3 #include "kerncompat.h"
4 #include "radix-tree.h"
7 #include "print-tree.h"
9 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
10 *root
, struct btrfs_path
*path
, int level
);
11 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
12 *root
, struct btrfs_path
*path
, int data_size
);
13 static int push_node_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
14 *root
, struct btrfs_buffer
*dst
, struct btrfs_buffer
16 static int balance_node_right(struct btrfs_trans_handle
*trans
, struct
17 btrfs_root
*root
, struct btrfs_buffer
*dst_buf
,
18 struct btrfs_buffer
*src_buf
);
19 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
20 struct btrfs_path
*path
, int level
, int slot
);
22 inline void btrfs_init_path(struct btrfs_path
*p
)
24 memset(p
, 0, sizeof(*p
));
27 void btrfs_release_path(struct btrfs_root
*root
, struct btrfs_path
*p
)
30 for (i
= 0; i
< BTRFS_MAX_LEVEL
; i
++) {
33 btrfs_block_release(root
, p
->nodes
[i
]);
35 memset(p
, 0, sizeof(*p
));
38 static int btrfs_cow_block(struct btrfs_trans_handle
*trans
, struct btrfs_root
39 *root
, struct btrfs_buffer
*buf
, struct btrfs_buffer
40 *parent
, int parent_slot
, struct btrfs_buffer
43 struct btrfs_buffer
*cow
;
45 if (!list_empty(&buf
->dirty
)) {
49 cow
= btrfs_alloc_free_block(trans
, root
);
50 memcpy(&cow
->node
, &buf
->node
, root
->blocksize
);
51 btrfs_set_header_blocknr(&cow
->node
.header
, cow
->blocknr
);
53 btrfs_inc_ref(trans
, root
, buf
);
54 if (buf
== root
->node
) {
57 if (buf
!= root
->commit_root
)
58 btrfs_free_extent(trans
, root
, buf
->blocknr
, 1, 1);
59 btrfs_block_release(root
, buf
);
61 btrfs_set_node_blockptr(&parent
->node
, parent_slot
,
63 BUG_ON(list_empty(&parent
->dirty
));
64 btrfs_free_extent(trans
, root
, buf
->blocknr
, 1, 1);
66 btrfs_block_release(root
, buf
);
71 * The leaf data grows from end-to-front in the node.
72 * this returns the address of the start of the last item,
73 * which is the stop of the leaf data stack
75 static inline unsigned int leaf_data_end(struct btrfs_root
*root
,
76 struct btrfs_leaf
*leaf
)
78 u32 nr
= btrfs_header_nritems(&leaf
->header
);
80 return BTRFS_LEAF_DATA_SIZE(root
);
81 return btrfs_item_offset(leaf
->items
+ nr
- 1);
85 * The space between the end of the leaf items and
86 * the start of the leaf data. IOW, how much room
87 * the leaf has left for both items and data
89 int btrfs_leaf_free_space(struct btrfs_root
*root
, struct btrfs_leaf
*leaf
)
91 int data_end
= leaf_data_end(root
, leaf
);
92 int nritems
= btrfs_header_nritems(&leaf
->header
);
93 char *items_end
= (char *)(leaf
->items
+ nritems
+ 1);
94 return (char *)(btrfs_leaf_data(leaf
) + data_end
) - (char *)items_end
;
98 * compare two keys in a memcmp fashion
100 static int comp_keys(struct btrfs_disk_key
*disk
, struct btrfs_key
*k2
)
104 btrfs_disk_key_to_cpu(&k1
, disk
);
106 if (k1
.objectid
> k2
->objectid
)
108 if (k1
.objectid
< k2
->objectid
)
110 if (k1
.flags
> k2
->flags
)
112 if (k1
.flags
< k2
->flags
)
114 if (k1
.offset
> k2
->offset
)
116 if (k1
.offset
< k2
->offset
)
121 static int check_node(struct btrfs_root
*root
, struct btrfs_path
*path
,
125 struct btrfs_node
*parent
= NULL
;
126 struct btrfs_node
*node
= &path
->nodes
[level
]->node
;
128 u32 nritems
= btrfs_header_nritems(&node
->header
);
130 if (path
->nodes
[level
+ 1])
131 parent
= &path
->nodes
[level
+ 1]->node
;
132 parent_slot
= path
->slots
[level
+ 1];
133 BUG_ON(nritems
== 0);
135 struct btrfs_disk_key
*parent_key
;
136 parent_key
= &parent
->ptrs
[parent_slot
].key
;
137 BUG_ON(memcmp(parent_key
, &node
->ptrs
[0].key
,
138 sizeof(struct btrfs_disk_key
)));
139 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
140 btrfs_header_blocknr(&node
->header
));
142 BUG_ON(nritems
> BTRFS_NODEPTRS_PER_BLOCK(root
));
143 for (i
= 0; nritems
> 1 && i
< nritems
- 2; i
++) {
144 struct btrfs_key cpukey
;
145 btrfs_disk_key_to_cpu(&cpukey
, &node
->ptrs
[i
+ 1].key
);
146 BUG_ON(comp_keys(&node
->ptrs
[i
].key
, &cpukey
) >= 0);
151 static int check_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
,
155 struct btrfs_leaf
*leaf
= &path
->nodes
[level
]->leaf
;
156 struct btrfs_node
*parent
= NULL
;
158 u32 nritems
= btrfs_header_nritems(&leaf
->header
);
160 if (path
->nodes
[level
+ 1])
161 parent
= &path
->nodes
[level
+ 1]->node
;
162 parent_slot
= path
->slots
[level
+ 1];
163 BUG_ON(btrfs_leaf_free_space(root
, leaf
) < 0);
169 struct btrfs_disk_key
*parent_key
;
170 parent_key
= &parent
->ptrs
[parent_slot
].key
;
171 BUG_ON(memcmp(parent_key
, &leaf
->items
[0].key
,
172 sizeof(struct btrfs_disk_key
)));
173 BUG_ON(btrfs_node_blockptr(parent
, parent_slot
) !=
174 btrfs_header_blocknr(&leaf
->header
));
176 for (i
= 0; nritems
> 1 && i
< nritems
- 2; i
++) {
177 struct btrfs_key cpukey
;
178 btrfs_disk_key_to_cpu(&cpukey
, &leaf
->items
[i
+ 1].key
);
179 BUG_ON(comp_keys(&leaf
->items
[i
].key
,
181 BUG_ON(btrfs_item_offset(leaf
->items
+ i
) !=
182 btrfs_item_end(leaf
->items
+ i
+ 1));
184 BUG_ON(btrfs_item_offset(leaf
->items
+ i
) +
185 btrfs_item_size(leaf
->items
+ i
) !=
186 BTRFS_LEAF_DATA_SIZE(root
));
192 static int check_block(struct btrfs_root
*root
, struct btrfs_path
*path
,
196 return check_leaf(root
, path
, level
);
197 return check_node(root
, path
, level
);
201 * search for key in the array p. items p are item_size apart
202 * and there are 'max' items in p
203 * the slot in the array is returned via slot, and it points to
204 * the place where you would insert key if it is not found in
207 * slot may point to max if the key is bigger than all of the keys
209 static int generic_bin_search(char *p
, int item_size
, struct btrfs_key
*key
,
216 struct btrfs_disk_key
*tmp
;
219 mid
= (low
+ high
) / 2;
220 tmp
= (struct btrfs_disk_key
*)(p
+ mid
* item_size
);
221 ret
= comp_keys(tmp
, key
);
237 * simple bin_search frontend that does the right thing for
240 static int bin_search(struct btrfs_node
*c
, struct btrfs_key
*key
, int *slot
)
242 if (btrfs_is_leaf(c
)) {
243 struct btrfs_leaf
*l
= (struct btrfs_leaf
*)c
;
244 return generic_bin_search((void *)l
->items
,
245 sizeof(struct btrfs_item
),
246 key
, btrfs_header_nritems(&c
->header
),
249 return generic_bin_search((void *)c
->ptrs
,
250 sizeof(struct btrfs_key_ptr
),
251 key
, btrfs_header_nritems(&c
->header
),
257 static struct btrfs_buffer
*read_node_slot(struct btrfs_root
*root
,
258 struct btrfs_buffer
*parent_buf
,
261 struct btrfs_node
*node
= &parent_buf
->node
;
264 if (slot
>= btrfs_header_nritems(&node
->header
))
266 return read_tree_block(root
, btrfs_node_blockptr(node
, slot
));
269 static int balance_level(struct btrfs_trans_handle
*trans
, struct btrfs_root
270 *root
, struct btrfs_path
*path
, int level
)
272 struct btrfs_buffer
*right_buf
;
273 struct btrfs_buffer
*mid_buf
;
274 struct btrfs_buffer
*left_buf
;
275 struct btrfs_buffer
*parent_buf
= NULL
;
276 struct btrfs_node
*right
= NULL
;
277 struct btrfs_node
*mid
;
278 struct btrfs_node
*left
= NULL
;
279 struct btrfs_node
*parent
= NULL
;
283 int orig_slot
= path
->slots
[level
];
289 mid_buf
= path
->nodes
[level
];
290 mid
= &mid_buf
->node
;
291 orig_ptr
= btrfs_node_blockptr(mid
, orig_slot
);
293 if (level
< BTRFS_MAX_LEVEL
- 1)
294 parent_buf
= path
->nodes
[level
+ 1];
295 pslot
= path
->slots
[level
+ 1];
298 struct btrfs_buffer
*child
;
299 u64 blocknr
= mid_buf
->blocknr
;
301 if (btrfs_header_nritems(&mid
->header
) != 1)
304 /* promote the child to a root */
305 child
= read_node_slot(root
, mid_buf
, 0);
308 path
->nodes
[level
] = NULL
;
309 /* once for the path */
310 btrfs_block_release(root
, mid_buf
);
311 /* once for the root ptr */
312 btrfs_block_release(root
, mid_buf
);
313 clean_tree_block(trans
, root
, mid_buf
);
314 return btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
316 parent
= &parent_buf
->node
;
318 if (btrfs_header_nritems(&mid
->header
) >
319 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
322 left_buf
= read_node_slot(root
, parent_buf
, pslot
- 1);
323 right_buf
= read_node_slot(root
, parent_buf
, pslot
+ 1);
325 /* first, try to make some room in the middle buffer */
327 btrfs_cow_block(trans
, root
, left_buf
, parent_buf
, pslot
- 1,
329 left
= &left_buf
->node
;
330 orig_slot
+= btrfs_header_nritems(&left
->header
);
331 wret
= push_node_left(trans
, root
, left_buf
, mid_buf
);
337 * then try to empty the right most buffer into the middle
340 btrfs_cow_block(trans
, root
, right_buf
, parent_buf
, pslot
+ 1,
342 right
= &right_buf
->node
;
343 wret
= push_node_left(trans
, root
, mid_buf
, right_buf
);
346 if (btrfs_header_nritems(&right
->header
) == 0) {
347 u64 blocknr
= right_buf
->blocknr
;
348 btrfs_block_release(root
, right_buf
);
349 clean_tree_block(trans
, root
, right_buf
);
352 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
+
356 wret
= btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
360 memcpy(&parent
->ptrs
[pslot
+ 1].key
,
362 sizeof(struct btrfs_disk_key
));
363 BUG_ON(list_empty(&parent_buf
->dirty
));
366 if (btrfs_header_nritems(&mid
->header
) == 1) {
368 * we're not allowed to leave a node with one item in the
369 * tree during a delete. A deletion from lower in the tree
370 * could try to delete the only pointer in this node.
371 * So, pull some keys from the left.
372 * There has to be a left pointer at this point because
373 * otherwise we would have pulled some pointers from the
377 wret
= balance_node_right(trans
, root
, mid_buf
, left_buf
);
382 if (btrfs_header_nritems(&mid
->header
) == 0) {
383 /* we've managed to empty the middle node, drop it */
384 u64 blocknr
= mid_buf
->blocknr
;
385 btrfs_block_release(root
, mid_buf
);
386 clean_tree_block(trans
, root
, mid_buf
);
389 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
);
392 wret
= btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
396 /* update the parent key to reflect our changes */
397 memcpy(&parent
->ptrs
[pslot
].key
, &mid
->ptrs
[0].key
,
398 sizeof(struct btrfs_disk_key
));
399 BUG_ON(list_empty(&parent_buf
->dirty
));
402 /* update the path */
404 if (btrfs_header_nritems(&left
->header
) > orig_slot
) {
405 left_buf
->count
++; // released below
406 path
->nodes
[level
] = left_buf
;
407 path
->slots
[level
+ 1] -= 1;
408 path
->slots
[level
] = orig_slot
;
410 btrfs_block_release(root
, mid_buf
);
412 orig_slot
-= btrfs_header_nritems(&left
->header
);
413 path
->slots
[level
] = orig_slot
;
416 /* double check we haven't messed things up */
417 check_block(root
, path
, level
);
418 if (orig_ptr
!= btrfs_node_blockptr(&path
->nodes
[level
]->node
,
423 btrfs_block_release(root
, right_buf
);
425 btrfs_block_release(root
, left_buf
);
430 * look for key in the tree. path is filled in with nodes along the way
431 * if key is found, we return zero and you can find the item in the leaf
432 * level of the path (level 0)
434 * If the key isn't found, the path points to the slot where it should
435 * be inserted, and 1 is returned. If there are other errors during the
436 * search a negative error number is returned.
438 * if ins_len > 0, nodes and leaves will be split as we walk down the
439 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
442 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
443 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
446 struct btrfs_buffer
*b
;
447 struct btrfs_buffer
*cow_buf
;
448 struct btrfs_node
*c
;
457 level
= btrfs_header_level(&b
->node
.header
);
460 wret
= btrfs_cow_block(trans
, root
, b
, p
->nodes
[level
+
461 1], p
->slots
[level
+ 1],
465 BUG_ON(!cow
&& ins_len
);
468 ret
= check_block(root
, p
, level
);
471 ret
= bin_search(c
, key
, &slot
);
472 if (!btrfs_is_leaf(c
)) {
475 p
->slots
[level
] = slot
;
476 if (ins_len
> 0 && btrfs_header_nritems(&c
->header
) ==
477 BTRFS_NODEPTRS_PER_BLOCK(root
)) {
478 int sret
= split_node(trans
, root
, p
, level
);
484 slot
= p
->slots
[level
];
485 } else if (ins_len
< 0) {
486 int sret
= balance_level(trans
, root
, p
,
494 slot
= p
->slots
[level
];
495 BUG_ON(btrfs_header_nritems(&c
->header
) == 1);
497 b
= read_tree_block(root
, btrfs_node_blockptr(c
, slot
));
499 struct btrfs_leaf
*l
= (struct btrfs_leaf
*)c
;
500 p
->slots
[level
] = slot
;
501 if (ins_len
> 0 && btrfs_leaf_free_space(root
, l
) <
502 sizeof(struct btrfs_item
) + ins_len
) {
503 int sret
= split_leaf(trans
, root
, p
, ins_len
);
508 BUG_ON(root
->node
->count
== 1);
512 BUG_ON(root
->node
->count
== 1);
517 * adjust the pointers going up the tree, starting at level
518 * making sure the right key of each node is points to 'key'.
519 * This is used after shifting pointers to the left, so it stops
520 * fixing up pointers when a given leaf/node is not in slot 0 of the
523 * If this fails to write a tree block, it returns -1, but continues
524 * fixing up the blocks in ram so the tree is consistent.
526 static int fixup_low_keys(struct btrfs_trans_handle
*trans
, struct btrfs_root
527 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
532 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
533 struct btrfs_node
*t
;
534 int tslot
= path
->slots
[i
];
537 t
= &path
->nodes
[i
]->node
;
538 memcpy(&t
->ptrs
[tslot
].key
, key
, sizeof(*key
));
539 BUG_ON(list_empty(&path
->nodes
[i
]->dirty
));
547 * try to push data from one node into the next node left in the
550 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
551 * error, and > 0 if there was no room in the left hand block.
553 static int push_node_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
554 *root
, struct btrfs_buffer
*dst_buf
, struct
555 btrfs_buffer
*src_buf
)
557 struct btrfs_node
*src
= &src_buf
->node
;
558 struct btrfs_node
*dst
= &dst_buf
->node
;
564 src_nritems
= btrfs_header_nritems(&src
->header
);
565 dst_nritems
= btrfs_header_nritems(&dst
->header
);
566 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
567 if (push_items
<= 0) {
571 if (src_nritems
< push_items
)
572 push_items
= src_nritems
;
574 memcpy(dst
->ptrs
+ dst_nritems
, src
->ptrs
,
575 push_items
* sizeof(struct btrfs_key_ptr
));
576 if (push_items
< src_nritems
) {
577 memmove(src
->ptrs
, src
->ptrs
+ push_items
,
578 (src_nritems
- push_items
) *
579 sizeof(struct btrfs_key_ptr
));
581 btrfs_set_header_nritems(&src
->header
, src_nritems
- push_items
);
582 btrfs_set_header_nritems(&dst
->header
, dst_nritems
+ push_items
);
583 BUG_ON(list_empty(&src_buf
->dirty
));
584 BUG_ON(list_empty(&dst_buf
->dirty
));
589 * try to push data from one node into the next node right in the
592 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
593 * error, and > 0 if there was no room in the right hand block.
595 * this will only push up to 1/2 the contents of the left node over
597 static int balance_node_right(struct btrfs_trans_handle
*trans
, struct
598 btrfs_root
*root
, struct btrfs_buffer
*dst_buf
,
599 struct btrfs_buffer
*src_buf
)
601 struct btrfs_node
*src
= &src_buf
->node
;
602 struct btrfs_node
*dst
= &dst_buf
->node
;
609 src_nritems
= btrfs_header_nritems(&src
->header
);
610 dst_nritems
= btrfs_header_nritems(&dst
->header
);
611 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
612 if (push_items
<= 0) {
616 max_push
= src_nritems
/ 2 + 1;
617 /* don't try to empty the node */
618 if (max_push
> src_nritems
)
620 if (max_push
< push_items
)
621 push_items
= max_push
;
623 memmove(dst
->ptrs
+ push_items
, dst
->ptrs
,
624 dst_nritems
* sizeof(struct btrfs_key_ptr
));
625 memcpy(dst
->ptrs
, src
->ptrs
+ src_nritems
- push_items
,
626 push_items
* sizeof(struct btrfs_key_ptr
));
628 btrfs_set_header_nritems(&src
->header
, src_nritems
- push_items
);
629 btrfs_set_header_nritems(&dst
->header
, dst_nritems
+ push_items
);
631 BUG_ON(list_empty(&src_buf
->dirty
));
632 BUG_ON(list_empty(&dst_buf
->dirty
));
637 * helper function to insert a new root level in the tree.
638 * A new node is allocated, and a single item is inserted to
639 * point to the existing root
641 * returns zero on success or < 0 on failure.
643 static int insert_new_root(struct btrfs_trans_handle
*trans
, struct btrfs_root
644 *root
, struct btrfs_path
*path
, int level
)
646 struct btrfs_buffer
*t
;
647 struct btrfs_node
*lower
;
648 struct btrfs_node
*c
;
649 struct btrfs_disk_key
*lower_key
;
651 BUG_ON(path
->nodes
[level
]);
652 BUG_ON(path
->nodes
[level
-1] != root
->node
);
654 t
= btrfs_alloc_free_block(trans
, root
);
656 memset(c
, 0, root
->blocksize
);
657 btrfs_set_header_nritems(&c
->header
, 1);
658 btrfs_set_header_level(&c
->header
, level
);
659 btrfs_set_header_blocknr(&c
->header
, t
->blocknr
);
660 btrfs_set_header_parentid(&c
->header
,
661 btrfs_header_parentid(&root
->node
->node
.header
));
662 lower
= &path
->nodes
[level
-1]->node
;
663 if (btrfs_is_leaf(lower
))
664 lower_key
= &((struct btrfs_leaf
*)lower
)->items
[0].key
;
666 lower_key
= &lower
->ptrs
[0].key
;
667 memcpy(&c
->ptrs
[0].key
, lower_key
, sizeof(struct btrfs_disk_key
));
668 btrfs_set_node_blockptr(c
, 0, path
->nodes
[level
- 1]->blocknr
);
669 /* the super has an extra ref to root->node */
670 btrfs_block_release(root
, root
->node
);
673 path
->nodes
[level
] = t
;
674 path
->slots
[level
] = 0;
679 * worker function to insert a single pointer in a node.
680 * the node should have enough room for the pointer already
682 * slot and level indicate where you want the key to go, and
683 * blocknr is the block the key points to.
685 * returns zero on success and < 0 on any error
687 static int insert_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
688 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
689 *key
, u64 blocknr
, int slot
, int level
)
691 struct btrfs_node
*lower
;
694 BUG_ON(!path
->nodes
[level
]);
695 lower
= &path
->nodes
[level
]->node
;
696 nritems
= btrfs_header_nritems(&lower
->header
);
699 if (nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
))
701 if (slot
!= nritems
) {
702 memmove(lower
->ptrs
+ slot
+ 1, lower
->ptrs
+ slot
,
703 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
705 memcpy(&lower
->ptrs
[slot
].key
, key
, sizeof(struct btrfs_disk_key
));
706 btrfs_set_node_blockptr(lower
, slot
, blocknr
);
707 btrfs_set_header_nritems(&lower
->header
, nritems
+ 1);
708 BUG_ON(list_empty(&path
->nodes
[level
]->dirty
));
713 * split the node at the specified level in path in two.
714 * The path is corrected to point to the appropriate node after the split
716 * Before splitting this tries to make some room in the node by pushing
717 * left and right, if either one works, it returns right away.
719 * returns 0 on success and < 0 on failure
721 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
722 *root
, struct btrfs_path
*path
, int level
)
724 struct btrfs_buffer
*t
;
725 struct btrfs_node
*c
;
726 struct btrfs_buffer
*split_buffer
;
727 struct btrfs_node
*split
;
733 t
= path
->nodes
[level
];
735 if (t
== root
->node
) {
736 /* trying to split the root, lets make a new one */
737 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
741 c_nritems
= btrfs_header_nritems(&c
->header
);
742 split_buffer
= btrfs_alloc_free_block(trans
, root
);
743 split
= &split_buffer
->node
;
744 btrfs_set_header_flags(&split
->header
, btrfs_header_flags(&c
->header
));
745 btrfs_set_header_blocknr(&split
->header
, split_buffer
->blocknr
);
746 btrfs_set_header_parentid(&split
->header
,
747 btrfs_header_parentid(&root
->node
->node
.header
));
748 mid
= (c_nritems
+ 1) / 2;
749 memcpy(split
->ptrs
, c
->ptrs
+ mid
,
750 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
751 btrfs_set_header_nritems(&split
->header
, c_nritems
- mid
);
752 btrfs_set_header_nritems(&c
->header
, mid
);
755 BUG_ON(list_empty(&t
->dirty
));
756 wret
= insert_ptr(trans
, root
, path
, &split
->ptrs
[0].key
,
757 split_buffer
->blocknr
, path
->slots
[level
+ 1] + 1,
762 if (path
->slots
[level
] >= mid
) {
763 path
->slots
[level
] -= mid
;
764 btrfs_block_release(root
, t
);
765 path
->nodes
[level
] = split_buffer
;
766 path
->slots
[level
+ 1] += 1;
768 btrfs_block_release(root
, split_buffer
);
774 * how many bytes are required to store the items in a leaf. start
775 * and nr indicate which items in the leaf to check. This totals up the
776 * space used both by the item structs and the item data
778 static int leaf_space_used(struct btrfs_leaf
*l
, int start
, int nr
)
781 int end
= start
+ nr
- 1;
785 data_len
= btrfs_item_end(l
->items
+ start
);
786 data_len
= data_len
- btrfs_item_offset(l
->items
+ end
);
787 data_len
+= sizeof(struct btrfs_item
) * nr
;
792 * push some data in the path leaf to the right, trying to free up at
793 * least data_size bytes. returns zero if the push worked, nonzero otherwise
795 * returns 1 if the push failed because the other node didn't have enough
796 * room, 0 if everything worked out and < 0 if there were major errors.
798 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
799 *root
, struct btrfs_path
*path
, int data_size
)
801 struct btrfs_buffer
*left_buf
= path
->nodes
[0];
802 struct btrfs_leaf
*left
= &left_buf
->leaf
;
803 struct btrfs_leaf
*right
;
804 struct btrfs_buffer
*right_buf
;
805 struct btrfs_buffer
*upper
;
811 struct btrfs_item
*item
;
815 slot
= path
->slots
[1];
816 if (!path
->nodes
[1]) {
819 upper
= path
->nodes
[1];
820 if (slot
>= btrfs_header_nritems(&upper
->node
.header
) - 1) {
823 right_buf
= read_tree_block(root
, btrfs_node_blockptr(&upper
->node
,
825 right
= &right_buf
->leaf
;
826 free_space
= btrfs_leaf_free_space(root
, right
);
827 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
828 btrfs_block_release(root
, right_buf
);
831 /* cow and double check */
832 btrfs_cow_block(trans
, root
, right_buf
, upper
, slot
+ 1, &right_buf
);
833 right
= &right_buf
->leaf
;
834 free_space
= btrfs_leaf_free_space(root
, right
);
835 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
836 btrfs_block_release(root
, right_buf
);
840 left_nritems
= btrfs_header_nritems(&left
->header
);
841 for (i
= left_nritems
- 1; i
>= 0; i
--) {
842 item
= left
->items
+ i
;
843 if (path
->slots
[0] == i
)
844 push_space
+= data_size
+ sizeof(*item
);
845 if (btrfs_item_size(item
) + sizeof(*item
) + push_space
>
849 push_space
+= btrfs_item_size(item
) + sizeof(*item
);
851 if (push_items
== 0) {
852 btrfs_block_release(root
, right_buf
);
855 right_nritems
= btrfs_header_nritems(&right
->header
);
856 /* push left to right */
857 push_space
= btrfs_item_end(left
->items
+ left_nritems
- push_items
);
858 push_space
-= leaf_data_end(root
, left
);
859 /* make room in the right data area */
860 memmove(btrfs_leaf_data(right
) + leaf_data_end(root
, right
) -
861 push_space
, btrfs_leaf_data(right
) + leaf_data_end(root
, right
),
862 BTRFS_LEAF_DATA_SIZE(root
) - leaf_data_end(root
, right
));
863 /* copy from the left data area */
864 memcpy(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
865 btrfs_leaf_data(left
) + leaf_data_end(root
, left
), push_space
);
866 memmove(right
->items
+ push_items
, right
->items
,
867 right_nritems
* sizeof(struct btrfs_item
));
868 /* copy the items from left to right */
869 memcpy(right
->items
, left
->items
+ left_nritems
- push_items
,
870 push_items
* sizeof(struct btrfs_item
));
872 /* update the item pointers */
873 right_nritems
+= push_items
;
874 btrfs_set_header_nritems(&right
->header
, right_nritems
);
875 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
876 for (i
= 0; i
< right_nritems
; i
++) {
877 btrfs_set_item_offset(right
->items
+ i
, push_space
-
878 btrfs_item_size(right
->items
+ i
));
879 push_space
= btrfs_item_offset(right
->items
+ i
);
881 left_nritems
-= push_items
;
882 btrfs_set_header_nritems(&left
->header
, left_nritems
);
884 BUG_ON(list_empty(&left_buf
->dirty
));
885 BUG_ON(list_empty(&right_buf
->dirty
));
886 memcpy(&upper
->node
.ptrs
[slot
+ 1].key
,
887 &right
->items
[0].key
, sizeof(struct btrfs_disk_key
));
888 BUG_ON(list_empty(&upper
->dirty
));
890 /* then fixup the leaf pointer in the path */
891 if (path
->slots
[0] >= left_nritems
) {
892 path
->slots
[0] -= left_nritems
;
893 btrfs_block_release(root
, path
->nodes
[0]);
894 path
->nodes
[0] = right_buf
;
897 btrfs_block_release(root
, right_buf
);
902 * push some data in the path leaf to the left, trying to free up at
903 * least data_size bytes. returns zero if the push worked, nonzero otherwise
905 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
906 *root
, struct btrfs_path
*path
, int data_size
)
908 struct btrfs_buffer
*right_buf
= path
->nodes
[0];
909 struct btrfs_leaf
*right
= &right_buf
->leaf
;
910 struct btrfs_buffer
*t
;
911 struct btrfs_leaf
*left
;
917 struct btrfs_item
*item
;
918 u32 old_left_nritems
;
922 slot
= path
->slots
[1];
926 if (!path
->nodes
[1]) {
929 t
= read_tree_block(root
, btrfs_node_blockptr(&path
->nodes
[1]->node
,
932 free_space
= btrfs_leaf_free_space(root
, left
);
933 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
934 btrfs_block_release(root
, t
);
938 /* cow and double check */
939 btrfs_cow_block(trans
, root
, t
, path
->nodes
[1], slot
- 1, &t
);
941 free_space
= btrfs_leaf_free_space(root
, left
);
942 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
943 btrfs_block_release(root
, t
);
947 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
948 item
= right
->items
+ i
;
949 if (path
->slots
[0] == i
)
950 push_space
+= data_size
+ sizeof(*item
);
951 if (btrfs_item_size(item
) + sizeof(*item
) + push_space
>
955 push_space
+= btrfs_item_size(item
) + sizeof(*item
);
957 if (push_items
== 0) {
958 btrfs_block_release(root
, t
);
961 /* push data from right to left */
962 memcpy(left
->items
+ btrfs_header_nritems(&left
->header
),
963 right
->items
, push_items
* sizeof(struct btrfs_item
));
964 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
965 btrfs_item_offset(right
->items
+ push_items
-1);
966 memcpy(btrfs_leaf_data(left
) + leaf_data_end(root
, left
) - push_space
,
967 btrfs_leaf_data(right
) +
968 btrfs_item_offset(right
->items
+ push_items
- 1),
970 old_left_nritems
= btrfs_header_nritems(&left
->header
);
971 BUG_ON(old_left_nritems
< 0);
973 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
974 u32 ioff
= btrfs_item_offset(left
->items
+ i
);
975 btrfs_set_item_offset(left
->items
+ i
, ioff
-
976 (BTRFS_LEAF_DATA_SIZE(root
) -
977 btrfs_item_offset(left
->items
+
978 old_left_nritems
- 1)));
980 btrfs_set_header_nritems(&left
->header
, old_left_nritems
+ push_items
);
982 /* fixup right node */
983 push_space
= btrfs_item_offset(right
->items
+ push_items
- 1) -
984 leaf_data_end(root
, right
);
985 memmove(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
986 push_space
, btrfs_leaf_data(right
) +
987 leaf_data_end(root
, right
), push_space
);
988 memmove(right
->items
, right
->items
+ push_items
,
989 (btrfs_header_nritems(&right
->header
) - push_items
) *
990 sizeof(struct btrfs_item
));
991 btrfs_set_header_nritems(&right
->header
,
992 btrfs_header_nritems(&right
->header
) -
994 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
996 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
997 btrfs_set_item_offset(right
->items
+ i
, push_space
-
998 btrfs_item_size(right
->items
+ i
));
999 push_space
= btrfs_item_offset(right
->items
+ i
);
1002 BUG_ON(list_empty(&t
->dirty
));
1003 BUG_ON(list_empty(&right_buf
->dirty
));
1005 wret
= fixup_low_keys(trans
, root
, path
, &right
->items
[0].key
, 1);
1009 /* then fixup the leaf pointer in the path */
1010 if (path
->slots
[0] < push_items
) {
1011 path
->slots
[0] += old_left_nritems
;
1012 btrfs_block_release(root
, path
->nodes
[0]);
1014 path
->slots
[1] -= 1;
1016 btrfs_block_release(root
, t
);
1017 path
->slots
[0] -= push_items
;
1019 BUG_ON(path
->slots
[0] < 0);
1024 * split the path's leaf in two, making sure there is at least data_size
1025 * available for the resulting leaf level of the path.
1027 * returns 0 if all went well and < 0 on failure.
1029 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
1030 *root
, struct btrfs_path
*path
, int data_size
)
1032 struct btrfs_buffer
*l_buf
;
1033 struct btrfs_leaf
*l
;
1037 struct btrfs_leaf
*right
;
1038 struct btrfs_buffer
*right_buffer
;
1039 int space_needed
= data_size
+ sizeof(struct btrfs_item
);
1046 wret
= push_leaf_left(trans
, root
, path
, data_size
);
1050 wret
= push_leaf_right(trans
, root
, path
, data_size
);
1054 l_buf
= path
->nodes
[0];
1057 /* did the pushes work? */
1058 if (btrfs_leaf_free_space(root
, l
) >=
1059 sizeof(struct btrfs_item
) + data_size
)
1062 if (!path
->nodes
[1]) {
1063 ret
= insert_new_root(trans
, root
, path
, 1);
1067 slot
= path
->slots
[0];
1068 nritems
= btrfs_header_nritems(&l
->header
);
1069 mid
= (nritems
+ 1)/ 2;
1070 right_buffer
= btrfs_alloc_free_block(trans
, root
);
1071 BUG_ON(!right_buffer
);
1072 BUG_ON(mid
== nritems
);
1073 right
= &right_buffer
->leaf
;
1074 memset(&right
->header
, 0, sizeof(right
->header
));
1076 /* FIXME, just alloc a new leaf here */
1077 if (leaf_space_used(l
, mid
, nritems
- mid
) + space_needed
>
1078 BTRFS_LEAF_DATA_SIZE(root
))
1081 /* FIXME, just alloc a new leaf here */
1082 if (leaf_space_used(l
, 0, mid
+ 1) + space_needed
>
1083 BTRFS_LEAF_DATA_SIZE(root
))
1086 btrfs_set_header_nritems(&right
->header
, nritems
- mid
);
1087 btrfs_set_header_blocknr(&right
->header
, right_buffer
->blocknr
);
1088 btrfs_set_header_level(&right
->header
, 0);
1089 btrfs_set_header_parentid(&right
->header
,
1090 btrfs_header_parentid(&root
->node
->node
.header
));
1091 data_copy_size
= btrfs_item_end(l
->items
+ mid
) -
1092 leaf_data_end(root
, l
);
1093 memcpy(right
->items
, l
->items
+ mid
,
1094 (nritems
- mid
) * sizeof(struct btrfs_item
));
1095 memcpy(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
1096 data_copy_size
, btrfs_leaf_data(l
) +
1097 leaf_data_end(root
, l
), data_copy_size
);
1098 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
1099 btrfs_item_end(l
->items
+ mid
);
1101 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
1102 u32 ioff
= btrfs_item_offset(right
->items
+ i
);
1103 btrfs_set_item_offset(right
->items
+ i
, ioff
+ rt_data_off
);
1106 btrfs_set_header_nritems(&l
->header
, mid
);
1108 wret
= insert_ptr(trans
, root
, path
, &right
->items
[0].key
,
1109 right_buffer
->blocknr
, path
->slots
[1] + 1, 1);
1112 BUG_ON(list_empty(&right_buffer
->dirty
));
1113 BUG_ON(list_empty(&l_buf
->dirty
));
1114 BUG_ON(path
->slots
[0] != slot
);
1116 btrfs_block_release(root
, path
->nodes
[0]);
1117 path
->nodes
[0] = right_buffer
;
1118 path
->slots
[0] -= mid
;
1119 path
->slots
[1] += 1;
1121 btrfs_block_release(root
, right_buffer
);
1122 BUG_ON(path
->slots
[0] < 0);
1127 * Given a key and some data, insert an item into the tree.
1128 * This does all the path init required, making room in the tree if needed.
1130 int btrfs_insert_empty_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
1131 *root
, struct btrfs_path
*path
, struct btrfs_key
1132 *cpu_key
, u32 data_size
)
1137 struct btrfs_leaf
*leaf
;
1138 struct btrfs_buffer
*leaf_buf
;
1140 unsigned int data_end
;
1141 struct btrfs_disk_key disk_key
;
1143 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
1145 /* create a root if there isn't one */
1148 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, data_size
, 1);
1150 btrfs_release_path(root
, path
);
1156 slot_orig
= path
->slots
[0];
1157 leaf_buf
= path
->nodes
[0];
1158 leaf
= &leaf_buf
->leaf
;
1160 nritems
= btrfs_header_nritems(&leaf
->header
);
1161 data_end
= leaf_data_end(root
, leaf
);
1163 if (btrfs_leaf_free_space(root
, leaf
) <
1164 sizeof(struct btrfs_item
) + data_size
)
1167 slot
= path
->slots
[0];
1169 if (slot
!= nritems
) {
1171 unsigned int old_data
= btrfs_item_end(leaf
->items
+ slot
);
1174 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1176 /* first correct the data pointers */
1177 for (i
= slot
; i
< nritems
; i
++) {
1178 u32 ioff
= btrfs_item_offset(leaf
->items
+ i
);
1179 btrfs_set_item_offset(leaf
->items
+ i
,
1183 /* shift the items */
1184 memmove(leaf
->items
+ slot
+ 1, leaf
->items
+ slot
,
1185 (nritems
- slot
) * sizeof(struct btrfs_item
));
1187 /* shift the data */
1188 memmove(btrfs_leaf_data(leaf
) + data_end
- data_size
,
1189 btrfs_leaf_data(leaf
) +
1190 data_end
, old_data
- data_end
);
1191 data_end
= old_data
;
1193 /* setup the item for the new data */
1194 memcpy(&leaf
->items
[slot
].key
, &disk_key
,
1195 sizeof(struct btrfs_disk_key
));
1196 btrfs_set_item_offset(leaf
->items
+ slot
, data_end
- data_size
);
1197 btrfs_set_item_size(leaf
->items
+ slot
, data_size
);
1198 btrfs_set_header_nritems(&leaf
->header
, nritems
+ 1);
1202 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
1204 BUG_ON(list_empty(&leaf_buf
->dirty
));
1205 if (btrfs_leaf_free_space(root
, leaf
) < 0)
1207 check_leaf(root
, path
, 0);
1213 * Given a key and some data, insert an item into the tree.
1214 * This does all the path init required, making room in the tree if needed.
1216 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
1217 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
1221 struct btrfs_path path
;
1224 btrfs_init_path(&path
);
1225 ret
= btrfs_insert_empty_item(trans
, root
, &path
, cpu_key
, data_size
);
1227 ptr
= btrfs_item_ptr(&path
.nodes
[0]->leaf
, path
.slots
[0], u8
);
1228 memcpy(ptr
, data
, data_size
);
1230 btrfs_release_path(root
, &path
);
1235 * delete the pointer from a given node.
1237 * If the delete empties a node, the node is removed from the tree,
1238 * continuing all the way the root if required. The root is converted into
1239 * a leaf if all the nodes are emptied.
1241 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1242 struct btrfs_path
*path
, int level
, int slot
)
1244 struct btrfs_node
*node
;
1245 struct btrfs_buffer
*parent
= path
->nodes
[level
];
1250 node
= &parent
->node
;
1251 nritems
= btrfs_header_nritems(&node
->header
);
1252 if (slot
!= nritems
-1) {
1253 memmove(node
->ptrs
+ slot
, node
->ptrs
+ slot
+ 1,
1254 sizeof(struct btrfs_key_ptr
) * (nritems
- slot
- 1));
1257 btrfs_set_header_nritems(&node
->header
, nritems
);
1258 if (nritems
== 0 && parent
== root
->node
) {
1259 BUG_ON(btrfs_header_level(&root
->node
->node
.header
) != 1);
1260 /* just turn the root into a leaf and break */
1261 btrfs_set_header_level(&root
->node
->node
.header
, 0);
1262 } else if (slot
== 0) {
1263 wret
= fixup_low_keys(trans
, root
, path
, &node
->ptrs
[0].key
,
1268 BUG_ON(list_empty(&parent
->dirty
));
1273 * delete the item at the leaf level in path. If that empties
1274 * the leaf, remove it from the tree
1276 int btrfs_del_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1277 struct btrfs_path
*path
)
1280 struct btrfs_leaf
*leaf
;
1281 struct btrfs_buffer
*leaf_buf
;
1288 leaf_buf
= path
->nodes
[0];
1289 leaf
= &leaf_buf
->leaf
;
1290 slot
= path
->slots
[0];
1291 doff
= btrfs_item_offset(leaf
->items
+ slot
);
1292 dsize
= btrfs_item_size(leaf
->items
+ slot
);
1293 nritems
= btrfs_header_nritems(&leaf
->header
);
1295 if (slot
!= nritems
- 1) {
1297 int data_end
= leaf_data_end(root
, leaf
);
1298 memmove(btrfs_leaf_data(leaf
) + data_end
+ dsize
,
1299 btrfs_leaf_data(leaf
) + data_end
,
1301 for (i
= slot
+ 1; i
< nritems
; i
++) {
1302 u32 ioff
= btrfs_item_offset(leaf
->items
+ i
);
1303 btrfs_set_item_offset(leaf
->items
+ i
, ioff
+ dsize
);
1305 memmove(leaf
->items
+ slot
, leaf
->items
+ slot
+ 1,
1306 sizeof(struct btrfs_item
) *
1307 (nritems
- slot
- 1));
1309 btrfs_set_header_nritems(&leaf
->header
, nritems
- 1);
1311 /* delete the leaf if we've emptied it */
1313 if (leaf_buf
== root
->node
) {
1314 btrfs_set_header_level(&leaf
->header
, 0);
1315 BUG_ON(list_empty(&leaf_buf
->dirty
));
1317 clean_tree_block(trans
, root
, leaf_buf
);
1318 wret
= del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
1321 wret
= btrfs_free_extent(trans
, root
,
1322 leaf_buf
->blocknr
, 1, 1);
1327 int used
= leaf_space_used(leaf
, 0, nritems
);
1329 wret
= fixup_low_keys(trans
, root
, path
,
1330 &leaf
->items
[0].key
, 1);
1334 BUG_ON(list_empty(&leaf_buf
->dirty
));
1336 /* delete the leaf if it is mostly empty */
1337 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
1338 /* push_leaf_left fixes the path.
1339 * make sure the path still points to our leaf
1340 * for possible call to del_ptr below
1342 slot
= path
->slots
[1];
1344 wret
= push_leaf_left(trans
, root
, path
, 1);
1347 if (path
->nodes
[0] == leaf_buf
&&
1348 btrfs_header_nritems(&leaf
->header
)) {
1349 wret
= push_leaf_right(trans
, root
, path
, 1);
1353 if (btrfs_header_nritems(&leaf
->header
) == 0) {
1354 u64 blocknr
= leaf_buf
->blocknr
;
1355 clean_tree_block(trans
, root
, leaf_buf
);
1356 wret
= del_ptr(trans
, root
, path
, 1, slot
);
1359 btrfs_block_release(root
, leaf_buf
);
1360 wret
= btrfs_free_extent(trans
, root
, blocknr
,
1365 btrfs_block_release(root
, leaf_buf
);
1373 * walk up the tree as far as required to find the next leaf.
1374 * returns 0 if it found something or 1 if there are no greater leaves.
1375 * returns < 0 on io errors.
1377 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
1382 struct btrfs_buffer
*c
;
1383 struct btrfs_buffer
*next
= NULL
;
1385 while(level
< BTRFS_MAX_LEVEL
) {
1386 if (!path
->nodes
[level
])
1388 slot
= path
->slots
[level
] + 1;
1389 c
= path
->nodes
[level
];
1390 if (slot
>= btrfs_header_nritems(&c
->node
.header
)) {
1394 blocknr
= btrfs_node_blockptr(&c
->node
, slot
);
1396 btrfs_block_release(root
, next
);
1397 next
= read_tree_block(root
, blocknr
);
1400 path
->slots
[level
] = slot
;
1403 c
= path
->nodes
[level
];
1404 btrfs_block_release(root
, c
);
1405 path
->nodes
[level
] = next
;
1406 path
->slots
[level
] = 0;
1409 next
= read_tree_block(root
,
1410 btrfs_node_blockptr(&next
->node
, 0));