
[0],
4342 struct btrfs_file_extent_item
);
4343 if (extent_len
!= btrfs_file_extent_num_bytes(leaf
, fi
))
4347 btrfs_set_path_blocking(path
);
4348 ret
= split_leaf(trans
, root
, &key
, path
, ins_len
, 1);
4352 path
->keep_locks
= 0;
4353 btrfs_unlock_up_safe(path
, 1);
4356 path
->keep_locks
= 0;
4360 static noinline
int split_item(struct btrfs_path
*path
,
4361 const struct btrfs_key
*new_key
,
4362 unsigned long split_offset
)
4364 struct extent_buffer
*leaf
;
4365 struct btrfs_item
*item
;
4366 struct btrfs_item
*new_item
;
4372 struct btrfs_disk_key disk_key
;
4374 leaf
= path
->nodes
[0];
4375 BUG_ON(btrfs_leaf_free_space(leaf
) < sizeof(struct btrfs_item
));
4377 btrfs_set_path_blocking(path
);
4379 item
= btrfs_item_nr(path
->slots
[0]);
4380 orig_offset
= btrfs_item_offset(leaf
, item
);
4381 item_size
= btrfs_item_size(leaf
, item
);
4383 buf
= kmalloc(item_size
, GFP_NOFS
);
4387 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
4388 path
->slots
[0]), item_size
);
4390 slot
= path
->slots
[0] + 1;
4391 nritems
= btrfs_header_nritems(leaf
);
4392 if (slot
!= nritems
) {
4393 /* shift the items */
4394 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
4395 btrfs_item_nr_offset(slot
),
4396 (nritems
- slot
) * sizeof(struct btrfs_item
));
4399 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
4400 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4402 new_item
= btrfs_item_nr(slot
);
4404 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
4405 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
4407 btrfs_set_item_offset(leaf
, item
,
4408 orig_offset
+ item_size
- split_offset
);
4409 btrfs_set_item_size(leaf
, item
, split_offset
);
4411 btrfs_set_header_nritems(leaf
, nritems
+ 1);
4413 /* write the data for the start of the original item */
4414 write_extent_buffer(leaf
, buf
,
4415 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4418 /* write the data for the new item */
4419 write_extent_buffer(leaf
, buf
+ split_offset
,
4420 btrfs_item_ptr_offset(leaf
, slot
),
4421 item_size
- split_offset
);
4422 btrfs_mark_buffer_dirty(leaf
);
4424 BUG_ON(btrfs_leaf_free_space(leaf
) < 0);
4430 * This function splits a single item into two items,
4431 * giving 'new_key' to the new item and splitting the
4432 * old one at split_offset (from the start of the item).
4434 * The path may be released by this operation. After
4435 * the split, the path is pointing to the old item. The
4436 * new item is going to be in the same node as the old one.
4438 * Note, the item being split must be smaller enough to live alone on
4439 * a tree block with room for one extra struct btrfs_item
4441 * This allows us to split the item in place, keeping a lock on the
4442 * leaf the entire time.
4444 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
4445 struct btrfs_root
*root
,
4446 struct btrfs_path
*path
,
4447 const struct btrfs_key
*new_key
,
4448 unsigned long split_offset
)
4451 ret
= setup_leaf_for_split(trans
, root
, path
,
4452 sizeof(struct btrfs_item
));
4456 ret
= split_item(path
, new_key
, split_offset
);
4461 * This function duplicate a item, giving 'new_key' to the new item.
4462 * It guarantees both items live in the same tree leaf and the new item
4463 * is contiguous with the original item.
4465 * This allows us to split file extent in place, keeping a lock on the
4466 * leaf the entire time.
4468 int btrfs_duplicate_item(struct btrfs_trans_handle
*trans
,
4469 struct btrfs_root
*root
,
4470 struct btrfs_path
*path
,
4471 const struct btrfs_key
*new_key
)
4473 struct extent_buffer
*leaf
;
4477 leaf
= path
->nodes
[0];
4478 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
4479 ret
= setup_leaf_for_split(trans
, root
, path
,
4480 item_size
+ sizeof(struct btrfs_item
));
4485 setup_items_for_insert(root
, path
, new_key
, &item_size
,
4486 item_size
, item_size
+
4487 sizeof(struct btrfs_item
), 1);
4488 leaf
= path
->nodes
[0];
4489 memcpy_extent_buffer(leaf
,
4490 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
4491 btrfs_item_ptr_offset(leaf
, path
->slots
[0] - 1),
4497 * make the item pointed to by the path smaller. new_size indicates
4498 * how small to make it, and from_end tells us if we just chop bytes
4499 * off the end of the item or if we shift the item to chop bytes off
4502 void btrfs_truncate_item(struct btrfs_path
*path
, u32 new_size
, int from_end
)
4505 struct extent_buffer
*leaf
;
4506 struct btrfs_item
*item
;
4508 unsigned int data_end
;
4509 unsigned int old_data_start
;
4510 unsigned int old_size
;
4511 unsigned int size_diff
;
4513 struct btrfs_map_token token
;
4515 leaf
= path
->nodes
[0];
4516 slot
= path
->slots
[0];
4518 old_size
= btrfs_item_size_nr(leaf
, slot
);
4519 if (old_size
== new_size
)
4522 nritems
= btrfs_header_nritems(leaf
);
4523 data_end
= leaf_data_end(leaf
);
4525 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
4527 size_diff
= old_size
- new_size
;
4530 BUG_ON(slot
>= nritems
);
4533 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4535 /* first correct the data pointers */
4536 btrfs_init_map_token(&token
, leaf
);
4537 for (i
= slot
; i
< nritems
; i
++) {
4539 item
= btrfs_item_nr(i
);
4541 ioff
= btrfs_token_item_offset(&token
, item
);
4542 btrfs_set_token_item_offset(&token
, item
, ioff
+ size_diff
);
4545 /* shift the data */
4547 memmove_extent_buffer(leaf
, BTRFS_LEAF_DATA_OFFSET
+
4548 data_end
+ size_diff
, BTRFS_LEAF_DATA_OFFSET
+
4549 data_end
, old_data_start
+ new_size
- data_end
);
4551 struct btrfs_disk_key disk_key
;
4554 btrfs_item_key(leaf
, &disk_key
, slot
);
4556 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
4558 struct btrfs_file_extent_item
*fi
;
4560 fi
= btrfs_item_ptr(leaf
, slot
,
4561 struct btrfs_file_extent_item
);
4562 fi
= (struct btrfs_file_extent_item
*)(
4563 (unsigned long)fi
- size_diff
);
4565 if (btrfs_file_extent_type(leaf
, fi
) ==
4566 BTRFS_FILE_EXTENT_INLINE
) {
4567 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
4568 memmove_extent_buffer(leaf
, ptr
,
4570 BTRFS_FILE_EXTENT_INLINE_DATA_START
);
4574 memmove_extent_buffer(leaf
, BTRFS_LEAF_DATA_OFFSET
+
4575 data_end
+ size_diff
, BTRFS_LEAF_DATA_OFFSET
+
4576 data_end
, old_data_start
- data_end
);
4578 offset
= btrfs_disk_key_offset(&disk_key
);
4579 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
4580 btrfs_set_item_key(leaf
, &disk_key
, slot
);
4582 fixup_low_keys(path
, &disk_key
, 1);
4585 item
= btrfs_item_nr(slot
);
4586 btrfs_set_item_size(leaf
, item
, new_size
);
4587 btrfs_mark_buffer_dirty(leaf
);
4589 if (btrfs_leaf_free_space(leaf
) < 0) {
4590 btrfs_print_leaf(leaf
);
4596 * make the item pointed to by the path bigger, data_size is the added size.
4598 void btrfs_extend_item(struct btrfs_path
*path
, u32 data_size
)
4601 struct extent_buffer
*leaf
;
4602 struct btrfs_item
*item
;
4604 unsigned int data_end
;
4605 unsigned int old_data
;
4606 unsigned int old_size
;
4608 struct btrfs_map_token token
;
4610 leaf
= path
->nodes
[0];
4612 nritems
= btrfs_header_nritems(leaf
);
4613 data_end
= leaf_data_end(leaf
);
4615 if (btrfs_leaf_free_space(leaf
) < data_size
) {
4616 btrfs_print_leaf(leaf
);
4619 slot
= path
->slots
[0];
4620 old_data
= btrfs_item_end_nr(leaf
, slot
);
4623 if (slot
>= nritems
) {
4624 btrfs_print_leaf(leaf
);
4625 btrfs_crit(leaf
->fs_info
, "slot %d too large, nritems %d",
4631 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4633 /* first correct the data pointers */
4634 btrfs_init_map_token(&token
, leaf
);
4635 for (i
= slot
; i
< nritems
; i
++) {
4637 item
= btrfs_item_nr(i
);
4639 ioff
= btrfs_token_item_offset(&token
, item
);
4640 btrfs_set_token_item_offset(&token
, item
, ioff
- data_size
);
4643 /* shift the data */
4644 memmove_extent_buffer(leaf
, BTRFS_LEAF_DATA_OFFSET
+
4645 data_end
- data_size
, BTRFS_LEAF_DATA_OFFSET
+
4646 data_end
, old_data
- data_end
);
4648 data_end
= old_data
;
4649 old_size
= btrfs_item_size_nr(leaf
, slot
);
4650 item
= btrfs_item_nr(slot
);
4651 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
4652 btrfs_mark_buffer_dirty(leaf
);
4654 if (btrfs_leaf_free_space(leaf
) < 0) {
4655 btrfs_print_leaf(leaf
);
4661 * this is a helper for btrfs_insert_empty_items, the main goal here is
4662 * to save stack depth by doing the bulk of the work in a function
4663 * that doesn't call btrfs_search_slot
4665 void setup_items_for_insert(struct btrfs_root
*root
, struct btrfs_path
*path
,
4666 const struct btrfs_key
*cpu_key
, u32
*data_size
,
4667 u32 total_data
, u32 total_size
, int nr
)
4669 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4670 struct btrfs_item
*item
;
4673 unsigned int data_end
;
4674 struct btrfs_disk_key disk_key
;
4675 struct extent_buffer
*leaf
;
4677 struct btrfs_map_token token
;
4679 if (path
->slots
[0] == 0) {
4680 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
4681 fixup_low_keys(path
, &disk_key
, 1);
4683 btrfs_unlock_up_safe(path
, 1);
4685 leaf
= path
->nodes
[0];
4686 slot
= path
->slots
[0];
4688 nritems
= btrfs_header_nritems(leaf
);
4689 data_end
= leaf_data_end(leaf
);
4691 if (btrfs_leaf_free_space(leaf
) < total_size
) {
4692 btrfs_print_leaf(leaf
);
4693 btrfs_crit(fs_info
, "not enough freespace need %u have %d",
4694 total_size
, btrfs_leaf_free_space(leaf
));
4698 btrfs_init_map_token(&token
, leaf
);
4699 if (slot
!= nritems
) {
4700 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
4702 if (old_data
< data_end
) {
4703 btrfs_print_leaf(leaf
);
4704 btrfs_crit(fs_info
, "slot %d old_data %d data_end %d",
4705 slot
, old_data
, data_end
);
4709 * item0..itemN ... dataN.offset..dataN.size .. data0.size
4711 /* first correct the data pointers */
4712 for (i
= slot
; i
< nritems
; i
++) {
4715 item
= btrfs_item_nr(i
);
4716 ioff
= btrfs_token_item_offset(&token
, item
);
4717 btrfs_set_token_item_offset(&token
, item
,
4720 /* shift the items */
4721 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
4722 btrfs_item_nr_offset(slot
),
4723 (nritems
- slot
) * sizeof(struct btrfs_item
));
4725 /* shift the data */
4726 memmove_extent_buffer(leaf
, BTRFS_LEAF_DATA_OFFSET
+
4727 data_end
- total_data
, BTRFS_LEAF_DATA_OFFSET
+
4728 data_end
, old_data
- data_end
);
4729 data_end
= old_data
;
4732 /* setup the item for the new data */
4733 for (i
= 0; i
< nr
; i
++) {
4734 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
4735 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
4736 item
= btrfs_item_nr(slot
+ i
);
4737 btrfs_set_token_item_offset(&token
, item
, data_end
- data_size
[i
]);
4738 data_end
-= data_size
[i
];
4739 btrfs_set_token_item_size(&token
, item
, data_size
[i
]);
4742 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
4743 btrfs_mark_buffer_dirty(leaf
);
4745 if (btrfs_leaf_free_space(leaf
) < 0) {
4746 btrfs_print_leaf(leaf
);
4752 * Given a key and some data, insert items into the tree.
4753 * This does all the path init required, making room in the tree if needed.
4755 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
4756 struct btrfs_root
*root
,
4757 struct btrfs_path
*path
,
4758 const struct btrfs_key
*cpu_key
, u32
*data_size
,
4767 for (i
= 0; i
< nr
; i
++)
4768 total_data
+= data_size
[i
];
4770 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
4771 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
4777 slot
= path
->slots
[0];
4780 setup_items_for_insert(root
, path
, cpu_key
, data_size
,
4781 total_data
, total_size
, nr
);
4786 * Given a key and some data, insert an item into the tree.
4787 * This does all the path init required, making room in the tree if needed.
4789 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4790 const struct btrfs_key
*cpu_key
, void *data
,
4794 struct btrfs_path
*path
;
4795 struct extent_buffer
*leaf
;
4798 path
= btrfs_alloc_path();
4801 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
4803 leaf
= path
->nodes
[0];
4804 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
4805 write_extent_buffer(leaf
, data
, ptr
, data_size
);
4806 btrfs_mark_buffer_dirty(leaf
);
4808 btrfs_free_path(path
);
4813 * delete the pointer from a given node.
4815 * the tree should have been previously balanced so the deletion does not
4818 static void del_ptr(struct btrfs_root
*root
, struct btrfs_path
*path
,
4819 int level
, int slot
)
4821 struct extent_buffer
*parent
= path
->nodes
[level
];
4825 nritems
= btrfs_header_nritems(parent
);
4826 if (slot
!= nritems
- 1) {
4828 ret
= tree_mod_log_insert_move(parent
, slot
, slot
+ 1,
4829 nritems
- slot
- 1);
4832 memmove_extent_buffer(parent
,
4833 btrfs_node_key_ptr_offset(slot
),
4834 btrfs_node_key_ptr_offset(slot
+ 1),
4835 sizeof(struct btrfs_key_ptr
) *
4836 (nritems
- slot
- 1));
4838 ret
= tree_mod_log_insert_key(parent
, slot
, MOD_LOG_KEY_REMOVE
,
4844 btrfs_set_header_nritems(parent
, nritems
);
4845 if (nritems
== 0 && parent
== root
->node
) {
4846 BUG_ON(btrfs_header_level(root
->node
) != 1);
4847 /* just turn the root into a leaf and break */
4848 btrfs_set_header_level(root
->node
, 0);
4849 } else if (slot
== 0) {
4850 struct btrfs_disk_key disk_key
;
4852 btrfs_node_key(parent
, &disk_key
, 0);
4853 fixup_low_keys(path
, &disk_key
, level
+ 1);
4855 btrfs_mark_buffer_dirty(parent
);
4859 * a helper function to delete the leaf pointed to by path->slots[1] and
4862 * This deletes the pointer in path->nodes[1] and frees the leaf
4863 * block extent. zero is returned if it all worked out, < 0 otherwise.
4865 * The path must have already been setup for deleting the leaf, including
4866 * all the proper balancing. path->nodes[1] must be locked.
4868 static noinline
void btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
4869 struct btrfs_root
*root
,
4870 struct btrfs_path
*path
,
4871 struct extent_buffer
*leaf
)
4873 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
4874 del_ptr(root
, path
, 1, path
->slots
[1]);
4877 * btrfs_free_extent is expensive, we want to make sure we
4878 * aren't holding any locks when we call it
4880 btrfs_unlock_up_safe(path
, 0);
4882 root_sub_used(root
, leaf
->len
);
4884 atomic_inc(&leaf
->refs
);
4885 btrfs_free_tree_block(trans
, root
, leaf
, 0, 1);
4886 free_extent_buffer_stale(leaf
);
4889 * delete the item at the leaf level in path. If that empties
4890 * the leaf, remove it from the tree
4892 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
4893 struct btrfs_path
*path
, int slot
, int nr
)
4895 struct btrfs_fs_info
*fs_info
= root
->fs_info
;
4896 struct extent_buffer
*leaf
;
4897 struct btrfs_item
*item
;
4905 leaf
= path
->nodes
[0];
4906 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
4908 for (i
= 0; i
< nr
; i
++)
4909 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
4911 nritems
= btrfs_header_nritems(leaf
);
4913 if (slot
+ nr
!= nritems
) {
4914 int data_end
= leaf_data_end(leaf
);
4915 struct btrfs_map_token token
;
4917 memmove_extent_buffer(leaf
, BTRFS_LEAF_DATA_OFFSET
+
4919 BTRFS_LEAF_DATA_OFFSET
+ data_end
,
4920 last_off
- data_end
);
4922 btrfs_init_map_token(&token
, leaf
);
4923 for (i
= slot
+ nr
; i
< nritems
; i
++) {
4926 item
= btrfs_item_nr(i
);
4927 ioff
= btrfs_token_item_offset(&token
, item
);
4928 btrfs_set_token_item_offset(&token
, item
, ioff
+ dsize
);
4931 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
4932 btrfs_item_nr_offset(slot
+ nr
),
4933 sizeof(struct btrfs_item
) *
4934 (nritems
- slot
- nr
));
4936 btrfs_set_header_nritems(leaf
, nritems
- nr
);
4939 /* delete the leaf if we've emptied it */
4941 if (leaf
== root
->node
) {
4942 btrfs_set_header_level(leaf
, 0);
4944 btrfs_set_path_blocking(path
);
4945 btrfs_clean_tree_block(leaf
);
4946 btrfs_del_leaf(trans
, root
, path
, leaf
);
4949 int used
= leaf_space_used(leaf
, 0, nritems
);
4951 struct btrfs_disk_key disk_key
;
4953 btrfs_item_key(leaf
, &disk_key
, 0);
4954 fixup_low_keys(path
, &disk_key
, 1);
4957 /* delete the leaf if it is mostly empty */
4958 if (used
< BTRFS_LEAF_DATA_SIZE(fs_info
) / 3) {
4959 /* push_leaf_left fixes the path.
4960 * make sure the path still points to our leaf
4961 * for possible call to del_ptr below
4963 slot
= path
->slots
[1];
4964 atomic_inc(&leaf
->refs
);
4966 btrfs_set_path_blocking(path
);
4967 wret
= push_leaf_left(trans
, root
, path
, 1, 1,
4969 if (wret
< 0 && wret
!= -ENOSPC
)
4972 if (path
->nodes
[0] == leaf
&&
4973 btrfs_header_nritems(leaf
)) {
4974 wret
= push_leaf_right(trans
, root
, path
, 1,
4976 if (wret
< 0 && wret
!= -ENOSPC
)
4980 if (btrfs_header_nritems(leaf
) == 0) {
4981 path
->slots
[1] = slot
;
4982 btrfs_del_leaf(trans
, root
, path
, leaf
);
4983 free_extent_buffer(leaf
);
4986 /* if we're still in the path, make sure
4987 * we're dirty. Otherwise, one of the
4988 * push_leaf functions must have already
4989 * dirtied this buffer
4991 if (path
->nodes
[0] == leaf
)
4992 btrfs_mark_buffer_dirty(leaf
);
4993 free_extent_buffer(leaf
);
4996 btrfs_mark_buffer_dirty(leaf
);
5003 * search the tree again to find a leaf with lesser keys
5004 * returns 0 if it found something or 1 if there are no lesser leaves.
5005 * returns < 0 on io errors.
5007 * This may release the path, and so you may lose any locks held at the
5010 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5012 struct btrfs_key key
;
5013 struct btrfs_disk_key found_key
;
5016 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
5018 if (key
.offset
> 0) {
5020 } else if (key
.type
> 0) {
5022 key
.offset
= (u64
)-1;
5023 } else if (key
.objectid
> 0) {
5026 key
.offset
= (u64
)-1;
5031 btrfs_release_path(path
);
5032 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5035 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
5036 ret
= comp_keys(&found_key
, &key
);
5038 * We might have had an item with the previous key in the tree right
5039 * before we released our path. And after we released our path, that
5040 * item might have been pushed to the first slot (0) of the leaf we
5041 * were holding due to a tree balance. Alternatively, an item with the
5042 * previous key can exist as the only element of a leaf (big fat item).
5043 * Therefore account for these 2 cases, so that our callers (like
5044 * btrfs_previous_item) don't miss an existing item with a key matching
5045 * the previous key we computed above.
5053 * A helper function to walk down the tree starting at min_key, and looking
5054 * for nodes or leaves that are have a minimum transaction id.
5055 * This is used by the btree defrag code, and tree logging
5057 * This does not cow, but it does stuff the starting key it finds back
5058 * into min_key, so you can call btrfs_search_slot with cow=1 on the
5059 * key and get a writable path.
5061 * This honors path->lowest_level to prevent descent past a given level
5064 * min_trans indicates the oldest transaction that you are interested
5065 * in walking through. Any nodes or leaves older than min_trans are
5066 * skipped over (without reading them).
5068 * returns zero if something useful was found, < 0 on error and 1 if there
5069 * was nothing in the tree that matched the search criteria.
5071 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
5072 struct btrfs_path
*path
,
5075 struct extent_buffer
*cur
;
5076 struct btrfs_key found_key
;
5082 int keep_locks
= path
->keep_locks
;
5084 path
->keep_locks
= 1;
5086 cur
= btrfs_read_lock_root_node(root
);
5087 level
= btrfs_header_level(cur
);
5088 WARN_ON(path
->nodes
[level
]);
5089 path
->nodes
[level
] = cur
;
5090 path
->locks
[level
] = BTRFS_READ_LOCK
;
5092 if (btrfs_header_generation(cur
) < min_trans
) {
5097 nritems
= btrfs_header_nritems(cur
);
5098 level
= btrfs_header_level(cur
);
5099 sret
= btrfs_bin_search(cur
, min_key
, &slot
);
5105 /* at the lowest level, we're done, setup the path and exit */
5106 if (level
== path
->lowest_level
) {
5107 if (slot
>= nritems
)
5110 path
->slots
[level
] = slot
;
5111 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
5114 if (sret
&& slot
> 0)
5117 * check this node pointer against the min_trans parameters.
5118 * If it is too old, old, skip to the next one.
5120 while (slot
< nritems
) {
5123 gen
= btrfs_node_ptr_generation(cur
, slot
);
5124 if (gen
< min_trans
) {
5132 * we didn't find a candidate key in this node, walk forward
5133 * and find another one
5135 if (slot
>= nritems
) {
5136 path
->slots
[level
] = slot
;
5137 btrfs_set_path_blocking(path
);
5138 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
5141 btrfs_release_path(path
);
5147 /* save our key for returning back */
5148 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
5149 path
->slots
[level
] = slot
;
5150 if (level
== path
->lowest_level
) {
5154 btrfs_set_path_blocking(path
);
5155 cur
= btrfs_read_node_slot(cur
, slot
);
5161 btrfs_tree_read_lock(cur
);
5163 path
->locks
[level
- 1] = BTRFS_READ_LOCK
;
5164 path
->nodes
[level
- 1] = cur
;
5165 unlock_up(path
, level
, 1, 0, NULL
);
5168 path
->keep_locks
= keep_locks
;
5170 btrfs_unlock_up_safe(path
, path
->lowest_level
+ 1);
5171 btrfs_set_path_blocking(path
);
5172 memcpy(min_key
, &found_key
, sizeof(found_key
));
5178 * this is similar to btrfs_next_leaf, but does not try to preserve
5179 * and fixup the path. It looks for and returns the next key in the
5180 * tree based on the current path and the min_trans parameters.
5182 * 0 is returned if another key is found, < 0 if there are any errors
5183 * and 1 is returned if there are no higher keys in the tree
5185 * path->keep_locks should be set to 1 on the search made before
5186 * calling this function.
5188 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
5189 struct btrfs_key
*key
, int level
, u64 min_trans
)
5192 struct extent_buffer
*c
;
5194 WARN_ON(!path
->keep_locks
&& !path
->skip_locking
);
5195 while (level
< BTRFS_MAX_LEVEL
) {
5196 if (!path
->nodes
[level
])
5199 slot
= path
->slots
[level
] + 1;
5200 c
= path
->nodes
[level
];
5202 if (slot
>= btrfs_header_nritems(c
)) {
5205 struct btrfs_key cur_key
;
5206 if (level
+ 1 >= BTRFS_MAX_LEVEL
||
5207 !path
->nodes
[level
+ 1])
5210 if (path
->locks
[level
+ 1] || path
->skip_locking
) {
5215 slot
= btrfs_header_nritems(c
) - 1;
5217 btrfs_item_key_to_cpu(c
, &cur_key
, slot
);
5219 btrfs_node_key_to_cpu(c
, &cur_key
, slot
);
5221 orig_lowest
= path
->lowest_level
;
5222 btrfs_release_path(path
);
5223 path
->lowest_level
= level
;
5224 ret
= btrfs_search_slot(NULL
, root
, &cur_key
, path
,
5226 path
->lowest_level
= orig_lowest
;
5230 c
= path
->nodes
[level
];
5231 slot
= path
->slots
[level
];
5238 btrfs_item_key_to_cpu(c
, key
, slot
);
5240 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
5242 if (gen
< min_trans
) {
5246 btrfs_node_key_to_cpu(c
, key
, slot
);
5254 * search the tree again to find a leaf with greater keys
5255 * returns 0 if it found something or 1 if there are no greater leaves.
5256 * returns < 0 on io errors.
5258 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
5260 return btrfs_next_old_leaf(root
, path
, 0);
5263 int btrfs_next_old_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
5268 struct extent_buffer
*c
;
5269 struct extent_buffer
*next
;
5270 struct btrfs_key key
;
5273 int old_spinning
= path
->leave_spinning
;
5274 int next_rw_lock
= 0;
5276 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5280 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
5285 btrfs_release_path(path
);
5287 path
->keep_locks
= 1;
5288 path
->leave_spinning
= 1;
5291 ret
= btrfs_search_old_slot(root
, &key
, path
, time_seq
);
5293 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
5294 path
->keep_locks
= 0;
5299 nritems
= btrfs_header_nritems(path
->nodes
[0]);
5301 * by releasing the path above we dropped all our locks. A balance
5302 * could have added more items next to the key that used to be
5303 * at the very end of the block. So, check again here and
5304 * advance the path if there are now more items available.
5306 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
5313 * So the above check misses one case:
5314 * - after releasing the path above, someone has removed the item that
5315 * used to be at the very end of the block, and balance between leafs
5316 * gets another one with bigger key.offset to replace it.
5318 * This one should be returned as well, or we can get leaf corruption
5319 * later(esp. in __btrfs_drop_extents()).
5321 * And a bit more explanation about this check,
5322 * with ret > 0, the key isn't found, the path points to the slot
5323 * where it should be inserted, so the path->slots[0] item must be the
5326 if (nritems
> 0 && ret
> 0 && path
->slots
[0] == nritems
- 1) {
5331 while (level
< BTRFS_MAX_LEVEL
) {
5332 if (!path
->nodes
[level
]) {
5337 slot
= path
->slots
[level
] + 1;
5338 c
= path
->nodes
[level
];
5339 if (slot
>= btrfs_header_nritems(c
)) {
5341 if (level
== BTRFS_MAX_LEVEL
) {
5349 btrfs_tree_unlock_rw(next
, next_rw_lock
);
5350 free_extent_buffer(next
);
5354 next_rw_lock
= path
->locks
[level
];
5355 ret
= read_block_for_search(root
, path
, &next
, level
,
5361 btrfs_release_path(path
);
5365 if (!path
->skip_locking
) {
5366 ret
= btrfs_try_tree_read_lock(next
);
5367 if (!ret
&& time_seq
) {
5369 * If we don't get the lock, we may be racing
5370 * with push_leaf_left, holding that lock while
5371 * itself waiting for the leaf we've currently
5372 * locked. To solve this situation, we give up
5373 * on our lock and cycle.
5375 free_extent_buffer(next
);
5376 btrfs_release_path(path
);
5381 btrfs_set_path_blocking(path
);
5382 btrfs_tree_read_lock(next
);
5384 next_rw_lock
= BTRFS_READ_LOCK
;
5388 path
->slots
[level
] = slot
;
5391 c
= path
->nodes
[level
];
5392 if (path
->locks
[level
])
5393 btrfs_tree_unlock_rw(c
, path
->locks
[level
]);
5395 free_extent_buffer(c
);
5396 path
->nodes
[level
] = next
;
5397 path
->slots
[level
] = 0;
5398 if (!path
->skip_locking
)
5399 path
->locks
[level
] = next_rw_lock
;
5403 ret
= read_block_for_search(root
, path
, &next
, level
,
5409 btrfs_release_path(path
);
5413 if (!path
->skip_locking
) {
5414 ret
= btrfs_try_tree_read_lock(next
);
5416 btrfs_set_path_blocking(path
);
5417 btrfs_tree_read_lock(next
);
5419 next_rw_lock
= BTRFS_READ_LOCK
;
5424 unlock_up(path
, 0, 1, 0, NULL
);
5425 path
->leave_spinning
= old_spinning
;
5427 btrfs_set_path_blocking(path
);
5433 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
5434 * searching until it gets past min_objectid or finds an item of 'type'
5436 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5438 int btrfs_previous_item(struct btrfs_root
*root
,
5439 struct btrfs_path
*path
, u64 min_objectid
,
5442 struct btrfs_key found_key
;
5443 struct extent_buffer
*leaf
;
5448 if (path
->slots
[0] == 0) {
5449 btrfs_set_path_blocking(path
);
5450 ret
= btrfs_prev_leaf(root
, path
);
5456 leaf
= path
->nodes
[0];
5457 nritems
= btrfs_header_nritems(leaf
);
5460 if (path
->slots
[0] == nritems
)
5463 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5464 if (found_key
.objectid
< min_objectid
)
5466 if (found_key
.type
== type
)
5468 if (found_key
.objectid
== min_objectid
&&
5469 found_key
.type
< type
)
5476 * search in extent tree to find a previous Metadata/Data extent item with
5479 * returns 0 if something is found, 1 if nothing was found and < 0 on error
5481 int btrfs_previous_extent_item(struct btrfs_root
*root
,
5482 struct btrfs_path
*path
, u64 min_objectid
)
5484 struct btrfs_key found_key
;
5485 struct extent_buffer
*leaf
;
5490 if (path
->slots
[0] == 0) {
5491 btrfs_set_path_blocking(path
);
5492 ret
= btrfs_prev_leaf(root
, path
);
5498 leaf
= path
->nodes
[0];
5499 nritems
= btrfs_header_nritems(leaf
);
5502 if (path
->slots
[0] == nritems
)
5505 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
5506 if (found_key
.objectid
< min_objectid
)
5508 if (found_key
.type
== BTRFS_EXTENT_ITEM_KEY
||
5509 found_key
.type
== BTRFS_METADATA_ITEM_KEY
)
5511 if (found_key
.objectid
== min_objectid
&&
5512 found_key
.type
< BTRFS_EXTENT_ITEM_KEY
)