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.
20 #include "transaction.h"
21 #include "print-tree.h"
23 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
24 *root
, struct btrfs_path
*path
, int level
);
25 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
26 *root
, struct btrfs_key
*ins_key
,
27 struct btrfs_path
*path
, int data_size
, int extend
);
28 static int push_node_left(struct btrfs_trans_handle
*trans
,
29 struct btrfs_root
*root
, struct extent_buffer
*dst
,
30 struct extent_buffer
*src
, int empty
);
31 static int balance_node_right(struct btrfs_trans_handle
*trans
,
32 struct btrfs_root
*root
,
33 struct extent_buffer
*dst_buf
,
34 struct extent_buffer
*src_buf
);
35 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
36 struct btrfs_path
*path
, int level
, int slot
);
38 inline void btrfs_init_path(struct btrfs_path
*p
)
40 memset(p
, 0, sizeof(*p
));
43 struct btrfs_path
*btrfs_alloc_path(void)
45 struct btrfs_path
*path
;
46 path
= kmalloc(sizeof(struct btrfs_path
), GFP_NOFS
);
48 btrfs_init_path(path
);
54 void btrfs_free_path(struct btrfs_path
*p
)
56 btrfs_release_path(NULL
, p
);
60 void btrfs_release_path(struct btrfs_root
*root
, struct btrfs_path
*p
)
63 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
66 free_extent_buffer(p
->nodes
[i
]);
68 memset(p
, 0, sizeof(*p
));
71 static void add_root_to_dirty_list(struct btrfs_root
*root
)
73 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
74 list_add(&root
->dirty_list
,
75 &root
->fs_info
->dirty_cowonly_roots
);
79 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
80 struct btrfs_root
*root
,
81 struct extent_buffer
*buf
,
82 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
84 struct extent_buffer
*cow
;
87 struct btrfs_root
*new_root
;
88 struct btrfs_disk_key disk_key
;
90 new_root
= kmalloc(sizeof(*new_root
), GFP_NOFS
);
94 memcpy(new_root
, root
, sizeof(*new_root
));
95 new_root
->root_key
.objectid
= new_root_objectid
;
97 WARN_ON(root
->ref_cows
&& trans
->transid
!=
98 root
->fs_info
->running_transaction
->transid
);
99 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
101 level
= btrfs_header_level(buf
);
103 btrfs_item_key(buf
, &disk_key
, 0);
105 btrfs_node_key(buf
, &disk_key
, 0);
106 cow
= btrfs_alloc_free_block(trans
, new_root
, buf
->len
,
107 new_root_objectid
, &disk_key
,
108 level
, buf
->start
, 0);
114 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
115 btrfs_set_header_bytenr(cow
, cow
->start
);
116 btrfs_set_header_generation(cow
, trans
->transid
);
117 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
118 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
119 BTRFS_HEADER_FLAG_RELOC
);
120 if (new_root_objectid
== BTRFS_TREE_RELOC_OBJECTID
)
121 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
123 btrfs_set_header_owner(cow
, new_root_objectid
);
125 write_extent_buffer(cow
, root
->fs_info
->fsid
,
126 (unsigned long)btrfs_header_fsid(cow
),
129 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
130 ret
= btrfs_inc_ref(trans
, new_root
, cow
, 0);
136 btrfs_mark_buffer_dirty(cow
);
142 * check if the tree block can be shared by multiple trees
144 int btrfs_block_can_be_shared(struct btrfs_root
*root
,
145 struct extent_buffer
*buf
)
148 * Tree blocks not in refernece counted trees and tree roots
149 * are never shared. If a block was allocated after the last
150 * snapshot and the block was not allocated by tree relocation,
151 * we know the block is not shared.
153 if (root
->ref_cows
&&
154 buf
!= root
->node
&& buf
!= root
->commit_root
&&
155 (btrfs_header_generation(buf
) <=
156 btrfs_root_last_snapshot(&root
->root_item
) ||
157 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
159 #ifdef BTRFS_COMPAT_EXTENT_TREE_V0
160 if (root
->ref_cows
&&
161 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
167 static noinline
int update_ref_for_cow(struct btrfs_trans_handle
*trans
,
168 struct btrfs_root
*root
,
169 struct extent_buffer
*buf
,
170 struct extent_buffer
*cow
)
179 * Backrefs update rules:
181 * Always use full backrefs for extent pointers in tree block
182 * allocated by tree relocation.
184 * If a shared tree block is no longer referenced by its owner
185 * tree (btrfs_header_owner(buf) == root->root_key.objectid),
186 * use full backrefs for extent pointers in tree block.
188 * If a tree block is been relocating
189 * (root->root_key.objectid == BTRFS_TREE_RELOC_OBJECTID),
190 * use full backrefs for extent pointers in tree block.
191 * The reason for this is some operations (such as drop tree)
192 * are only allowed for blocks use full backrefs.
195 if (btrfs_block_can_be_shared(root
, buf
)) {
196 ret
= btrfs_lookup_extent_info(trans
, root
, buf
->start
,
197 buf
->len
, &refs
, &flags
);
202 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
||
203 btrfs_header_backref_rev(buf
) < BTRFS_MIXED_BACKREF_REV
)
204 flags
= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
209 owner
= btrfs_header_owner(buf
);
210 BUG_ON(!(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) &&
211 owner
== BTRFS_TREE_RELOC_OBJECTID
);
214 if ((owner
== root
->root_key
.objectid
||
215 root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) &&
216 !(flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
)) {
217 ret
= btrfs_inc_ref(trans
, root
, buf
, 1);
220 if (root
->root_key
.objectid
==
221 BTRFS_TREE_RELOC_OBJECTID
) {
222 ret
= btrfs_dec_ref(trans
, root
, buf
, 0);
224 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
227 new_flags
|= BTRFS_BLOCK_FLAG_FULL_BACKREF
;
230 if (root
->root_key
.objectid
==
231 BTRFS_TREE_RELOC_OBJECTID
)
232 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
234 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
237 if (new_flags
!= 0) {
238 ret
= btrfs_set_block_flags(trans
, root
, buf
->start
,
239 buf
->len
, new_flags
);
243 if (flags
& BTRFS_BLOCK_FLAG_FULL_BACKREF
) {
244 if (root
->root_key
.objectid
==
245 BTRFS_TREE_RELOC_OBJECTID
)
246 ret
= btrfs_inc_ref(trans
, root
, cow
, 1);
248 ret
= btrfs_inc_ref(trans
, root
, cow
, 0);
250 ret
= btrfs_dec_ref(trans
, root
, buf
, 1);
253 clean_tree_block(trans
, root
, buf
);
258 int __btrfs_cow_block(struct btrfs_trans_handle
*trans
,
259 struct btrfs_root
*root
,
260 struct extent_buffer
*buf
,
261 struct extent_buffer
*parent
, int parent_slot
,
262 struct extent_buffer
**cow_ret
,
263 u64 search_start
, u64 empty_size
)
265 struct extent_buffer
*cow
;
266 struct btrfs_disk_key disk_key
;
269 WARN_ON(root
->ref_cows
&& trans
->transid
!=
270 root
->fs_info
->running_transaction
->transid
);
271 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
273 level
= btrfs_header_level(buf
);
276 btrfs_item_key(buf
, &disk_key
, 0);
278 btrfs_node_key(buf
, &disk_key
, 0);
280 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
,
281 root
->root_key
.objectid
, &disk_key
,
282 level
, search_start
, empty_size
);
286 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
287 btrfs_set_header_bytenr(cow
, cow
->start
);
288 btrfs_set_header_generation(cow
, trans
->transid
);
289 btrfs_set_header_backref_rev(cow
, BTRFS_MIXED_BACKREF_REV
);
290 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
|
291 BTRFS_HEADER_FLAG_RELOC
);
292 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
)
293 btrfs_set_header_flag(cow
, BTRFS_HEADER_FLAG_RELOC
);
295 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
297 write_extent_buffer(cow
, root
->fs_info
->fsid
,
298 (unsigned long)btrfs_header_fsid(cow
),
301 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
303 update_ref_for_cow(trans
, root
, buf
, cow
);
305 if (buf
== root
->node
) {
307 extent_buffer_get(cow
);
309 btrfs_free_extent(trans
, root
, buf
->start
, buf
->len
,
310 0, root
->root_key
.objectid
, level
, 0);
311 free_extent_buffer(buf
);
312 add_root_to_dirty_list(root
);
314 btrfs_set_node_blockptr(parent
, parent_slot
,
316 WARN_ON(trans
->transid
== 0);
317 btrfs_set_node_ptr_generation(parent
, parent_slot
,
319 btrfs_mark_buffer_dirty(parent
);
320 WARN_ON(btrfs_header_generation(parent
) != trans
->transid
);
322 btrfs_free_extent(trans
, root
, buf
->start
, buf
->len
,
323 0, root
->root_key
.objectid
, level
, 1);
325 free_extent_buffer(buf
);
326 btrfs_mark_buffer_dirty(cow
);
331 static inline int should_cow_block(struct btrfs_trans_handle
*trans
,
332 struct btrfs_root
*root
,
333 struct extent_buffer
*buf
)
335 if (btrfs_header_generation(buf
) == trans
->transid
&&
336 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
) &&
337 !(root
->root_key
.objectid
!= BTRFS_TREE_RELOC_OBJECTID
&&
338 btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_RELOC
)))
343 int btrfs_cow_block(struct btrfs_trans_handle
*trans
,
344 struct btrfs_root
*root
, struct extent_buffer
*buf
,
345 struct extent_buffer
*parent
, int parent_slot
,
346 struct extent_buffer
**cow_ret
)
351 if (trans->transaction != root->fs_info->running_transaction) {
352 printk(KERN_CRIT "trans %Lu running %Lu\n", trans->transid,
353 root->fs_info->running_transaction->transid);
357 if (trans
->transid
!= root
->fs_info
->generation
) {
358 printk(KERN_CRIT
"trans %llu running %llu\n",
359 (unsigned long long)trans
->transid
,
360 (unsigned long long)root
->fs_info
->generation
);
363 if (!should_cow_block(trans
, root
, buf
)) {
368 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
369 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
370 parent_slot
, cow_ret
, search_start
, 0);
375 static int close_blocks(u64 blocknr, u64 other, u32 blocksize)
377 if (blocknr < other && other - (blocknr + blocksize) < 32768)
379 if (blocknr > other && blocknr - (other + blocksize) < 32768)
386 * compare two keys in a memcmp fashion
388 int btrfs_comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
392 btrfs_disk_key_to_cpu(&k1
, disk
);
394 if (k1
.objectid
> k2
->objectid
)
396 if (k1
.objectid
< k2
->objectid
)
398 if (k1
.type
> k2
->type
)
400 if (k1
.type
< k2
->type
)
402 if (k1
.offset
> k2
->offset
)
404 if (k1
.offset
< k2
->offset
)
411 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
412 struct btrfs_root
*root
, struct extent_buffer
*parent
,
413 int start_slot
, int cache_only
, u64
*last_ret
,
414 struct btrfs_key
*progress
)
416 struct extent_buffer
*cur
;
417 struct extent_buffer
*tmp
;
420 u64 search_start
= *last_ret
;
430 int progress_passed
= 0;
431 struct btrfs_disk_key disk_key
;
433 parent_level
= btrfs_header_level(parent
);
434 if (cache_only
&& parent_level
!= 1)
437 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
438 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
439 root
->fs_info
->running_transaction
->transid
);
442 if (trans
->transid
!= root
->fs_info
->generation
) {
443 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
444 root
->fs_info
->generation
);
448 parent_nritems
= btrfs_header_nritems(parent
);
449 blocksize
= btrfs_level_size(root
, parent_level
- 1);
450 end_slot
= parent_nritems
;
452 if (parent_nritems
== 1)
455 for (i
= start_slot
; i
< end_slot
; i
++) {
458 if (!parent
->map_token
) {
459 map_extent_buffer(parent
,
460 btrfs_node_key_ptr_offset(i
),
461 sizeof(struct btrfs_key_ptr
),
462 &parent
->map_token
, &parent
->kaddr
,
463 &parent
->map_start
, &parent
->map_len
,
466 btrfs_node_key(parent
, &disk_key
, i
);
467 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
471 blocknr
= btrfs_node_blockptr(parent
, i
);
472 gen
= btrfs_node_ptr_generation(parent
, i
);
474 last_block
= blocknr
;
477 other
= btrfs_node_blockptr(parent
, i
- 1);
478 close
= close_blocks(blocknr
, other
, blocksize
);
480 if (close
&& i
< end_slot
- 2) {
481 other
= btrfs_node_blockptr(parent
, i
+ 1);
482 close
= close_blocks(blocknr
, other
, blocksize
);
485 last_block
= blocknr
;
488 if (parent
->map_token
) {
489 unmap_extent_buffer(parent
, parent
->map_token
,
491 parent
->map_token
= NULL
;
494 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
496 uptodate
= btrfs_buffer_uptodate(cur
, gen
);
499 if (!cur
|| !uptodate
) {
501 free_extent_buffer(cur
);
505 cur
= read_tree_block(root
, blocknr
,
507 } else if (!uptodate
) {
508 btrfs_read_buffer(cur
, gen
);
511 if (search_start
== 0)
512 search_start
= last_block
;
514 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
517 (end_slot
- i
) * blocksize
));
519 free_extent_buffer(cur
);
522 search_start
= tmp
->start
;
523 last_block
= tmp
->start
;
524 *last_ret
= search_start
;
525 if (parent_level
== 1)
526 btrfs_clear_buffer_defrag(tmp
);
527 free_extent_buffer(tmp
);
529 if (parent
->map_token
) {
530 unmap_extent_buffer(parent
, parent
->map_token
,
532 parent
->map_token
= NULL
;
539 * The leaf data grows from end-to-front in the node.
540 * this returns the address of the start of the last item,
541 * which is the stop of the leaf data stack
543 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
544 struct extent_buffer
*leaf
)
546 u32 nr
= btrfs_header_nritems(leaf
);
548 return BTRFS_LEAF_DATA_SIZE(root
);
549 return btrfs_item_offset_nr(leaf
, nr
- 1);
552 static int check_node(struct btrfs_root
*root
, struct btrfs_path
*path
,
555 struct extent_buffer
*parent
= NULL
;
556 struct extent_buffer
*node
= path
->nodes
[level
];
557 struct btrfs_disk_key parent_key
;
558 struct btrfs_disk_key node_key
;
561 struct btrfs_key cpukey
;
562 u32 nritems
= btrfs_header_nritems(node
);
564 if (path
->nodes
[level
+ 1])
565 parent
= path
->nodes
[level
+ 1];
567 slot
= path
->slots
[level
];
568 BUG_ON(nritems
== 0);
570 parent_slot
= path
->slots
[level
+ 1];
571 btrfs_node_key(parent
, &parent_key
, parent_slot
);
572 btrfs_node_key(node
, &node_key
, 0);
573 BUG_ON(memcmp(&parent_key
, &node_key
,
574 sizeof(struct btrfs_disk_key
)));
575 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
576 btrfs_header_bytenr(node
));
578 BUG_ON(nritems
> BTRFS_NODEPTRS_PER_BLOCK(root
));
580 btrfs_node_key_to_cpu(node
, &cpukey
, slot
- 1);
581 btrfs_node_key(node
, &node_key
, slot
);
582 BUG_ON(btrfs_comp_keys(&node_key
, &cpukey
) <= 0);
584 if (slot
< nritems
- 1) {
585 btrfs_node_key_to_cpu(node
, &cpukey
, slot
+ 1);
586 btrfs_node_key(node
, &node_key
, slot
);
587 BUG_ON(btrfs_comp_keys(&node_key
, &cpukey
) >= 0);
592 static int check_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
595 struct extent_buffer
*leaf
= path
->nodes
[level
];
596 struct extent_buffer
*parent
= NULL
;
598 struct btrfs_key cpukey
;
599 struct btrfs_disk_key parent_key
;
600 struct btrfs_disk_key leaf_key
;
601 int slot
= path
->slots
[0];
603 u32 nritems
= btrfs_header_nritems(leaf
);
605 if (path
->nodes
[level
+ 1])
606 parent
= path
->nodes
[level
+ 1];
612 parent_slot
= path
->slots
[level
+ 1];
613 btrfs_node_key(parent
, &parent_key
, parent_slot
);
614 btrfs_item_key(leaf
, &leaf_key
, 0);
616 BUG_ON(memcmp(&parent_key
, &leaf_key
,
617 sizeof(struct btrfs_disk_key
)));
618 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
619 btrfs_header_bytenr(leaf
));
622 for (i
= 0; nritems
> 1 && i
< nritems
- 2; i
++) {
623 btrfs_item_key_to_cpu(leaf
, &cpukey
, i
+ 1);
624 btrfs_item_key(leaf
, &leaf_key
, i
);
625 if (comp_keys(&leaf_key
, &cpukey
) >= 0) {
626 btrfs_print_leaf(root
, leaf
);
627 printk("slot %d offset bad key\n", i
);
630 if (btrfs_item_offset_nr(leaf
, i
) !=
631 btrfs_item_end_nr(leaf
, i
+ 1)) {
632 btrfs_print_leaf(root
, leaf
);
633 printk("slot %d offset bad\n", i
);
637 if (btrfs_item_offset_nr(leaf
, i
) +
638 btrfs_item_size_nr(leaf
, i
) !=
639 BTRFS_LEAF_DATA_SIZE(root
)) {
640 btrfs_print_leaf(root
, leaf
);
641 printk("slot %d first offset bad\n", i
);
647 if (btrfs_item_size_nr(leaf
, nritems
- 1) > 4096) {
648 btrfs_print_leaf(root
, leaf
);
649 printk("slot %d bad size \n", nritems
- 1);
654 if (slot
!= 0 && slot
< nritems
- 1) {
655 btrfs_item_key(leaf
, &leaf_key
, slot
);
656 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
- 1);
657 if (btrfs_comp_keys(&leaf_key
, &cpukey
) <= 0) {
658 btrfs_print_leaf(root
, leaf
);
659 printk("slot %d offset bad key\n", slot
);
662 if (btrfs_item_offset_nr(leaf
, slot
- 1) !=
663 btrfs_item_end_nr(leaf
, slot
)) {
664 btrfs_print_leaf(root
, leaf
);
665 printk("slot %d offset bad\n", slot
);
669 if (slot
< nritems
- 1) {
670 btrfs_item_key(leaf
, &leaf_key
, slot
);
671 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
+ 1);
672 BUG_ON(btrfs_comp_keys(&leaf_key
, &cpukey
) >= 0);
673 if (btrfs_item_offset_nr(leaf
, slot
) !=
674 btrfs_item_end_nr(leaf
, slot
+ 1)) {
675 btrfs_print_leaf(root
, leaf
);
676 printk("slot %d offset bad\n", slot
);
680 BUG_ON(btrfs_item_offset_nr(leaf
, 0) +
681 btrfs_item_size_nr(leaf
, 0) != BTRFS_LEAF_DATA_SIZE(root
));
685 static int noinline
check_block(struct btrfs_root
*root
,
686 struct btrfs_path
*path
, int level
)
690 struct extent_buffer
*buf
= path
->nodes
[level
];
692 if (memcmp_extent_buffer(buf
, root
->fs_info
->fsid
,
693 (unsigned long)btrfs_header_fsid(buf
),
695 printk("warning bad block %Lu\n", buf
->start
);
700 return check_leaf(root
, path
, level
);
701 return check_node(root
, path
, level
);
705 * search for key in the extent_buffer. The items start at offset p,
706 * and they are item_size apart. There are 'max' items in p.
708 * the slot in the array is returned via slot, and it points to
709 * the place where you would insert key if it is not found in
712 * slot may point to max if the key is bigger than all of the keys
714 static int generic_bin_search(struct extent_buffer
*eb
, unsigned long p
,
715 int item_size
, struct btrfs_key
*key
,
722 unsigned long offset
;
723 struct btrfs_disk_key
*tmp
;
726 mid
= (low
+ high
) / 2;
727 offset
= p
+ mid
* item_size
;
729 tmp
= (struct btrfs_disk_key
*)(eb
->data
+ offset
);
730 ret
= btrfs_comp_keys(tmp
, key
);
746 * simple bin_search frontend that does the right thing for
749 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
750 int level
, int *slot
)
753 return generic_bin_search(eb
,
754 offsetof(struct btrfs_leaf
, items
),
755 sizeof(struct btrfs_item
),
756 key
, btrfs_header_nritems(eb
),
759 return generic_bin_search(eb
,
760 offsetof(struct btrfs_node
, ptrs
),
761 sizeof(struct btrfs_key_ptr
),
762 key
, btrfs_header_nritems(eb
),
768 static struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
769 struct extent_buffer
*parent
, int slot
)
771 int level
= btrfs_header_level(parent
);
774 if (slot
>= btrfs_header_nritems(parent
))
779 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
780 btrfs_level_size(root
, level
- 1),
781 btrfs_node_ptr_generation(parent
, slot
));
784 static int balance_level(struct btrfs_trans_handle
*trans
,
785 struct btrfs_root
*root
,
786 struct btrfs_path
*path
, int level
)
788 struct extent_buffer
*right
= NULL
;
789 struct extent_buffer
*mid
;
790 struct extent_buffer
*left
= NULL
;
791 struct extent_buffer
*parent
= NULL
;
795 int orig_slot
= path
->slots
[level
];
801 mid
= path
->nodes
[level
];
802 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
804 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
806 if (level
< BTRFS_MAX_LEVEL
- 1)
807 parent
= path
->nodes
[level
+ 1];
808 pslot
= path
->slots
[level
+ 1];
811 * deal with the case where there is only one pointer in the root
812 * by promoting the node below to a root
815 struct extent_buffer
*child
;
817 if (btrfs_header_nritems(mid
) != 1)
820 /* promote the child to a root */
821 child
= read_node_slot(root
, mid
, 0);
823 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
);
827 add_root_to_dirty_list(root
);
828 path
->nodes
[level
] = NULL
;
829 clean_tree_block(trans
, root
, mid
);
830 wait_on_tree_block_writeback(root
, mid
);
831 /* once for the path */
832 free_extent_buffer(mid
);
834 ret
= btrfs_free_extent(trans
, root
, mid
->start
, mid
->len
,
835 0, root
->root_key
.objectid
,
837 /* once for the root ptr */
838 free_extent_buffer(mid
);
841 if (btrfs_header_nritems(mid
) >
842 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
845 left
= read_node_slot(root
, parent
, pslot
- 1);
847 wret
= btrfs_cow_block(trans
, root
, left
,
848 parent
, pslot
- 1, &left
);
854 right
= read_node_slot(root
, parent
, pslot
+ 1);
856 wret
= btrfs_cow_block(trans
, root
, right
,
857 parent
, pslot
+ 1, &right
);
864 /* first, try to make some room in the middle buffer */
866 orig_slot
+= btrfs_header_nritems(left
);
867 wret
= push_node_left(trans
, root
, left
, mid
, 1);
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 u32 blocksize
= right
->len
;
883 clean_tree_block(trans
, root
, right
);
884 wait_on_tree_block_writeback(root
, right
);
885 free_extent_buffer(right
);
887 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
+
891 wret
= btrfs_free_extent(trans
, root
, bytenr
,
893 root
->root_key
.objectid
,
898 struct btrfs_disk_key right_key
;
899 btrfs_node_key(right
, &right_key
, 0);
900 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
901 btrfs_mark_buffer_dirty(parent
);
904 if (btrfs_header_nritems(mid
) == 1) {
906 * we're not allowed to leave a node with one item in the
907 * tree during a delete. A deletion from lower in the tree
908 * could try to delete the only pointer in this node.
909 * So, pull some keys from the left.
910 * There has to be a left pointer at this point because
911 * otherwise we would have pulled some pointers from the
915 wret
= balance_node_right(trans
, root
, mid
, left
);
921 wret
= push_node_left(trans
, root
, left
, mid
, 1);
927 if (btrfs_header_nritems(mid
) == 0) {
928 /* we've managed to empty the middle node, drop it */
929 u64 bytenr
= mid
->start
;
930 u32 blocksize
= mid
->len
;
931 clean_tree_block(trans
, root
, mid
);
932 wait_on_tree_block_writeback(root
, mid
);
933 free_extent_buffer(mid
);
935 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
);
938 wret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
939 0, root
->root_key
.objectid
,
944 /* update the parent key to reflect our changes */
945 struct btrfs_disk_key mid_key
;
946 btrfs_node_key(mid
, &mid_key
, 0);
947 btrfs_set_node_key(parent
, &mid_key
, pslot
);
948 btrfs_mark_buffer_dirty(parent
);
951 /* update the path */
953 if (btrfs_header_nritems(left
) > orig_slot
) {
954 extent_buffer_get(left
);
955 path
->nodes
[level
] = left
;
956 path
->slots
[level
+ 1] -= 1;
957 path
->slots
[level
] = orig_slot
;
959 free_extent_buffer(mid
);
961 orig_slot
-= btrfs_header_nritems(left
);
962 path
->slots
[level
] = orig_slot
;
965 /* double check we haven't messed things up */
966 check_block(root
, path
, level
);
968 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
972 free_extent_buffer(right
);
974 free_extent_buffer(left
);
978 /* returns zero if the push worked, non-zero otherwise */
979 static int noinline
push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
980 struct btrfs_root
*root
,
981 struct btrfs_path
*path
, int level
)
983 struct extent_buffer
*right
= NULL
;
984 struct extent_buffer
*mid
;
985 struct extent_buffer
*left
= NULL
;
986 struct extent_buffer
*parent
= NULL
;
990 int orig_slot
= path
->slots
[level
];
995 mid
= path
->nodes
[level
];
996 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
998 if (level
< BTRFS_MAX_LEVEL
- 1)
999 parent
= path
->nodes
[level
+ 1];
1000 pslot
= path
->slots
[level
+ 1];
1005 left
= read_node_slot(root
, parent
, pslot
- 1);
1007 /* first, try to make some room in the middle buffer */
1010 left_nr
= btrfs_header_nritems(left
);
1011 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1014 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1019 wret
= push_node_left(trans
, root
,
1026 struct btrfs_disk_key disk_key
;
1027 orig_slot
+= left_nr
;
1028 btrfs_node_key(mid
, &disk_key
, 0);
1029 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1030 btrfs_mark_buffer_dirty(parent
);
1031 if (btrfs_header_nritems(left
) > orig_slot
) {
1032 path
->nodes
[level
] = left
;
1033 path
->slots
[level
+ 1] -= 1;
1034 path
->slots
[level
] = orig_slot
;
1035 free_extent_buffer(mid
);
1038 btrfs_header_nritems(left
);
1039 path
->slots
[level
] = orig_slot
;
1040 free_extent_buffer(left
);
1044 free_extent_buffer(left
);
1046 right
= read_node_slot(root
, parent
, pslot
+ 1);
1049 * then try to empty the right most buffer into the middle
1053 right_nr
= btrfs_header_nritems(right
);
1054 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1057 ret
= btrfs_cow_block(trans
, root
, right
,
1063 wret
= balance_node_right(trans
, root
,
1070 struct btrfs_disk_key disk_key
;
1072 btrfs_node_key(right
, &disk_key
, 0);
1073 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
1074 btrfs_mark_buffer_dirty(parent
);
1076 if (btrfs_header_nritems(mid
) <= orig_slot
) {
1077 path
->nodes
[level
] = right
;
1078 path
->slots
[level
+ 1] += 1;
1079 path
->slots
[level
] = orig_slot
-
1080 btrfs_header_nritems(mid
);
1081 free_extent_buffer(mid
);
1083 free_extent_buffer(right
);
1087 free_extent_buffer(right
);
1093 * readahead one full node of leaves
1095 static void reada_for_search(struct btrfs_root
*root
, struct btrfs_path
*path
,
1096 int level
, int slot
, u64 objectid
)
1098 struct extent_buffer
*node
;
1099 struct btrfs_disk_key disk_key
;
1105 int direction
= path
->reada
;
1106 struct extent_buffer
*eb
;
1114 if (!path
->nodes
[level
])
1117 node
= path
->nodes
[level
];
1118 search
= btrfs_node_blockptr(node
, slot
);
1119 blocksize
= btrfs_level_size(root
, level
- 1);
1120 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
1122 free_extent_buffer(eb
);
1126 highest_read
= search
;
1127 lowest_read
= search
;
1129 nritems
= btrfs_header_nritems(node
);
1132 if (direction
< 0) {
1136 } else if (direction
> 0) {
1141 if (path
->reada
< 0 && objectid
) {
1142 btrfs_node_key(node
, &disk_key
, nr
);
1143 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
1146 search
= btrfs_node_blockptr(node
, nr
);
1147 if ((search
>= lowest_read
&& search
<= highest_read
) ||
1148 (search
< lowest_read
&& lowest_read
- search
<= 32768) ||
1149 (search
> highest_read
&& search
- highest_read
<= 32768)) {
1150 readahead_tree_block(root
, search
, blocksize
,
1151 btrfs_node_ptr_generation(node
, nr
));
1155 if (path
->reada
< 2 && (nread
> (256 * 1024) || nscan
> 32))
1157 if(nread
> (1024 * 1024) || nscan
> 128)
1160 if (search
< lowest_read
)
1161 lowest_read
= search
;
1162 if (search
> highest_read
)
1163 highest_read
= search
;
1168 * look for key in the tree. path is filled in with nodes along the way
1169 * if key is found, we return zero and you can find the item in the leaf
1170 * level of the path (level 0)
1172 * If the key isn't found, the path points to the slot where it should
1173 * be inserted, and 1 is returned. If there are other errors during the
1174 * search a negative error number is returned.
1176 * if ins_len > 0, nodes and leaves will be split as we walk down the
1177 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1180 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
1181 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
1184 struct extent_buffer
*b
;
1188 int should_reada
= p
->reada
;
1189 u8 lowest_level
= 0;
1191 lowest_level
= p
->lowest_level
;
1192 WARN_ON(lowest_level
&& ins_len
);
1193 WARN_ON(p
->nodes
[0] != NULL
);
1195 WARN_ON(!mutex_is_locked(&root->fs_info->fs_mutex));
1199 extent_buffer_get(b
);
1201 level
= btrfs_header_level(b
);
1204 wret
= btrfs_cow_block(trans
, root
, b
,
1205 p
->nodes
[level
+ 1],
1206 p
->slots
[level
+ 1],
1209 free_extent_buffer(b
);
1213 BUG_ON(!cow
&& ins_len
);
1214 if (level
!= btrfs_header_level(b
))
1216 level
= btrfs_header_level(b
);
1217 p
->nodes
[level
] = b
;
1218 ret
= check_block(root
, p
, level
);
1221 ret
= bin_search(b
, key
, level
, &slot
);
1223 if (ret
&& slot
> 0)
1225 p
->slots
[level
] = slot
;
1226 if ((p
->search_for_split
|| ins_len
> 0) &&
1227 btrfs_header_nritems(b
) >=
1228 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
1229 int sret
= split_node(trans
, root
, p
, level
);
1233 b
= p
->nodes
[level
];
1234 slot
= p
->slots
[level
];
1235 } else if (ins_len
< 0) {
1236 int sret
= balance_level(trans
, root
, p
,
1240 b
= p
->nodes
[level
];
1242 btrfs_release_path(NULL
, p
);
1245 slot
= p
->slots
[level
];
1246 BUG_ON(btrfs_header_nritems(b
) == 1);
1248 /* this is only true while dropping a snapshot */
1249 if (level
== lowest_level
)
1253 reada_for_search(root
, p
, level
, slot
,
1256 b
= read_node_slot(root
, b
, slot
);
1258 p
->slots
[level
] = slot
;
1260 ins_len
> btrfs_leaf_free_space(root
, b
)) {
1261 int sret
= split_leaf(trans
, root
, key
,
1262 p
, ins_len
, ret
== 0);
1274 * adjust the pointers going up the tree, starting at level
1275 * making sure the right key of each node is points to 'key'.
1276 * This is used after shifting pointers to the left, so it stops
1277 * fixing up pointers when a given leaf/node is not in slot 0 of the
1280 * If this fails to write a tree block, it returns -1, but continues
1281 * fixing up the blocks in ram so the tree is consistent.
1283 static int fixup_low_keys(struct btrfs_trans_handle
*trans
,
1284 struct btrfs_root
*root
, struct btrfs_path
*path
,
1285 struct btrfs_disk_key
*key
, int level
)
1289 struct extent_buffer
*t
;
1291 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
1292 int tslot
= path
->slots
[i
];
1293 if (!path
->nodes
[i
])
1296 btrfs_set_node_key(t
, key
, tslot
);
1297 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
1307 * This function isn't completely safe. It's the caller's responsibility
1308 * that the new key won't break the order
1310 int btrfs_set_item_key_safe(struct btrfs_trans_handle
*trans
,
1311 struct btrfs_root
*root
, struct btrfs_path
*path
,
1312 struct btrfs_key
*new_key
)
1314 struct btrfs_disk_key disk_key
;
1315 struct extent_buffer
*eb
;
1318 eb
= path
->nodes
[0];
1319 slot
= path
->slots
[0];
1321 btrfs_item_key(eb
, &disk_key
, slot
- 1);
1322 if (btrfs_comp_keys(&disk_key
, new_key
) >= 0)
1325 if (slot
< btrfs_header_nritems(eb
) - 1) {
1326 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
1327 if (btrfs_comp_keys(&disk_key
, new_key
) <= 0)
1331 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
1332 btrfs_set_item_key(eb
, &disk_key
, slot
);
1333 btrfs_mark_buffer_dirty(eb
);
1335 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
1340 * try to push data from one node into the next node left in the
1343 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1344 * error, and > 0 if there was no room in the left hand block.
1346 static int push_node_left(struct btrfs_trans_handle
*trans
,
1347 struct btrfs_root
*root
, struct extent_buffer
*dst
,
1348 struct extent_buffer
*src
, int empty
)
1355 src_nritems
= btrfs_header_nritems(src
);
1356 dst_nritems
= btrfs_header_nritems(dst
);
1357 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
1358 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1359 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
1361 if (!empty
&& src_nritems
<= 8)
1364 if (push_items
<= 0) {
1369 push_items
= min(src_nritems
, push_items
);
1370 if (push_items
< src_nritems
) {
1371 /* leave at least 8 pointers in the node if
1372 * we aren't going to empty it
1374 if (src_nritems
- push_items
< 8) {
1375 if (push_items
<= 8)
1381 push_items
= min(src_nritems
- 8, push_items
);
1383 copy_extent_buffer(dst
, src
,
1384 btrfs_node_key_ptr_offset(dst_nritems
),
1385 btrfs_node_key_ptr_offset(0),
1386 push_items
* sizeof(struct btrfs_key_ptr
));
1388 if (push_items
< src_nritems
) {
1389 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
1390 btrfs_node_key_ptr_offset(push_items
),
1391 (src_nritems
- push_items
) *
1392 sizeof(struct btrfs_key_ptr
));
1394 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1395 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1396 btrfs_mark_buffer_dirty(src
);
1397 btrfs_mark_buffer_dirty(dst
);
1403 * try to push data from one node into the next node right in the
1406 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1407 * error, and > 0 if there was no room in the right hand block.
1409 * this will only push up to 1/2 the contents of the left node over
1411 static int balance_node_right(struct btrfs_trans_handle
*trans
,
1412 struct btrfs_root
*root
,
1413 struct extent_buffer
*dst
,
1414 struct extent_buffer
*src
)
1422 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1423 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
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 if (push_items
<= 0) {
1432 if (src_nritems
< 4) {
1436 max_push
= src_nritems
/ 2 + 1;
1437 /* don't try to empty the node */
1438 if (max_push
>= src_nritems
) {
1442 if (max_push
< push_items
)
1443 push_items
= max_push
;
1445 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
1446 btrfs_node_key_ptr_offset(0),
1448 sizeof(struct btrfs_key_ptr
));
1450 copy_extent_buffer(dst
, src
,
1451 btrfs_node_key_ptr_offset(0),
1452 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
1453 push_items
* sizeof(struct btrfs_key_ptr
));
1455 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1456 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1458 btrfs_mark_buffer_dirty(src
);
1459 btrfs_mark_buffer_dirty(dst
);
1465 * helper function to insert a new root level in the tree.
1466 * A new node is allocated, and a single item is inserted to
1467 * point to the existing root
1469 * returns zero on success or < 0 on failure.
1471 static int noinline
insert_new_root(struct btrfs_trans_handle
*trans
,
1472 struct btrfs_root
*root
,
1473 struct btrfs_path
*path
, int level
)
1476 struct extent_buffer
*lower
;
1477 struct extent_buffer
*c
;
1478 struct extent_buffer
*old
;
1479 struct btrfs_disk_key lower_key
;
1481 BUG_ON(path
->nodes
[level
]);
1482 BUG_ON(path
->nodes
[level
-1] != root
->node
);
1484 lower
= path
->nodes
[level
-1];
1486 btrfs_item_key(lower
, &lower_key
, 0);
1488 btrfs_node_key(lower
, &lower_key
, 0);
1490 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1491 root
->root_key
.objectid
, &lower_key
,
1492 level
, root
->node
->start
, 0);
1497 memset_extent_buffer(c
, 0, 0, sizeof(struct btrfs_header
));
1498 btrfs_set_header_nritems(c
, 1);
1499 btrfs_set_header_level(c
, level
);
1500 btrfs_set_header_bytenr(c
, c
->start
);
1501 btrfs_set_header_generation(c
, trans
->transid
);
1502 btrfs_set_header_backref_rev(c
, BTRFS_MIXED_BACKREF_REV
);
1503 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
1505 write_extent_buffer(c
, root
->fs_info
->fsid
,
1506 (unsigned long)btrfs_header_fsid(c
),
1509 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
1510 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
1513 btrfs_set_node_key(c
, &lower_key
, 0);
1514 btrfs_set_node_blockptr(c
, 0, lower
->start
);
1515 lower_gen
= btrfs_header_generation(lower
);
1516 WARN_ON(lower_gen
!= trans
->transid
);
1518 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
1520 btrfs_mark_buffer_dirty(c
);
1525 /* the super has an extra ref to root->node */
1526 free_extent_buffer(old
);
1528 add_root_to_dirty_list(root
);
1529 extent_buffer_get(c
);
1530 path
->nodes
[level
] = c
;
1531 path
->slots
[level
] = 0;
1536 * worker function to insert a single pointer in a node.
1537 * the node should have enough room for the pointer already
1539 * slot and level indicate where you want the key to go, and
1540 * blocknr is the block the key points to.
1542 * returns zero on success and < 0 on any error
1544 static int insert_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
1545 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
1546 *key
, u64 bytenr
, int slot
, int level
)
1548 struct extent_buffer
*lower
;
1551 BUG_ON(!path
->nodes
[level
]);
1552 lower
= path
->nodes
[level
];
1553 nritems
= btrfs_header_nritems(lower
);
1556 if (nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
))
1558 if (slot
!= nritems
) {
1559 memmove_extent_buffer(lower
,
1560 btrfs_node_key_ptr_offset(slot
+ 1),
1561 btrfs_node_key_ptr_offset(slot
),
1562 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
1564 btrfs_set_node_key(lower
, key
, slot
);
1565 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
1566 WARN_ON(trans
->transid
== 0);
1567 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
1568 btrfs_set_header_nritems(lower
, nritems
+ 1);
1569 btrfs_mark_buffer_dirty(lower
);
1574 * split the node at the specified level in path in two.
1575 * The path is corrected to point to the appropriate node after the split
1577 * Before splitting this tries to make some room in the node by pushing
1578 * left and right, if either one works, it returns right away.
1580 * returns 0 on success and < 0 on failure
1582 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
1583 *root
, struct btrfs_path
*path
, int level
)
1585 struct extent_buffer
*c
;
1586 struct extent_buffer
*split
;
1587 struct btrfs_disk_key disk_key
;
1593 c
= path
->nodes
[level
];
1594 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
1595 if (c
== root
->node
) {
1596 /* trying to split the root, lets make a new one */
1597 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
1601 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
1602 c
= path
->nodes
[level
];
1603 if (!ret
&& btrfs_header_nritems(c
) <
1604 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
1610 c_nritems
= btrfs_header_nritems(c
);
1611 mid
= (c_nritems
+ 1) / 2;
1612 btrfs_node_key(c
, &disk_key
, mid
);
1614 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
1615 root
->root_key
.objectid
,
1616 &disk_key
, level
, c
->start
, 0);
1618 return PTR_ERR(split
);
1620 memset_extent_buffer(split
, 0, 0, sizeof(struct btrfs_header
));
1621 btrfs_set_header_level(split
, btrfs_header_level(c
));
1622 btrfs_set_header_bytenr(split
, split
->start
);
1623 btrfs_set_header_generation(split
, trans
->transid
);
1624 btrfs_set_header_backref_rev(split
, BTRFS_MIXED_BACKREF_REV
);
1625 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
1626 write_extent_buffer(split
, root
->fs_info
->fsid
,
1627 (unsigned long)btrfs_header_fsid(split
),
1629 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
1630 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
1634 copy_extent_buffer(split
, c
,
1635 btrfs_node_key_ptr_offset(0),
1636 btrfs_node_key_ptr_offset(mid
),
1637 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
1638 btrfs_set_header_nritems(split
, c_nritems
- mid
);
1639 btrfs_set_header_nritems(c
, mid
);
1642 btrfs_mark_buffer_dirty(c
);
1643 btrfs_mark_buffer_dirty(split
);
1645 wret
= insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
1646 path
->slots
[level
+ 1] + 1,
1651 if (path
->slots
[level
] >= mid
) {
1652 path
->slots
[level
] -= mid
;
1653 free_extent_buffer(c
);
1654 path
->nodes
[level
] = split
;
1655 path
->slots
[level
+ 1] += 1;
1657 free_extent_buffer(split
);
1663 * how many bytes are required to store the items in a leaf. start
1664 * and nr indicate which items in the leaf to check. This totals up the
1665 * space used both by the item structs and the item data
1667 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
1670 int nritems
= btrfs_header_nritems(l
);
1671 int end
= min(nritems
, start
+ nr
) - 1;
1675 data_len
= btrfs_item_end_nr(l
, start
);
1676 data_len
= data_len
- btrfs_item_offset_nr(l
, end
);
1677 data_len
+= sizeof(struct btrfs_item
) * nr
;
1678 WARN_ON(data_len
< 0);
1683 * The space between the end of the leaf items and
1684 * the start of the leaf data. IOW, how much room
1685 * the leaf has left for both items and data
1687 int btrfs_leaf_free_space(struct btrfs_root
*root
, struct extent_buffer
*leaf
)
1689 int nritems
= btrfs_header_nritems(leaf
);
1691 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
1693 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
1694 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
1695 leaf_space_used(leaf
, 0, nritems
), nritems
);
1701 * push some data in the path leaf to the right, trying to free up at
1702 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1704 * returns 1 if the push failed because the other node didn't have enough
1705 * room, 0 if everything worked out and < 0 if there were major errors.
1707 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
1708 *root
, struct btrfs_path
*path
, int data_size
,
1711 struct extent_buffer
*left
= path
->nodes
[0];
1712 struct extent_buffer
*right
;
1713 struct extent_buffer
*upper
;
1714 struct btrfs_disk_key disk_key
;
1720 struct btrfs_item
*item
;
1728 slot
= path
->slots
[1];
1729 if (!path
->nodes
[1]) {
1732 upper
= path
->nodes
[1];
1733 if (slot
>= btrfs_header_nritems(upper
) - 1)
1736 right
= read_node_slot(root
, upper
, slot
+ 1);
1737 free_space
= btrfs_leaf_free_space(root
, right
);
1738 if (free_space
< data_size
) {
1739 free_extent_buffer(right
);
1743 /* cow and double check */
1744 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
1747 free_extent_buffer(right
);
1750 free_space
= btrfs_leaf_free_space(root
, right
);
1751 if (free_space
< data_size
) {
1752 free_extent_buffer(right
);
1756 left_nritems
= btrfs_header_nritems(left
);
1757 if (left_nritems
== 0) {
1758 free_extent_buffer(right
);
1767 i
= left_nritems
- 1;
1769 item
= btrfs_item_nr(left
, i
);
1771 if (path
->slots
[0] == i
)
1772 push_space
+= data_size
+ sizeof(*item
);
1774 this_item_size
= btrfs_item_size(left
, item
);
1775 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
1778 push_space
+= this_item_size
+ sizeof(*item
);
1784 if (push_items
== 0) {
1785 free_extent_buffer(right
);
1789 if (!empty
&& push_items
== left_nritems
)
1792 /* push left to right */
1793 right_nritems
= btrfs_header_nritems(right
);
1795 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
1796 push_space
-= leaf_data_end(root
, left
);
1798 /* make room in the right data area */
1799 data_end
= leaf_data_end(root
, right
);
1800 memmove_extent_buffer(right
,
1801 btrfs_leaf_data(right
) + data_end
- push_space
,
1802 btrfs_leaf_data(right
) + data_end
,
1803 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
1805 /* copy from the left data area */
1806 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
1807 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
1808 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
1811 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
1812 btrfs_item_nr_offset(0),
1813 right_nritems
* sizeof(struct btrfs_item
));
1815 /* copy the items from left to right */
1816 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
1817 btrfs_item_nr_offset(left_nritems
- push_items
),
1818 push_items
* sizeof(struct btrfs_item
));
1820 /* update the item pointers */
1821 right_nritems
+= push_items
;
1822 btrfs_set_header_nritems(right
, right_nritems
);
1823 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
1824 for (i
= 0; i
< right_nritems
; i
++) {
1825 item
= btrfs_item_nr(right
, i
);
1826 push_space
-= btrfs_item_size(right
, item
);
1827 btrfs_set_item_offset(right
, item
, push_space
);
1830 left_nritems
-= push_items
;
1831 btrfs_set_header_nritems(left
, left_nritems
);
1834 btrfs_mark_buffer_dirty(left
);
1835 btrfs_mark_buffer_dirty(right
);
1837 btrfs_item_key(right
, &disk_key
, 0);
1838 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
1839 btrfs_mark_buffer_dirty(upper
);
1841 /* then fixup the leaf pointer in the path */
1842 if (path
->slots
[0] >= left_nritems
) {
1843 path
->slots
[0] -= left_nritems
;
1844 free_extent_buffer(path
->nodes
[0]);
1845 path
->nodes
[0] = right
;
1846 path
->slots
[1] += 1;
1848 free_extent_buffer(right
);
1853 * push some data in the path leaf to the left, trying to free up at
1854 * least data_size bytes. returns zero if the push worked, nonzero otherwise
1856 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
1857 *root
, struct btrfs_path
*path
, int data_size
,
1860 struct btrfs_disk_key disk_key
;
1861 struct extent_buffer
*right
= path
->nodes
[0];
1862 struct extent_buffer
*left
;
1868 struct btrfs_item
*item
;
1869 u32 old_left_nritems
;
1875 u32 old_left_item_size
;
1877 slot
= path
->slots
[1];
1880 if (!path
->nodes
[1])
1883 right_nritems
= btrfs_header_nritems(right
);
1884 if (right_nritems
== 0) {
1888 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
1889 free_space
= btrfs_leaf_free_space(root
, left
);
1890 if (free_space
< data_size
) {
1891 free_extent_buffer(left
);
1895 /* cow and double check */
1896 ret
= btrfs_cow_block(trans
, root
, left
,
1897 path
->nodes
[1], slot
- 1, &left
);
1899 /* we hit -ENOSPC, but it isn't fatal here */
1900 free_extent_buffer(left
);
1904 free_space
= btrfs_leaf_free_space(root
, left
);
1905 if (free_space
< data_size
) {
1906 free_extent_buffer(left
);
1913 nr
= right_nritems
- 1;
1915 for (i
= 0; i
< nr
; i
++) {
1916 item
= btrfs_item_nr(right
, i
);
1918 if (path
->slots
[0] == i
)
1919 push_space
+= data_size
+ sizeof(*item
);
1921 this_item_size
= btrfs_item_size(right
, item
);
1922 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
1926 push_space
+= this_item_size
+ sizeof(*item
);
1929 if (push_items
== 0) {
1930 free_extent_buffer(left
);
1933 if (!empty
&& push_items
== btrfs_header_nritems(right
))
1936 /* push data from right to left */
1937 copy_extent_buffer(left
, right
,
1938 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
1939 btrfs_item_nr_offset(0),
1940 push_items
* sizeof(struct btrfs_item
));
1942 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
1943 btrfs_item_offset_nr(right
, push_items
-1);
1945 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
1946 leaf_data_end(root
, left
) - push_space
,
1947 btrfs_leaf_data(right
) +
1948 btrfs_item_offset_nr(right
, push_items
- 1),
1950 old_left_nritems
= btrfs_header_nritems(left
);
1951 BUG_ON(old_left_nritems
< 0);
1953 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
1954 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
1957 item
= btrfs_item_nr(left
, i
);
1958 ioff
= btrfs_item_offset(left
, item
);
1959 btrfs_set_item_offset(left
, item
,
1960 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
));
1962 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
1964 /* fixup right node */
1965 if (push_items
> right_nritems
) {
1966 printk("push items %d nr %u\n", push_items
, right_nritems
);
1970 if (push_items
< right_nritems
) {
1971 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
1972 leaf_data_end(root
, right
);
1973 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
1974 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
1975 btrfs_leaf_data(right
) +
1976 leaf_data_end(root
, right
), push_space
);
1978 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
1979 btrfs_item_nr_offset(push_items
),
1980 (btrfs_header_nritems(right
) - push_items
) *
1981 sizeof(struct btrfs_item
));
1983 right_nritems
-= push_items
;
1984 btrfs_set_header_nritems(right
, right_nritems
);
1985 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
1986 for (i
= 0; i
< right_nritems
; i
++) {
1987 item
= btrfs_item_nr(right
, i
);
1988 push_space
= push_space
- btrfs_item_size(right
, item
);
1989 btrfs_set_item_offset(right
, item
, push_space
);
1992 btrfs_mark_buffer_dirty(left
);
1994 btrfs_mark_buffer_dirty(right
);
1996 btrfs_item_key(right
, &disk_key
, 0);
1997 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2001 /* then fixup the leaf pointer in the path */
2002 if (path
->slots
[0] < push_items
) {
2003 path
->slots
[0] += old_left_nritems
;
2004 free_extent_buffer(path
->nodes
[0]);
2005 path
->nodes
[0] = left
;
2006 path
->slots
[1] -= 1;
2008 free_extent_buffer(left
);
2009 path
->slots
[0] -= push_items
;
2011 BUG_ON(path
->slots
[0] < 0);
2016 * split the path's leaf in two, making sure there is at least data_size
2017 * available for the resulting leaf level of the path.
2019 * returns 0 if all went well and < 0 on failure.
2021 static noinline
int copy_for_split(struct btrfs_trans_handle
*trans
,
2022 struct btrfs_root
*root
,
2023 struct btrfs_path
*path
,
2024 struct extent_buffer
*l
,
2025 struct extent_buffer
*right
,
2026 int slot
, int mid
, int nritems
)
2033 struct btrfs_disk_key disk_key
;
2035 nritems
= nritems
- mid
;
2036 btrfs_set_header_nritems(right
, nritems
);
2037 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
2039 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
2040 btrfs_item_nr_offset(mid
),
2041 nritems
* sizeof(struct btrfs_item
));
2043 copy_extent_buffer(right
, l
,
2044 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
2045 data_copy_size
, btrfs_leaf_data(l
) +
2046 leaf_data_end(root
, l
), data_copy_size
);
2048 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
2049 btrfs_item_end_nr(l
, mid
);
2051 for (i
= 0; i
< nritems
; i
++) {
2052 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
2053 u32 ioff
= btrfs_item_offset(right
, item
);
2054 btrfs_set_item_offset(right
, item
, ioff
+ rt_data_off
);
2057 btrfs_set_header_nritems(l
, mid
);
2059 btrfs_item_key(right
, &disk_key
, 0);
2060 wret
= insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
2061 path
->slots
[1] + 1, 1);
2065 btrfs_mark_buffer_dirty(right
);
2066 btrfs_mark_buffer_dirty(l
);
2067 BUG_ON(path
->slots
[0] != slot
);
2070 free_extent_buffer(path
->nodes
[0]);
2071 path
->nodes
[0] = right
;
2072 path
->slots
[0] -= mid
;
2073 path
->slots
[1] += 1;
2075 free_extent_buffer(right
);
2078 BUG_ON(path
->slots
[0] < 0);
2084 * split the path's leaf in two, making sure there is at least data_size
2085 * available for the resulting leaf level of the path.
2087 * returns 0 if all went well and < 0 on failure.
2089 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
2090 struct btrfs_root
*root
,
2091 struct btrfs_key
*ins_key
,
2092 struct btrfs_path
*path
, int data_size
,
2095 struct btrfs_disk_key disk_key
;
2096 struct extent_buffer
*l
;
2100 struct extent_buffer
*right
;
2104 int num_doubles
= 0;
2106 /* first try to make some room by pushing left and right */
2107 if (data_size
&& ins_key
->type
!= BTRFS_DIR_ITEM_KEY
) {
2108 wret
= push_leaf_right(trans
, root
, path
, data_size
, 0);
2112 wret
= push_leaf_left(trans
, root
, path
, data_size
, 0);
2118 /* did the pushes work? */
2119 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
2123 if (!path
->nodes
[1]) {
2124 ret
= insert_new_root(trans
, root
, path
, 1);
2131 slot
= path
->slots
[0];
2132 nritems
= btrfs_header_nritems(l
);
2133 mid
= (nritems
+ 1) / 2;
2137 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
2138 BTRFS_LEAF_DATA_SIZE(root
)) {
2139 if (slot
>= nritems
) {
2143 if (mid
!= nritems
&&
2144 leaf_space_used(l
, mid
, nritems
- mid
) +
2145 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
2151 if (leaf_space_used(l
, 0, mid
) + data_size
>
2152 BTRFS_LEAF_DATA_SIZE(root
)) {
2153 if (!extend
&& data_size
&& slot
== 0) {
2155 } else if ((extend
|| !data_size
) && slot
== 0) {
2159 if (mid
!= nritems
&&
2160 leaf_space_used(l
, mid
, nritems
- mid
) +
2161 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
2169 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
2171 btrfs_item_key(l
, &disk_key
, mid
);
2173 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
2174 root
->root_key
.objectid
,
2175 &disk_key
, 0, l
->start
, 0);
2176 if (IS_ERR(right
)) {
2178 return PTR_ERR(right
);
2181 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
2182 btrfs_set_header_bytenr(right
, right
->start
);
2183 btrfs_set_header_generation(right
, trans
->transid
);
2184 btrfs_set_header_backref_rev(right
, BTRFS_MIXED_BACKREF_REV
);
2185 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
2186 btrfs_set_header_level(right
, 0);
2187 write_extent_buffer(right
, root
->fs_info
->fsid
,
2188 (unsigned long)btrfs_header_fsid(right
),
2191 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
2192 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
2197 btrfs_set_header_nritems(right
, 0);
2198 wret
= insert_ptr(trans
, root
, path
,
2199 &disk_key
, right
->start
,
2200 path
->slots
[1] + 1, 1);
2204 free_extent_buffer(path
->nodes
[0]);
2205 path
->nodes
[0] = right
;
2207 path
->slots
[1] += 1;
2209 btrfs_set_header_nritems(right
, 0);
2210 wret
= insert_ptr(trans
, root
, path
,
2216 free_extent_buffer(path
->nodes
[0]);
2217 path
->nodes
[0] = right
;
2219 if (path
->slots
[1] == 0) {
2220 wret
= fixup_low_keys(trans
, root
,
2221 path
, &disk_key
, 1);
2226 btrfs_mark_buffer_dirty(right
);
2230 ret
= copy_for_split(trans
, root
, path
, l
, right
, slot
, mid
, nritems
);
2234 BUG_ON(num_doubles
!= 0);
2243 * This function splits a single item into two items,
2244 * giving 'new_key' to the new item and splitting the
2245 * old one at split_offset (from the start of the item).
2247 * The path may be released by this operation. After
2248 * the split, the path is pointing to the old item. The
2249 * new item is going to be in the same node as the old one.
2251 * Note, the item being split must be smaller enough to live alone on
2252 * a tree block with room for one extra struct btrfs_item
2254 * This allows us to split the item in place, keeping a lock on the
2255 * leaf the entire time.
2257 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
2258 struct btrfs_root
*root
,
2259 struct btrfs_path
*path
,
2260 struct btrfs_key
*new_key
,
2261 unsigned long split_offset
)
2264 struct extent_buffer
*leaf
;
2265 struct btrfs_key orig_key
;
2266 struct btrfs_item
*item
;
2267 struct btrfs_item
*new_item
;
2272 struct btrfs_disk_key disk_key
;
2275 leaf
= path
->nodes
[0];
2276 btrfs_item_key_to_cpu(leaf
, &orig_key
, path
->slots
[0]);
2277 if (btrfs_leaf_free_space(root
, leaf
) >= sizeof(struct btrfs_item
))
2280 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2281 btrfs_release_path(root
, path
);
2283 path
->search_for_split
= 1;
2285 ret
= btrfs_search_slot(trans
, root
, &orig_key
, path
, 0, 1);
2286 path
->search_for_split
= 0;
2288 /* if our item isn't there or got smaller, return now */
2289 if (ret
!= 0 || item_size
!= btrfs_item_size_nr(path
->nodes
[0],
2294 ret
= split_leaf(trans
, root
, &orig_key
, path
, 0, 0);
2297 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
2298 leaf
= path
->nodes
[0];
2301 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
2302 orig_offset
= btrfs_item_offset(leaf
, item
);
2303 item_size
= btrfs_item_size(leaf
, item
);
2306 buf
= kmalloc(item_size
, GFP_NOFS
);
2307 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
2308 path
->slots
[0]), item_size
);
2309 slot
= path
->slots
[0] + 1;
2310 leaf
= path
->nodes
[0];
2312 nritems
= btrfs_header_nritems(leaf
);
2314 if (slot
!= nritems
) {
2315 /* shift the items */
2316 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
2317 btrfs_item_nr_offset(slot
),
2318 (nritems
- slot
) * sizeof(struct btrfs_item
));
2322 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2323 btrfs_set_item_key(leaf
, &disk_key
, slot
);
2325 new_item
= btrfs_item_nr(leaf
, slot
);
2327 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
2328 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
2330 btrfs_set_item_offset(leaf
, item
,
2331 orig_offset
+ item_size
- split_offset
);
2332 btrfs_set_item_size(leaf
, item
, split_offset
);
2334 btrfs_set_header_nritems(leaf
, nritems
+ 1);
2336 /* write the data for the start of the original item */
2337 write_extent_buffer(leaf
, buf
,
2338 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
2341 /* write the data for the new item */
2342 write_extent_buffer(leaf
, buf
+ split_offset
,
2343 btrfs_item_ptr_offset(leaf
, slot
),
2344 item_size
- split_offset
);
2345 btrfs_mark_buffer_dirty(leaf
);
2348 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2349 btrfs_print_leaf(root
, leaf
);
2356 int btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
2357 struct btrfs_root
*root
,
2358 struct btrfs_path
*path
,
2359 u32 new_size
, int from_end
)
2363 struct extent_buffer
*leaf
;
2364 struct btrfs_item
*item
;
2366 unsigned int data_end
;
2367 unsigned int old_data_start
;
2368 unsigned int old_size
;
2369 unsigned int size_diff
;
2372 leaf
= path
->nodes
[0];
2373 slot
= path
->slots
[0];
2375 old_size
= btrfs_item_size_nr(leaf
, slot
);
2376 if (old_size
== new_size
)
2379 nritems
= btrfs_header_nritems(leaf
);
2380 data_end
= leaf_data_end(root
, leaf
);
2382 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
2384 size_diff
= old_size
- new_size
;
2387 BUG_ON(slot
>= nritems
);
2390 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2392 /* first correct the data pointers */
2393 for (i
= slot
; i
< nritems
; i
++) {
2395 item
= btrfs_item_nr(leaf
, i
);
2396 ioff
= btrfs_item_offset(leaf
, item
);
2397 btrfs_set_item_offset(leaf
, item
, ioff
+ size_diff
);
2400 /* shift the data */
2402 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2403 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
2404 data_end
, old_data_start
+ new_size
- data_end
);
2406 struct btrfs_disk_key disk_key
;
2409 btrfs_item_key(leaf
, &disk_key
, slot
);
2411 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
2413 struct btrfs_file_extent_item
*fi
;
2415 fi
= btrfs_item_ptr(leaf
, slot
,
2416 struct btrfs_file_extent_item
);
2417 fi
= (struct btrfs_file_extent_item
*)(
2418 (unsigned long)fi
- size_diff
);
2420 if (btrfs_file_extent_type(leaf
, fi
) ==
2421 BTRFS_FILE_EXTENT_INLINE
) {
2422 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
2423 memmove_extent_buffer(leaf
, ptr
,
2425 offsetof(struct btrfs_file_extent_item
,
2430 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2431 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
2432 data_end
, old_data_start
- data_end
);
2434 offset
= btrfs_disk_key_offset(&disk_key
);
2435 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
2436 btrfs_set_item_key(leaf
, &disk_key
, slot
);
2438 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2441 item
= btrfs_item_nr(leaf
, slot
);
2442 btrfs_set_item_size(leaf
, item
, new_size
);
2443 btrfs_mark_buffer_dirty(leaf
);
2446 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2447 btrfs_print_leaf(root
, leaf
);
2453 int btrfs_extend_item(struct btrfs_trans_handle
*trans
,
2454 struct btrfs_root
*root
, struct btrfs_path
*path
,
2459 struct extent_buffer
*leaf
;
2460 struct btrfs_item
*item
;
2462 unsigned int data_end
;
2463 unsigned int old_data
;
2464 unsigned int old_size
;
2467 leaf
= path
->nodes
[0];
2469 nritems
= btrfs_header_nritems(leaf
);
2470 data_end
= leaf_data_end(root
, leaf
);
2472 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
2473 btrfs_print_leaf(root
, leaf
);
2476 slot
= path
->slots
[0];
2477 old_data
= btrfs_item_end_nr(leaf
, slot
);
2480 if (slot
>= nritems
) {
2481 btrfs_print_leaf(root
, leaf
);
2482 printk("slot %d too large, nritems %d\n", slot
, nritems
);
2487 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2489 /* first correct the data pointers */
2490 for (i
= slot
; i
< nritems
; i
++) {
2492 item
= btrfs_item_nr(leaf
, i
);
2493 ioff
= btrfs_item_offset(leaf
, item
);
2494 btrfs_set_item_offset(leaf
, item
, ioff
- data_size
);
2497 /* shift the data */
2498 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2499 data_end
- data_size
, btrfs_leaf_data(leaf
) +
2500 data_end
, old_data
- data_end
);
2502 data_end
= old_data
;
2503 old_size
= btrfs_item_size_nr(leaf
, slot
);
2504 item
= btrfs_item_nr(leaf
, slot
);
2505 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
2506 btrfs_mark_buffer_dirty(leaf
);
2509 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2510 btrfs_print_leaf(root
, leaf
);
2517 * Given a key and some data, insert an item into the tree.
2518 * This does all the path init required, making room in the tree if needed.
2520 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
2521 struct btrfs_root
*root
,
2522 struct btrfs_path
*path
,
2523 struct btrfs_key
*cpu_key
, u32
*data_size
,
2526 struct extent_buffer
*leaf
;
2527 struct btrfs_item
*item
;
2534 unsigned int data_end
;
2535 struct btrfs_disk_key disk_key
;
2537 for (i
= 0; i
< nr
; i
++) {
2538 total_data
+= data_size
[i
];
2541 /* create a root if there isn't one */
2545 total_size
= total_data
+ nr
* sizeof(struct btrfs_item
);
2546 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
2553 leaf
= path
->nodes
[0];
2555 nritems
= btrfs_header_nritems(leaf
);
2556 data_end
= leaf_data_end(root
, leaf
);
2558 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
2559 btrfs_print_leaf(root
, leaf
);
2560 printk("not enough freespace need %u have %d\n",
2561 total_size
, btrfs_leaf_free_space(root
, leaf
));
2565 slot
= path
->slots
[0];
2568 if (slot
!= nritems
) {
2570 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
2572 if (old_data
< data_end
) {
2573 btrfs_print_leaf(root
, leaf
);
2574 printk("slot %d old_data %d data_end %d\n",
2575 slot
, old_data
, data_end
);
2579 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2581 /* first correct the data pointers */
2582 for (i
= slot
; i
< nritems
; i
++) {
2585 item
= btrfs_item_nr(leaf
, i
);
2586 ioff
= btrfs_item_offset(leaf
, item
);
2587 btrfs_set_item_offset(leaf
, item
, ioff
- total_data
);
2590 /* shift the items */
2591 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
2592 btrfs_item_nr_offset(slot
),
2593 (nritems
- slot
) * sizeof(struct btrfs_item
));
2595 /* shift the data */
2596 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2597 data_end
- total_data
, btrfs_leaf_data(leaf
) +
2598 data_end
, old_data
- data_end
);
2599 data_end
= old_data
;
2602 /* setup the item for the new data */
2603 for (i
= 0; i
< nr
; i
++) {
2604 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
2605 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
2606 item
= btrfs_item_nr(leaf
, slot
+ i
);
2607 btrfs_set_item_offset(leaf
, item
, data_end
- data_size
[i
]);
2608 data_end
-= data_size
[i
];
2609 btrfs_set_item_size(leaf
, item
, data_size
[i
]);
2611 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
2612 btrfs_mark_buffer_dirty(leaf
);
2616 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
2617 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2620 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2621 btrfs_print_leaf(root
, leaf
);
2630 * Given a key and some data, insert an item into the tree.
2631 * This does all the path init required, making room in the tree if needed.
2633 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
2634 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
2638 struct btrfs_path
*path
;
2639 struct extent_buffer
*leaf
;
2642 path
= btrfs_alloc_path();
2644 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
2646 leaf
= path
->nodes
[0];
2647 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
2648 write_extent_buffer(leaf
, data
, ptr
, data_size
);
2649 btrfs_mark_buffer_dirty(leaf
);
2651 btrfs_free_path(path
);
2656 * delete the pointer from a given node.
2658 * If the delete empties a node, the node is removed from the tree,
2659 * continuing all the way the root if required. The root is converted into
2660 * a leaf if all the nodes are emptied.
2662 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2663 struct btrfs_path
*path
, int level
, int slot
)
2665 struct extent_buffer
*parent
= path
->nodes
[level
];
2670 nritems
= btrfs_header_nritems(parent
);
2671 if (slot
!= nritems
-1) {
2672 memmove_extent_buffer(parent
,
2673 btrfs_node_key_ptr_offset(slot
),
2674 btrfs_node_key_ptr_offset(slot
+ 1),
2675 sizeof(struct btrfs_key_ptr
) *
2676 (nritems
- slot
- 1));
2679 btrfs_set_header_nritems(parent
, nritems
);
2680 if (nritems
== 0 && parent
== root
->node
) {
2681 BUG_ON(btrfs_header_level(root
->node
) != 1);
2682 /* just turn the root into a leaf and break */
2683 btrfs_set_header_level(root
->node
, 0);
2684 } else if (slot
== 0) {
2685 struct btrfs_disk_key disk_key
;
2687 btrfs_node_key(parent
, &disk_key
, 0);
2688 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
2692 btrfs_mark_buffer_dirty(parent
);
2697 * a helper function to delete the leaf pointed to by path->slots[1] and
2700 * This deletes the pointer in path->nodes[1] and frees the leaf
2701 * block extent. zero is returned if it all worked out, < 0 otherwise.
2703 * The path must have already been setup for deleting the leaf, including
2704 * all the proper balancing. path->nodes[1] must be locked.
2706 static noinline
int btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
2707 struct btrfs_root
*root
,
2708 struct btrfs_path
*path
,
2709 struct extent_buffer
*leaf
)
2713 WARN_ON(btrfs_header_generation(leaf
) != trans
->transid
);
2714 ret
= del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
2718 ret
= btrfs_free_extent(trans
, root
, leaf
->start
, leaf
->len
,
2719 0, root
->root_key
.objectid
, 0, 0);
2724 * delete the item at the leaf level in path. If that empties
2725 * the leaf, remove it from the tree
2727 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
2728 struct btrfs_path
*path
, int slot
, int nr
)
2730 struct extent_buffer
*leaf
;
2731 struct btrfs_item
*item
;
2739 leaf
= path
->nodes
[0];
2740 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
2742 for (i
= 0; i
< nr
; i
++)
2743 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
2745 nritems
= btrfs_header_nritems(leaf
);
2747 if (slot
+ nr
!= nritems
) {
2749 int data_end
= leaf_data_end(root
, leaf
);
2751 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
2753 btrfs_leaf_data(leaf
) + data_end
,
2754 last_off
- data_end
);
2756 for (i
= slot
+ nr
; i
< nritems
; i
++) {
2759 item
= btrfs_item_nr(leaf
, i
);
2760 ioff
= btrfs_item_offset(leaf
, item
);
2761 btrfs_set_item_offset(leaf
, item
, ioff
+ dsize
);
2764 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
2765 btrfs_item_nr_offset(slot
+ nr
),
2766 sizeof(struct btrfs_item
) *
2767 (nritems
- slot
- nr
));
2769 btrfs_set_header_nritems(leaf
, nritems
- nr
);
2772 /* delete the leaf if we've emptied it */
2774 if (leaf
== root
->node
) {
2775 btrfs_set_header_level(leaf
, 0);
2777 clean_tree_block(trans
, root
, leaf
);
2778 wait_on_tree_block_writeback(root
, leaf
);
2780 wret
= btrfs_del_leaf(trans
, root
, path
, leaf
);
2786 int used
= leaf_space_used(leaf
, 0, nritems
);
2788 struct btrfs_disk_key disk_key
;
2790 btrfs_item_key(leaf
, &disk_key
, 0);
2791 wret
= fixup_low_keys(trans
, root
, path
,
2797 /* delete the leaf if it is mostly empty */
2798 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 4) {
2799 /* push_leaf_left fixes the path.
2800 * make sure the path still points to our leaf
2801 * for possible call to del_ptr below
2803 slot
= path
->slots
[1];
2804 extent_buffer_get(leaf
);
2806 wret
= push_leaf_left(trans
, root
, path
, 1, 1);
2807 if (wret
< 0 && wret
!= -ENOSPC
)
2810 if (path
->nodes
[0] == leaf
&&
2811 btrfs_header_nritems(leaf
)) {
2812 wret
= push_leaf_right(trans
, root
, path
, 1, 1);
2813 if (wret
< 0 && wret
!= -ENOSPC
)
2817 if (btrfs_header_nritems(leaf
) == 0) {
2818 clean_tree_block(trans
, root
, leaf
);
2819 wait_on_tree_block_writeback(root
, leaf
);
2821 path
->slots
[1] = slot
;
2822 ret
= btrfs_del_leaf(trans
, root
, path
, leaf
);
2824 free_extent_buffer(leaf
);
2827 btrfs_mark_buffer_dirty(leaf
);
2828 free_extent_buffer(leaf
);
2831 btrfs_mark_buffer_dirty(leaf
);
2838 * walk up the tree as far as required to find the previous leaf.
2839 * returns 0 if it found something or 1 if there are no lesser leaves.
2840 * returns < 0 on io errors.
2842 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
2846 struct extent_buffer
*c
;
2847 struct extent_buffer
*next
= NULL
;
2849 while(level
< BTRFS_MAX_LEVEL
) {
2850 if (!path
->nodes
[level
])
2853 slot
= path
->slots
[level
];
2854 c
= path
->nodes
[level
];
2857 if (level
== BTRFS_MAX_LEVEL
)
2864 free_extent_buffer(next
);
2866 next
= read_node_slot(root
, c
, slot
);
2869 path
->slots
[level
] = slot
;
2872 c
= path
->nodes
[level
];
2873 free_extent_buffer(c
);
2874 slot
= btrfs_header_nritems(next
);
2877 path
->nodes
[level
] = next
;
2878 path
->slots
[level
] = slot
;
2881 next
= read_node_slot(root
, next
, slot
);
2887 * walk up the tree as far as required to find the next leaf.
2888 * returns 0 if it found something or 1 if there are no greater leaves.
2889 * returns < 0 on io errors.
2891 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
2895 struct extent_buffer
*c
;
2896 struct extent_buffer
*next
= NULL
;
2898 while(level
< BTRFS_MAX_LEVEL
) {
2899 if (!path
->nodes
[level
])
2902 slot
= path
->slots
[level
] + 1;
2903 c
= path
->nodes
[level
];
2904 if (slot
>= btrfs_header_nritems(c
)) {
2906 if (level
== BTRFS_MAX_LEVEL
)
2912 free_extent_buffer(next
);
2915 reada_for_search(root
, path
, level
, slot
, 0);
2917 next
= read_node_slot(root
, c
, slot
);
2920 path
->slots
[level
] = slot
;
2923 c
= path
->nodes
[level
];
2924 free_extent_buffer(c
);
2925 path
->nodes
[level
] = next
;
2926 path
->slots
[level
] = 0;
2930 reada_for_search(root
, path
, level
, 0, 0);
2931 next
= read_node_slot(root
, next
, 0);
2936 int btrfs_previous_item(struct btrfs_root
*root
,
2937 struct btrfs_path
*path
, u64 min_objectid
,
2940 struct btrfs_key found_key
;
2941 struct extent_buffer
*leaf
;
2945 if (path
->slots
[0] == 0) {
2946 ret
= btrfs_prev_leaf(root
, path
);
2952 leaf
= path
->nodes
[0];
2953 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
2954 if (found_key
.type
== type
)