2 * Copyright (C) 2007,2008 Oracle. All rights reserved.
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of the GNU General Public
6 * License v2 as published by the Free Software Foundation.
8 * This program is distributed in the hope that it will be useful,
9 * but WITHOUT ANY WARRANTY; without even the implied warranty of
10 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
11 * General Public License for more details.
13 * You should have received a copy of the GNU General Public
14 * License along with this program; if not, write to the
15 * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
16 * Boston, MA 021110-1307, USA.
19 #include <linux/sched.h>
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 /* this also releases the path */
58 void btrfs_free_path(struct btrfs_path
*p
)
60 btrfs_release_path(NULL
, p
);
61 kmem_cache_free(btrfs_path_cachep
, p
);
65 * path release drops references on the extent buffers in the path
66 * and it drops any locks held by this path
68 * It is safe to call this on paths that no locks or extent buffers held.
70 void noinline
btrfs_release_path(struct btrfs_root
*root
, struct btrfs_path
*p
)
74 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
79 btrfs_tree_unlock(p
->nodes
[i
]);
82 free_extent_buffer(p
->nodes
[i
]);
88 * safely gets a reference on the root node of a tree. A lock
89 * is not taken, so a concurrent writer may put a different node
90 * at the root of the tree. See btrfs_lock_root_node for the
93 * The extent buffer returned by this has a reference taken, so
94 * it won't disappear. It may stop being the root of the tree
95 * at any time because there are no locks held.
97 struct extent_buffer
*btrfs_root_node(struct btrfs_root
*root
)
99 struct extent_buffer
*eb
;
100 spin_lock(&root
->node_lock
);
102 extent_buffer_get(eb
);
103 spin_unlock(&root
->node_lock
);
107 /* loop around taking references on and locking the root node of the
108 * tree until you end up with a lock on the root. A locked buffer
109 * is returned, with a reference held.
111 struct extent_buffer
*btrfs_lock_root_node(struct btrfs_root
*root
)
113 struct extent_buffer
*eb
;
116 eb
= btrfs_root_node(root
);
119 spin_lock(&root
->node_lock
);
120 if (eb
== root
->node
) {
121 spin_unlock(&root
->node_lock
);
124 spin_unlock(&root
->node_lock
);
126 btrfs_tree_unlock(eb
);
127 free_extent_buffer(eb
);
132 /* cowonly root (everything not a reference counted cow subvolume), just get
133 * put onto a simple dirty list. transaction.c walks this to make sure they
134 * get properly updated on disk.
136 static void add_root_to_dirty_list(struct btrfs_root
*root
)
138 if (root
->track_dirty
&& list_empty(&root
->dirty_list
)) {
139 list_add(&root
->dirty_list
,
140 &root
->fs_info
->dirty_cowonly_roots
);
145 * used by snapshot creation to make a copy of a root for a tree with
146 * a given objectid. The buffer with the new root node is returned in
147 * cow_ret, and this func returns zero on success or a negative error code.
149 int btrfs_copy_root(struct btrfs_trans_handle
*trans
,
150 struct btrfs_root
*root
,
151 struct extent_buffer
*buf
,
152 struct extent_buffer
**cow_ret
, u64 new_root_objectid
)
154 struct extent_buffer
*cow
;
158 struct btrfs_root
*new_root
;
160 new_root
= kmalloc(sizeof(*new_root
), GFP_NOFS
);
164 memcpy(new_root
, root
, sizeof(*new_root
));
165 new_root
->root_key
.objectid
= new_root_objectid
;
167 WARN_ON(root
->ref_cows
&& trans
->transid
!=
168 root
->fs_info
->running_transaction
->transid
);
169 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
171 level
= btrfs_header_level(buf
);
172 nritems
= btrfs_header_nritems(buf
);
174 cow
= btrfs_alloc_free_block(trans
, new_root
, buf
->len
, 0,
175 new_root_objectid
, trans
->transid
,
176 level
, buf
->start
, 0);
182 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
183 btrfs_set_header_bytenr(cow
, cow
->start
);
184 btrfs_set_header_generation(cow
, trans
->transid
);
185 btrfs_set_header_owner(cow
, new_root_objectid
);
186 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
);
188 write_extent_buffer(cow
, root
->fs_info
->fsid
,
189 (unsigned long)btrfs_header_fsid(cow
),
192 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
193 ret
= btrfs_inc_ref(trans
, new_root
, buf
, cow
, NULL
);
199 btrfs_mark_buffer_dirty(cow
);
205 * does the dirty work in cow of a single block. The parent block
206 * (if supplied) is updated to point to the new cow copy. The new
207 * buffer is marked dirty and returned locked. If you modify the block
208 * it needs to be marked dirty again.
210 * search_start -- an allocation hint for the new block
212 * empty_size -- a hint that you plan on doing more cow. This is the size in bytes
213 * the allocator should try to find free next to the block it returns. This is
214 * just a hint and may be ignored by the allocator.
216 * prealloc_dest -- if you have already reserved a destination for the cow,
217 * this uses that block instead of allocating a new one. btrfs_alloc_reserved_extent
218 * is used to finish the allocation.
220 static int noinline
__btrfs_cow_block(struct btrfs_trans_handle
*trans
,
221 struct btrfs_root
*root
,
222 struct extent_buffer
*buf
,
223 struct extent_buffer
*parent
, int parent_slot
,
224 struct extent_buffer
**cow_ret
,
225 u64 search_start
, u64 empty_size
,
229 struct extent_buffer
*cow
;
238 WARN_ON(!btrfs_tree_locked(buf
));
241 parent_start
= parent
->start
;
245 WARN_ON(root
->ref_cows
&& trans
->transid
!=
246 root
->fs_info
->running_transaction
->transid
);
247 WARN_ON(root
->ref_cows
&& trans
->transid
!= root
->last_trans
);
249 level
= btrfs_header_level(buf
);
250 nritems
= btrfs_header_nritems(buf
);
253 struct btrfs_key ins
;
255 ins
.objectid
= prealloc_dest
;
256 ins
.offset
= buf
->len
;
257 ins
.type
= BTRFS_EXTENT_ITEM_KEY
;
259 ret
= btrfs_alloc_reserved_extent(trans
, root
, parent_start
,
260 root
->root_key
.objectid
,
261 trans
->transid
, level
, &ins
);
263 cow
= btrfs_init_new_buffer(trans
, root
, prealloc_dest
,
266 cow
= btrfs_alloc_free_block(trans
, root
, buf
->len
,
268 root
->root_key
.objectid
,
269 trans
->transid
, level
,
270 search_start
, empty_size
);
275 copy_extent_buffer(cow
, buf
, 0, 0, cow
->len
);
276 btrfs_set_header_bytenr(cow
, cow
->start
);
277 btrfs_set_header_generation(cow
, trans
->transid
);
278 btrfs_set_header_owner(cow
, root
->root_key
.objectid
);
279 btrfs_clear_header_flag(cow
, BTRFS_HEADER_FLAG_WRITTEN
);
281 write_extent_buffer(cow
, root
->fs_info
->fsid
,
282 (unsigned long)btrfs_header_fsid(cow
),
285 WARN_ON(btrfs_header_generation(buf
) > trans
->transid
);
286 if (btrfs_header_generation(buf
) != trans
->transid
) {
288 ret
= btrfs_inc_ref(trans
, root
, buf
, cow
, &nr_extents
);
292 ret
= btrfs_cache_ref(trans
, root
, buf
, nr_extents
);
294 } else if (btrfs_header_owner(buf
) == BTRFS_TREE_RELOC_OBJECTID
) {
296 * There are only two places that can drop reference to
297 * tree blocks owned by living reloc trees, one is here,
298 * the other place is btrfs_drop_subtree. In both places,
299 * we check reference count while tree block is locked.
300 * Furthermore, if reference count is one, it won't get
301 * increased by someone else.
304 ret
= btrfs_lookup_extent_ref(trans
, root
, buf
->start
,
308 ret
= btrfs_update_ref(trans
, root
, buf
, cow
,
310 clean_tree_block(trans
, root
, buf
);
312 ret
= btrfs_inc_ref(trans
, root
, buf
, cow
, NULL
);
316 ret
= btrfs_update_ref(trans
, root
, buf
, cow
, 0, nritems
);
319 clean_tree_block(trans
, root
, buf
);
322 if (root
->root_key
.objectid
== BTRFS_TREE_RELOC_OBJECTID
) {
323 ret
= btrfs_reloc_tree_cache_ref(trans
, root
, cow
, buf
->start
);
327 if (buf
== root
->node
) {
328 WARN_ON(parent
&& parent
!= buf
);
330 spin_lock(&root
->node_lock
);
332 extent_buffer_get(cow
);
333 spin_unlock(&root
->node_lock
);
335 if (buf
!= root
->commit_root
) {
336 btrfs_free_extent(trans
, root
, buf
->start
,
337 buf
->len
, buf
->start
,
338 root
->root_key
.objectid
,
339 btrfs_header_generation(buf
),
342 free_extent_buffer(buf
);
343 add_root_to_dirty_list(root
);
345 btrfs_set_node_blockptr(parent
, parent_slot
,
347 WARN_ON(trans
->transid
== 0);
348 btrfs_set_node_ptr_generation(parent
, parent_slot
,
350 btrfs_mark_buffer_dirty(parent
);
351 WARN_ON(btrfs_header_generation(parent
) != trans
->transid
);
352 btrfs_free_extent(trans
, root
, buf
->start
, buf
->len
,
353 parent_start
, btrfs_header_owner(parent
),
354 btrfs_header_generation(parent
), level
, 1);
357 btrfs_tree_unlock(buf
);
358 free_extent_buffer(buf
);
359 btrfs_mark_buffer_dirty(cow
);
365 * cows a single block, see __btrfs_cow_block for the real work.
366 * This version of it has extra checks so that a block isn't cow'd more than
367 * once per transaction, as long as it hasn't been written yet
369 int noinline
btrfs_cow_block(struct btrfs_trans_handle
*trans
,
370 struct btrfs_root
*root
, struct extent_buffer
*buf
,
371 struct extent_buffer
*parent
, int parent_slot
,
372 struct extent_buffer
**cow_ret
, u64 prealloc_dest
)
377 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
378 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
379 root
->fs_info
->running_transaction
->transid
);
382 if (trans
->transid
!= root
->fs_info
->generation
) {
383 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
384 root
->fs_info
->generation
);
388 spin_lock(&root
->fs_info
->hash_lock
);
389 if (btrfs_header_generation(buf
) == trans
->transid
&&
390 btrfs_header_owner(buf
) == root
->root_key
.objectid
&&
391 !btrfs_header_flag(buf
, BTRFS_HEADER_FLAG_WRITTEN
)) {
393 spin_unlock(&root
->fs_info
->hash_lock
);
394 WARN_ON(prealloc_dest
);
397 spin_unlock(&root
->fs_info
->hash_lock
);
398 search_start
= buf
->start
& ~((u64
)(1024 * 1024 * 1024) - 1);
399 ret
= __btrfs_cow_block(trans
, root
, buf
, parent
,
400 parent_slot
, cow_ret
, search_start
, 0,
406 * helper function for defrag to decide if two blocks pointed to by a
407 * node are actually close by
409 static int close_blocks(u64 blocknr
, u64 other
, u32 blocksize
)
411 if (blocknr
< other
&& other
- (blocknr
+ blocksize
) < 32768)
413 if (blocknr
> other
&& blocknr
- (other
+ blocksize
) < 32768)
419 * compare two keys in a memcmp fashion
421 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
425 btrfs_disk_key_to_cpu(&k1
, disk
);
427 if (k1
.objectid
> k2
->objectid
)
429 if (k1
.objectid
< k2
->objectid
)
431 if (k1
.type
> k2
->type
)
433 if (k1
.type
< k2
->type
)
435 if (k1
.offset
> k2
->offset
)
437 if (k1
.offset
< k2
->offset
)
443 * same as comp_keys only with two btrfs_key's
445 static int comp_cpu_keys(struct btrfs_key
*k1
, struct btrfs_key
*k2
)
447 if (k1
->objectid
> k2
->objectid
)
449 if (k1
->objectid
< k2
->objectid
)
451 if (k1
->type
> k2
->type
)
453 if (k1
->type
< k2
->type
)
455 if (k1
->offset
> k2
->offset
)
457 if (k1
->offset
< k2
->offset
)
463 * this is used by the defrag code to go through all the
464 * leaves pointed to by a node and reallocate them so that
465 * disk order is close to key order
467 int btrfs_realloc_node(struct btrfs_trans_handle
*trans
,
468 struct btrfs_root
*root
, struct extent_buffer
*parent
,
469 int start_slot
, int cache_only
, u64
*last_ret
,
470 struct btrfs_key
*progress
)
472 struct extent_buffer
*cur
;
475 u64 search_start
= *last_ret
;
485 int progress_passed
= 0;
486 struct btrfs_disk_key disk_key
;
488 parent_level
= btrfs_header_level(parent
);
489 if (cache_only
&& parent_level
!= 1)
492 if (trans
->transaction
!= root
->fs_info
->running_transaction
) {
493 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
494 root
->fs_info
->running_transaction
->transid
);
497 if (trans
->transid
!= root
->fs_info
->generation
) {
498 printk(KERN_CRIT
"trans %Lu running %Lu\n", trans
->transid
,
499 root
->fs_info
->generation
);
503 parent_nritems
= btrfs_header_nritems(parent
);
504 blocksize
= btrfs_level_size(root
, parent_level
- 1);
505 end_slot
= parent_nritems
;
507 if (parent_nritems
== 1)
510 for (i
= start_slot
; i
< end_slot
; i
++) {
513 if (!parent
->map_token
) {
514 map_extent_buffer(parent
,
515 btrfs_node_key_ptr_offset(i
),
516 sizeof(struct btrfs_key_ptr
),
517 &parent
->map_token
, &parent
->kaddr
,
518 &parent
->map_start
, &parent
->map_len
,
521 btrfs_node_key(parent
, &disk_key
, i
);
522 if (!progress_passed
&& comp_keys(&disk_key
, progress
) < 0)
526 blocknr
= btrfs_node_blockptr(parent
, i
);
527 gen
= btrfs_node_ptr_generation(parent
, i
);
529 last_block
= blocknr
;
532 other
= btrfs_node_blockptr(parent
, i
- 1);
533 close
= close_blocks(blocknr
, other
, blocksize
);
535 if (!close
&& i
< end_slot
- 2) {
536 other
= btrfs_node_blockptr(parent
, i
+ 1);
537 close
= close_blocks(blocknr
, other
, blocksize
);
540 last_block
= blocknr
;
543 if (parent
->map_token
) {
544 unmap_extent_buffer(parent
, parent
->map_token
,
546 parent
->map_token
= NULL
;
549 cur
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
551 uptodate
= btrfs_buffer_uptodate(cur
, gen
);
554 if (!cur
|| !uptodate
) {
556 free_extent_buffer(cur
);
560 cur
= read_tree_block(root
, blocknr
,
562 } else if (!uptodate
) {
563 btrfs_read_buffer(cur
, gen
);
566 if (search_start
== 0)
567 search_start
= last_block
;
569 btrfs_tree_lock(cur
);
570 err
= __btrfs_cow_block(trans
, root
, cur
, parent
, i
,
573 (end_slot
- i
) * blocksize
), 0);
575 btrfs_tree_unlock(cur
);
576 free_extent_buffer(cur
);
579 search_start
= cur
->start
;
580 last_block
= cur
->start
;
581 *last_ret
= search_start
;
582 btrfs_tree_unlock(cur
);
583 free_extent_buffer(cur
);
585 if (parent
->map_token
) {
586 unmap_extent_buffer(parent
, parent
->map_token
,
588 parent
->map_token
= NULL
;
594 * The leaf data grows from end-to-front in the node.
595 * this returns the address of the start of the last item,
596 * which is the stop of the leaf data stack
598 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
599 struct extent_buffer
*leaf
)
601 u32 nr
= btrfs_header_nritems(leaf
);
603 return BTRFS_LEAF_DATA_SIZE(root
);
604 return btrfs_item_offset_nr(leaf
, nr
- 1);
608 * extra debugging checks to make sure all the items in a key are
609 * well formed and in the proper order
611 static int check_node(struct btrfs_root
*root
, struct btrfs_path
*path
,
614 struct extent_buffer
*parent
= NULL
;
615 struct extent_buffer
*node
= path
->nodes
[level
];
616 struct btrfs_disk_key parent_key
;
617 struct btrfs_disk_key node_key
;
620 struct btrfs_key cpukey
;
621 u32 nritems
= btrfs_header_nritems(node
);
623 if (path
->nodes
[level
+ 1])
624 parent
= path
->nodes
[level
+ 1];
626 slot
= path
->slots
[level
];
627 BUG_ON(nritems
== 0);
629 parent_slot
= path
->slots
[level
+ 1];
630 btrfs_node_key(parent
, &parent_key
, parent_slot
);
631 btrfs_node_key(node
, &node_key
, 0);
632 BUG_ON(memcmp(&parent_key
, &node_key
,
633 sizeof(struct btrfs_disk_key
)));
634 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
635 btrfs_header_bytenr(node
));
637 BUG_ON(nritems
> BTRFS_NODEPTRS_PER_BLOCK(root
));
639 btrfs_node_key_to_cpu(node
, &cpukey
, slot
- 1);
640 btrfs_node_key(node
, &node_key
, slot
);
641 BUG_ON(comp_keys(&node_key
, &cpukey
) <= 0);
643 if (slot
< nritems
- 1) {
644 btrfs_node_key_to_cpu(node
, &cpukey
, slot
+ 1);
645 btrfs_node_key(node
, &node_key
, slot
);
646 BUG_ON(comp_keys(&node_key
, &cpukey
) >= 0);
652 * extra checking to make sure all the items in a leaf are
653 * well formed and in the proper order
655 static int check_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
658 struct extent_buffer
*leaf
= path
->nodes
[level
];
659 struct extent_buffer
*parent
= NULL
;
661 struct btrfs_key cpukey
;
662 struct btrfs_disk_key parent_key
;
663 struct btrfs_disk_key leaf_key
;
664 int slot
= path
->slots
[0];
666 u32 nritems
= btrfs_header_nritems(leaf
);
668 if (path
->nodes
[level
+ 1])
669 parent
= path
->nodes
[level
+ 1];
675 parent_slot
= path
->slots
[level
+ 1];
676 btrfs_node_key(parent
, &parent_key
, parent_slot
);
677 btrfs_item_key(leaf
, &leaf_key
, 0);
679 BUG_ON(memcmp(&parent_key
, &leaf_key
,
680 sizeof(struct btrfs_disk_key
)));
681 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
682 btrfs_header_bytenr(leaf
));
685 for (i
= 0; nritems
> 1 && i
< nritems
- 2; i
++) {
686 btrfs_item_key_to_cpu(leaf
, &cpukey
, i
+ 1);
687 btrfs_item_key(leaf
, &leaf_key
, i
);
688 if (comp_keys(&leaf_key
, &cpukey
) >= 0) {
689 btrfs_print_leaf(root
, leaf
);
690 printk("slot %d offset bad key\n", i
);
693 if (btrfs_item_offset_nr(leaf
, i
) !=
694 btrfs_item_end_nr(leaf
, i
+ 1)) {
695 btrfs_print_leaf(root
, leaf
);
696 printk("slot %d offset bad\n", i
);
700 if (btrfs_item_offset_nr(leaf
, i
) +
701 btrfs_item_size_nr(leaf
, i
) !=
702 BTRFS_LEAF_DATA_SIZE(root
)) {
703 btrfs_print_leaf(root
, leaf
);
704 printk("slot %d first offset bad\n", i
);
710 if (btrfs_item_size_nr(leaf
, nritems
- 1) > 4096) {
711 btrfs_print_leaf(root
, leaf
);
712 printk("slot %d bad size \n", nritems
- 1);
717 if (slot
!= 0 && slot
< nritems
- 1) {
718 btrfs_item_key(leaf
, &leaf_key
, slot
);
719 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
- 1);
720 if (comp_keys(&leaf_key
, &cpukey
) <= 0) {
721 btrfs_print_leaf(root
, leaf
);
722 printk("slot %d offset bad key\n", slot
);
725 if (btrfs_item_offset_nr(leaf
, slot
- 1) !=
726 btrfs_item_end_nr(leaf
, slot
)) {
727 btrfs_print_leaf(root
, leaf
);
728 printk("slot %d offset bad\n", slot
);
732 if (slot
< nritems
- 1) {
733 btrfs_item_key(leaf
, &leaf_key
, slot
);
734 btrfs_item_key_to_cpu(leaf
, &cpukey
, slot
+ 1);
735 BUG_ON(comp_keys(&leaf_key
, &cpukey
) >= 0);
736 if (btrfs_item_offset_nr(leaf
, slot
) !=
737 btrfs_item_end_nr(leaf
, slot
+ 1)) {
738 btrfs_print_leaf(root
, leaf
);
739 printk("slot %d offset bad\n", slot
);
743 BUG_ON(btrfs_item_offset_nr(leaf
, 0) +
744 btrfs_item_size_nr(leaf
, 0) != BTRFS_LEAF_DATA_SIZE(root
));
748 static int noinline
check_block(struct btrfs_root
*root
,
749 struct btrfs_path
*path
, int level
)
753 if (btrfs_header_level(path
->nodes
[level
]) != level
)
754 printk("warning: bad level %Lu wanted %d found %d\n",
755 path
->nodes
[level
]->start
, level
,
756 btrfs_header_level(path
->nodes
[level
]));
757 found_start
= btrfs_header_bytenr(path
->nodes
[level
]);
758 if (found_start
!= path
->nodes
[level
]->start
) {
759 printk("warning: bad bytentr %Lu found %Lu\n",
760 path
->nodes
[level
]->start
, found_start
);
763 struct extent_buffer
*buf
= path
->nodes
[level
];
765 if (memcmp_extent_buffer(buf
, root
->fs_info
->fsid
,
766 (unsigned long)btrfs_header_fsid(buf
),
768 printk("warning bad block %Lu\n", buf
->start
);
773 return check_leaf(root
, path
, level
);
774 return check_node(root
, path
, level
);
778 * search for key in the extent_buffer. The items start at offset p,
779 * and they are item_size apart. There are 'max' items in p.
781 * the slot in the array is returned via slot, and it points to
782 * the place where you would insert key if it is not found in
785 * slot may point to max if the key is bigger than all of the keys
787 static noinline
int generic_bin_search(struct extent_buffer
*eb
,
789 int item_size
, struct btrfs_key
*key
,
796 struct btrfs_disk_key
*tmp
= NULL
;
797 struct btrfs_disk_key unaligned
;
798 unsigned long offset
;
799 char *map_token
= NULL
;
801 unsigned long map_start
= 0;
802 unsigned long map_len
= 0;
806 mid
= (low
+ high
) / 2;
807 offset
= p
+ mid
* item_size
;
809 if (!map_token
|| offset
< map_start
||
810 (offset
+ sizeof(struct btrfs_disk_key
)) >
811 map_start
+ map_len
) {
813 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
817 err
= map_private_extent_buffer(eb
, offset
,
818 sizeof(struct btrfs_disk_key
),
820 &map_start
, &map_len
, KM_USER0
);
823 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
826 read_extent_buffer(eb
, &unaligned
,
827 offset
, sizeof(unaligned
));
832 tmp
= (struct btrfs_disk_key
*)(kaddr
+ offset
-
835 ret
= comp_keys(tmp
, key
);
844 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
850 unmap_extent_buffer(eb
, map_token
, KM_USER0
);
855 * simple bin_search frontend that does the right thing for
858 static int bin_search(struct extent_buffer
*eb
, struct btrfs_key
*key
,
859 int level
, int *slot
)
862 return generic_bin_search(eb
,
863 offsetof(struct btrfs_leaf
, items
),
864 sizeof(struct btrfs_item
),
865 key
, btrfs_header_nritems(eb
),
868 return generic_bin_search(eb
,
869 offsetof(struct btrfs_node
, ptrs
),
870 sizeof(struct btrfs_key_ptr
),
871 key
, btrfs_header_nritems(eb
),
877 /* given a node and slot number, this reads the blocks it points to. The
878 * extent buffer is returned with a reference taken (but unlocked).
879 * NULL is returned on error.
881 static noinline
struct extent_buffer
*read_node_slot(struct btrfs_root
*root
,
882 struct extent_buffer
*parent
, int slot
)
884 int level
= btrfs_header_level(parent
);
887 if (slot
>= btrfs_header_nritems(parent
))
892 return read_tree_block(root
, btrfs_node_blockptr(parent
, slot
),
893 btrfs_level_size(root
, level
- 1),
894 btrfs_node_ptr_generation(parent
, slot
));
898 * node level balancing, used to make sure nodes are in proper order for
899 * item deletion. We balance from the top down, so we have to make sure
900 * that a deletion won't leave an node completely empty later on.
902 static noinline
int balance_level(struct btrfs_trans_handle
*trans
,
903 struct btrfs_root
*root
,
904 struct btrfs_path
*path
, int level
)
906 struct extent_buffer
*right
= NULL
;
907 struct extent_buffer
*mid
;
908 struct extent_buffer
*left
= NULL
;
909 struct extent_buffer
*parent
= NULL
;
913 int orig_slot
= path
->slots
[level
];
914 int err_on_enospc
= 0;
920 mid
= path
->nodes
[level
];
921 WARN_ON(!path
->locks
[level
]);
922 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
924 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
926 if (level
< BTRFS_MAX_LEVEL
- 1)
927 parent
= path
->nodes
[level
+ 1];
928 pslot
= path
->slots
[level
+ 1];
931 * deal with the case where there is only one pointer in the root
932 * by promoting the node below to a root
935 struct extent_buffer
*child
;
937 if (btrfs_header_nritems(mid
) != 1)
940 /* promote the child to a root */
941 child
= read_node_slot(root
, mid
, 0);
942 btrfs_tree_lock(child
);
944 ret
= btrfs_cow_block(trans
, root
, child
, mid
, 0, &child
, 0);
947 spin_lock(&root
->node_lock
);
949 spin_unlock(&root
->node_lock
);
951 ret
= btrfs_update_extent_ref(trans
, root
, child
->start
,
952 mid
->start
, child
->start
,
953 root
->root_key
.objectid
,
954 trans
->transid
, level
- 1);
957 add_root_to_dirty_list(root
);
958 btrfs_tree_unlock(child
);
959 path
->locks
[level
] = 0;
960 path
->nodes
[level
] = NULL
;
961 clean_tree_block(trans
, root
, mid
);
962 btrfs_tree_unlock(mid
);
963 /* once for the path */
964 free_extent_buffer(mid
);
965 ret
= btrfs_free_extent(trans
, root
, mid
->start
, mid
->len
,
966 mid
->start
, root
->root_key
.objectid
,
967 btrfs_header_generation(mid
),
969 /* once for the root ptr */
970 free_extent_buffer(mid
);
973 if (btrfs_header_nritems(mid
) >
974 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
977 if (btrfs_header_nritems(mid
) < 2)
980 left
= read_node_slot(root
, parent
, pslot
- 1);
982 btrfs_tree_lock(left
);
983 wret
= btrfs_cow_block(trans
, root
, left
,
984 parent
, pslot
- 1, &left
, 0);
990 right
= read_node_slot(root
, parent
, pslot
+ 1);
992 btrfs_tree_lock(right
);
993 wret
= btrfs_cow_block(trans
, root
, right
,
994 parent
, pslot
+ 1, &right
, 0);
1001 /* first, try to make some room in the middle buffer */
1003 orig_slot
+= btrfs_header_nritems(left
);
1004 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1007 if (btrfs_header_nritems(mid
) < 2)
1012 * then try to empty the right most buffer into the middle
1015 wret
= push_node_left(trans
, root
, mid
, right
, 1);
1016 if (wret
< 0 && wret
!= -ENOSPC
)
1018 if (btrfs_header_nritems(right
) == 0) {
1019 u64 bytenr
= right
->start
;
1020 u64 generation
= btrfs_header_generation(parent
);
1021 u32 blocksize
= right
->len
;
1023 clean_tree_block(trans
, root
, right
);
1024 btrfs_tree_unlock(right
);
1025 free_extent_buffer(right
);
1027 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
+
1031 wret
= btrfs_free_extent(trans
, root
, bytenr
,
1032 blocksize
, parent
->start
,
1033 btrfs_header_owner(parent
),
1034 generation
, level
, 1);
1038 struct btrfs_disk_key right_key
;
1039 btrfs_node_key(right
, &right_key
, 0);
1040 btrfs_set_node_key(parent
, &right_key
, pslot
+ 1);
1041 btrfs_mark_buffer_dirty(parent
);
1044 if (btrfs_header_nritems(mid
) == 1) {
1046 * we're not allowed to leave a node with one item in the
1047 * tree during a delete. A deletion from lower in the tree
1048 * could try to delete the only pointer in this node.
1049 * So, pull some keys from the left.
1050 * There has to be a left pointer at this point because
1051 * otherwise we would have pulled some pointers from the
1055 wret
= balance_node_right(trans
, root
, mid
, left
);
1061 wret
= push_node_left(trans
, root
, left
, mid
, 1);
1067 if (btrfs_header_nritems(mid
) == 0) {
1068 /* we've managed to empty the middle node, drop it */
1069 u64 root_gen
= btrfs_header_generation(parent
);
1070 u64 bytenr
= mid
->start
;
1071 u32 blocksize
= mid
->len
;
1073 clean_tree_block(trans
, root
, mid
);
1074 btrfs_tree_unlock(mid
);
1075 free_extent_buffer(mid
);
1077 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
);
1080 wret
= btrfs_free_extent(trans
, root
, bytenr
, blocksize
,
1082 btrfs_header_owner(parent
),
1083 root_gen
, level
, 1);
1087 /* update the parent key to reflect our changes */
1088 struct btrfs_disk_key mid_key
;
1089 btrfs_node_key(mid
, &mid_key
, 0);
1090 btrfs_set_node_key(parent
, &mid_key
, pslot
);
1091 btrfs_mark_buffer_dirty(parent
);
1094 /* update the path */
1096 if (btrfs_header_nritems(left
) > orig_slot
) {
1097 extent_buffer_get(left
);
1098 /* left was locked after cow */
1099 path
->nodes
[level
] = left
;
1100 path
->slots
[level
+ 1] -= 1;
1101 path
->slots
[level
] = orig_slot
;
1103 btrfs_tree_unlock(mid
);
1104 free_extent_buffer(mid
);
1107 orig_slot
-= btrfs_header_nritems(left
);
1108 path
->slots
[level
] = orig_slot
;
1111 /* double check we haven't messed things up */
1112 check_block(root
, path
, level
);
1114 btrfs_node_blockptr(path
->nodes
[level
], path
->slots
[level
]))
1118 btrfs_tree_unlock(right
);
1119 free_extent_buffer(right
);
1122 if (path
->nodes
[level
] != left
)
1123 btrfs_tree_unlock(left
);
1124 free_extent_buffer(left
);
1129 /* Node balancing for insertion. Here we only split or push nodes around
1130 * when they are completely full. This is also done top down, so we
1131 * have to be pessimistic.
1133 static int noinline
push_nodes_for_insert(struct btrfs_trans_handle
*trans
,
1134 struct btrfs_root
*root
,
1135 struct btrfs_path
*path
, int level
)
1137 struct extent_buffer
*right
= NULL
;
1138 struct extent_buffer
*mid
;
1139 struct extent_buffer
*left
= NULL
;
1140 struct extent_buffer
*parent
= NULL
;
1144 int orig_slot
= path
->slots
[level
];
1150 mid
= path
->nodes
[level
];
1151 WARN_ON(btrfs_header_generation(mid
) != trans
->transid
);
1152 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
1154 if (level
< BTRFS_MAX_LEVEL
- 1)
1155 parent
= path
->nodes
[level
+ 1];
1156 pslot
= path
->slots
[level
+ 1];
1161 left
= read_node_slot(root
, parent
, pslot
- 1);
1163 /* first, try to make some room in the middle buffer */
1167 btrfs_tree_lock(left
);
1168 left_nr
= btrfs_header_nritems(left
);
1169 if (left_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1172 ret
= btrfs_cow_block(trans
, root
, left
, parent
,
1173 pslot
- 1, &left
, 0);
1177 wret
= push_node_left(trans
, root
,
1184 struct btrfs_disk_key disk_key
;
1185 orig_slot
+= left_nr
;
1186 btrfs_node_key(mid
, &disk_key
, 0);
1187 btrfs_set_node_key(parent
, &disk_key
, pslot
);
1188 btrfs_mark_buffer_dirty(parent
);
1189 if (btrfs_header_nritems(left
) > orig_slot
) {
1190 path
->nodes
[level
] = left
;
1191 path
->slots
[level
+ 1] -= 1;
1192 path
->slots
[level
] = orig_slot
;
1193 btrfs_tree_unlock(mid
);
1194 free_extent_buffer(mid
);
1197 btrfs_header_nritems(left
);
1198 path
->slots
[level
] = orig_slot
;
1199 btrfs_tree_unlock(left
);
1200 free_extent_buffer(left
);
1204 btrfs_tree_unlock(left
);
1205 free_extent_buffer(left
);
1207 right
= read_node_slot(root
, parent
, pslot
+ 1);
1210 * then try to empty the right most buffer into the middle
1214 btrfs_tree_lock(right
);
1215 right_nr
= btrfs_header_nritems(right
);
1216 if (right_nr
>= BTRFS_NODEPTRS_PER_BLOCK(root
) - 1) {
1219 ret
= btrfs_cow_block(trans
, root
, right
,
1225 wret
= balance_node_right(trans
, root
,
1232 struct btrfs_disk_key disk_key
;
1234 btrfs_node_key(right
, &disk_key
, 0);
1235 btrfs_set_node_key(parent
, &disk_key
, pslot
+ 1);
1236 btrfs_mark_buffer_dirty(parent
);
1238 if (btrfs_header_nritems(mid
) <= orig_slot
) {
1239 path
->nodes
[level
] = right
;
1240 path
->slots
[level
+ 1] += 1;
1241 path
->slots
[level
] = orig_slot
-
1242 btrfs_header_nritems(mid
);
1243 btrfs_tree_unlock(mid
);
1244 free_extent_buffer(mid
);
1246 btrfs_tree_unlock(right
);
1247 free_extent_buffer(right
);
1251 btrfs_tree_unlock(right
);
1252 free_extent_buffer(right
);
1258 * readahead one full node of leaves, finding things that are close
1259 * to the block in 'slot', and triggering ra on them.
1261 static noinline
void reada_for_search(struct btrfs_root
*root
,
1262 struct btrfs_path
*path
,
1263 int level
, int slot
, u64 objectid
)
1265 struct extent_buffer
*node
;
1266 struct btrfs_disk_key disk_key
;
1272 int direction
= path
->reada
;
1273 struct extent_buffer
*eb
;
1281 if (!path
->nodes
[level
])
1284 node
= path
->nodes
[level
];
1286 search
= btrfs_node_blockptr(node
, slot
);
1287 blocksize
= btrfs_level_size(root
, level
- 1);
1288 eb
= btrfs_find_tree_block(root
, search
, blocksize
);
1290 free_extent_buffer(eb
);
1294 highest_read
= search
;
1295 lowest_read
= search
;
1297 nritems
= btrfs_header_nritems(node
);
1300 if (direction
< 0) {
1304 } else if (direction
> 0) {
1309 if (path
->reada
< 0 && objectid
) {
1310 btrfs_node_key(node
, &disk_key
, nr
);
1311 if (btrfs_disk_key_objectid(&disk_key
) != objectid
)
1314 search
= btrfs_node_blockptr(node
, nr
);
1315 if ((search
>= lowest_read
&& search
<= highest_read
) ||
1316 (search
< lowest_read
&& lowest_read
- search
<= 16384) ||
1317 (search
> highest_read
&& search
- highest_read
<= 16384)) {
1318 readahead_tree_block(root
, search
, blocksize
,
1319 btrfs_node_ptr_generation(node
, nr
));
1323 if (path
->reada
< 2 && (nread
> (64 * 1024) || nscan
> 32))
1325 if(nread
> (256 * 1024) || nscan
> 128)
1328 if (search
< lowest_read
)
1329 lowest_read
= search
;
1330 if (search
> highest_read
)
1331 highest_read
= search
;
1336 * when we walk down the tree, it is usually safe to unlock the higher layers in
1337 * the tree. The exceptions are when our path goes through slot 0, because operations
1338 * on the tree might require changing key pointers higher up in the tree.
1340 * callers might also have set path->keep_locks, which tells this code to
1341 * keep the lock if the path points to the last slot in the block. This is
1342 * part of walking through the tree, and selecting the next slot in the higher
1345 * lowest_unlock sets the lowest level in the tree we're allowed to unlock.
1346 * so if lowest_unlock is 1, level 0 won't be unlocked
1348 static noinline
void unlock_up(struct btrfs_path
*path
, int level
,
1352 int skip_level
= level
;
1354 struct extent_buffer
*t
;
1356 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
1357 if (!path
->nodes
[i
])
1359 if (!path
->locks
[i
])
1361 if (!no_skips
&& path
->slots
[i
] == 0) {
1365 if (!no_skips
&& path
->keep_locks
) {
1368 nritems
= btrfs_header_nritems(t
);
1369 if (nritems
< 1 || path
->slots
[i
] >= nritems
- 1) {
1374 if (skip_level
< i
&& i
>= lowest_unlock
)
1378 if (i
>= lowest_unlock
&& i
> skip_level
&& path
->locks
[i
]) {
1379 btrfs_tree_unlock(t
);
1386 * look for key in the tree. path is filled in with nodes along the way
1387 * if key is found, we return zero and you can find the item in the leaf
1388 * level of the path (level 0)
1390 * If the key isn't found, the path points to the slot where it should
1391 * be inserted, and 1 is returned. If there are other errors during the
1392 * search a negative error number is returned.
1394 * if ins_len > 0, nodes and leaves will be split as we walk down the
1395 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
1398 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
1399 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
1402 struct extent_buffer
*b
;
1403 struct extent_buffer
*tmp
;
1407 int should_reada
= p
->reada
;
1408 int lowest_unlock
= 1;
1410 u8 lowest_level
= 0;
1413 struct btrfs_key prealloc_block
;
1415 lowest_level
= p
->lowest_level
;
1416 WARN_ON(lowest_level
&& ins_len
> 0);
1417 WARN_ON(p
->nodes
[0] != NULL
);
1422 prealloc_block
.objectid
= 0;
1425 if (p
->skip_locking
)
1426 b
= btrfs_root_node(root
);
1428 b
= btrfs_lock_root_node(root
);
1431 level
= btrfs_header_level(b
);
1434 * setup the path here so we can release it under lock
1435 * contention with the cow code
1437 p
->nodes
[level
] = b
;
1438 if (!p
->skip_locking
)
1439 p
->locks
[level
] = 1;
1444 /* is a cow on this block not required */
1445 spin_lock(&root
->fs_info
->hash_lock
);
1446 if (btrfs_header_generation(b
) == trans
->transid
&&
1447 btrfs_header_owner(b
) == root
->root_key
.objectid
&&
1448 !btrfs_header_flag(b
, BTRFS_HEADER_FLAG_WRITTEN
)) {
1449 spin_unlock(&root
->fs_info
->hash_lock
);
1452 spin_unlock(&root
->fs_info
->hash_lock
);
1454 /* ok, we have to cow, is our old prealloc the right
1457 if (prealloc_block
.objectid
&&
1458 prealloc_block
.offset
!= b
->len
) {
1459 btrfs_free_reserved_extent(root
,
1460 prealloc_block
.objectid
,
1461 prealloc_block
.offset
);
1462 prealloc_block
.objectid
= 0;
1466 * for higher level blocks, try not to allocate blocks
1467 * with the block and the parent locks held.
1469 if (level
> 1 && !prealloc_block
.objectid
&&
1470 btrfs_path_lock_waiting(p
, level
)) {
1472 u64 hint
= b
->start
;
1474 btrfs_release_path(root
, p
);
1475 ret
= btrfs_reserve_extent(trans
, root
,
1478 &prealloc_block
, 0);
1483 wret
= btrfs_cow_block(trans
, root
, b
,
1484 p
->nodes
[level
+ 1],
1485 p
->slots
[level
+ 1],
1486 &b
, prealloc_block
.objectid
);
1487 prealloc_block
.objectid
= 0;
1489 free_extent_buffer(b
);
1495 BUG_ON(!cow
&& ins_len
);
1496 if (level
!= btrfs_header_level(b
))
1498 level
= btrfs_header_level(b
);
1500 p
->nodes
[level
] = b
;
1501 if (!p
->skip_locking
)
1502 p
->locks
[level
] = 1;
1504 ret
= check_block(root
, p
, level
);
1510 ret
= bin_search(b
, key
, level
, &slot
);
1512 if (ret
&& slot
> 0)
1514 p
->slots
[level
] = slot
;
1515 if ((p
->search_for_split
|| ins_len
> 0) &&
1516 btrfs_header_nritems(b
) >=
1517 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3) {
1518 int sret
= split_node(trans
, root
, p
, level
);
1524 b
= p
->nodes
[level
];
1525 slot
= p
->slots
[level
];
1526 } else if (ins_len
< 0) {
1527 int sret
= balance_level(trans
, root
, p
,
1533 b
= p
->nodes
[level
];
1535 btrfs_release_path(NULL
, p
);
1538 slot
= p
->slots
[level
];
1539 BUG_ON(btrfs_header_nritems(b
) == 1);
1541 unlock_up(p
, level
, lowest_unlock
);
1543 /* this is only true while dropping a snapshot */
1544 if (level
== lowest_level
) {
1549 blocknr
= btrfs_node_blockptr(b
, slot
);
1550 gen
= btrfs_node_ptr_generation(b
, slot
);
1551 blocksize
= btrfs_level_size(root
, level
- 1);
1553 tmp
= btrfs_find_tree_block(root
, blocknr
, blocksize
);
1554 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
)) {
1558 * reduce lock contention at high levels
1559 * of the btree by dropping locks before
1563 btrfs_release_path(NULL
, p
);
1565 free_extent_buffer(tmp
);
1567 reada_for_search(root
, p
,
1571 tmp
= read_tree_block(root
, blocknr
,
1574 free_extent_buffer(tmp
);
1578 free_extent_buffer(tmp
);
1580 reada_for_search(root
, p
,
1583 b
= read_node_slot(root
, b
, slot
);
1586 if (!p
->skip_locking
)
1589 p
->slots
[level
] = slot
;
1591 btrfs_leaf_free_space(root
, b
) < ins_len
) {
1592 int sret
= split_leaf(trans
, root
, key
,
1593 p
, ins_len
, ret
== 0);
1600 if (!p
->search_for_split
)
1601 unlock_up(p
, level
, lowest_unlock
);
1607 if (prealloc_block
.objectid
) {
1608 btrfs_free_reserved_extent(root
,
1609 prealloc_block
.objectid
,
1610 prealloc_block
.offset
);
1616 int btrfs_merge_path(struct btrfs_trans_handle
*trans
,
1617 struct btrfs_root
*root
,
1618 struct btrfs_key
*node_keys
,
1619 u64
*nodes
, int lowest_level
)
1621 struct extent_buffer
*eb
;
1622 struct extent_buffer
*parent
;
1623 struct btrfs_key key
;
1632 eb
= btrfs_lock_root_node(root
);
1633 ret
= btrfs_cow_block(trans
, root
, eb
, NULL
, 0, &eb
, 0);
1638 level
= btrfs_header_level(parent
);
1639 if (level
== 0 || level
<= lowest_level
)
1642 ret
= bin_search(parent
, &node_keys
[lowest_level
], level
,
1644 if (ret
&& slot
> 0)
1647 bytenr
= btrfs_node_blockptr(parent
, slot
);
1648 if (nodes
[level
- 1] == bytenr
)
1651 blocksize
= btrfs_level_size(root
, level
- 1);
1652 generation
= btrfs_node_ptr_generation(parent
, slot
);
1653 btrfs_node_key_to_cpu(eb
, &key
, slot
);
1654 key_match
= !memcmp(&key
, &node_keys
[level
- 1], sizeof(key
));
1656 if (generation
== trans
->transid
) {
1657 eb
= read_tree_block(root
, bytenr
, blocksize
,
1659 btrfs_tree_lock(eb
);
1663 * if node keys match and node pointer hasn't been modified
1664 * in the running transaction, we can merge the path. for
1665 * blocks owened by reloc trees, the node pointer check is
1666 * skipped, this is because these blocks are fully controlled
1667 * by the space balance code, no one else can modify them.
1669 if (!nodes
[level
- 1] || !key_match
||
1670 (generation
== trans
->transid
&&
1671 btrfs_header_owner(eb
) != BTRFS_TREE_RELOC_OBJECTID
)) {
1672 if (level
== 1 || level
== lowest_level
+ 1) {
1673 if (generation
== trans
->transid
) {
1674 btrfs_tree_unlock(eb
);
1675 free_extent_buffer(eb
);
1680 if (generation
!= trans
->transid
) {
1681 eb
= read_tree_block(root
, bytenr
, blocksize
,
1683 btrfs_tree_lock(eb
);
1686 ret
= btrfs_cow_block(trans
, root
, eb
, parent
, slot
,
1690 if (root
->root_key
.objectid
==
1691 BTRFS_TREE_RELOC_OBJECTID
) {
1692 if (!nodes
[level
- 1]) {
1693 nodes
[level
- 1] = eb
->start
;
1694 memcpy(&node_keys
[level
- 1], &key
,
1695 sizeof(node_keys
[0]));
1701 btrfs_tree_unlock(parent
);
1702 free_extent_buffer(parent
);
1707 btrfs_set_node_blockptr(parent
, slot
, nodes
[level
- 1]);
1708 btrfs_set_node_ptr_generation(parent
, slot
, trans
->transid
);
1709 btrfs_mark_buffer_dirty(parent
);
1711 ret
= btrfs_inc_extent_ref(trans
, root
,
1713 blocksize
, parent
->start
,
1714 btrfs_header_owner(parent
),
1715 btrfs_header_generation(parent
),
1720 * If the block was created in the running transaction,
1721 * it's possible this is the last reference to it, so we
1722 * should drop the subtree.
1724 if (generation
== trans
->transid
) {
1725 ret
= btrfs_drop_subtree(trans
, root
, eb
, parent
);
1727 btrfs_tree_unlock(eb
);
1728 free_extent_buffer(eb
);
1730 ret
= btrfs_free_extent(trans
, root
, bytenr
,
1731 blocksize
, parent
->start
,
1732 btrfs_header_owner(parent
),
1733 btrfs_header_generation(parent
),
1739 btrfs_tree_unlock(parent
);
1740 free_extent_buffer(parent
);
1745 * adjust the pointers going up the tree, starting at level
1746 * making sure the right key of each node is points to 'key'.
1747 * This is used after shifting pointers to the left, so it stops
1748 * fixing up pointers when a given leaf/node is not in slot 0 of the
1751 * If this fails to write a tree block, it returns -1, but continues
1752 * fixing up the blocks in ram so the tree is consistent.
1754 static int fixup_low_keys(struct btrfs_trans_handle
*trans
,
1755 struct btrfs_root
*root
, struct btrfs_path
*path
,
1756 struct btrfs_disk_key
*key
, int level
)
1760 struct extent_buffer
*t
;
1762 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
1763 int tslot
= path
->slots
[i
];
1764 if (!path
->nodes
[i
])
1767 btrfs_set_node_key(t
, key
, tslot
);
1768 btrfs_mark_buffer_dirty(path
->nodes
[i
]);
1778 * This function isn't completely safe. It's the caller's responsibility
1779 * that the new key won't break the order
1781 int btrfs_set_item_key_safe(struct btrfs_trans_handle
*trans
,
1782 struct btrfs_root
*root
, struct btrfs_path
*path
,
1783 struct btrfs_key
*new_key
)
1785 struct btrfs_disk_key disk_key
;
1786 struct extent_buffer
*eb
;
1789 eb
= path
->nodes
[0];
1790 slot
= path
->slots
[0];
1792 btrfs_item_key(eb
, &disk_key
, slot
- 1);
1793 if (comp_keys(&disk_key
, new_key
) >= 0)
1796 if (slot
< btrfs_header_nritems(eb
) - 1) {
1797 btrfs_item_key(eb
, &disk_key
, slot
+ 1);
1798 if (comp_keys(&disk_key
, new_key
) <= 0)
1802 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
1803 btrfs_set_item_key(eb
, &disk_key
, slot
);
1804 btrfs_mark_buffer_dirty(eb
);
1806 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
1811 * try to push data from one node into the next node left in the
1814 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
1815 * error, and > 0 if there was no room in the left hand block.
1817 static int push_node_left(struct btrfs_trans_handle
*trans
,
1818 struct btrfs_root
*root
, struct extent_buffer
*dst
,
1819 struct extent_buffer
*src
, int empty
)
1826 src_nritems
= btrfs_header_nritems(src
);
1827 dst_nritems
= btrfs_header_nritems(dst
);
1828 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
1829 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1830 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
1832 if (!empty
&& src_nritems
<= 8)
1835 if (push_items
<= 0) {
1840 push_items
= min(src_nritems
, push_items
);
1841 if (push_items
< src_nritems
) {
1842 /* leave at least 8 pointers in the node if
1843 * we aren't going to empty it
1845 if (src_nritems
- push_items
< 8) {
1846 if (push_items
<= 8)
1852 push_items
= min(src_nritems
- 8, push_items
);
1854 copy_extent_buffer(dst
, src
,
1855 btrfs_node_key_ptr_offset(dst_nritems
),
1856 btrfs_node_key_ptr_offset(0),
1857 push_items
* sizeof(struct btrfs_key_ptr
));
1859 if (push_items
< src_nritems
) {
1860 memmove_extent_buffer(src
, btrfs_node_key_ptr_offset(0),
1861 btrfs_node_key_ptr_offset(push_items
),
1862 (src_nritems
- push_items
) *
1863 sizeof(struct btrfs_key_ptr
));
1865 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1866 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1867 btrfs_mark_buffer_dirty(src
);
1868 btrfs_mark_buffer_dirty(dst
);
1870 ret
= btrfs_update_ref(trans
, root
, src
, dst
, dst_nritems
, push_items
);
1877 * try to push data from one node into the next node right in the
1880 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
1881 * error, and > 0 if there was no room in the right hand block.
1883 * this will only push up to 1/2 the contents of the left node over
1885 static int balance_node_right(struct btrfs_trans_handle
*trans
,
1886 struct btrfs_root
*root
,
1887 struct extent_buffer
*dst
,
1888 struct extent_buffer
*src
)
1896 WARN_ON(btrfs_header_generation(src
) != trans
->transid
);
1897 WARN_ON(btrfs_header_generation(dst
) != trans
->transid
);
1899 src_nritems
= btrfs_header_nritems(src
);
1900 dst_nritems
= btrfs_header_nritems(dst
);
1901 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
1902 if (push_items
<= 0) {
1906 if (src_nritems
< 4) {
1910 max_push
= src_nritems
/ 2 + 1;
1911 /* don't try to empty the node */
1912 if (max_push
>= src_nritems
) {
1916 if (max_push
< push_items
)
1917 push_items
= max_push
;
1919 memmove_extent_buffer(dst
, btrfs_node_key_ptr_offset(push_items
),
1920 btrfs_node_key_ptr_offset(0),
1922 sizeof(struct btrfs_key_ptr
));
1924 copy_extent_buffer(dst
, src
,
1925 btrfs_node_key_ptr_offset(0),
1926 btrfs_node_key_ptr_offset(src_nritems
- push_items
),
1927 push_items
* sizeof(struct btrfs_key_ptr
));
1929 btrfs_set_header_nritems(src
, src_nritems
- push_items
);
1930 btrfs_set_header_nritems(dst
, dst_nritems
+ push_items
);
1932 btrfs_mark_buffer_dirty(src
);
1933 btrfs_mark_buffer_dirty(dst
);
1935 ret
= btrfs_update_ref(trans
, root
, src
, dst
, 0, push_items
);
1942 * helper function to insert a new root level in the tree.
1943 * A new node is allocated, and a single item is inserted to
1944 * point to the existing root
1946 * returns zero on success or < 0 on failure.
1948 static int noinline
insert_new_root(struct btrfs_trans_handle
*trans
,
1949 struct btrfs_root
*root
,
1950 struct btrfs_path
*path
, int level
)
1953 struct extent_buffer
*lower
;
1954 struct extent_buffer
*c
;
1955 struct extent_buffer
*old
;
1956 struct btrfs_disk_key lower_key
;
1959 BUG_ON(path
->nodes
[level
]);
1960 BUG_ON(path
->nodes
[level
-1] != root
->node
);
1962 lower
= path
->nodes
[level
-1];
1964 btrfs_item_key(lower
, &lower_key
, 0);
1966 btrfs_node_key(lower
, &lower_key
, 0);
1968 c
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
, 0,
1969 root
->root_key
.objectid
, trans
->transid
,
1970 level
, root
->node
->start
, 0);
1974 memset_extent_buffer(c
, 0, 0, root
->nodesize
);
1975 btrfs_set_header_nritems(c
, 1);
1976 btrfs_set_header_level(c
, level
);
1977 btrfs_set_header_bytenr(c
, c
->start
);
1978 btrfs_set_header_generation(c
, trans
->transid
);
1979 btrfs_set_header_owner(c
, root
->root_key
.objectid
);
1981 write_extent_buffer(c
, root
->fs_info
->fsid
,
1982 (unsigned long)btrfs_header_fsid(c
),
1985 write_extent_buffer(c
, root
->fs_info
->chunk_tree_uuid
,
1986 (unsigned long)btrfs_header_chunk_tree_uuid(c
),
1989 btrfs_set_node_key(c
, &lower_key
, 0);
1990 btrfs_set_node_blockptr(c
, 0, lower
->start
);
1991 lower_gen
= btrfs_header_generation(lower
);
1992 WARN_ON(lower_gen
!= trans
->transid
);
1994 btrfs_set_node_ptr_generation(c
, 0, lower_gen
);
1996 btrfs_mark_buffer_dirty(c
);
1998 spin_lock(&root
->node_lock
);
2001 spin_unlock(&root
->node_lock
);
2003 ret
= btrfs_update_extent_ref(trans
, root
, lower
->start
,
2004 lower
->start
, c
->start
,
2005 root
->root_key
.objectid
,
2006 trans
->transid
, level
- 1);
2009 /* the super has an extra ref to root->node */
2010 free_extent_buffer(old
);
2012 add_root_to_dirty_list(root
);
2013 extent_buffer_get(c
);
2014 path
->nodes
[level
] = c
;
2015 path
->locks
[level
] = 1;
2016 path
->slots
[level
] = 0;
2021 * worker function to insert a single pointer in a node.
2022 * the node should have enough room for the pointer already
2024 * slot and level indicate where you want the key to go, and
2025 * blocknr is the block the key points to.
2027 * returns zero on success and < 0 on any error
2029 static int insert_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
2030 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
2031 *key
, u64 bytenr
, int slot
, int level
)
2033 struct extent_buffer
*lower
;
2036 BUG_ON(!path
->nodes
[level
]);
2037 lower
= path
->nodes
[level
];
2038 nritems
= btrfs_header_nritems(lower
);
2041 if (nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
))
2043 if (slot
!= nritems
) {
2044 memmove_extent_buffer(lower
,
2045 btrfs_node_key_ptr_offset(slot
+ 1),
2046 btrfs_node_key_ptr_offset(slot
),
2047 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
2049 btrfs_set_node_key(lower
, key
, slot
);
2050 btrfs_set_node_blockptr(lower
, slot
, bytenr
);
2051 WARN_ON(trans
->transid
== 0);
2052 btrfs_set_node_ptr_generation(lower
, slot
, trans
->transid
);
2053 btrfs_set_header_nritems(lower
, nritems
+ 1);
2054 btrfs_mark_buffer_dirty(lower
);
2059 * split the node at the specified level in path in two.
2060 * The path is corrected to point to the appropriate node after the split
2062 * Before splitting this tries to make some room in the node by pushing
2063 * left and right, if either one works, it returns right away.
2065 * returns 0 on success and < 0 on failure
2067 static noinline
int split_node(struct btrfs_trans_handle
*trans
,
2068 struct btrfs_root
*root
,
2069 struct btrfs_path
*path
, int level
)
2071 struct extent_buffer
*c
;
2072 struct extent_buffer
*split
;
2073 struct btrfs_disk_key disk_key
;
2079 c
= path
->nodes
[level
];
2080 WARN_ON(btrfs_header_generation(c
) != trans
->transid
);
2081 if (c
== root
->node
) {
2082 /* trying to split the root, lets make a new one */
2083 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
2087 ret
= push_nodes_for_insert(trans
, root
, path
, level
);
2088 c
= path
->nodes
[level
];
2089 if (!ret
&& btrfs_header_nritems(c
) <
2090 BTRFS_NODEPTRS_PER_BLOCK(root
) - 3)
2096 c_nritems
= btrfs_header_nritems(c
);
2098 split
= btrfs_alloc_free_block(trans
, root
, root
->nodesize
,
2099 path
->nodes
[level
+ 1]->start
,
2100 root
->root_key
.objectid
,
2101 trans
->transid
, level
, c
->start
, 0);
2103 return PTR_ERR(split
);
2105 btrfs_set_header_flags(split
, btrfs_header_flags(c
));
2106 btrfs_set_header_level(split
, btrfs_header_level(c
));
2107 btrfs_set_header_bytenr(split
, split
->start
);
2108 btrfs_set_header_generation(split
, trans
->transid
);
2109 btrfs_set_header_owner(split
, root
->root_key
.objectid
);
2110 btrfs_set_header_flags(split
, 0);
2111 write_extent_buffer(split
, root
->fs_info
->fsid
,
2112 (unsigned long)btrfs_header_fsid(split
),
2114 write_extent_buffer(split
, root
->fs_info
->chunk_tree_uuid
,
2115 (unsigned long)btrfs_header_chunk_tree_uuid(split
),
2118 mid
= (c_nritems
+ 1) / 2;
2120 copy_extent_buffer(split
, c
,
2121 btrfs_node_key_ptr_offset(0),
2122 btrfs_node_key_ptr_offset(mid
),
2123 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
2124 btrfs_set_header_nritems(split
, c_nritems
- mid
);
2125 btrfs_set_header_nritems(c
, mid
);
2128 btrfs_mark_buffer_dirty(c
);
2129 btrfs_mark_buffer_dirty(split
);
2131 btrfs_node_key(split
, &disk_key
, 0);
2132 wret
= insert_ptr(trans
, root
, path
, &disk_key
, split
->start
,
2133 path
->slots
[level
+ 1] + 1,
2138 ret
= btrfs_update_ref(trans
, root
, c
, split
, 0, c_nritems
- mid
);
2141 if (path
->slots
[level
] >= mid
) {
2142 path
->slots
[level
] -= mid
;
2143 btrfs_tree_unlock(c
);
2144 free_extent_buffer(c
);
2145 path
->nodes
[level
] = split
;
2146 path
->slots
[level
+ 1] += 1;
2148 btrfs_tree_unlock(split
);
2149 free_extent_buffer(split
);
2155 * how many bytes are required to store the items in a leaf. start
2156 * and nr indicate which items in the leaf to check. This totals up the
2157 * space used both by the item structs and the item data
2159 static int leaf_space_used(struct extent_buffer
*l
, int start
, int nr
)
2162 int nritems
= btrfs_header_nritems(l
);
2163 int end
= min(nritems
, start
+ nr
) - 1;
2167 data_len
= btrfs_item_end_nr(l
, start
);
2168 data_len
= data_len
- btrfs_item_offset_nr(l
, end
);
2169 data_len
+= sizeof(struct btrfs_item
) * nr
;
2170 WARN_ON(data_len
< 0);
2175 * The space between the end of the leaf items and
2176 * the start of the leaf data. IOW, how much room
2177 * the leaf has left for both items and data
2179 int noinline
btrfs_leaf_free_space(struct btrfs_root
*root
,
2180 struct extent_buffer
*leaf
)
2182 int nritems
= btrfs_header_nritems(leaf
);
2184 ret
= BTRFS_LEAF_DATA_SIZE(root
) - leaf_space_used(leaf
, 0, nritems
);
2186 printk("leaf free space ret %d, leaf data size %lu, used %d nritems %d\n",
2187 ret
, (unsigned long) BTRFS_LEAF_DATA_SIZE(root
),
2188 leaf_space_used(leaf
, 0, nritems
), nritems
);
2194 * push some data in the path leaf to the right, trying to free up at
2195 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2197 * returns 1 if the push failed because the other node didn't have enough
2198 * room, 0 if everything worked out and < 0 if there were major errors.
2200 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
2201 *root
, struct btrfs_path
*path
, int data_size
,
2204 struct extent_buffer
*left
= path
->nodes
[0];
2205 struct extent_buffer
*right
;
2206 struct extent_buffer
*upper
;
2207 struct btrfs_disk_key disk_key
;
2213 struct btrfs_item
*item
;
2221 slot
= path
->slots
[1];
2222 if (!path
->nodes
[1]) {
2225 upper
= path
->nodes
[1];
2226 if (slot
>= btrfs_header_nritems(upper
) - 1)
2229 WARN_ON(!btrfs_tree_locked(path
->nodes
[1]));
2231 right
= read_node_slot(root
, upper
, slot
+ 1);
2232 btrfs_tree_lock(right
);
2233 free_space
= btrfs_leaf_free_space(root
, right
);
2234 if (free_space
< data_size
)
2237 /* cow and double check */
2238 ret
= btrfs_cow_block(trans
, root
, right
, upper
,
2239 slot
+ 1, &right
, 0);
2243 free_space
= btrfs_leaf_free_space(root
, right
);
2244 if (free_space
< data_size
)
2247 left_nritems
= btrfs_header_nritems(left
);
2248 if (left_nritems
== 0)
2256 if (path
->slots
[0] >= left_nritems
)
2257 push_space
+= data_size
;
2259 i
= left_nritems
- 1;
2261 item
= btrfs_item_nr(left
, i
);
2263 if (!empty
&& push_items
> 0) {
2264 if (path
->slots
[0] > i
)
2266 if (path
->slots
[0] == i
) {
2267 int space
= btrfs_leaf_free_space(root
, left
);
2268 if (space
+ push_space
* 2 > free_space
)
2273 if (path
->slots
[0] == i
)
2274 push_space
+= data_size
;
2276 if (!left
->map_token
) {
2277 map_extent_buffer(left
, (unsigned long)item
,
2278 sizeof(struct btrfs_item
),
2279 &left
->map_token
, &left
->kaddr
,
2280 &left
->map_start
, &left
->map_len
,
2284 this_item_size
= btrfs_item_size(left
, item
);
2285 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
2289 push_space
+= this_item_size
+ sizeof(*item
);
2294 if (left
->map_token
) {
2295 unmap_extent_buffer(left
, left
->map_token
, KM_USER1
);
2296 left
->map_token
= NULL
;
2299 if (push_items
== 0)
2302 if (!empty
&& push_items
== left_nritems
)
2305 /* push left to right */
2306 right_nritems
= btrfs_header_nritems(right
);
2308 push_space
= btrfs_item_end_nr(left
, left_nritems
- push_items
);
2309 push_space
-= leaf_data_end(root
, left
);
2311 /* make room in the right data area */
2312 data_end
= leaf_data_end(root
, right
);
2313 memmove_extent_buffer(right
,
2314 btrfs_leaf_data(right
) + data_end
- push_space
,
2315 btrfs_leaf_data(right
) + data_end
,
2316 BTRFS_LEAF_DATA_SIZE(root
) - data_end
);
2318 /* copy from the left data area */
2319 copy_extent_buffer(right
, left
, btrfs_leaf_data(right
) +
2320 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
2321 btrfs_leaf_data(left
) + leaf_data_end(root
, left
),
2324 memmove_extent_buffer(right
, btrfs_item_nr_offset(push_items
),
2325 btrfs_item_nr_offset(0),
2326 right_nritems
* sizeof(struct btrfs_item
));
2328 /* copy the items from left to right */
2329 copy_extent_buffer(right
, left
, btrfs_item_nr_offset(0),
2330 btrfs_item_nr_offset(left_nritems
- push_items
),
2331 push_items
* sizeof(struct btrfs_item
));
2333 /* update the item pointers */
2334 right_nritems
+= push_items
;
2335 btrfs_set_header_nritems(right
, right_nritems
);
2336 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
2337 for (i
= 0; i
< right_nritems
; i
++) {
2338 item
= btrfs_item_nr(right
, i
);
2339 if (!right
->map_token
) {
2340 map_extent_buffer(right
, (unsigned long)item
,
2341 sizeof(struct btrfs_item
),
2342 &right
->map_token
, &right
->kaddr
,
2343 &right
->map_start
, &right
->map_len
,
2346 push_space
-= btrfs_item_size(right
, item
);
2347 btrfs_set_item_offset(right
, item
, push_space
);
2350 if (right
->map_token
) {
2351 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2352 right
->map_token
= NULL
;
2354 left_nritems
-= push_items
;
2355 btrfs_set_header_nritems(left
, left_nritems
);
2358 btrfs_mark_buffer_dirty(left
);
2359 btrfs_mark_buffer_dirty(right
);
2361 ret
= btrfs_update_ref(trans
, root
, left
, right
, 0, push_items
);
2364 btrfs_item_key(right
, &disk_key
, 0);
2365 btrfs_set_node_key(upper
, &disk_key
, slot
+ 1);
2366 btrfs_mark_buffer_dirty(upper
);
2368 /* then fixup the leaf pointer in the path */
2369 if (path
->slots
[0] >= left_nritems
) {
2370 path
->slots
[0] -= left_nritems
;
2371 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
2372 clean_tree_block(trans
, root
, path
->nodes
[0]);
2373 btrfs_tree_unlock(path
->nodes
[0]);
2374 free_extent_buffer(path
->nodes
[0]);
2375 path
->nodes
[0] = right
;
2376 path
->slots
[1] += 1;
2378 btrfs_tree_unlock(right
);
2379 free_extent_buffer(right
);
2384 btrfs_tree_unlock(right
);
2385 free_extent_buffer(right
);
2390 * push some data in the path leaf to the left, trying to free up at
2391 * least data_size bytes. returns zero if the push worked, nonzero otherwise
2393 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
2394 *root
, struct btrfs_path
*path
, int data_size
,
2397 struct btrfs_disk_key disk_key
;
2398 struct extent_buffer
*right
= path
->nodes
[0];
2399 struct extent_buffer
*left
;
2405 struct btrfs_item
*item
;
2406 u32 old_left_nritems
;
2412 u32 old_left_item_size
;
2414 slot
= path
->slots
[1];
2417 if (!path
->nodes
[1])
2420 right_nritems
= btrfs_header_nritems(right
);
2421 if (right_nritems
== 0) {
2425 WARN_ON(!btrfs_tree_locked(path
->nodes
[1]));
2427 left
= read_node_slot(root
, path
->nodes
[1], slot
- 1);
2428 btrfs_tree_lock(left
);
2429 free_space
= btrfs_leaf_free_space(root
, left
);
2430 if (free_space
< data_size
) {
2435 /* cow and double check */
2436 ret
= btrfs_cow_block(trans
, root
, left
,
2437 path
->nodes
[1], slot
- 1, &left
, 0);
2439 /* we hit -ENOSPC, but it isn't fatal here */
2444 free_space
= btrfs_leaf_free_space(root
, left
);
2445 if (free_space
< data_size
) {
2453 nr
= right_nritems
- 1;
2455 for (i
= 0; i
< nr
; i
++) {
2456 item
= btrfs_item_nr(right
, i
);
2457 if (!right
->map_token
) {
2458 map_extent_buffer(right
, (unsigned long)item
,
2459 sizeof(struct btrfs_item
),
2460 &right
->map_token
, &right
->kaddr
,
2461 &right
->map_start
, &right
->map_len
,
2465 if (!empty
&& push_items
> 0) {
2466 if (path
->slots
[0] < i
)
2468 if (path
->slots
[0] == i
) {
2469 int space
= btrfs_leaf_free_space(root
, right
);
2470 if (space
+ push_space
* 2 > free_space
)
2475 if (path
->slots
[0] == i
)
2476 push_space
+= data_size
;
2478 this_item_size
= btrfs_item_size(right
, item
);
2479 if (this_item_size
+ sizeof(*item
) + push_space
> free_space
)
2483 push_space
+= this_item_size
+ sizeof(*item
);
2486 if (right
->map_token
) {
2487 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2488 right
->map_token
= NULL
;
2491 if (push_items
== 0) {
2495 if (!empty
&& push_items
== btrfs_header_nritems(right
))
2498 /* push data from right to left */
2499 copy_extent_buffer(left
, right
,
2500 btrfs_item_nr_offset(btrfs_header_nritems(left
)),
2501 btrfs_item_nr_offset(0),
2502 push_items
* sizeof(struct btrfs_item
));
2504 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
2505 btrfs_item_offset_nr(right
, push_items
-1);
2507 copy_extent_buffer(left
, right
, btrfs_leaf_data(left
) +
2508 leaf_data_end(root
, left
) - push_space
,
2509 btrfs_leaf_data(right
) +
2510 btrfs_item_offset_nr(right
, push_items
- 1),
2512 old_left_nritems
= btrfs_header_nritems(left
);
2513 BUG_ON(old_left_nritems
<= 0);
2515 old_left_item_size
= btrfs_item_offset_nr(left
, old_left_nritems
- 1);
2516 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
2519 item
= btrfs_item_nr(left
, i
);
2520 if (!left
->map_token
) {
2521 map_extent_buffer(left
, (unsigned long)item
,
2522 sizeof(struct btrfs_item
),
2523 &left
->map_token
, &left
->kaddr
,
2524 &left
->map_start
, &left
->map_len
,
2528 ioff
= btrfs_item_offset(left
, item
);
2529 btrfs_set_item_offset(left
, item
,
2530 ioff
- (BTRFS_LEAF_DATA_SIZE(root
) - old_left_item_size
));
2532 btrfs_set_header_nritems(left
, old_left_nritems
+ push_items
);
2533 if (left
->map_token
) {
2534 unmap_extent_buffer(left
, left
->map_token
, KM_USER1
);
2535 left
->map_token
= NULL
;
2538 /* fixup right node */
2539 if (push_items
> right_nritems
) {
2540 printk("push items %d nr %u\n", push_items
, right_nritems
);
2544 if (push_items
< right_nritems
) {
2545 push_space
= btrfs_item_offset_nr(right
, push_items
- 1) -
2546 leaf_data_end(root
, right
);
2547 memmove_extent_buffer(right
, btrfs_leaf_data(right
) +
2548 BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
2549 btrfs_leaf_data(right
) +
2550 leaf_data_end(root
, right
), push_space
);
2552 memmove_extent_buffer(right
, btrfs_item_nr_offset(0),
2553 btrfs_item_nr_offset(push_items
),
2554 (btrfs_header_nritems(right
) - push_items
) *
2555 sizeof(struct btrfs_item
));
2557 right_nritems
-= push_items
;
2558 btrfs_set_header_nritems(right
, right_nritems
);
2559 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
2560 for (i
= 0; i
< right_nritems
; i
++) {
2561 item
= btrfs_item_nr(right
, i
);
2563 if (!right
->map_token
) {
2564 map_extent_buffer(right
, (unsigned long)item
,
2565 sizeof(struct btrfs_item
),
2566 &right
->map_token
, &right
->kaddr
,
2567 &right
->map_start
, &right
->map_len
,
2571 push_space
= push_space
- btrfs_item_size(right
, item
);
2572 btrfs_set_item_offset(right
, item
, push_space
);
2574 if (right
->map_token
) {
2575 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2576 right
->map_token
= NULL
;
2579 btrfs_mark_buffer_dirty(left
);
2581 btrfs_mark_buffer_dirty(right
);
2583 ret
= btrfs_update_ref(trans
, root
, right
, left
,
2584 old_left_nritems
, push_items
);
2587 btrfs_item_key(right
, &disk_key
, 0);
2588 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
2592 /* then fixup the leaf pointer in the path */
2593 if (path
->slots
[0] < push_items
) {
2594 path
->slots
[0] += old_left_nritems
;
2595 if (btrfs_header_nritems(path
->nodes
[0]) == 0)
2596 clean_tree_block(trans
, root
, path
->nodes
[0]);
2597 btrfs_tree_unlock(path
->nodes
[0]);
2598 free_extent_buffer(path
->nodes
[0]);
2599 path
->nodes
[0] = left
;
2600 path
->slots
[1] -= 1;
2602 btrfs_tree_unlock(left
);
2603 free_extent_buffer(left
);
2604 path
->slots
[0] -= push_items
;
2606 BUG_ON(path
->slots
[0] < 0);
2609 btrfs_tree_unlock(left
);
2610 free_extent_buffer(left
);
2615 * split the path's leaf in two, making sure there is at least data_size
2616 * available for the resulting leaf level of the path.
2618 * returns 0 if all went well and < 0 on failure.
2620 static noinline
int split_leaf(struct btrfs_trans_handle
*trans
,
2621 struct btrfs_root
*root
,
2622 struct btrfs_key
*ins_key
,
2623 struct btrfs_path
*path
, int data_size
,
2626 struct extent_buffer
*l
;
2630 struct extent_buffer
*right
;
2637 int num_doubles
= 0;
2638 struct btrfs_disk_key disk_key
;
2640 /* first try to make some room by pushing left and right */
2641 if (data_size
&& ins_key
->type
!= BTRFS_DIR_ITEM_KEY
) {
2642 wret
= push_leaf_right(trans
, root
, path
, data_size
, 0);
2647 wret
= push_leaf_left(trans
, root
, path
, data_size
, 0);
2653 /* did the pushes work? */
2654 if (btrfs_leaf_free_space(root
, l
) >= data_size
)
2658 if (!path
->nodes
[1]) {
2659 ret
= insert_new_root(trans
, root
, path
, 1);
2666 slot
= path
->slots
[0];
2667 nritems
= btrfs_header_nritems(l
);
2668 mid
= (nritems
+ 1)/ 2;
2670 right
= btrfs_alloc_free_block(trans
, root
, root
->leafsize
,
2671 path
->nodes
[1]->start
,
2672 root
->root_key
.objectid
,
2673 trans
->transid
, 0, l
->start
, 0);
2674 if (IS_ERR(right
)) {
2676 return PTR_ERR(right
);
2679 memset_extent_buffer(right
, 0, 0, sizeof(struct btrfs_header
));
2680 btrfs_set_header_bytenr(right
, right
->start
);
2681 btrfs_set_header_generation(right
, trans
->transid
);
2682 btrfs_set_header_owner(right
, root
->root_key
.objectid
);
2683 btrfs_set_header_level(right
, 0);
2684 write_extent_buffer(right
, root
->fs_info
->fsid
,
2685 (unsigned long)btrfs_header_fsid(right
),
2688 write_extent_buffer(right
, root
->fs_info
->chunk_tree_uuid
,
2689 (unsigned long)btrfs_header_chunk_tree_uuid(right
),
2693 leaf_space_used(l
, mid
, nritems
- mid
) + data_size
>
2694 BTRFS_LEAF_DATA_SIZE(root
)) {
2695 if (slot
>= nritems
) {
2696 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
2697 btrfs_set_header_nritems(right
, 0);
2698 wret
= insert_ptr(trans
, root
, path
,
2699 &disk_key
, right
->start
,
2700 path
->slots
[1] + 1, 1);
2704 btrfs_tree_unlock(path
->nodes
[0]);
2705 free_extent_buffer(path
->nodes
[0]);
2706 path
->nodes
[0] = right
;
2708 path
->slots
[1] += 1;
2709 btrfs_mark_buffer_dirty(right
);
2713 if (mid
!= nritems
&&
2714 leaf_space_used(l
, mid
, nritems
- mid
) +
2715 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
2720 if (leaf_space_used(l
, 0, mid
) + data_size
>
2721 BTRFS_LEAF_DATA_SIZE(root
)) {
2722 if (!extend
&& data_size
&& slot
== 0) {
2723 btrfs_cpu_key_to_disk(&disk_key
, ins_key
);
2724 btrfs_set_header_nritems(right
, 0);
2725 wret
= insert_ptr(trans
, root
, path
,
2731 btrfs_tree_unlock(path
->nodes
[0]);
2732 free_extent_buffer(path
->nodes
[0]);
2733 path
->nodes
[0] = right
;
2735 if (path
->slots
[1] == 0) {
2736 wret
= fixup_low_keys(trans
, root
,
2737 path
, &disk_key
, 1);
2741 btrfs_mark_buffer_dirty(right
);
2743 } else if ((extend
|| !data_size
) && slot
== 0) {
2747 if (mid
!= nritems
&&
2748 leaf_space_used(l
, mid
, nritems
- mid
) +
2749 data_size
> BTRFS_LEAF_DATA_SIZE(root
)) {
2755 nritems
= nritems
- mid
;
2756 btrfs_set_header_nritems(right
, nritems
);
2757 data_copy_size
= btrfs_item_end_nr(l
, mid
) - leaf_data_end(root
, l
);
2759 copy_extent_buffer(right
, l
, btrfs_item_nr_offset(0),
2760 btrfs_item_nr_offset(mid
),
2761 nritems
* sizeof(struct btrfs_item
));
2763 copy_extent_buffer(right
, l
,
2764 btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
2765 data_copy_size
, btrfs_leaf_data(l
) +
2766 leaf_data_end(root
, l
), data_copy_size
);
2768 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
2769 btrfs_item_end_nr(l
, mid
);
2771 for (i
= 0; i
< nritems
; i
++) {
2772 struct btrfs_item
*item
= btrfs_item_nr(right
, i
);
2775 if (!right
->map_token
) {
2776 map_extent_buffer(right
, (unsigned long)item
,
2777 sizeof(struct btrfs_item
),
2778 &right
->map_token
, &right
->kaddr
,
2779 &right
->map_start
, &right
->map_len
,
2783 ioff
= btrfs_item_offset(right
, item
);
2784 btrfs_set_item_offset(right
, item
, ioff
+ rt_data_off
);
2787 if (right
->map_token
) {
2788 unmap_extent_buffer(right
, right
->map_token
, KM_USER1
);
2789 right
->map_token
= NULL
;
2792 btrfs_set_header_nritems(l
, mid
);
2794 btrfs_item_key(right
, &disk_key
, 0);
2795 wret
= insert_ptr(trans
, root
, path
, &disk_key
, right
->start
,
2796 path
->slots
[1] + 1, 1);
2800 btrfs_mark_buffer_dirty(right
);
2801 btrfs_mark_buffer_dirty(l
);
2802 BUG_ON(path
->slots
[0] != slot
);
2804 ret
= btrfs_update_ref(trans
, root
, l
, right
, 0, nritems
);
2808 btrfs_tree_unlock(path
->nodes
[0]);
2809 free_extent_buffer(path
->nodes
[0]);
2810 path
->nodes
[0] = right
;
2811 path
->slots
[0] -= mid
;
2812 path
->slots
[1] += 1;
2814 btrfs_tree_unlock(right
);
2815 free_extent_buffer(right
);
2818 BUG_ON(path
->slots
[0] < 0);
2821 BUG_ON(num_doubles
!= 0);
2829 * This function splits a single item into two items,
2830 * giving 'new_key' to the new item and splitting the
2831 * old one at split_offset (from the start of the item).
2833 * The path may be released by this operation. After
2834 * the split, the path is pointing to the old item. The
2835 * new item is going to be in the same node as the old one.
2837 * Note, the item being split must be smaller enough to live alone on
2838 * a tree block with room for one extra struct btrfs_item
2840 * This allows us to split the item in place, keeping a lock on the
2841 * leaf the entire time.
2843 int btrfs_split_item(struct btrfs_trans_handle
*trans
,
2844 struct btrfs_root
*root
,
2845 struct btrfs_path
*path
,
2846 struct btrfs_key
*new_key
,
2847 unsigned long split_offset
)
2850 struct extent_buffer
*leaf
;
2851 struct btrfs_key orig_key
;
2852 struct btrfs_item
*item
;
2853 struct btrfs_item
*new_item
;
2858 struct btrfs_disk_key disk_key
;
2861 leaf
= path
->nodes
[0];
2862 btrfs_item_key_to_cpu(leaf
, &orig_key
, path
->slots
[0]);
2863 if (btrfs_leaf_free_space(root
, leaf
) >= sizeof(struct btrfs_item
))
2866 item_size
= btrfs_item_size_nr(leaf
, path
->slots
[0]);
2867 btrfs_release_path(root
, path
);
2869 path
->search_for_split
= 1;
2870 path
->keep_locks
= 1;
2872 ret
= btrfs_search_slot(trans
, root
, &orig_key
, path
, 0, 1);
2873 path
->search_for_split
= 0;
2875 /* if our item isn't there or got smaller, return now */
2876 if (ret
!= 0 || item_size
!= btrfs_item_size_nr(path
->nodes
[0],
2878 path
->keep_locks
= 0;
2882 ret
= split_leaf(trans
, root
, &orig_key
, path
,
2883 sizeof(struct btrfs_item
), 1);
2884 path
->keep_locks
= 0;
2887 leaf
= path
->nodes
[0];
2888 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < sizeof(struct btrfs_item
));
2891 item
= btrfs_item_nr(leaf
, path
->slots
[0]);
2892 orig_offset
= btrfs_item_offset(leaf
, item
);
2893 item_size
= btrfs_item_size(leaf
, item
);
2896 buf
= kmalloc(item_size
, GFP_NOFS
);
2897 read_extent_buffer(leaf
, buf
, btrfs_item_ptr_offset(leaf
,
2898 path
->slots
[0]), item_size
);
2899 slot
= path
->slots
[0] + 1;
2900 leaf
= path
->nodes
[0];
2902 nritems
= btrfs_header_nritems(leaf
);
2904 if (slot
!= nritems
) {
2905 /* shift the items */
2906 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ 1),
2907 btrfs_item_nr_offset(slot
),
2908 (nritems
- slot
) * sizeof(struct btrfs_item
));
2912 btrfs_cpu_key_to_disk(&disk_key
, new_key
);
2913 btrfs_set_item_key(leaf
, &disk_key
, slot
);
2915 new_item
= btrfs_item_nr(leaf
, slot
);
2917 btrfs_set_item_offset(leaf
, new_item
, orig_offset
);
2918 btrfs_set_item_size(leaf
, new_item
, item_size
- split_offset
);
2920 btrfs_set_item_offset(leaf
, item
,
2921 orig_offset
+ item_size
- split_offset
);
2922 btrfs_set_item_size(leaf
, item
, split_offset
);
2924 btrfs_set_header_nritems(leaf
, nritems
+ 1);
2926 /* write the data for the start of the original item */
2927 write_extent_buffer(leaf
, buf
,
2928 btrfs_item_ptr_offset(leaf
, path
->slots
[0]),
2931 /* write the data for the new item */
2932 write_extent_buffer(leaf
, buf
+ split_offset
,
2933 btrfs_item_ptr_offset(leaf
, slot
),
2934 item_size
- split_offset
);
2935 btrfs_mark_buffer_dirty(leaf
);
2938 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
2939 btrfs_print_leaf(root
, leaf
);
2947 * make the item pointed to by the path smaller. new_size indicates
2948 * how small to make it, and from_end tells us if we just chop bytes
2949 * off the end of the item or if we shift the item to chop bytes off
2952 int btrfs_truncate_item(struct btrfs_trans_handle
*trans
,
2953 struct btrfs_root
*root
,
2954 struct btrfs_path
*path
,
2955 u32 new_size
, int from_end
)
2960 struct extent_buffer
*leaf
;
2961 struct btrfs_item
*item
;
2963 unsigned int data_end
;
2964 unsigned int old_data_start
;
2965 unsigned int old_size
;
2966 unsigned int size_diff
;
2969 slot_orig
= path
->slots
[0];
2970 leaf
= path
->nodes
[0];
2971 slot
= path
->slots
[0];
2973 old_size
= btrfs_item_size_nr(leaf
, slot
);
2974 if (old_size
== new_size
)
2977 nritems
= btrfs_header_nritems(leaf
);
2978 data_end
= leaf_data_end(root
, leaf
);
2980 old_data_start
= btrfs_item_offset_nr(leaf
, slot
);
2982 size_diff
= old_size
- new_size
;
2985 BUG_ON(slot
>= nritems
);
2988 * item0..itemN ... dataN.offset..dataN.size .. data0.size
2990 /* first correct the data pointers */
2991 for (i
= slot
; i
< nritems
; i
++) {
2993 item
= btrfs_item_nr(leaf
, i
);
2995 if (!leaf
->map_token
) {
2996 map_extent_buffer(leaf
, (unsigned long)item
,
2997 sizeof(struct btrfs_item
),
2998 &leaf
->map_token
, &leaf
->kaddr
,
2999 &leaf
->map_start
, &leaf
->map_len
,
3003 ioff
= btrfs_item_offset(leaf
, item
);
3004 btrfs_set_item_offset(leaf
, item
, ioff
+ size_diff
);
3007 if (leaf
->map_token
) {
3008 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
3009 leaf
->map_token
= NULL
;
3012 /* shift the data */
3014 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3015 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
3016 data_end
, old_data_start
+ new_size
- data_end
);
3018 struct btrfs_disk_key disk_key
;
3021 btrfs_item_key(leaf
, &disk_key
, slot
);
3023 if (btrfs_disk_key_type(&disk_key
) == BTRFS_EXTENT_DATA_KEY
) {
3025 struct btrfs_file_extent_item
*fi
;
3027 fi
= btrfs_item_ptr(leaf
, slot
,
3028 struct btrfs_file_extent_item
);
3029 fi
= (struct btrfs_file_extent_item
*)(
3030 (unsigned long)fi
- size_diff
);
3032 if (btrfs_file_extent_type(leaf
, fi
) ==
3033 BTRFS_FILE_EXTENT_INLINE
) {
3034 ptr
= btrfs_item_ptr_offset(leaf
, slot
);
3035 memmove_extent_buffer(leaf
, ptr
,
3037 offsetof(struct btrfs_file_extent_item
,
3042 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3043 data_end
+ size_diff
, btrfs_leaf_data(leaf
) +
3044 data_end
, old_data_start
- data_end
);
3046 offset
= btrfs_disk_key_offset(&disk_key
);
3047 btrfs_set_disk_key_offset(&disk_key
, offset
+ size_diff
);
3048 btrfs_set_item_key(leaf
, &disk_key
, slot
);
3050 fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3053 item
= btrfs_item_nr(leaf
, slot
);
3054 btrfs_set_item_size(leaf
, item
, new_size
);
3055 btrfs_mark_buffer_dirty(leaf
);
3058 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
3059 btrfs_print_leaf(root
, leaf
);
3066 * make the item pointed to by the path bigger, data_size is the new size.
3068 int btrfs_extend_item(struct btrfs_trans_handle
*trans
,
3069 struct btrfs_root
*root
, struct btrfs_path
*path
,
3075 struct extent_buffer
*leaf
;
3076 struct btrfs_item
*item
;
3078 unsigned int data_end
;
3079 unsigned int old_data
;
3080 unsigned int old_size
;
3083 slot_orig
= path
->slots
[0];
3084 leaf
= path
->nodes
[0];
3086 nritems
= btrfs_header_nritems(leaf
);
3087 data_end
= leaf_data_end(root
, leaf
);
3089 if (btrfs_leaf_free_space(root
, leaf
) < data_size
) {
3090 btrfs_print_leaf(root
, leaf
);
3093 slot
= path
->slots
[0];
3094 old_data
= btrfs_item_end_nr(leaf
, slot
);
3097 if (slot
>= nritems
) {
3098 btrfs_print_leaf(root
, leaf
);
3099 printk("slot %d too large, nritems %d\n", slot
, nritems
);
3104 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3106 /* first correct the data pointers */
3107 for (i
= slot
; i
< nritems
; i
++) {
3109 item
= btrfs_item_nr(leaf
, i
);
3111 if (!leaf
->map_token
) {
3112 map_extent_buffer(leaf
, (unsigned long)item
,
3113 sizeof(struct btrfs_item
),
3114 &leaf
->map_token
, &leaf
->kaddr
,
3115 &leaf
->map_start
, &leaf
->map_len
,
3118 ioff
= btrfs_item_offset(leaf
, item
);
3119 btrfs_set_item_offset(leaf
, item
, ioff
- data_size
);
3122 if (leaf
->map_token
) {
3123 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
3124 leaf
->map_token
= NULL
;
3127 /* shift the data */
3128 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3129 data_end
- data_size
, btrfs_leaf_data(leaf
) +
3130 data_end
, old_data
- data_end
);
3132 data_end
= old_data
;
3133 old_size
= btrfs_item_size_nr(leaf
, slot
);
3134 item
= btrfs_item_nr(leaf
, slot
);
3135 btrfs_set_item_size(leaf
, item
, old_size
+ data_size
);
3136 btrfs_mark_buffer_dirty(leaf
);
3139 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
3140 btrfs_print_leaf(root
, leaf
);
3147 * Given a key and some data, insert items into the tree.
3148 * This does all the path init required, making room in the tree if needed.
3149 * Returns the number of keys that were inserted.
3151 int btrfs_insert_some_items(struct btrfs_trans_handle
*trans
,
3152 struct btrfs_root
*root
,
3153 struct btrfs_path
*path
,
3154 struct btrfs_key
*cpu_key
, u32
*data_size
,
3157 struct extent_buffer
*leaf
;
3158 struct btrfs_item
*item
;
3165 unsigned int data_end
;
3166 struct btrfs_disk_key disk_key
;
3167 struct btrfs_key found_key
;
3169 for (i
= 0; i
< nr
; i
++) {
3170 if (total_size
+ data_size
[i
] + sizeof(struct btrfs_item
) >
3171 BTRFS_LEAF_DATA_SIZE(root
)) {
3175 total_data
+= data_size
[i
];
3176 total_size
+= data_size
[i
] + sizeof(struct btrfs_item
);
3180 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
3186 leaf
= path
->nodes
[0];
3188 nritems
= btrfs_header_nritems(leaf
);
3189 data_end
= leaf_data_end(root
, leaf
);
3191 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
3192 for (i
= nr
; i
>= 0; i
--) {
3193 total_data
-= data_size
[i
];
3194 total_size
-= data_size
[i
] + sizeof(struct btrfs_item
);
3195 if (total_size
< btrfs_leaf_free_space(root
, leaf
))
3201 slot
= path
->slots
[0];
3204 if (slot
!= nritems
) {
3205 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
3207 item
= btrfs_item_nr(leaf
, slot
);
3208 btrfs_item_key_to_cpu(leaf
, &found_key
, slot
);
3210 /* figure out how many keys we can insert in here */
3211 total_data
= data_size
[0];
3212 for (i
= 1; i
< nr
; i
++) {
3213 if (comp_cpu_keys(&found_key
, cpu_key
+ i
) <= 0)
3215 total_data
+= data_size
[i
];
3219 if (old_data
< data_end
) {
3220 btrfs_print_leaf(root
, leaf
);
3221 printk("slot %d old_data %d data_end %d\n",
3222 slot
, old_data
, data_end
);
3226 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3228 /* first correct the data pointers */
3229 WARN_ON(leaf
->map_token
);
3230 for (i
= slot
; i
< nritems
; i
++) {
3233 item
= btrfs_item_nr(leaf
, i
);
3234 if (!leaf
->map_token
) {
3235 map_extent_buffer(leaf
, (unsigned long)item
,
3236 sizeof(struct btrfs_item
),
3237 &leaf
->map_token
, &leaf
->kaddr
,
3238 &leaf
->map_start
, &leaf
->map_len
,
3242 ioff
= btrfs_item_offset(leaf
, item
);
3243 btrfs_set_item_offset(leaf
, item
, ioff
- total_data
);
3245 if (leaf
->map_token
) {
3246 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
3247 leaf
->map_token
= NULL
;
3250 /* shift the items */
3251 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
3252 btrfs_item_nr_offset(slot
),
3253 (nritems
- slot
) * sizeof(struct btrfs_item
));
3255 /* shift the data */
3256 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3257 data_end
- total_data
, btrfs_leaf_data(leaf
) +
3258 data_end
, old_data
- data_end
);
3259 data_end
= old_data
;
3262 * this sucks but it has to be done, if we are inserting at
3263 * the end of the leaf only insert 1 of the items, since we
3264 * have no way of knowing whats on the next leaf and we'd have
3265 * to drop our current locks to figure it out
3270 /* setup the item for the new data */
3271 for (i
= 0; i
< nr
; i
++) {
3272 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
3273 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
3274 item
= btrfs_item_nr(leaf
, slot
+ i
);
3275 btrfs_set_item_offset(leaf
, item
, data_end
- data_size
[i
]);
3276 data_end
-= data_size
[i
];
3277 btrfs_set_item_size(leaf
, item
, data_size
[i
]);
3279 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
3280 btrfs_mark_buffer_dirty(leaf
);
3284 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
3285 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3288 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
3289 btrfs_print_leaf(root
, leaf
);
3299 * Given a key and some data, insert items into the tree.
3300 * This does all the path init required, making room in the tree if needed.
3302 int btrfs_insert_empty_items(struct btrfs_trans_handle
*trans
,
3303 struct btrfs_root
*root
,
3304 struct btrfs_path
*path
,
3305 struct btrfs_key
*cpu_key
, u32
*data_size
,
3308 struct extent_buffer
*leaf
;
3309 struct btrfs_item
*item
;
3317 unsigned int data_end
;
3318 struct btrfs_disk_key disk_key
;
3320 for (i
= 0; i
< nr
; i
++) {
3321 total_data
+= data_size
[i
];
3324 total_size
= total_data
+ (nr
* sizeof(struct btrfs_item
));
3325 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, total_size
, 1);
3331 slot_orig
= path
->slots
[0];
3332 leaf
= path
->nodes
[0];
3334 nritems
= btrfs_header_nritems(leaf
);
3335 data_end
= leaf_data_end(root
, leaf
);
3337 if (btrfs_leaf_free_space(root
, leaf
) < total_size
) {
3338 btrfs_print_leaf(root
, leaf
);
3339 printk("not enough freespace need %u have %d\n",
3340 total_size
, btrfs_leaf_free_space(root
, leaf
));
3344 slot
= path
->slots
[0];
3347 if (slot
!= nritems
) {
3348 unsigned int old_data
= btrfs_item_end_nr(leaf
, slot
);
3350 if (old_data
< data_end
) {
3351 btrfs_print_leaf(root
, leaf
);
3352 printk("slot %d old_data %d data_end %d\n",
3353 slot
, old_data
, data_end
);
3357 * item0..itemN ... dataN.offset..dataN.size .. data0.size
3359 /* first correct the data pointers */
3360 WARN_ON(leaf
->map_token
);
3361 for (i
= slot
; i
< nritems
; i
++) {
3364 item
= btrfs_item_nr(leaf
, i
);
3365 if (!leaf
->map_token
) {
3366 map_extent_buffer(leaf
, (unsigned long)item
,
3367 sizeof(struct btrfs_item
),
3368 &leaf
->map_token
, &leaf
->kaddr
,
3369 &leaf
->map_start
, &leaf
->map_len
,
3373 ioff
= btrfs_item_offset(leaf
, item
);
3374 btrfs_set_item_offset(leaf
, item
, ioff
- total_data
);
3376 if (leaf
->map_token
) {
3377 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
3378 leaf
->map_token
= NULL
;
3381 /* shift the items */
3382 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
+ nr
),
3383 btrfs_item_nr_offset(slot
),
3384 (nritems
- slot
) * sizeof(struct btrfs_item
));
3386 /* shift the data */
3387 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3388 data_end
- total_data
, btrfs_leaf_data(leaf
) +
3389 data_end
, old_data
- data_end
);
3390 data_end
= old_data
;
3393 /* setup the item for the new data */
3394 for (i
= 0; i
< nr
; i
++) {
3395 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
+ i
);
3396 btrfs_set_item_key(leaf
, &disk_key
, slot
+ i
);
3397 item
= btrfs_item_nr(leaf
, slot
+ i
);
3398 btrfs_set_item_offset(leaf
, item
, data_end
- data_size
[i
]);
3399 data_end
-= data_size
[i
];
3400 btrfs_set_item_size(leaf
, item
, data_size
[i
]);
3402 btrfs_set_header_nritems(leaf
, nritems
+ nr
);
3403 btrfs_mark_buffer_dirty(leaf
);
3407 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
3408 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
3411 if (btrfs_leaf_free_space(root
, leaf
) < 0) {
3412 btrfs_print_leaf(root
, leaf
);
3420 * Given a key and some data, insert an item into the tree.
3421 * This does all the path init required, making room in the tree if needed.
3423 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
3424 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
3428 struct btrfs_path
*path
;
3429 struct extent_buffer
*leaf
;
3432 path
= btrfs_alloc_path();
3434 ret
= btrfs_insert_empty_item(trans
, root
, path
, cpu_key
, data_size
);
3436 leaf
= path
->nodes
[0];
3437 ptr
= btrfs_item_ptr_offset(leaf
, path
->slots
[0]);
3438 write_extent_buffer(leaf
, data
, ptr
, data_size
);
3439 btrfs_mark_buffer_dirty(leaf
);
3441 btrfs_free_path(path
);
3446 * delete the pointer from a given node.
3448 * the tree should have been previously balanced so the deletion does not
3451 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3452 struct btrfs_path
*path
, int level
, int slot
)
3454 struct extent_buffer
*parent
= path
->nodes
[level
];
3459 nritems
= btrfs_header_nritems(parent
);
3460 if (slot
!= nritems
-1) {
3461 memmove_extent_buffer(parent
,
3462 btrfs_node_key_ptr_offset(slot
),
3463 btrfs_node_key_ptr_offset(slot
+ 1),
3464 sizeof(struct btrfs_key_ptr
) *
3465 (nritems
- slot
- 1));
3468 btrfs_set_header_nritems(parent
, nritems
);
3469 if (nritems
== 0 && parent
== root
->node
) {
3470 BUG_ON(btrfs_header_level(root
->node
) != 1);
3471 /* just turn the root into a leaf and break */
3472 btrfs_set_header_level(root
->node
, 0);
3473 } else if (slot
== 0) {
3474 struct btrfs_disk_key disk_key
;
3476 btrfs_node_key(parent
, &disk_key
, 0);
3477 wret
= fixup_low_keys(trans
, root
, path
, &disk_key
, level
+ 1);
3481 btrfs_mark_buffer_dirty(parent
);
3486 * a helper function to delete the leaf pointed to by path->slots[1] and
3487 * path->nodes[1]. bytenr is the node block pointer, but since the callers
3488 * already know it, it is faster to have them pass it down than to
3489 * read it out of the node again.
3491 * This deletes the pointer in path->nodes[1] and frees the leaf
3492 * block extent. zero is returned if it all worked out, < 0 otherwise.
3494 * The path must have already been setup for deleting the leaf, including
3495 * all the proper balancing. path->nodes[1] must be locked.
3497 noinline
int btrfs_del_leaf(struct btrfs_trans_handle
*trans
,
3498 struct btrfs_root
*root
,
3499 struct btrfs_path
*path
, u64 bytenr
)
3502 u64 root_gen
= btrfs_header_generation(path
->nodes
[1]);
3504 ret
= del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
3508 ret
= btrfs_free_extent(trans
, root
, bytenr
,
3509 btrfs_level_size(root
, 0),
3510 path
->nodes
[1]->start
,
3511 btrfs_header_owner(path
->nodes
[1]),
3516 * delete the item at the leaf level in path. If that empties
3517 * the leaf, remove it from the tree
3519 int btrfs_del_items(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
3520 struct btrfs_path
*path
, int slot
, int nr
)
3522 struct extent_buffer
*leaf
;
3523 struct btrfs_item
*item
;
3531 leaf
= path
->nodes
[0];
3532 last_off
= btrfs_item_offset_nr(leaf
, slot
+ nr
- 1);
3534 for (i
= 0; i
< nr
; i
++)
3535 dsize
+= btrfs_item_size_nr(leaf
, slot
+ i
);
3537 nritems
= btrfs_header_nritems(leaf
);
3539 if (slot
+ nr
!= nritems
) {
3540 int data_end
= leaf_data_end(root
, leaf
);
3542 memmove_extent_buffer(leaf
, btrfs_leaf_data(leaf
) +
3544 btrfs_leaf_data(leaf
) + data_end
,
3545 last_off
- data_end
);
3547 for (i
= slot
+ nr
; i
< nritems
; i
++) {
3550 item
= btrfs_item_nr(leaf
, i
);
3551 if (!leaf
->map_token
) {
3552 map_extent_buffer(leaf
, (unsigned long)item
,
3553 sizeof(struct btrfs_item
),
3554 &leaf
->map_token
, &leaf
->kaddr
,
3555 &leaf
->map_start
, &leaf
->map_len
,
3558 ioff
= btrfs_item_offset(leaf
, item
);
3559 btrfs_set_item_offset(leaf
, item
, ioff
+ dsize
);
3562 if (leaf
->map_token
) {
3563 unmap_extent_buffer(leaf
, leaf
->map_token
, KM_USER1
);
3564 leaf
->map_token
= NULL
;
3567 memmove_extent_buffer(leaf
, btrfs_item_nr_offset(slot
),
3568 btrfs_item_nr_offset(slot
+ nr
),
3569 sizeof(struct btrfs_item
) *
3570 (nritems
- slot
- nr
));
3572 btrfs_set_header_nritems(leaf
, nritems
- nr
);
3575 /* delete the leaf if we've emptied it */
3577 if (leaf
== root
->node
) {
3578 btrfs_set_header_level(leaf
, 0);
3580 ret
= btrfs_del_leaf(trans
, root
, path
, leaf
->start
);
3584 int used
= leaf_space_used(leaf
, 0, nritems
);
3586 struct btrfs_disk_key disk_key
;
3588 btrfs_item_key(leaf
, &disk_key
, 0);
3589 wret
= fixup_low_keys(trans
, root
, path
,
3595 /* delete the leaf if it is mostly empty */
3596 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 4) {
3597 /* push_leaf_left fixes the path.
3598 * make sure the path still points to our leaf
3599 * for possible call to del_ptr below
3601 slot
= path
->slots
[1];
3602 extent_buffer_get(leaf
);
3604 wret
= push_leaf_left(trans
, root
, path
, 1, 1);
3605 if (wret
< 0 && wret
!= -ENOSPC
)
3608 if (path
->nodes
[0] == leaf
&&
3609 btrfs_header_nritems(leaf
)) {
3610 wret
= push_leaf_right(trans
, root
, path
, 1, 1);
3611 if (wret
< 0 && wret
!= -ENOSPC
)
3615 if (btrfs_header_nritems(leaf
) == 0) {
3616 path
->slots
[1] = slot
;
3617 ret
= btrfs_del_leaf(trans
, root
, path
, leaf
->start
);
3619 free_extent_buffer(leaf
);
3621 /* if we're still in the path, make sure
3622 * we're dirty. Otherwise, one of the
3623 * push_leaf functions must have already
3624 * dirtied this buffer
3626 if (path
->nodes
[0] == leaf
)
3627 btrfs_mark_buffer_dirty(leaf
);
3628 free_extent_buffer(leaf
);
3631 btrfs_mark_buffer_dirty(leaf
);
3638 * search the tree again to find a leaf with lesser keys
3639 * returns 0 if it found something or 1 if there are no lesser leaves.
3640 * returns < 0 on io errors.
3642 * This may release the path, and so you may lose any locks held at the
3645 int btrfs_prev_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
3647 struct btrfs_key key
;
3648 struct btrfs_disk_key found_key
;
3651 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, 0);
3655 else if (key
.type
> 0)
3657 else if (key
.objectid
> 0)
3662 btrfs_release_path(root
, path
);
3663 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3666 btrfs_item_key(path
->nodes
[0], &found_key
, 0);
3667 ret
= comp_keys(&found_key
, &key
);
3674 * A helper function to walk down the tree starting at min_key, and looking
3675 * for nodes or leaves that are either in cache or have a minimum
3676 * transaction id. This is used by the btree defrag code, and tree logging
3678 * This does not cow, but it does stuff the starting key it finds back
3679 * into min_key, so you can call btrfs_search_slot with cow=1 on the
3680 * key and get a writable path.
3682 * This does lock as it descends, and path->keep_locks should be set
3683 * to 1 by the caller.
3685 * This honors path->lowest_level to prevent descent past a given level
3688 * min_trans indicates the oldest transaction that you are interested
3689 * in walking through. Any nodes or leaves older than min_trans are
3690 * skipped over (without reading them).
3692 * returns zero if something useful was found, < 0 on error and 1 if there
3693 * was nothing in the tree that matched the search criteria.
3695 int btrfs_search_forward(struct btrfs_root
*root
, struct btrfs_key
*min_key
,
3696 struct btrfs_key
*max_key
,
3697 struct btrfs_path
*path
, int cache_only
,
3700 struct extent_buffer
*cur
;
3701 struct btrfs_key found_key
;
3708 WARN_ON(!path
->keep_locks
);
3710 cur
= btrfs_lock_root_node(root
);
3711 level
= btrfs_header_level(cur
);
3712 WARN_ON(path
->nodes
[level
]);
3713 path
->nodes
[level
] = cur
;
3714 path
->locks
[level
] = 1;
3716 if (btrfs_header_generation(cur
) < min_trans
) {
3721 nritems
= btrfs_header_nritems(cur
);
3722 level
= btrfs_header_level(cur
);
3723 sret
= bin_search(cur
, min_key
, level
, &slot
);
3725 /* at the lowest level, we're done, setup the path and exit */
3726 if (level
== path
->lowest_level
) {
3727 if (slot
>= nritems
)
3730 path
->slots
[level
] = slot
;
3731 btrfs_item_key_to_cpu(cur
, &found_key
, slot
);
3734 if (sret
&& slot
> 0)
3737 * check this node pointer against the cache_only and
3738 * min_trans parameters. If it isn't in cache or is too
3739 * old, skip to the next one.
3741 while(slot
< nritems
) {
3744 struct extent_buffer
*tmp
;
3745 struct btrfs_disk_key disk_key
;
3747 blockptr
= btrfs_node_blockptr(cur
, slot
);
3748 gen
= btrfs_node_ptr_generation(cur
, slot
);
3749 if (gen
< min_trans
) {
3757 btrfs_node_key(cur
, &disk_key
, slot
);
3758 if (comp_keys(&disk_key
, max_key
) >= 0) {
3764 tmp
= btrfs_find_tree_block(root
, blockptr
,
3765 btrfs_level_size(root
, level
- 1));
3767 if (tmp
&& btrfs_buffer_uptodate(tmp
, gen
)) {
3768 free_extent_buffer(tmp
);
3772 free_extent_buffer(tmp
);
3777 * we didn't find a candidate key in this node, walk forward
3778 * and find another one
3780 if (slot
>= nritems
) {
3781 path
->slots
[level
] = slot
;
3782 sret
= btrfs_find_next_key(root
, path
, min_key
, level
,
3783 cache_only
, min_trans
);
3785 btrfs_release_path(root
, path
);
3791 /* save our key for returning back */
3792 btrfs_node_key_to_cpu(cur
, &found_key
, slot
);
3793 path
->slots
[level
] = slot
;
3794 if (level
== path
->lowest_level
) {
3796 unlock_up(path
, level
, 1);
3799 cur
= read_node_slot(root
, cur
, slot
);
3801 btrfs_tree_lock(cur
);
3802 path
->locks
[level
- 1] = 1;
3803 path
->nodes
[level
- 1] = cur
;
3804 unlock_up(path
, level
, 1);
3808 memcpy(min_key
, &found_key
, sizeof(found_key
));
3813 * this is similar to btrfs_next_leaf, but does not try to preserve
3814 * and fixup the path. It looks for and returns the next key in the
3815 * tree based on the current path and the cache_only and min_trans
3818 * 0 is returned if another key is found, < 0 if there are any errors
3819 * and 1 is returned if there are no higher keys in the tree
3821 * path->keep_locks should be set to 1 on the search made before
3822 * calling this function.
3824 int btrfs_find_next_key(struct btrfs_root
*root
, struct btrfs_path
*path
,
3825 struct btrfs_key
*key
, int lowest_level
,
3826 int cache_only
, u64 min_trans
)
3828 int level
= lowest_level
;
3830 struct extent_buffer
*c
;
3832 WARN_ON(!path
->keep_locks
);
3833 while(level
< BTRFS_MAX_LEVEL
) {
3834 if (!path
->nodes
[level
])
3837 slot
= path
->slots
[level
] + 1;
3838 c
= path
->nodes
[level
];
3840 if (slot
>= btrfs_header_nritems(c
)) {
3842 if (level
== BTRFS_MAX_LEVEL
) {
3848 btrfs_item_key_to_cpu(c
, key
, slot
);
3850 u64 blockptr
= btrfs_node_blockptr(c
, slot
);
3851 u64 gen
= btrfs_node_ptr_generation(c
, slot
);
3854 struct extent_buffer
*cur
;
3855 cur
= btrfs_find_tree_block(root
, blockptr
,
3856 btrfs_level_size(root
, level
- 1));
3857 if (!cur
|| !btrfs_buffer_uptodate(cur
, gen
)) {
3860 free_extent_buffer(cur
);
3863 free_extent_buffer(cur
);
3865 if (gen
< min_trans
) {
3869 btrfs_node_key_to_cpu(c
, key
, slot
);
3877 * search the tree again to find a leaf with greater keys
3878 * returns 0 if it found something or 1 if there are no greater leaves.
3879 * returns < 0 on io errors.
3881 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
3885 struct extent_buffer
*c
;
3886 struct extent_buffer
*next
= NULL
;
3887 struct btrfs_key key
;
3891 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3896 btrfs_item_key_to_cpu(path
->nodes
[0], &key
, nritems
- 1);
3898 btrfs_release_path(root
, path
);
3899 path
->keep_locks
= 1;
3900 ret
= btrfs_search_slot(NULL
, root
, &key
, path
, 0, 0);
3901 path
->keep_locks
= 0;
3906 nritems
= btrfs_header_nritems(path
->nodes
[0]);
3908 * by releasing the path above we dropped all our locks. A balance
3909 * could have added more items next to the key that used to be
3910 * at the very end of the block. So, check again here and
3911 * advance the path if there are now more items available.
3913 if (nritems
> 0 && path
->slots
[0] < nritems
- 1) {
3918 while(level
< BTRFS_MAX_LEVEL
) {
3919 if (!path
->nodes
[level
])
3922 slot
= path
->slots
[level
] + 1;
3923 c
= path
->nodes
[level
];
3924 if (slot
>= btrfs_header_nritems(c
)) {
3926 if (level
== BTRFS_MAX_LEVEL
) {
3933 btrfs_tree_unlock(next
);
3934 free_extent_buffer(next
);
3937 if (level
== 1 && (path
->locks
[1] || path
->skip_locking
) &&
3939 reada_for_search(root
, path
, level
, slot
, 0);
3941 next
= read_node_slot(root
, c
, slot
);
3942 if (!path
->skip_locking
) {
3943 WARN_ON(!btrfs_tree_locked(c
));
3944 btrfs_tree_lock(next
);
3948 path
->slots
[level
] = slot
;
3951 c
= path
->nodes
[level
];
3952 if (path
->locks
[level
])
3953 btrfs_tree_unlock(c
);
3954 free_extent_buffer(c
);
3955 path
->nodes
[level
] = next
;
3956 path
->slots
[level
] = 0;
3957 if (!path
->skip_locking
)
3958 path
->locks
[level
] = 1;
3961 if (level
== 1 && path
->locks
[1] && path
->reada
)
3962 reada_for_search(root
, path
, level
, slot
, 0);
3963 next
= read_node_slot(root
, next
, 0);
3964 if (!path
->skip_locking
) {
3965 WARN_ON(!btrfs_tree_locked(path
->nodes
[level
]));
3966 btrfs_tree_lock(next
);
3970 unlock_up(path
, 0, 1);
3975 * this uses btrfs_prev_leaf to walk backwards in the tree, and keeps
3976 * searching until it gets past min_objectid or finds an item of 'type'
3978 * returns 0 if something is found, 1 if nothing was found and < 0 on error
3980 int btrfs_previous_item(struct btrfs_root
*root
,
3981 struct btrfs_path
*path
, u64 min_objectid
,
3984 struct btrfs_key found_key
;
3985 struct extent_buffer
*leaf
;
3990 if (path
->slots
[0] == 0) {
3991 ret
= btrfs_prev_leaf(root
, path
);
3997 leaf
= path
->nodes
[0];
3998 nritems
= btrfs_header_nritems(leaf
);
4001 if (path
->slots
[0] == nritems
)
4004 btrfs_item_key_to_cpu(leaf
, &found_key
, path
->slots
[0]);
4005 if (found_key
.type
== type
)
4007 if (found_key
.objectid
< min_objectid
)
4009 if (found_key
.objectid
== min_objectid
&&
4010 found_key
.type
< type
)