2 * Copyright (C) 2007 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
22 #include "transaction.h"
23 #include "print-tree.h"
26 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
27 *root
, struct btrfs_path
*path
, int level
);
28 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
29 *root
, struct btrfs_key
*ins_key
,
30 struct btrfs_path
*path
, int data_size
, int extend
);
31 static int push_node_left(struct btrfs_trans_handle
*trans
,
32 struct btrfs_root
*root
, struct extent_buffer
*dst
,
33 struct extent_buffer
*src
, int empty
);
34 static int balance_node_right(struct btrfs_trans_handle
*trans
,
35 struct btrfs_root
*root
,
36 struct extent_buffer
*dst_buf
,
37 struct extent_buffer
*src_buf
);
38 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
39 struct btrfs_path
*path
, int level
, int slot
);
41 inline void btrfs_init_path(struct btrfs_path
*p
)
43 memset(p
, 0, sizeof(*p
));
46 struct btrfs_path
*btrfs_alloc_path(void)
48 struct btrfs_path
*path
;
49 path
= kmem_cache_alloc(btrfs_path_cachep
, GFP_NOFS
);
51 btrfs_init_path(path
);
57 void btrfs_free_path(struct btrfs_path
*p
)
59 btrfs_release_path(NULL
, p
);
60 kmem_cache_free(btrfs_path_cachep
, p
);
63 void btrfs_release_path(struct btrfs_root
*root
, struct btrfs_path
*p
)
66 int keep
= p
->keep_locks
;
67 int skip
= p
->skip_locking
;
69 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
73 btrfs_tree_unlock(p
->nodes
[i
]);
76 free_extent_buffer(p
->nodes
[i
]);
78 memset(p
, 0, sizeof(*p
));
80 p
->skip_locking
= skip
;
83 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
85 struct extent_buffer
*eb
;
86 spin_lock(&root
->node_lock
);
88 extent_buffer_get(eb
);
89 spin_unlock(&root
->node_lock
);
93 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
95 struct extent_buffer
*eb
;
98 eb
= btrfs_root_node(root
);
101 spin_lock(&root
->node_lock
);
102 if (eb
== root
->node
) {
103 spin_unlock(&root
->node_lock
);
106 spin_unlock(&root
->node_lock
);
108 btrfs_tree_unlock(eb
);
109 free_extent_buffer(eb
);
114 static void add_root_to_dirty_list(struct btrfs_root
*root
)
116 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
117 list_add(&root
->dirty_list
,
118 &root
->fs_info
->dirty_cowonly_roots
);
122 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
123 struct btrfs_root
*root
,
124 struct extent_buffer
*buf
,
125 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
127 struct extent_buffer
*cow
;
131 struct btrfs_key first_key
;
132 struct btrfs_root
*new_root
;
134 new_root
= kmalloc(sizeof(*new_root
), GFP_NOFS
);
138 memcpy(new_root
, root
, sizeof(*new_root
));
139 new_root
->root_key
.objectid
= new_root_objectid
;
141 WARN_ON(root
->ref_cows
&& trans
->transid
!=
142 root
->fs_info
->running_transaction
->transid
);
143 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
145 level
= btrfs_header_level(buf
);
146 nritems
= btrfs_header_nritems(buf
);
149 btrfs_item_key_to_cpu(buf
, &first_key
, 0);
151 btrfs_node_key_to_cpu(buf
, &first_key
, 0);
153 first_key
.objectid
= 0;
155 cow
= btrfs_alloc_free_block(trans
, new_root
, buf
->len
,
157 trans
->transid
, first_key
.objectid
,
158 level
, buf
->start
, 0);
164 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
165 btrfs_set_header_bytenr(cow
, cow
->start
);
166 btrfs_set_header_generation(cow
, trans
->transid
);
167 btrfs_set_header_owner(cow
, new_root_objectid
);
168 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
);
170 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
171 ret
= btrfs_inc_ref(trans
, new_root
, buf
);
177 btrfs_mark_buffer_dirty(cow
);
182 int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
183 struct btrfs_root
*root
,
184 struct extent_buffer
*buf
,
185 struct extent_buffer
*parent
, int parent_slot
,
186 struct extent_buffer
**cow_ret
,
187 u64 search_start
, u64 empty_size
)
190 struct extent_buffer
*cow
;
193 int different_trans
= 0;
196 struct btrfs_key first_key
;
201 WARN_ON(!btrfs_tree_locked(buf
));
203 if (root
->ref_cows
) {
204 root_gen
= trans
->transid
;
208 WARN_ON(root
->ref_cows
&& trans
->transid
!=
209 root
->fs_info
->running_transaction
->transid
);
210 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
212 level
= btrfs_header_level(buf
);
213 nritems
= btrfs_header_nritems(buf
);
216 btrfs_item_key_to_cpu(buf
, &first_key
, 0);
218 btrfs_node_key_to_cpu(buf
, &first_key
, 0);
220 first_key
.objectid
= 0;
222 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
,
223 root
->root_key
.objectid
,
224 root_gen
, first_key
.objectid
, level
,
225 search_start
, empty_size
);
229 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
230 btrfs_set_header_bytenr(cow
, cow
->start
);
231 btrfs_set_header_generation(cow
, trans
->transid
);
232 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
233 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
);
235 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
236 if (btrfs_header_generation(buf
) != trans
->transid
) {
238 ret
= btrfs_inc_ref(trans
, root
, buf
);
242 clean_tree_block(trans
, root
, buf
);
245 if (buf
== root
->node
) {
246 WARN_ON(parent
&& parent
!= buf
);
247 root_gen
= btrfs_header_generation(buf
);
249 spin_lock(&root
->node_lock
);
251 extent_buffer_get(cow
);
252 spin_unlock(&root
->node_lock
);
254 if (buf
!= root
->commit_root
) {
255 btrfs_free_extent(trans
, root
, buf
->start
,
256 buf
->len
, root
->root_key
.objectid
,
259 free_extent_buffer(buf
);
260 add_root_to_dirty_list(root
);
262 root_gen
= btrfs_header_generation(parent
);
263 btrfs_set_node_blockptr(parent
, parent_slot
,
265 WARN_ON(trans
->transid
== 0);
266 btrfs_set_node_ptr_generation(parent
, parent_slot
,
268 btrfs_mark_buffer_dirty(parent
);
269 WARN_ON(btrfs_header_generation(parent
) != trans
->transid
);
270 btrfs_free_extent(trans
, root
, buf
->start
, buf
->len
,
271 btrfs_header_owner(parent
), root_gen
,
275 btrfs_tree_unlock(buf
);
276 free_extent_buffer(buf
);
277 btrfs_mark_buffer_dirty(cow
);
282 int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
283 struct btrfs_root
*root
, struct extent_buffer
*buf
,
284 struct extent_buffer
*parent
, int parent_slot
,
285 struct extent_buffer
**cow_ret
)
291 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
292 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
293 root
->fs_info
->running_transaction
->transid
);
296 if (trans
->transid
!= root
->fs_info
->generation
) {
297 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
298 root
->fs_info
->generation
);
302 header_trans
= btrfs_header_generation(buf
);
303 spin_lock(&root
->fs_info
->hash_lock
);
304 if (header_trans
== trans
->transid
&&
305 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
307 spin_unlock(&root
->fs_info
->hash_lock
);
310 spin_unlock(&root
->fs_info
->hash_lock
);
311 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
312 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
313 parent_slot
, cow_ret
, search_start
, 0);
317 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
319 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
321 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
327 * compare two keys in a memcmp fashion
329 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
333 btrfs_disk_key_to_cpu(&k1
, disk
);
335 if (k1
.objectid
> k2
->objectid
)
337 if (k1
.objectid
< k2
->objectid
)
339 if (k1
.type
> k2
->type
)
341 if (k1
.type
< k2
->type
)
343 if (k1
.offset
> k2
->offset
)
345 if (k1
.offset
< k2
->offset
)
351 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
352 struct btrfs_root
*root
, struct extent_buffer
*parent
,
353 int start_slot
, int cache_only
, u64
*last_ret
,
354 struct btrfs_key
*progress
)
356 struct extent_buffer
*cur
;
357 struct extent_buffer
*tmp
;
360 u64 search_start
= *last_ret
;
370 int progress_passed
= 0;
371 struct btrfs_disk_key disk_key
;
373 /* FIXME this code needs locking */
376 parent_level
= btrfs_header_level(parent
);
377 if (cache_only
&& parent_level
!= 1)
380 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
381 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
382 root
->fs_info
->running_transaction
->transid
);
385 if (trans
->transid
!= root
->fs_info
->generation
) {
386 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
387 root
->fs_info
->generation
);
391 parent_nritems
= btrfs_header_nritems(parent
);
392 blocksize
= btrfs_level_size(root
, parent_level
- 1);
393 end_slot
= parent_nritems
;
395 if (parent_nritems
== 1)
398 for (i
= start_slot
; i
< end_slot
; i
++) {
401 if (!parent
->map_token
) {
402 map_extent_buffer(parent
,
403 btrfs_node_key_ptr_offset(i
),
404 sizeof(struct btrfs_key_ptr
),
405 &parent
->map_token
, &parent
->kaddr
,
406 &parent
->map_start
, &parent
->map_len
,
409 btrfs_node_key(parent
, &disk_key
, i
);
410 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
414 blocknr
= btrfs_node_blockptr(parent
, i
);
415 gen
= btrfs_node_ptr_generation(parent
, i
);
417 last_block
= blocknr
;
420 other
= btrfs_node_blockptr(parent
, i
- 1);
421 close
= close_blocks(blocknr
, other
, blocksize
);
423 if (!close
&& i
< end_slot
- 2) {
424 other
= btrfs_node_blockptr(parent
, i
+ 1);
425 close
= close_blocks(blocknr
, other
, blocksize
);
428 last_block
= blocknr
;
431 if (parent
->map_token
) {
432 unmap_extent_buffer(parent
, parent
->map_token
,
434 parent
->map_token
= NULL
;
437 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
439 uptodate
= btrfs_buffer_uptodate(cur
, gen
);
442 if (!cur
|| !uptodate
) {
444 free_extent_buffer(cur
);
448 cur
= read_tree_block(root
, blocknr
,
450 } else if (!uptodate
) {
451 btrfs_read_buffer(cur
, gen
);
454 if (search_start
== 0)
455 search_start
= last_block
;
457 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
460 (end_slot
- i
) * blocksize
));
462 free_extent_buffer(cur
);
465 search_start
= tmp
->start
;
466 last_block
= tmp
->start
;
467 *last_ret
= search_start
;
468 if (parent_level
== 1)
469 btrfs_clear_buffer_defrag(tmp
);
470 free_extent_buffer(tmp
);
472 if (parent
->map_token
) {
473 unmap_extent_buffer(parent
, parent
->map_token
,
475 parent
->map_token
= NULL
;
481 * The leaf data grows from end-to-front in the node.
482 * this returns the address of the start of the last item,
483 * which is the stop of the leaf data stack
485 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
486 struct extent_buffer
*leaf
)
488 u32 nr
= btrfs_header_nritems(leaf
);
490 return BTRFS_LEAF_DATA_SIZE(root
);
491 return btrfs_item_offset_nr(leaf
, nr
- 1);
494 static int check_node(struct btrfs_root
*root
, struct btrfs_path
*path
,
497 struct extent_buffer
*parent
= NULL
;
498 struct extent_buffer
*node
= path
->nodes
[level
];
499 struct btrfs_disk_key parent_key
;
500 struct btrfs_disk_key node_key
;
503 struct btrfs_key cpukey
;
504 u32 nritems
= btrfs_header_nritems(node
);
506 if (path
->nodes
[level
+ 1])
507 parent
= path
->nodes
[level
+ 1];
509 slot
= path
->slots
[level
];
510 BUG_ON(nritems
== 0);
512 parent_slot
= path
->slots
[level
+ 1];
513 btrfs_node_key(parent
, &parent_key
, parent_slot
);
514 btrfs_node_key(node
, &node_key
, 0);
515 BUG_ON(memcmp(&parent_key
, &node_key
,
516 sizeof(struct btrfs_disk_key
)));
517 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
518 btrfs_header_bytenr(node
));
520 BUG_ON(nritems
> BTRFS_NODEPTRS_PER_BLOCK(root
));
522 btrfs_node_key_to_cpu(node
, &cpukey
, slot
- 1);
523 btrfs_node_key(node
, &node_key
, slot
);
524 BUG_ON(comp_keys(&node_key
, &cpukey
) <= 0);
526 if (slot
< nritems
- 1) {
527 btrfs_node_key_to_cpu(node
, &cpukey
, slot
+ 1);
528 btrfs_node_key(node
, &node_key
, slot
);
529 BUG_ON(comp_keys(&node_key
, &cpukey
) >= 0);
534 static int check_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
537 struct extent_buffer
*leaf
= path
->nodes
[level
];
538 struct extent_buffer
*parent
= NULL
;
540 struct btrfs_key cpukey
;
541 struct btrfs_disk_key parent_key
;
542 struct btrfs_disk_key leaf_key
;
543 int slot
= path
->slots
[0];
545 u32 nritems
= btrfs_header_nritems(leaf
);
547 if (path
->nodes
[level
+ 1])
548 parent
= path
->nodes
[level
+ 1];
554 parent_slot
= path
->slots
[level
+ 1];
555 btrfs_node_key(parent
, &parent_key
, parent_slot
);
556 btrfs_item_key(leaf
, &leaf_key
, 0);
558 BUG_ON(memcmp(&parent_key
, &leaf_key
,
559 sizeof(struct btrfs_disk_key
)));
560 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
561 btrfs_header_bytenr(leaf
));
564 for (i
= 0; nritems
> 1 && i
< nritems
- 2; i
++) {
565 btrfs_item_key_to_cpu(leaf
, &cpukey
, i
+ 1);
566 btrfs_item_key(leaf
, &leaf_key
, i
);
567 if (comp_keys(&leaf_key
, &cpukey
) >= 0) {
568 btrfs_print_leaf(root
, leaf
);
569 printk("slot %d offset bad key\n", i
);
572 if (btrfs_item_offset_nr(leaf
, i
) !=
573 btrfs_item_end_nr(leaf
, i
+ 1)) {
574 btrfs_print_leaf(root
, leaf
);
575 printk("slot %d offset bad\n", i
);
579 if (btrfs_item_offset_nr(leaf
, i
) +
580 btrfs_item_size_nr(leaf
, i
) !=
581 BTRFS_LEAF_DATA_SIZE(root
)) {
582 btrfs_print_leaf(root
, leaf
);
583 printk("slot %d first offset bad\n", i
);
589 if (btrfs_item_size_nr(leaf
, nritems
- 1) > 4096) {
590 btrfs_print_leaf(root
, leaf
);
591 printk("slot %d bad size \n", nritems
- 1);
596 if (slot
!= 0 && slot
< nritems
- 1) {
597 btrfs_item_key(leaf
, &leaf_key
, slot
);
598 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
- 1);
599 if (comp_keys(&leaf_key
, &cpukey
) <= 0) {
600 btrfs_print_leaf(root
, leaf
);
601 printk("slot %d offset bad key\n", slot
);
604 if (btrfs_item_offset_nr(leaf
, slot
- 1) !=
605 btrfs_item_end_nr(leaf
, slot
)) {
606 btrfs_print_leaf(root
, leaf
);
607 printk("slot %d offset bad\n", slot
);
611 if (slot
< nritems
- 1) {
612 btrfs_item_key(leaf
, &leaf_key
, slot
);
613 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
+ 1);
614 BUG_ON(comp_keys(&leaf_key
, &cpukey
) >= 0);
615 if (btrfs_item_offset_nr(leaf
, slot
) !=
616 btrfs_item_end_nr(leaf
, slot
+ 1)) {
617 btrfs_print_leaf(root
, leaf
);
618 printk("slot %d offset bad\n", slot
);
622 BUG_ON(btrfs_item_offset_nr(leaf
, 0) +
623 btrfs_item_size_nr(leaf
, 0) != BTRFS_LEAF_DATA_SIZE(root
));
627 static int noinline
check_block(struct btrfs_root
*root
,
628 struct btrfs_path
*path
, int level
)
632 if (btrfs_header_level(path
->nodes
[level
]) != level
)
633 printk("warning: bad level %Lu wanted %d found %d\n",
634 path
->nodes
[level
]->start
, level
,
635 btrfs_header_level(path
->nodes
[level
]));
636 found_start
= btrfs_header_bytenr(path
->nodes
[level
]);
637 if (found_start
!= path
->nodes
[level
]->start
) {
638 printk("warning: bad bytentr %Lu found %Lu\n",
639 path
->nodes
[level
]->start
, found_start
);
642 struct extent_buffer
*buf
= path
->nodes
[level
];
644 if (memcmp_extent_buffer(buf
, root
->fs_info
->fsid
,
645 (unsigned long)btrfs_header_fsid(buf
),
647 printk("warning bad block %Lu\n", buf
->start
);
652 return check_leaf(root
, path
, level
);
653 return check_node(root
, path
, level
);
657 * search for key in the extent_buffer. The items start at offset p,
658 * and they are item_size apart. There are 'max' items in p.
660 * the slot in the array is returned via slot, and it points to
661 * the place where you would insert key if it is not found in
664 * slot may point to max if the key is bigger than all of the keys
666 static int generic_bin_search(struct extent_buffer
*eb
, unsigned long p
,
667 int item_size
, struct btrfs_key
*key
,
674 struct btrfs_disk_key
*tmp
= NULL
;
675 struct btrfs_disk_key unaligned
;
676 unsigned long offset
;
677 char *map_token
= NULL
;
679 unsigned long map_start
= 0;
680 unsigned long map_len
= 0;
684 mid
= (low
+ high
) / 2;
685 offset
= p
+ mid
* item_size
;
687 if (!map_token
|| offset
< map_start
||
688 (offset
+ sizeof(struct btrfs_disk_key
)) >
689 map_start
+ map_len
) {
691 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
694 err
= map_extent_buffer(eb
, offset
,
695 sizeof(struct btrfs_disk_key
),
697 &map_start
, &map_len
, KM_USER0
);
700 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
703 read_extent_buffer(eb
, &unaligned
,
704 offset
, sizeof(unaligned
));
709 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
712 ret
= comp_keys(tmp
, key
);
721 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
727 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
732 * simple bin_search frontend that does the right thing for
735 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
736 int level
, int *slot
)
739 return generic_bin_search(eb
,
740 offsetof(struct btrfs_leaf
, items
),
741 sizeof(struct btrfs_item
),
742 key
, btrfs_header_nritems(eb
),
745 return generic_bin_search(eb
,
746 offsetof(struct btrfs_node
, ptrs
),
747 sizeof(struct btrfs_key_ptr
),
748 key
, btrfs_header_nritems(eb
),
754 static struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
755 struct extent_buffer
*parent
, int slot
)
757 int level
= btrfs_header_level(parent
);
760 if (slot
>= btrfs_header_nritems(parent
))
765 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
766 btrfs_level_size(root
, level
- 1),
767 btrfs_node_ptr_generation(parent
, slot
));
770 static int balance_level(struct btrfs_trans_handle
*trans
,
771 struct btrfs_root
*root
,
772 struct btrfs_path
*path
, int level
)
774 struct extent_buffer
*right
= NULL
;
775 struct extent_buffer
*mid
;
776 struct extent_buffer
*left
= NULL
;
777 struct extent_buffer
*parent
= NULL
;
781 int orig_slot
= path
->slots
[level
];
782 int err_on_enospc
= 0;
788 mid
= path
->nodes
[level
];
789 WARN_ON(!path
->locks
[level
]);
790 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
792 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
794 if (level
< BTRFS_MAX_LEVEL
- 1)
795 parent
= path
->nodes
[level
+ 1];
796 pslot
= path
->slots
[level
+ 1];
799 * deal with the case where there is only one pointer in the root
800 * by promoting the node below to a root
803 struct extent_buffer
*child
;
805 if (btrfs_header_nritems(mid
) != 1)
808 /* promote the child to a root */
809 child
= read_node_slot(root
, mid
, 0);
810 btrfs_tree_lock(child
);
812 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
815 spin_lock(&root
->node_lock
);
817 spin_unlock(&root
->node_lock
);
819 add_root_to_dirty_list(root
);
820 btrfs_tree_unlock(child
);
821 path
->locks
[level
] = 0;
822 path
->nodes
[level
] = NULL
;
823 clean_tree_block(trans
, root
, mid
);
824 btrfs_tree_unlock(mid
);
825 /* once for the path */
826 free_extent_buffer(mid
);
827 ret
= btrfs_free_extent(trans
, root
, mid
->start
, mid
->len
,
828 root
->root_key
.objectid
,
829 btrfs_header_generation(mid
), 0, 0, 1);
830 /* once for the root ptr */
831 free_extent_buffer(mid
);
834 if (btrfs_header_nritems(mid
) >
835 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
838 if (btrfs_header_nritems(mid
) < 2)
841 left
= read_node_slot(root
, parent
, pslot
- 1);
843 btrfs_tree_lock(left
);
844 wret
= btrfs_cow_block(trans
, root
, left
,
845 parent
, pslot
- 1, &left
);
851 right
= read_node_slot(root
, parent
, pslot
+ 1);
853 btrfs_tree_lock(right
);
854 wret
= btrfs_cow_block(trans
, root
, right
,
855 parent
, pslot
+ 1, &right
);
862 /* first, try to make some room in the middle buffer */
864 orig_slot
+= btrfs_header_nritems(left
);
865 wret
= push_node_left(trans
, root
, left
, mid
, 1);
868 if (btrfs_header_nritems(mid
) < 2)
873 * then try to empty the right most buffer into the middle
876 wret
= push_node_left(trans
, root
, mid
, right
, 1);
877 if (wret
< 0 && wret
!= -ENOSPC
)
879 if (btrfs_header_nritems(right
) == 0) {
880 u64 bytenr
= right
->start
;
881 u64 generation
= btrfs_header_generation(parent
);
882 u32 blocksize
= right
->len
;
884 clean_tree_block(trans
, root
, right
);
885 btrfs_tree_unlock(right
);
886 free_extent_buffer(right
);
888 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
+
892 wret
= btrfs_free_extent(trans
, root
, bytenr
,
894 btrfs_header_owner(parent
),
895 generation
, 0, 0, 1);
899 struct btrfs_disk_key right_key
;
900 btrfs_node_key(right
, &right_key
, 0);
901 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
902 btrfs_mark_buffer_dirty(parent
);
905 if (btrfs_header_nritems(mid
) == 1) {
907 * we're not allowed to leave a node with one item in the
908 * tree during a delete. A deletion from lower in the tree
909 * could try to delete the only pointer in this node.
910 * So, pull some keys from the left.
911 * There has to be a left pointer at this point because
912 * otherwise we would have pulled some pointers from the
916 wret
= balance_node_right(trans
, root
, mid
, left
);
922 wret
= push_node_left(trans
, root
, left
, mid
, 1);
928 if (btrfs_header_nritems(mid
) == 0) {
929 /* we've managed to empty the middle node, drop it */
930 u64 root_gen
= btrfs_header_generation(parent
);
931 u64 bytenr
= mid
->start
;
932 u32 blocksize
= mid
->len
;
934 clean_tree_block(trans
, root
, mid
);
935 btrfs_tree_unlock(mid
);
936 free_extent_buffer(mid
);
938 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
);
941 wret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
942 btrfs_header_owner(parent
),
947 /* update the parent key to reflect our changes */
948 struct btrfs_disk_key mid_key
;
949 btrfs_node_key(mid
, &mid_key
, 0);
950 btrfs_set_node_key(parent
, &mid_key
, pslot
);
951 btrfs_mark_buffer_dirty(parent
);
954 /* update the path */
956 if (btrfs_header_nritems(left
) > orig_slot
) {
957 extent_buffer_get(left
);
958 /* left was locked after cow */
959 path
->nodes
[level
] = left
;
960 path
->slots
[level
+ 1] -= 1;
961 path
->slots
[level
] = orig_slot
;
963 btrfs_tree_unlock(mid
);
964 free_extent_buffer(mid
);
967 orig_slot
-= btrfs_header_nritems(left
);
968 path
->slots
[level
] = orig_slot
;
971 /* double check we haven't messed things up */
972 check_block(root
, path
, level
);
974 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
978 btrfs_tree_unlock(right
);
979 free_extent_buffer(right
);
982 if (path
->nodes
[level
] != left
)
983 btrfs_tree_unlock(left
);
984 free_extent_buffer(left
);
989 /* returns zero if the push worked, non-zero otherwise */
990 static int noinline
push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
991 struct btrfs_root
*root
,
992 struct btrfs_path
*path
, int level
)
994 struct extent_buffer
*right
= NULL
;
995 struct extent_buffer
*mid
;
996 struct extent_buffer
*left
= NULL
;
997 struct extent_buffer
*parent
= NULL
;
1001 int orig_slot
= path
->slots
[level
];
1007 mid
= path
->nodes
[level
];
1008 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1009 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1011 if (level
< BTRFS_MAX_LEVEL
- 1)
1012 parent
= path
->nodes
[level
+ 1];
1013 pslot
= path
->slots
[level
+ 1];
1018 left
= read_node_slot(root
, parent
, pslot
- 1);
1020 /* first, try to make some room in the middle buffer */
1024 btrfs_tree_lock(left
);
1025 left_nr
= btrfs_header_nritems(left
);
1026 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1029 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1034 wret
= push_node_left(trans
, root
,
1041 struct btrfs_disk_key disk_key
;
1042 orig_slot
+= left_nr
;
1043 btrfs_node_key(mid
, &disk_key
, 0);
1044 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1045 btrfs_mark_buffer_dirty(parent
);
1046 if (btrfs_header_nritems(left
) > orig_slot
) {
1047 path
->nodes
[level
] = left
;
1048 path
->slots
[level
+ 1] -= 1;
1049 path
->slots
[level
] = orig_slot
;
1050 btrfs_tree_unlock(mid
);
1051 free_extent_buffer(mid
);
1054 btrfs_header_nritems(left
);
1055 path
->slots
[level
] = orig_slot
;
1056 btrfs_tree_unlock(left
);
1057 free_extent_buffer(left
);
1061 btrfs_tree_unlock(left
);
1062 free_extent_buffer(left
);
1064 right
= read_node_slot(root
, parent
, pslot
+ 1);
1067 * then try to empty the right most buffer into the middle
1071 btrfs_tree_lock(right
);
1072 right_nr
= btrfs_header_nritems(right
);
1073 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1076 ret
= btrfs_cow_block(trans
, root
, right
,
1082 wret
= balance_node_right(trans
, root
,
1089 struct btrfs_disk_key disk_key
;
1091 btrfs_node_key(right
, &disk_key
, 0);
1092 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
1093 btrfs_mark_buffer_dirty(parent
);
1095 if (btrfs_header_nritems(mid
) <= orig_slot
) {
1096 path
->nodes
[level
] = right
;
1097 path
->slots
[level
+ 1] += 1;
1098 path
->slots
[level
] = orig_slot
-
1099 btrfs_header_nritems(mid
);
1100 btrfs_tree_unlock(mid
);
1101 free_extent_buffer(mid
);
1103 btrfs_tree_unlock(right
);
1104 free_extent_buffer(right
);
1108 btrfs_tree_unlock(right
);
1109 free_extent_buffer(right
);
1115 * readahead one full node of leaves
1117 static void reada_for_search(struct btrfs_root
*root
, struct btrfs_path
*path
,
1118 int level
, int slot
, u64 objectid
)
1120 struct extent_buffer
*node
;
1121 struct btrfs_disk_key disk_key
;
1127 int direction
= path
->reada
;
1128 struct extent_buffer
*eb
;
1136 if (!path
->nodes
[level
])
1139 node
= path
->nodes
[level
];
1141 search
= btrfs_node_blockptr(node
, slot
);
1142 blocksize
= btrfs_level_size(root
, level
- 1);
1143 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
1145 free_extent_buffer(eb
);
1149 highest_read
= search
;
1150 lowest_read
= search
;
1152 nritems
= btrfs_header_nritems(node
);
1155 if (direction
< 0) {
1159 } else if (direction
> 0) {
1164 if (path
->reada
< 0 && objectid
) {
1165 btrfs_node_key(node
, &disk_key
, nr
);
1166 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
1169 search
= btrfs_node_blockptr(node
, nr
);
1170 if ((search
>= lowest_read
&& search
<= highest_read
) ||
1171 (search
< lowest_read
&& lowest_read
- search
<= 32768) ||
1172 (search
> highest_read
&& search
- highest_read
<= 32768)) {
1173 readahead_tree_block(root
, search
, blocksize
,
1174 btrfs_node_ptr_generation(node
, nr
));
1178 if (path
->reada
< 2 && (nread
> (256 * 1024) || nscan
> 32))
1180 if(nread
> (1024 * 1024) || nscan
> 128)
1183 if (search
< lowest_read
)
1184 lowest_read
= search
;
1185 if (search
> highest_read
)
1186 highest_read
= search
;
1190 static void unlock_up(struct btrfs_path
*path
, int level
, int lowest_unlock
)
1193 int skip_level
= level
;
1195 struct extent_buffer
*t
;
1197 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
1198 if (!path
->nodes
[i
])
1200 if (!path
->locks
[i
])
1202 if (!no_skips
&& path
->slots
[i
] == 0) {
1206 if (!no_skips
&& path
->keep_locks
) {
1209 nritems
= btrfs_header_nritems(t
);
1210 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
1215 if (skip_level
< i
&& i
>= lowest_unlock
)
1219 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
1220 btrfs_tree_unlock(t
);
1227 * look for key in the tree. path is filled in with nodes along the way
1228 * if key is found, we return zero and you can find the item in the leaf
1229 * level of the path (level 0)
1231 * If the key isn't found, the path points to the slot where it should
1232 * be inserted, and 1 is returned. If there are other errors during the
1233 * search a negative error number is returned.
1235 * if ins_len > 0, nodes and leaves will be split as we walk down the
1236 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1239 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
1240 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
1243 struct extent_buffer
*b
;
1244 struct extent_buffer
*tmp
;
1248 int should_reada
= p
->reada
;
1249 int lowest_unlock
= 1;
1251 u8 lowest_level
= 0;
1255 lowest_level
= p
->lowest_level
;
1256 WARN_ON(lowest_level
&& ins_len
);
1257 WARN_ON(p
->nodes
[0] != NULL
);
1258 WARN_ON(cow
&& root
== root
->fs_info
->extent_root
&&
1259 !mutex_is_locked(&root
->fs_info
->alloc_mutex
));
1260 WARN_ON(root
== root
->fs_info
->chunk_root
&&
1261 !mutex_is_locked(&root
->fs_info
->chunk_mutex
));
1262 WARN_ON(root
== root
->fs_info
->dev_root
&&
1263 !mutex_is_locked(&root
->fs_info
->chunk_mutex
));
1267 if (p
->skip_locking
)
1268 b
= btrfs_root_node(root
);
1270 b
= btrfs_lock_root_node(root
);
1273 level
= btrfs_header_level(b
);
1276 wret
= btrfs_cow_block(trans
, root
, b
,
1277 p
->nodes
[level
+ 1],
1278 p
->slots
[level
+ 1],
1281 free_extent_buffer(b
);
1285 BUG_ON(!cow
&& ins_len
);
1286 if (level
!= btrfs_header_level(b
))
1288 level
= btrfs_header_level(b
);
1289 p
->nodes
[level
] = b
;
1290 if (!p
->skip_locking
)
1291 p
->locks
[level
] = 1;
1292 ret
= check_block(root
, p
, level
);
1296 ret
= bin_search(b
, key
, level
, &slot
);
1298 if (ret
&& slot
> 0)
1300 p
->slots
[level
] = slot
;
1301 if (ins_len
> 0 && btrfs_header_nritems(b
) >=
1302 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
1303 int sret
= split_node(trans
, root
, p
, level
);
1307 b
= p
->nodes
[level
];
1308 slot
= p
->slots
[level
];
1309 } else if (ins_len
< 0) {
1310 int sret
= balance_level(trans
, root
, p
,
1314 b
= p
->nodes
[level
];
1316 btrfs_release_path(NULL
, p
);
1319 slot
= p
->slots
[level
];
1320 BUG_ON(btrfs_header_nritems(b
) == 1);
1322 /* this is only true while dropping a snapshot */
1323 if (level
== lowest_level
) {
1324 unlock_up(p
, level
, lowest_unlock
);
1329 reada_for_search(root
, p
, level
, slot
,
1332 blocknr
= btrfs_node_blockptr(b
, slot
);
1333 gen
= btrfs_node_ptr_generation(b
, slot
);
1334 blocksize
= btrfs_level_size(root
, level
- 1);
1336 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1337 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
)) {
1341 * reduce lock contention at high levels
1342 * of the btree by dropping locks before
1346 btrfs_release_path(NULL
, p
);
1348 free_extent_buffer(tmp
);
1349 tmp
= read_tree_block(root
, blocknr
,
1352 free_extent_buffer(tmp
);
1355 b
= read_node_slot(root
, b
, slot
);
1358 if (!p
->skip_locking
)
1360 unlock_up(p
, level
, lowest_unlock
);
1362 p
->slots
[level
] = slot
;
1363 if (ins_len
> 0 && btrfs_leaf_free_space(root
, b
) <
1364 sizeof(struct btrfs_item
) + ins_len
) {
1365 int sret
= split_leaf(trans
, root
, key
,
1366 p
, ins_len
, ret
== 0);
1371 unlock_up(p
, level
, lowest_unlock
);
1379 * adjust the pointers going up the tree, starting at level
1380 * making sure the right key of each node is points to 'key'.
1381 * This is used after shifting pointers to the left, so it stops
1382 * fixing up pointers when a given leaf/node is not in slot 0 of the
1385 * If this fails to write a tree block, it returns -1, but continues
1386 * fixing up the blocks in ram so the tree is consistent.
1388 static int fixup_low_keys(struct btrfs_trans_handle
*trans
,
1389 struct btrfs_root
*root
, struct btrfs_path
*path
,
1390 struct btrfs_disk_key
*key
, int level
)
1394 struct extent_buffer
*t
;
1396 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
1397 int tslot
= path
->slots
[i
];
1398 if (!path
->nodes
[i
])
1401 btrfs_set_node_key(t
, key
, tslot
);
1402 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
1410 * try to push data from one node into the next node left in the
1413 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1414 * error, and > 0 if there was no room in the left hand block.
1416 static int push_node_left(struct btrfs_trans_handle
*trans
,
1417 struct btrfs_root
*root
, struct extent_buffer
*dst
,
1418 struct extent_buffer
*src
, int empty
)
1425 src_nritems
= btrfs_header_nritems(src
);
1426 dst_nritems
= btrfs_header_nritems(dst
);
1427 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
1428 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1429 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
1431 if (!empty
&& src_nritems
<= 8)
1434 if (push_items
<= 0) {
1439 push_items
= min(src_nritems
, push_items
);
1440 if (push_items
< src_nritems
) {
1441 /* leave at least 8 pointers in the node if
1442 * we aren't going to empty it
1444 if (src_nritems
- push_items
< 8) {
1445 if (push_items
<= 8)
1451 push_items
= min(src_nritems
- 8, push_items
);
1453 copy_extent_buffer(dst
, src
,
1454 btrfs_node_key_ptr_offset(dst_nritems
),
1455 btrfs_node_key_ptr_offset(0),
1456 push_items
* sizeof(struct btrfs_key_ptr
));
1458 if (push_items
< src_nritems
) {
1459 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
1460 btrfs_node_key_ptr_offset(push_items
),
1461 (src_nritems
- push_items
) *
1462 sizeof(struct btrfs_key_ptr
));
1464 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1465 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1466 btrfs_mark_buffer_dirty(src
);
1467 btrfs_mark_buffer_dirty(dst
);
1472 * try to push data from one node into the next node right in the
1475 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1476 * error, and > 0 if there was no room in the right hand block.
1478 * this will only push up to 1/2 the contents of the left node over
1480 static int balance_node_right(struct btrfs_trans_handle
*trans
,
1481 struct btrfs_root
*root
,
1482 struct extent_buffer
*dst
,
1483 struct extent_buffer
*src
)
1491 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1492 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
1494 src_nritems
= btrfs_header_nritems(src
);
1495 dst_nritems
= btrfs_header_nritems(dst
);
1496 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
1497 if (push_items
<= 0) {
1501 if (src_nritems
< 4) {
1505 max_push
= src_nritems
/ 2 + 1;
1506 /* don't try to empty the node */
1507 if (max_push
>= src_nritems
) {
1511 if (max_push
< push_items
)
1512 push_items
= max_push
;
1514 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
1515 btrfs_node_key_ptr_offset(0),
1517 sizeof(struct btrfs_key_ptr
));
1519 copy_extent_buffer(dst
, src
,
1520 btrfs_node_key_ptr_offset(0),
1521 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
1522 push_items
* sizeof(struct btrfs_key_ptr
));
1524 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1525 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1527 btrfs_mark_buffer_dirty(src
);
1528 btrfs_mark_buffer_dirty(dst
);
1533 * helper function to insert a new root level in the tree.
1534 * A new node is allocated, and a single item is inserted to
1535 * point to the existing root
1537 * returns zero on success or < 0 on failure.
1539 static int noinline
insert_new_root(struct btrfs_trans_handle
*trans
,
1540 struct btrfs_root
*root
,
1541 struct btrfs_path
*path
, int level
)
1545 struct extent_buffer
*lower
;
1546 struct extent_buffer
*c
;
1547 struct extent_buffer
*old
;
1548 struct btrfs_disk_key lower_key
;
1550 BUG_ON(path
->nodes
[level
]);
1551 BUG_ON(path
->nodes
[level
-1] != root
->node
);
1554 root_gen
= trans
->transid
;
1558 lower
= path
->nodes
[level
-1];
1560 btrfs_item_key(lower
, &lower_key
, 0);
1562 btrfs_node_key(lower
, &lower_key
, 0);
1564 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1565 root
->root_key
.objectid
,
1566 root_gen
, lower_key
.objectid
, level
,
1567 root
->node
->start
, 0);
1571 memset_extent_buffer(c
, 0, 0, root
->nodesize
);
1572 btrfs_set_header_nritems(c
, 1);
1573 btrfs_set_header_level(c
, level
);
1574 btrfs_set_header_bytenr(c
, c
->start
);
1575 btrfs_set_header_generation(c
, trans
->transid
);
1576 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
1578 write_extent_buffer(c
, root
->fs_info
->fsid
,
1579 (unsigned long)btrfs_header_fsid(c
),
1582 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
1583 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
1586 btrfs_set_node_key(c
, &lower_key
, 0);
1587 btrfs_set_node_blockptr(c
, 0, lower
->start
);
1588 lower_gen
= btrfs_header_generation(lower
);
1589 WARN_ON(lower_gen
== 0);
1591 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
1593 btrfs_mark_buffer_dirty(c
);
1595 spin_lock(&root
->node_lock
);
1598 spin_unlock(&root
->node_lock
);
1600 /* the super has an extra ref to root->node */
1601 free_extent_buffer(old
);
1603 add_root_to_dirty_list(root
);
1604 extent_buffer_get(c
);
1605 path
->nodes
[level
] = c
;
1606 path
->locks
[level
] = 1;
1607 path
->slots
[level
] = 0;
1609 if (root
->ref_cows
&& lower_gen
!= trans
->transid
) {
1610 struct btrfs_path
*back_path
= btrfs_alloc_path();
1612 mutex_lock(&root
->fs_info
->alloc_mutex
);
1613 ret
= btrfs_insert_extent_backref(trans
,
1614 root
->fs_info
->extent_root
,
1616 root
->root_key
.objectid
,
1617 trans
->transid
, 0, 0);
1619 mutex_unlock(&root
->fs_info
->alloc_mutex
);
1620 btrfs_free_path(back_path
);
1626 * worker function to insert a single pointer in a node.
1627 * the node should have enough room for the pointer already
1629 * slot and level indicate where you want the key to go, and
1630 * blocknr is the block the key points to.
1632 * returns zero on success and < 0 on any error
1634 static int insert_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
1635 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
1636 *key
, u64 bytenr
, int slot
, int level
)
1638 struct extent_buffer
*lower
;
1641 BUG_ON(!path
->nodes
[level
]);
1642 lower
= path
->nodes
[level
];
1643 nritems
= btrfs_header_nritems(lower
);
1646 if (nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
))
1648 if (slot
!= nritems
) {
1649 memmove_extent_buffer(lower
,
1650 btrfs_node_key_ptr_offset(slot
+ 1),
1651 btrfs_node_key_ptr_offset(slot
),
1652 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
1654 btrfs_set_node_key(lower
, key
, slot
);
1655 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
1656 WARN_ON(trans
->transid
== 0);
1657 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
1658 btrfs_set_header_nritems(lower
, nritems
+ 1);
1659 btrfs_mark_buffer_dirty(lower
);
1664 * split the node at the specified level in path in two.
1665 * The path is corrected to point to the appropriate node after the split
1667 * Before splitting this tries to make some room in the node by pushing
1668 * left and right, if either one works, it returns right away.
1670 * returns 0 on success and < 0 on failure
1672 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
1673 *root
, struct btrfs_path
*path
, int level
)
1676 struct extent_buffer
*c
;
1677 struct extent_buffer
*split
;
1678 struct btrfs_disk_key disk_key
;
1684 c
= path
->nodes
[level
];
1685 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
1686 if (c
== root
->node
) {
1687 /* trying to split the root, lets make a new one */
1688 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
1692 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
1693 c
= path
->nodes
[level
];
1694 if (!ret
&& btrfs_header_nritems(c
) <
1695 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
1701 c_nritems
= btrfs_header_nritems(c
);
1703 root_gen
= trans
->transid
;
1707 btrfs_node_key(c
, &disk_key
, 0);
1708 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1709 root
->root_key
.objectid
,
1711 btrfs_disk_key_objectid(&disk_key
),
1712 level
, c
->start
, 0);
1714 return PTR_ERR(split
);
1716 btrfs_set_header_flags(split
, btrfs_header_flags(c
));
1717 btrfs_set_header_level(split
, btrfs_header_level(c
));
1718 btrfs_set_header_bytenr(split
, split
->start
);
1719 btrfs_set_header_generation(split
, trans
->transid
);
1720 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
1721 btrfs_set_header_flags(split
, 0);
1722 write_extent_buffer(split
, root
->fs_info
->fsid
,
1723 (unsigned long)btrfs_header_fsid(split
),
1725 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
1726 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
1729 mid
= (c_nritems
+ 1) / 2;
1731 copy_extent_buffer(split
, c
,
1732 btrfs_node_key_ptr_offset(0),
1733 btrfs_node_key_ptr_offset(mid
),
1734 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
1735 btrfs_set_header_nritems(split
, c_nritems
- mid
);
1736 btrfs_set_header_nritems(c
, mid
);
1739 btrfs_mark_buffer_dirty(c
);
1740 btrfs_mark_buffer_dirty(split
);
1742 btrfs_node_key(split
, &disk_key
, 0);
1743 wret
= insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
1744 path
->slots
[level
+ 1] + 1,
1749 if (path
->slots
[level
] >= mid
) {
1750 path
->slots
[level
] -= mid
;
1751 btrfs_tree_unlock(c
);
1752 free_extent_buffer(c
);
1753 path
->nodes
[level
] = split
;
1754 path
->slots
[level
+ 1] += 1;
1756 btrfs_tree_unlock(split
);
1757 free_extent_buffer(split
);
1763 * how many bytes are required to store the items in a leaf. start
1764 * and nr indicate which items in the leaf to check. This totals up the
1765 * space used both by the item structs and the item data
1767 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
1770 int nritems
= btrfs_header_nritems(l
);
1771 int end
= min(nritems
, start
+ nr
) - 1;
1775 data_len
= btrfs_item_end_nr(l
, start
);
1776 data_len
= data_len
- btrfs_item_offset_nr(l
, end
);
1777 data_len
+= sizeof(struct btrfs_item
) * nr
;
1778 WARN_ON(data_len
< 0);
1783 * The space between the end of the leaf items and
1784 * the start of the leaf data. IOW, how much room
1785 * the leaf has left for both items and data
1787 int btrfs_leaf_free_space(struct btrfs_root
*root
, struct extent_buffer
*leaf
)
1789 int nritems
= btrfs_header_nritems(leaf
);
1791 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
1793 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1794 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
1795 leaf_space_used(leaf
, 0, nritems
), nritems
);
1801 * push some data in the path leaf to the right, trying to free up at
1802 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1804 * returns 1 if the push failed because the other node didn't have enough
1805 * room, 0 if everything worked out and < 0 if there were major errors.
1807 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
1808 *root
, struct btrfs_path
*path
, int data_size
,
1811 struct extent_buffer
*left
= path
->nodes
[0];
1812 struct extent_buffer
*right
;
1813 struct extent_buffer
*upper
;
1814 struct btrfs_disk_key disk_key
;
1820 struct btrfs_item
*item
;
1828 slot
= path
->slots
[1];
1829 if (!path
->nodes
[1]) {
1832 upper
= path
->nodes
[1];
1833 if (slot
>= btrfs_header_nritems(upper
) - 1)
1836 WARN_ON(!btrfs_tree_locked(path
->nodes
[1]));
1838 right
= read_node_slot(root
, upper
, slot
+ 1);
1839 btrfs_tree_lock(right
);
1840 free_space
= btrfs_leaf_free_space(root
, right
);
1841 if (free_space
< data_size
+ sizeof(struct btrfs_item
))
1844 /* cow and double check */
1845 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
1850 free_space
= btrfs_leaf_free_space(root
, right
);
1851 if (free_space
< data_size
+ sizeof(struct btrfs_item
))
1854 left_nritems
= btrfs_header_nritems(left
);
1855 if (left_nritems
== 0)
1863 i
= left_nritems
- 1;
1865 item
= btrfs_item_nr(left
, i
);
1867 if (path
->slots
[0] == i
)
1868 push_space
+= data_size
+ sizeof(*item
);
1870 if (!left
->map_token
) {
1871 map_extent_buffer(left
, (unsigned long)item
,
1872 sizeof(struct btrfs_item
),
1873 &left
->map_token
, &left
->kaddr
,
1874 &left
->map_start
, &left
->map_len
,
1878 this_item_size
= btrfs_item_size(left
, item
);
1879 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
1882 push_space
+= this_item_size
+ sizeof(*item
);
1887 if (left
->map_token
) {
1888 unmap_extent_buffer(left
, left
->map_token
, KM_USER1
);
1889 left
->map_token
= NULL
;
1892 if (push_items
== 0)
1895 if (!empty
&& push_items
== left_nritems
)
1898 /* push left to right */
1899 right_nritems
= btrfs_header_nritems(right
);
1901 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
1902 push_space
-= leaf_data_end(root
, left
);
1904 /* make room in the right data area */
1905 data_end
= leaf_data_end(root
, right
);
1906 memmove_extent_buffer(right
,
1907 btrfs_leaf_data(right
) + data_end
- push_space
,
1908 btrfs_leaf_data(right
) + data_end
,
1909 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
1911 /* copy from the left data area */
1912 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
1913 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
1914 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
1917 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
1918 btrfs_item_nr_offset(0),
1919 right_nritems
* sizeof(struct btrfs_item
));
1921 /* copy the items from left to right */
1922 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
1923 btrfs_item_nr_offset(left_nritems
- push_items
),
1924 push_items
* sizeof(struct btrfs_item
));
1926 /* update the item pointers */
1927 right_nritems
+= push_items
;
1928 btrfs_set_header_nritems(right
, right_nritems
);
1929 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
1930 for (i
= 0; i
< right_nritems
; i
++) {
1931 item
= btrfs_item_nr(right
, i
);
1932 if (!right
->map_token
) {
1933 map_extent_buffer(right
, (unsigned long)item
,
1934 sizeof(struct btrfs_item
),
1935 &right
->map_token
, &right
->kaddr
,
1936 &right
->map_start
, &right
->map_len
,
1939 push_space
-= btrfs_item_size(right
, item
);
1940 btrfs_set_item_offset(right
, item
, push_space
);
1943 if (right
->map_token
) {
1944 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
1945 right
->map_token
= NULL
;
1947 left_nritems
-= push_items
;
1948 btrfs_set_header_nritems(left
, left_nritems
);
1951 btrfs_mark_buffer_dirty(left
);
1952 btrfs_mark_buffer_dirty(right
);
1954 btrfs_item_key(right
, &disk_key
, 0);
1955 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
1956 btrfs_mark_buffer_dirty(upper
);
1958 /* then fixup the leaf pointer in the path */
1959 if (path
->slots
[0] >= left_nritems
) {
1960 path
->slots
[0] -= left_nritems
;
1961 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
1962 clean_tree_block(trans
, root
, path
->nodes
[0]);
1963 btrfs_tree_unlock(path
->nodes
[0]);
1964 free_extent_buffer(path
->nodes
[0]);
1965 path
->nodes
[0] = right
;
1966 path
->slots
[1] += 1;
1968 btrfs_tree_unlock(right
);
1969 free_extent_buffer(right
);
1974 btrfs_tree_unlock(right
);
1975 free_extent_buffer(right
);
1980 * push some data in the path leaf to the left, trying to free up at
1981 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1983 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
1984 *root
, struct btrfs_path
*path
, int data_size
,
1987 struct btrfs_disk_key disk_key
;
1988 struct extent_buffer
*right
= path
->nodes
[0];
1989 struct extent_buffer
*left
;
1995 struct btrfs_item
*item
;
1996 u32 old_left_nritems
;
2002 u32 old_left_item_size
;
2004 slot
= path
->slots
[1];
2007 if (!path
->nodes
[1])
2010 right_nritems
= btrfs_header_nritems(right
);
2011 if (right_nritems
== 0) {
2015 WARN_ON(!btrfs_tree_locked(path
->nodes
[1]));
2017 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
2018 btrfs_tree_lock(left
);
2019 free_space
= btrfs_leaf_free_space(root
, left
);
2020 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
2025 /* cow and double check */
2026 ret
= btrfs_cow_block(trans
, root
, left
,
2027 path
->nodes
[1], slot
- 1, &left
);
2029 /* we hit -ENOSPC, but it isn't fatal here */
2034 free_space
= btrfs_leaf_free_space(root
, left
);
2035 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
2043 nr
= right_nritems
- 1;
2045 for (i
= 0; i
< nr
; i
++) {
2046 item
= btrfs_item_nr(right
, i
);
2047 if (!right
->map_token
) {
2048 map_extent_buffer(right
, (unsigned long)item
,
2049 sizeof(struct btrfs_item
),
2050 &right
->map_token
, &right
->kaddr
,
2051 &right
->map_start
, &right
->map_len
,
2055 if (path
->slots
[0] == i
)
2056 push_space
+= data_size
+ sizeof(*item
);
2058 this_item_size
= btrfs_item_size(right
, item
);
2059 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
2063 push_space
+= this_item_size
+ sizeof(*item
);
2066 if (right
->map_token
) {
2067 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2068 right
->map_token
= NULL
;
2071 if (push_items
== 0) {
2075 if (!empty
&& push_items
== btrfs_header_nritems(right
))
2078 /* push data from right to left */
2079 copy_extent_buffer(left
, right
,
2080 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
2081 btrfs_item_nr_offset(0),
2082 push_items
* sizeof(struct btrfs_item
));
2084 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
2085 btrfs_item_offset_nr(right
, push_items
-1);
2087 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
2088 leaf_data_end(root
, left
) - push_space
,
2089 btrfs_leaf_data(right
) +
2090 btrfs_item_offset_nr(right
, push_items
- 1),
2092 old_left_nritems
= btrfs_header_nritems(left
);
2093 BUG_ON(old_left_nritems
< 0);
2095 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
2096 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
2099 item
= btrfs_item_nr(left
, i
);
2100 if (!left
->map_token
) {
2101 map_extent_buffer(left
, (unsigned long)item
,
2102 sizeof(struct btrfs_item
),
2103 &left
->map_token
, &left
->kaddr
,
2104 &left
->map_start
, &left
->map_len
,
2108 ioff
= btrfs_item_offset(left
, item
);
2109 btrfs_set_item_offset(left
, item
,
2110 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
));
2112 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
2113 if (left
->map_token
) {
2114 unmap_extent_buffer(left
, left
->map_token
, KM_USER1
);
2115 left
->map_token
= NULL
;
2118 /* fixup right node */
2119 if (push_items
> right_nritems
) {
2120 printk("push items %d nr %u\n", push_items
, right_nritems
);
2124 if (push_items
< right_nritems
) {
2125 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
2126 leaf_data_end(root
, right
);
2127 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
2128 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
2129 btrfs_leaf_data(right
) +
2130 leaf_data_end(root
, right
), push_space
);
2132 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
2133 btrfs_item_nr_offset(push_items
),
2134 (btrfs_header_nritems(right
) - push_items
) *
2135 sizeof(struct btrfs_item
));
2137 right_nritems
-= push_items
;
2138 btrfs_set_header_nritems(right
, right_nritems
);
2139 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
2140 for (i
= 0; i
< right_nritems
; i
++) {
2141 item
= btrfs_item_nr(right
, i
);
2143 if (!right
->map_token
) {
2144 map_extent_buffer(right
, (unsigned long)item
,
2145 sizeof(struct btrfs_item
),
2146 &right
->map_token
, &right
->kaddr
,
2147 &right
->map_start
, &right
->map_len
,
2151 push_space
= push_space
- btrfs_item_size(right
, item
);
2152 btrfs_set_item_offset(right
, item
, push_space
);
2154 if (right
->map_token
) {
2155 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2156 right
->map_token
= NULL
;
2159 btrfs_mark_buffer_dirty(left
);
2161 btrfs_mark_buffer_dirty(right
);
2163 btrfs_item_key(right
, &disk_key
, 0);
2164 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2168 /* then fixup the leaf pointer in the path */
2169 if (path
->slots
[0] < push_items
) {
2170 path
->slots
[0] += old_left_nritems
;
2171 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
2172 clean_tree_block(trans
, root
, path
->nodes
[0]);
2173 btrfs_tree_unlock(path
->nodes
[0]);
2174 free_extent_buffer(path
->nodes
[0]);
2175 path
->nodes
[0] = left
;
2176 path
->slots
[1] -= 1;
2178 btrfs_tree_unlock(left
);
2179 free_extent_buffer(left
);
2180 path
->slots
[0] -= push_items
;
2182 BUG_ON(path
->slots
[0] < 0);
2185 btrfs_tree_unlock(left
);
2186 free_extent_buffer(left
);
2191 * split the path's leaf in two, making sure there is at least data_size
2192 * available for the resulting leaf level of the path.
2194 * returns 0 if all went well and < 0 on failure.
2196 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
2197 *root
, struct btrfs_key
*ins_key
,
2198 struct btrfs_path
*path
, int data_size
, int extend
)
2201 struct extent_buffer
*l
;
2205 struct extent_buffer
*right
;
2206 int space_needed
= data_size
+ sizeof(struct btrfs_item
);
2213 int num_doubles
= 0;
2214 struct btrfs_disk_key disk_key
;
2217 space_needed
= data_size
;
2220 root_gen
= trans
->transid
;
2224 /* first try to make some room by pushing left and right */
2225 if (ins_key
->type
!= BTRFS_DIR_ITEM_KEY
) {
2226 wret
= push_leaf_right(trans
, root
, path
, data_size
, 0);
2231 wret
= push_leaf_left(trans
, root
, path
, data_size
, 0);
2237 /* did the pushes work? */
2238 if (btrfs_leaf_free_space(root
, l
) >= space_needed
)
2242 if (!path
->nodes
[1]) {
2243 ret
= insert_new_root(trans
, root
, path
, 1);
2250 slot
= path
->slots
[0];
2251 nritems
= btrfs_header_nritems(l
);
2252 mid
= (nritems
+ 1)/ 2;
2254 btrfs_item_key(l
, &disk_key
, 0);
2256 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
2257 root
->root_key
.objectid
,
2258 root_gen
, disk_key
.objectid
, 0,
2260 if (IS_ERR(right
)) {
2262 return PTR_ERR(right
);
2265 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
2266 btrfs_set_header_bytenr(right
, right
->start
);
2267 btrfs_set_header_generation(right
, trans
->transid
);
2268 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
2269 btrfs_set_header_level(right
, 0);
2270 write_extent_buffer(right
, root
->fs_info
->fsid
,
2271 (unsigned long)btrfs_header_fsid(right
),
2274 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
2275 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
2279 leaf_space_used(l
, mid
, nritems
- mid
) + space_needed
>
2280 BTRFS_LEAF_DATA_SIZE(root
)) {
2281 if (slot
>= nritems
) {
2282 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
2283 btrfs_set_header_nritems(right
, 0);
2284 wret
= insert_ptr(trans
, root
, path
,
2285 &disk_key
, right
->start
,
2286 path
->slots
[1] + 1, 1);
2290 btrfs_tree_unlock(path
->nodes
[0]);
2291 free_extent_buffer(path
->nodes
[0]);
2292 path
->nodes
[0] = right
;
2294 path
->slots
[1] += 1;
2295 btrfs_mark_buffer_dirty(right
);
2299 if (mid
!= nritems
&&
2300 leaf_space_used(l
, mid
, nritems
- mid
) +
2301 space_needed
> BTRFS_LEAF_DATA_SIZE(root
)) {
2306 if (leaf_space_used(l
, 0, mid
+ 1) + space_needed
>
2307 BTRFS_LEAF_DATA_SIZE(root
)) {
2308 if (!extend
&& slot
== 0) {
2309 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
2310 btrfs_set_header_nritems(right
, 0);
2311 wret
= insert_ptr(trans
, root
, path
,
2317 btrfs_tree_unlock(path
->nodes
[0]);
2318 free_extent_buffer(path
->nodes
[0]);
2319 path
->nodes
[0] = right
;
2321 if (path
->slots
[1] == 0) {
2322 wret
= fixup_low_keys(trans
, root
,
2323 path
, &disk_key
, 1);
2327 btrfs_mark_buffer_dirty(right
);
2329 } else if (extend
&& slot
== 0) {
2333 if (mid
!= nritems
&&
2334 leaf_space_used(l
, mid
, nritems
- mid
) +
2335 space_needed
> BTRFS_LEAF_DATA_SIZE(root
)) {
2341 nritems
= nritems
- mid
;
2342 btrfs_set_header_nritems(right
, nritems
);
2343 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
2345 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
2346 btrfs_item_nr_offset(mid
),
2347 nritems
* sizeof(struct btrfs_item
));
2349 copy_extent_buffer(right
, l
,
2350 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
2351 data_copy_size
, btrfs_leaf_data(l
) +
2352 leaf_data_end(root
, l
), data_copy_size
);
2354 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
2355 btrfs_item_end_nr(l
, mid
);
2357 for (i
= 0; i
< nritems
; i
++) {
2358 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
2361 if (!right
->map_token
) {
2362 map_extent_buffer(right
, (unsigned long)item
,
2363 sizeof(struct btrfs_item
),
2364 &right
->map_token
, &right
->kaddr
,
2365 &right
->map_start
, &right
->map_len
,
2369 ioff
= btrfs_item_offset(right
, item
);
2370 btrfs_set_item_offset(right
, item
, ioff
+ rt_data_off
);
2373 if (right
->map_token
) {
2374 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2375 right
->map_token
= NULL
;
2378 btrfs_set_header_nritems(l
, mid
);
2380 btrfs_item_key(right
, &disk_key
, 0);
2381 wret
= insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
2382 path
->slots
[1] + 1, 1);
2386 btrfs_mark_buffer_dirty(right
);
2387 btrfs_mark_buffer_dirty(l
);
2388 BUG_ON(path
->slots
[0] != slot
);
2391 btrfs_tree_unlock(path
->nodes
[0]);
2392 free_extent_buffer(path
->nodes
[0]);
2393 path
->nodes
[0] = right
;
2394 path
->slots
[0] -= mid
;
2395 path
->slots
[1] += 1;
2397 btrfs_tree_unlock(right
);
2398 free_extent_buffer(right
);
2401 BUG_ON(path
->slots
[0] < 0);
2404 BUG_ON(num_doubles
!= 0);
2411 int btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
2412 struct btrfs_root
*root
,
2413 struct btrfs_path
*path
,
2414 u32 new_size
, int from_end
)
2419 struct extent_buffer
*leaf
;
2420 struct btrfs_item
*item
;
2422 unsigned int data_end
;
2423 unsigned int old_data_start
;
2424 unsigned int old_size
;
2425 unsigned int size_diff
;
2428 slot_orig
= path
->slots
[0];
2429 leaf
= path
->nodes
[0];
2430 slot
= path
->slots
[0];
2432 old_size
= btrfs_item_size_nr(leaf
, slot
);
2433 if (old_size
== new_size
)
2436 nritems
= btrfs_header_nritems(leaf
);
2437 data_end
= leaf_data_end(root
, leaf
);
2439 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
2441 size_diff
= old_size
- new_size
;
2444 BUG_ON(slot
>= nritems
);
2447 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2449 /* first correct the data pointers */
2450 for (i
= slot
; i
< nritems
; i
++) {
2452 item
= btrfs_item_nr(leaf
, i
);
2454 if (!leaf
->map_token
) {
2455 map_extent_buffer(leaf
, (unsigned long)item
,
2456 sizeof(struct btrfs_item
),
2457 &leaf
->map_token
, &leaf
->kaddr
,
2458 &leaf
->map_start
, &leaf
->map_len
,
2462 ioff
= btrfs_item_offset(leaf
, item
);
2463 btrfs_set_item_offset(leaf
, item
, ioff
+ size_diff
);
2466 if (leaf
->map_token
) {
2467 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2468 leaf
->map_token
= NULL
;
2471 /* shift the data */
2473 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2474 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
2475 data_end
, old_data_start
+ new_size
- data_end
);
2477 struct btrfs_disk_key disk_key
;
2480 btrfs_item_key(leaf
, &disk_key
, slot
);
2482 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
2484 struct btrfs_file_extent_item
*fi
;
2486 fi
= btrfs_item_ptr(leaf
, slot
,
2487 struct btrfs_file_extent_item
);
2488 fi
= (struct btrfs_file_extent_item
*)(
2489 (unsigned long)fi
- size_diff
);
2491 if (btrfs_file_extent_type(leaf
, fi
) ==
2492 BTRFS_FILE_EXTENT_INLINE
) {
2493 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
2494 memmove_extent_buffer(leaf
, ptr
,
2496 offsetof(struct btrfs_file_extent_item
,
2501 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2502 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
2503 data_end
, old_data_start
- data_end
);
2505 offset
= btrfs_disk_key_offset(&disk_key
);
2506 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
2507 btrfs_set_item_key(leaf
, &disk_key
, slot
);
2509 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2512 item
= btrfs_item_nr(leaf
, slot
);
2513 btrfs_set_item_size(leaf
, item
, new_size
);
2514 btrfs_mark_buffer_dirty(leaf
);
2517 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2518 btrfs_print_leaf(root
, leaf
);
2524 int btrfs_extend_item(struct btrfs_trans_handle
*trans
,
2525 struct btrfs_root
*root
, struct btrfs_path
*path
,
2531 struct extent_buffer
*leaf
;
2532 struct btrfs_item
*item
;
2534 unsigned int data_end
;
2535 unsigned int old_data
;
2536 unsigned int old_size
;
2539 slot_orig
= path
->slots
[0];
2540 leaf
= path
->nodes
[0];
2542 nritems
= btrfs_header_nritems(leaf
);
2543 data_end
= leaf_data_end(root
, leaf
);
2545 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
2546 btrfs_print_leaf(root
, leaf
);
2549 slot
= path
->slots
[0];
2550 old_data
= btrfs_item_end_nr(leaf
, slot
);
2553 if (slot
>= nritems
) {
2554 btrfs_print_leaf(root
, leaf
);
2555 printk("slot %d too large, nritems %d\n", slot
, nritems
);
2560 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2562 /* first correct the data pointers */
2563 for (i
= slot
; i
< nritems
; i
++) {
2565 item
= btrfs_item_nr(leaf
, i
);
2567 if (!leaf
->map_token
) {
2568 map_extent_buffer(leaf
, (unsigned long)item
,
2569 sizeof(struct btrfs_item
),
2570 &leaf
->map_token
, &leaf
->kaddr
,
2571 &leaf
->map_start
, &leaf
->map_len
,
2574 ioff
= btrfs_item_offset(leaf
, item
);
2575 btrfs_set_item_offset(leaf
, item
, ioff
- data_size
);
2578 if (leaf
->map_token
) {
2579 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2580 leaf
->map_token
= NULL
;
2583 /* shift the data */
2584 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2585 data_end
- data_size
, btrfs_leaf_data(leaf
) +
2586 data_end
, old_data
- data_end
);
2588 data_end
= old_data
;
2589 old_size
= btrfs_item_size_nr(leaf
, slot
);
2590 item
= btrfs_item_nr(leaf
, slot
);
2591 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
2592 btrfs_mark_buffer_dirty(leaf
);
2595 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2596 btrfs_print_leaf(root
, leaf
);
2603 * Given a key and some data, insert an item into the tree.
2604 * This does all the path init required, making room in the tree if needed.
2606 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
2607 struct btrfs_root
*root
,
2608 struct btrfs_path
*path
,
2609 struct btrfs_key
*cpu_key
, u32
*data_size
,
2612 struct extent_buffer
*leaf
;
2613 struct btrfs_item
*item
;
2621 unsigned int data_end
;
2622 struct btrfs_disk_key disk_key
;
2624 for (i
= 0; i
< nr
; i
++) {
2625 total_data
+= data_size
[i
];
2628 total_size
= total_data
+ (nr
- 1) * sizeof(struct btrfs_item
);
2629 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
2636 slot_orig
= path
->slots
[0];
2637 leaf
= path
->nodes
[0];
2639 nritems
= btrfs_header_nritems(leaf
);
2640 data_end
= leaf_data_end(root
, leaf
);
2642 if (btrfs_leaf_free_space(root
, leaf
) <
2643 sizeof(struct btrfs_item
) + total_size
) {
2644 btrfs_print_leaf(root
, leaf
);
2645 printk("not enough freespace need %u have %d\n",
2646 total_size
, btrfs_leaf_free_space(root
, leaf
));
2650 slot
= path
->slots
[0];
2653 if (slot
!= nritems
) {
2655 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
2657 if (old_data
< data_end
) {
2658 btrfs_print_leaf(root
, leaf
);
2659 printk("slot %d old_data %d data_end %d\n",
2660 slot
, old_data
, data_end
);
2664 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2666 /* first correct the data pointers */
2667 WARN_ON(leaf
->map_token
);
2668 for (i
= slot
; i
< nritems
; i
++) {
2671 item
= btrfs_item_nr(leaf
, i
);
2672 if (!leaf
->map_token
) {
2673 map_extent_buffer(leaf
, (unsigned long)item
,
2674 sizeof(struct btrfs_item
),
2675 &leaf
->map_token
, &leaf
->kaddr
,
2676 &leaf
->map_start
, &leaf
->map_len
,
2680 ioff
= btrfs_item_offset(leaf
, item
);
2681 btrfs_set_item_offset(leaf
, item
, ioff
- total_data
);
2683 if (leaf
->map_token
) {
2684 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2685 leaf
->map_token
= NULL
;
2688 /* shift the items */
2689 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
2690 btrfs_item_nr_offset(slot
),
2691 (nritems
- slot
) * sizeof(struct btrfs_item
));
2693 /* shift the data */
2694 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2695 data_end
- total_data
, btrfs_leaf_data(leaf
) +
2696 data_end
, old_data
- data_end
);
2697 data_end
= old_data
;
2700 /* setup the item for the new data */
2701 for (i
= 0; i
< nr
; i
++) {
2702 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
2703 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
2704 item
= btrfs_item_nr(leaf
, slot
+ i
);
2705 btrfs_set_item_offset(leaf
, item
, data_end
- data_size
[i
]);
2706 data_end
-= data_size
[i
];
2707 btrfs_set_item_size(leaf
, item
, data_size
[i
]);
2709 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
2710 btrfs_mark_buffer_dirty(leaf
);
2714 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
2715 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2718 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2719 btrfs_print_leaf(root
, leaf
);
2727 * Given a key and some data, insert an item into the tree.
2728 * This does all the path init required, making room in the tree if needed.
2730 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
2731 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
2735 struct btrfs_path
*path
;
2736 struct extent_buffer
*leaf
;
2739 path
= btrfs_alloc_path();
2741 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
2743 leaf
= path
->nodes
[0];
2744 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2745 write_extent_buffer(leaf
, data
, ptr
, data_size
);
2746 btrfs_mark_buffer_dirty(leaf
);
2748 btrfs_free_path(path
);
2753 * delete the pointer from a given node.
2755 * If the delete empties a node, the node is removed from the tree,
2756 * continuing all the way the root if required. The root is converted into
2757 * a leaf if all the nodes are emptied.
2759 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2760 struct btrfs_path
*path
, int level
, int slot
)
2762 struct extent_buffer
*parent
= path
->nodes
[level
];
2767 nritems
= btrfs_header_nritems(parent
);
2768 if (slot
!= nritems
-1) {
2769 memmove_extent_buffer(parent
,
2770 btrfs_node_key_ptr_offset(slot
),
2771 btrfs_node_key_ptr_offset(slot
+ 1),
2772 sizeof(struct btrfs_key_ptr
) *
2773 (nritems
- slot
- 1));
2776 btrfs_set_header_nritems(parent
, nritems
);
2777 if (nritems
== 0 && parent
== root
->node
) {
2778 BUG_ON(btrfs_header_level(root
->node
) != 1);
2779 /* just turn the root into a leaf and break */
2780 btrfs_set_header_level(root
->node
, 0);
2781 } else if (slot
== 0) {
2782 struct btrfs_disk_key disk_key
;
2784 btrfs_node_key(parent
, &disk_key
, 0);
2785 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
2789 btrfs_mark_buffer_dirty(parent
);
2794 * delete the item at the leaf level in path. If that empties
2795 * the leaf, remove it from the tree
2797 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2798 struct btrfs_path
*path
, int slot
, int nr
)
2800 struct extent_buffer
*leaf
;
2801 struct btrfs_item
*item
;
2809 leaf
= path
->nodes
[0];
2810 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
2812 for (i
= 0; i
< nr
; i
++)
2813 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
2815 nritems
= btrfs_header_nritems(leaf
);
2817 if (slot
+ nr
!= nritems
) {
2819 int data_end
= leaf_data_end(root
, leaf
);
2821 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2823 btrfs_leaf_data(leaf
) + data_end
,
2824 last_off
- data_end
);
2826 for (i
= slot
+ nr
; i
< nritems
; i
++) {
2829 item
= btrfs_item_nr(leaf
, i
);
2830 if (!leaf
->map_token
) {
2831 map_extent_buffer(leaf
, (unsigned long)item
,
2832 sizeof(struct btrfs_item
),
2833 &leaf
->map_token
, &leaf
->kaddr
,
2834 &leaf
->map_start
, &leaf
->map_len
,
2837 ioff
= btrfs_item_offset(leaf
, item
);
2838 btrfs_set_item_offset(leaf
, item
, ioff
+ dsize
);
2841 if (leaf
->map_token
) {
2842 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
2843 leaf
->map_token
= NULL
;
2846 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
2847 btrfs_item_nr_offset(slot
+ nr
),
2848 sizeof(struct btrfs_item
) *
2849 (nritems
- slot
- nr
));
2851 btrfs_set_header_nritems(leaf
, nritems
- nr
);
2854 /* delete the leaf if we've emptied it */
2856 if (leaf
== root
->node
) {
2857 btrfs_set_header_level(leaf
, 0);
2859 u64 root_gen
= btrfs_header_generation(path
->nodes
[1]);
2860 wret
= del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
2863 wret
= btrfs_free_extent(trans
, root
,
2864 leaf
->start
, leaf
->len
,
2865 btrfs_header_owner(path
->nodes
[1]),
2871 int used
= leaf_space_used(leaf
, 0, nritems
);
2873 struct btrfs_disk_key disk_key
;
2875 btrfs_item_key(leaf
, &disk_key
, 0);
2876 wret
= fixup_low_keys(trans
, root
, path
,
2882 /* delete the leaf if it is mostly empty */
2883 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 4) {
2884 /* push_leaf_left fixes the path.
2885 * make sure the path still points to our leaf
2886 * for possible call to del_ptr below
2888 slot
= path
->slots
[1];
2889 extent_buffer_get(leaf
);
2891 wret
= push_leaf_left(trans
, root
, path
, 1, 1);
2892 if (wret
< 0 && wret
!= -ENOSPC
)
2895 if (path
->nodes
[0] == leaf
&&
2896 btrfs_header_nritems(leaf
)) {
2897 wret
= push_leaf_right(trans
, root
, path
, 1, 1);
2898 if (wret
< 0 && wret
!= -ENOSPC
)
2902 if (btrfs_header_nritems(leaf
) == 0) {
2904 u64 bytenr
= leaf
->start
;
2905 u32 blocksize
= leaf
->len
;
2907 root_gen
= btrfs_header_generation(
2910 wret
= del_ptr(trans
, root
, path
, 1, slot
);
2914 free_extent_buffer(leaf
);
2915 wret
= btrfs_free_extent(trans
, root
, bytenr
,
2917 btrfs_header_owner(path
->nodes
[1]),
2922 /* if we're still in the path, make sure
2923 * we're dirty. Otherwise, one of the
2924 * push_leaf functions must have already
2925 * dirtied this buffer
2927 if (path
->nodes
[0] == leaf
)
2928 btrfs_mark_buffer_dirty(leaf
);
2929 free_extent_buffer(leaf
);
2932 btrfs_mark_buffer_dirty(leaf
);
2939 * search the tree again to find a leaf with lesser keys
2940 * returns 0 if it found something or 1 if there are no lesser leaves.
2941 * returns < 0 on io errors.
2943 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
2945 struct btrfs_key key
;
2946 struct btrfs_disk_key found_key
;
2949 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
2953 else if (key
.type
> 0)
2955 else if (key
.objectid
> 0)
2960 btrfs_release_path(root
, path
);
2961 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2964 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
2965 ret
= comp_keys(&found_key
, &key
);
2972 * search the tree again to find a leaf with greater keys
2973 * returns 0 if it found something or 1 if there are no greater leaves.
2974 * returns < 0 on io errors.
2976 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
2980 struct extent_buffer
*c
;
2981 struct extent_buffer
*next
= NULL
;
2982 struct btrfs_key key
;
2986 nritems
= btrfs_header_nritems(path
->nodes
[0]);
2991 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
2993 btrfs_release_path(root
, path
);
2994 path
->keep_locks
= 1;
2995 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
2996 path
->keep_locks
= 0;
3001 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3003 * by releasing the path above we dropped all our locks. A balance
3004 * could have added more items next to the key that used to be
3005 * at the very end of the block. So, check again here and
3006 * advance the path if there are now more items available.
3008 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
3013 while(level
< BTRFS_MAX_LEVEL
) {
3014 if (!path
->nodes
[level
])
3017 slot
= path
->slots
[level
] + 1;
3018 c
= path
->nodes
[level
];
3019 if (slot
>= btrfs_header_nritems(c
)) {
3021 if (level
== BTRFS_MAX_LEVEL
) {
3028 btrfs_tree_unlock(next
);
3029 free_extent_buffer(next
);
3032 if (level
== 1 && path
->locks
[1] && path
->reada
)
3033 reada_for_search(root
, path
, level
, slot
, 0);
3035 next
= read_node_slot(root
, c
, slot
);
3036 if (!path
->skip_locking
) {
3037 WARN_ON(!btrfs_tree_locked(c
));
3038 btrfs_tree_lock(next
);
3042 path
->slots
[level
] = slot
;
3045 c
= path
->nodes
[level
];
3046 if (path
->locks
[level
])
3047 btrfs_tree_unlock(c
);
3048 free_extent_buffer(c
);
3049 path
->nodes
[level
] = next
;
3050 path
->slots
[level
] = 0;
3051 path
->locks
[level
] = 1;
3054 if (level
== 1 && path
->locks
[1] && path
->reada
)
3055 reada_for_search(root
, path
, level
, slot
, 0);
3056 next
= read_node_slot(root
, next
, 0);
3057 if (!path
->skip_locking
) {
3058 WARN_ON(!btrfs_tree_locked(path
->nodes
[level
]));
3059 btrfs_tree_lock(next
);
3063 unlock_up(path
, 0, 1);
3067 int btrfs_previous_item(struct btrfs_root
*root
,
3068 struct btrfs_path
*path
, u64 min_objectid
,
3071 struct btrfs_key found_key
;
3072 struct extent_buffer
*leaf
;
3076 if (path
->slots
[0] == 0) {
3077 ret
= btrfs_prev_leaf(root
, path
);
3083 leaf
= path
->nodes
[0];
3084 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
3085 if (found_key
.type
== type
)