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 * deal with the case where there is only one pointer in the root
299 * by promoting the node below to a root
302 struct btrfs_buffer
*child
;
303 u64 blocknr
= mid_buf
->blocknr
;
305 if (btrfs_header_nritems(&mid
->header
) != 1)
308 /* promote the child to a root */
309 child
= read_node_slot(root
, mid_buf
, 0);
312 path
->nodes
[level
] = NULL
;
313 /* once for the path */
314 btrfs_block_release(root
, mid_buf
);
315 /* once for the root ptr */
316 btrfs_block_release(root
, mid_buf
);
317 clean_tree_block(trans
, root
, mid_buf
);
318 return btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
320 parent
= &parent_buf
->node
;
322 if (btrfs_header_nritems(&mid
->header
) >
323 BTRFS_NODEPTRS_PER_BLOCK(root
) / 4)
326 left_buf
= read_node_slot(root
, parent_buf
, pslot
- 1);
327 right_buf
= read_node_slot(root
, parent_buf
, pslot
+ 1);
329 /* first, try to make some room in the middle buffer */
331 btrfs_cow_block(trans
, root
, left_buf
, parent_buf
, pslot
- 1,
333 left
= &left_buf
->node
;
334 orig_slot
+= btrfs_header_nritems(&left
->header
);
335 wret
= push_node_left(trans
, root
, left_buf
, mid_buf
);
341 * then try to empty the right most buffer into the middle
344 btrfs_cow_block(trans
, root
, right_buf
, parent_buf
, pslot
+ 1,
346 right
= &right_buf
->node
;
347 wret
= push_node_left(trans
, root
, mid_buf
, right_buf
);
350 if (btrfs_header_nritems(&right
->header
) == 0) {
351 u64 blocknr
= right_buf
->blocknr
;
352 btrfs_block_release(root
, right_buf
);
353 clean_tree_block(trans
, root
, right_buf
);
356 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
+
360 wret
= btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
364 memcpy(&parent
->ptrs
[pslot
+ 1].key
,
366 sizeof(struct btrfs_disk_key
));
367 BUG_ON(list_empty(&parent_buf
->dirty
));
370 if (btrfs_header_nritems(&mid
->header
) == 1) {
372 * we're not allowed to leave a node with one item in the
373 * tree during a delete. A deletion from lower in the tree
374 * could try to delete the only pointer in this node.
375 * So, pull some keys from the left.
376 * There has to be a left pointer at this point because
377 * otherwise we would have pulled some pointers from the
381 wret
= balance_node_right(trans
, root
, mid_buf
, left_buf
);
386 if (btrfs_header_nritems(&mid
->header
) == 0) {
387 /* we've managed to empty the middle node, drop it */
388 u64 blocknr
= mid_buf
->blocknr
;
389 btrfs_block_release(root
, mid_buf
);
390 clean_tree_block(trans
, root
, mid_buf
);
393 wret
= del_ptr(trans
, root
, path
, level
+ 1, pslot
);
396 wret
= btrfs_free_extent(trans
, root
, blocknr
, 1, 1);
400 /* update the parent key to reflect our changes */
401 memcpy(&parent
->ptrs
[pslot
].key
, &mid
->ptrs
[0].key
,
402 sizeof(struct btrfs_disk_key
));
403 BUG_ON(list_empty(&parent_buf
->dirty
));
406 /* update the path */
408 if (btrfs_header_nritems(&left
->header
) > orig_slot
) {
409 left_buf
->count
++; // released below
410 path
->nodes
[level
] = left_buf
;
411 path
->slots
[level
+ 1] -= 1;
412 path
->slots
[level
] = orig_slot
;
414 btrfs_block_release(root
, mid_buf
);
416 orig_slot
-= btrfs_header_nritems(&left
->header
);
417 path
->slots
[level
] = orig_slot
;
420 /* double check we haven't messed things up */
421 check_block(root
, path
, level
);
422 if (orig_ptr
!= btrfs_node_blockptr(&path
->nodes
[level
]->node
,
427 btrfs_block_release(root
, right_buf
);
429 btrfs_block_release(root
, left_buf
);
434 * look for key in the tree. path is filled in with nodes along the way
435 * if key is found, we return zero and you can find the item in the leaf
436 * level of the path (level 0)
438 * If the key isn't found, the path points to the slot where it should
439 * be inserted, and 1 is returned. If there are other errors during the
440 * search a negative error number is returned.
442 * if ins_len > 0, nodes and leaves will be split as we walk down the
443 * tree. if ins_len < 0, nodes will be merged as we walk down the tree (if
446 int btrfs_search_slot(struct btrfs_trans_handle
*trans
, struct btrfs_root
447 *root
, struct btrfs_key
*key
, struct btrfs_path
*p
, int
450 struct btrfs_buffer
*b
;
451 struct btrfs_buffer
*cow_buf
;
452 struct btrfs_node
*c
;
461 level
= btrfs_header_level(&b
->node
.header
);
464 wret
= btrfs_cow_block(trans
, root
, b
, p
->nodes
[level
+
465 1], p
->slots
[level
+ 1],
469 BUG_ON(!cow
&& ins_len
);
472 ret
= check_block(root
, p
, level
);
475 ret
= bin_search(c
, key
, &slot
);
476 if (!btrfs_is_leaf(c
)) {
479 p
->slots
[level
] = slot
;
480 if (ins_len
> 0 && btrfs_header_nritems(&c
->header
) ==
481 BTRFS_NODEPTRS_PER_BLOCK(root
)) {
482 int sret
= split_node(trans
, root
, p
, level
);
488 slot
= p
->slots
[level
];
489 } else if (ins_len
< 0) {
490 int sret
= balance_level(trans
, root
, p
,
498 slot
= p
->slots
[level
];
499 BUG_ON(btrfs_header_nritems(&c
->header
) == 1);
501 b
= read_tree_block(root
, btrfs_node_blockptr(c
, slot
));
503 struct btrfs_leaf
*l
= (struct btrfs_leaf
*)c
;
504 p
->slots
[level
] = slot
;
505 if (ins_len
> 0 && btrfs_leaf_free_space(root
, l
) <
506 sizeof(struct btrfs_item
) + ins_len
) {
507 int sret
= split_leaf(trans
, root
, p
, ins_len
);
512 BUG_ON(root
->node
->count
== 1);
516 BUG_ON(root
->node
->count
== 1);
521 * adjust the pointers going up the tree, starting at level
522 * making sure the right key of each node is points to 'key'.
523 * This is used after shifting pointers to the left, so it stops
524 * fixing up pointers when a given leaf/node is not in slot 0 of the
527 * If this fails to write a tree block, it returns -1, but continues
528 * fixing up the blocks in ram so the tree is consistent.
530 static int fixup_low_keys(struct btrfs_trans_handle
*trans
, struct btrfs_root
531 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
536 for (i
= level
; i
< BTRFS_MAX_LEVEL
; i
++) {
537 struct btrfs_node
*t
;
538 int tslot
= path
->slots
[i
];
541 t
= &path
->nodes
[i
]->node
;
542 memcpy(&t
->ptrs
[tslot
].key
, key
, sizeof(*key
));
543 BUG_ON(list_empty(&path
->nodes
[i
]->dirty
));
551 * try to push data from one node into the next node left in the
554 * returns 0 if some ptrs were pushed left, < 0 if there was some horrible
555 * error, and > 0 if there was no room in the left hand block.
557 static int push_node_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
558 *root
, struct btrfs_buffer
*dst_buf
, struct
559 btrfs_buffer
*src_buf
)
561 struct btrfs_node
*src
= &src_buf
->node
;
562 struct btrfs_node
*dst
= &dst_buf
->node
;
568 src_nritems
= btrfs_header_nritems(&src
->header
);
569 dst_nritems
= btrfs_header_nritems(&dst
->header
);
570 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
571 if (push_items
<= 0) {
575 if (src_nritems
< push_items
)
576 push_items
= src_nritems
;
578 memcpy(dst
->ptrs
+ dst_nritems
, src
->ptrs
,
579 push_items
* sizeof(struct btrfs_key_ptr
));
580 if (push_items
< src_nritems
) {
581 memmove(src
->ptrs
, src
->ptrs
+ push_items
,
582 (src_nritems
- push_items
) *
583 sizeof(struct btrfs_key_ptr
));
585 btrfs_set_header_nritems(&src
->header
, src_nritems
- push_items
);
586 btrfs_set_header_nritems(&dst
->header
, dst_nritems
+ push_items
);
587 BUG_ON(list_empty(&src_buf
->dirty
));
588 BUG_ON(list_empty(&dst_buf
->dirty
));
593 * try to push data from one node into the next node right in the
596 * returns 0 if some ptrs were pushed, < 0 if there was some horrible
597 * error, and > 0 if there was no room in the right hand block.
599 * this will only push up to 1/2 the contents of the left node over
601 static int balance_node_right(struct btrfs_trans_handle
*trans
, struct
602 btrfs_root
*root
, struct btrfs_buffer
*dst_buf
,
603 struct btrfs_buffer
*src_buf
)
605 struct btrfs_node
*src
= &src_buf
->node
;
606 struct btrfs_node
*dst
= &dst_buf
->node
;
613 src_nritems
= btrfs_header_nritems(&src
->header
);
614 dst_nritems
= btrfs_header_nritems(&dst
->header
);
615 push_items
= BTRFS_NODEPTRS_PER_BLOCK(root
) - dst_nritems
;
616 if (push_items
<= 0) {
620 max_push
= src_nritems
/ 2 + 1;
621 /* don't try to empty the node */
622 if (max_push
> src_nritems
)
624 if (max_push
< push_items
)
625 push_items
= max_push
;
627 memmove(dst
->ptrs
+ push_items
, dst
->ptrs
,
628 dst_nritems
* sizeof(struct btrfs_key_ptr
));
629 memcpy(dst
->ptrs
, src
->ptrs
+ src_nritems
- push_items
,
630 push_items
* sizeof(struct btrfs_key_ptr
));
632 btrfs_set_header_nritems(&src
->header
, src_nritems
- push_items
);
633 btrfs_set_header_nritems(&dst
->header
, dst_nritems
+ push_items
);
635 BUG_ON(list_empty(&src_buf
->dirty
));
636 BUG_ON(list_empty(&dst_buf
->dirty
));
641 * helper function to insert a new root level in the tree.
642 * A new node is allocated, and a single item is inserted to
643 * point to the existing root
645 * returns zero on success or < 0 on failure.
647 static int insert_new_root(struct btrfs_trans_handle
*trans
, struct btrfs_root
648 *root
, struct btrfs_path
*path
, int level
)
650 struct btrfs_buffer
*t
;
651 struct btrfs_node
*lower
;
652 struct btrfs_node
*c
;
653 struct btrfs_disk_key
*lower_key
;
655 BUG_ON(path
->nodes
[level
]);
656 BUG_ON(path
->nodes
[level
-1] != root
->node
);
658 t
= btrfs_alloc_free_block(trans
, root
);
660 memset(c
, 0, root
->blocksize
);
661 btrfs_set_header_nritems(&c
->header
, 1);
662 btrfs_set_header_level(&c
->header
, level
);
663 btrfs_set_header_blocknr(&c
->header
, t
->blocknr
);
664 btrfs_set_header_parentid(&c
->header
,
665 btrfs_header_parentid(&root
->node
->node
.header
));
666 lower
= &path
->nodes
[level
-1]->node
;
667 if (btrfs_is_leaf(lower
))
668 lower_key
= &((struct btrfs_leaf
*)lower
)->items
[0].key
;
670 lower_key
= &lower
->ptrs
[0].key
;
671 memcpy(&c
->ptrs
[0].key
, lower_key
, sizeof(struct btrfs_disk_key
));
672 btrfs_set_node_blockptr(c
, 0, path
->nodes
[level
- 1]->blocknr
);
673 /* the super has an extra ref to root->node */
674 btrfs_block_release(root
, root
->node
);
677 path
->nodes
[level
] = t
;
678 path
->slots
[level
] = 0;
683 * worker function to insert a single pointer in a node.
684 * the node should have enough room for the pointer already
686 * slot and level indicate where you want the key to go, and
687 * blocknr is the block the key points to.
689 * returns zero on success and < 0 on any error
691 static int insert_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
692 *root
, struct btrfs_path
*path
, struct btrfs_disk_key
693 *key
, u64 blocknr
, int slot
, int level
)
695 struct btrfs_node
*lower
;
698 BUG_ON(!path
->nodes
[level
]);
699 lower
= &path
->nodes
[level
]->node
;
700 nritems
= btrfs_header_nritems(&lower
->header
);
703 if (nritems
== BTRFS_NODEPTRS_PER_BLOCK(root
))
705 if (slot
!= nritems
) {
706 memmove(lower
->ptrs
+ slot
+ 1, lower
->ptrs
+ slot
,
707 (nritems
- slot
) * sizeof(struct btrfs_key_ptr
));
709 memcpy(&lower
->ptrs
[slot
].key
, key
, sizeof(struct btrfs_disk_key
));
710 btrfs_set_node_blockptr(lower
, slot
, blocknr
);
711 btrfs_set_header_nritems(&lower
->header
, nritems
+ 1);
712 BUG_ON(list_empty(&path
->nodes
[level
]->dirty
));
717 * split the node at the specified level in path in two.
718 * The path is corrected to point to the appropriate node after the split
720 * Before splitting this tries to make some room in the node by pushing
721 * left and right, if either one works, it returns right away.
723 * returns 0 on success and < 0 on failure
725 static int split_node(struct btrfs_trans_handle
*trans
, struct btrfs_root
726 *root
, struct btrfs_path
*path
, int level
)
728 struct btrfs_buffer
*t
;
729 struct btrfs_node
*c
;
730 struct btrfs_buffer
*split_buffer
;
731 struct btrfs_node
*split
;
737 t
= path
->nodes
[level
];
739 if (t
== root
->node
) {
740 /* trying to split the root, lets make a new one */
741 ret
= insert_new_root(trans
, root
, path
, level
+ 1);
745 c_nritems
= btrfs_header_nritems(&c
->header
);
746 split_buffer
= btrfs_alloc_free_block(trans
, root
);
747 split
= &split_buffer
->node
;
748 btrfs_set_header_flags(&split
->header
, btrfs_header_flags(&c
->header
));
749 btrfs_set_header_level(&split
->header
, btrfs_header_level(&c
->header
));
750 btrfs_set_header_blocknr(&split
->header
, split_buffer
->blocknr
);
751 btrfs_set_header_parentid(&split
->header
,
752 btrfs_header_parentid(&root
->node
->node
.header
));
753 mid
= (c_nritems
+ 1) / 2;
754 memcpy(split
->ptrs
, c
->ptrs
+ mid
,
755 (c_nritems
- mid
) * sizeof(struct btrfs_key_ptr
));
756 btrfs_set_header_nritems(&split
->header
, c_nritems
- mid
);
757 btrfs_set_header_nritems(&c
->header
, mid
);
760 BUG_ON(list_empty(&t
->dirty
));
761 wret
= insert_ptr(trans
, root
, path
, &split
->ptrs
[0].key
,
762 split_buffer
->blocknr
, path
->slots
[level
+ 1] + 1,
767 if (path
->slots
[level
] >= mid
) {
768 path
->slots
[level
] -= mid
;
769 btrfs_block_release(root
, t
);
770 path
->nodes
[level
] = split_buffer
;
771 path
->slots
[level
+ 1] += 1;
773 btrfs_block_release(root
, split_buffer
);
779 * how many bytes are required to store the items in a leaf. start
780 * and nr indicate which items in the leaf to check. This totals up the
781 * space used both by the item structs and the item data
783 static int leaf_space_used(struct btrfs_leaf
*l
, int start
, int nr
)
786 int end
= start
+ nr
- 1;
790 data_len
= btrfs_item_end(l
->items
+ start
);
791 data_len
= data_len
- btrfs_item_offset(l
->items
+ end
);
792 data_len
+= sizeof(struct btrfs_item
) * nr
;
797 * push some data in the path leaf to the right, trying to free up at
798 * least data_size bytes. returns zero if the push worked, nonzero otherwise
800 * returns 1 if the push failed because the other node didn't have enough
801 * room, 0 if everything worked out and < 0 if there were major errors.
803 static int push_leaf_right(struct btrfs_trans_handle
*trans
, struct btrfs_root
804 *root
, struct btrfs_path
*path
, int data_size
)
806 struct btrfs_buffer
*left_buf
= path
->nodes
[0];
807 struct btrfs_leaf
*left
= &left_buf
->leaf
;
808 struct btrfs_leaf
*right
;
809 struct btrfs_buffer
*right_buf
;
810 struct btrfs_buffer
*upper
;
816 struct btrfs_item
*item
;
820 slot
= path
->slots
[1];
821 if (!path
->nodes
[1]) {
824 upper
= path
->nodes
[1];
825 if (slot
>= btrfs_header_nritems(&upper
->node
.header
) - 1) {
828 right_buf
= read_tree_block(root
, btrfs_node_blockptr(&upper
->node
,
830 right
= &right_buf
->leaf
;
831 free_space
= btrfs_leaf_free_space(root
, right
);
832 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
833 btrfs_block_release(root
, right_buf
);
836 /* cow and double check */
837 btrfs_cow_block(trans
, root
, right_buf
, upper
, slot
+ 1, &right_buf
);
838 right
= &right_buf
->leaf
;
839 free_space
= btrfs_leaf_free_space(root
, right
);
840 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
841 btrfs_block_release(root
, right_buf
);
845 left_nritems
= btrfs_header_nritems(&left
->header
);
846 for (i
= left_nritems
- 1; i
>= 0; i
--) {
847 item
= left
->items
+ i
;
848 if (path
->slots
[0] == i
)
849 push_space
+= data_size
+ sizeof(*item
);
850 if (btrfs_item_size(item
) + sizeof(*item
) + push_space
>
854 push_space
+= btrfs_item_size(item
) + sizeof(*item
);
856 if (push_items
== 0) {
857 btrfs_block_release(root
, right_buf
);
860 right_nritems
= btrfs_header_nritems(&right
->header
);
861 /* push left to right */
862 push_space
= btrfs_item_end(left
->items
+ left_nritems
- push_items
);
863 push_space
-= leaf_data_end(root
, left
);
864 /* make room in the right data area */
865 memmove(btrfs_leaf_data(right
) + leaf_data_end(root
, right
) -
866 push_space
, btrfs_leaf_data(right
) + leaf_data_end(root
, right
),
867 BTRFS_LEAF_DATA_SIZE(root
) - leaf_data_end(root
, right
));
868 /* copy from the left data area */
869 memcpy(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) - push_space
,
870 btrfs_leaf_data(left
) + leaf_data_end(root
, left
), push_space
);
871 memmove(right
->items
+ push_items
, right
->items
,
872 right_nritems
* sizeof(struct btrfs_item
));
873 /* copy the items from left to right */
874 memcpy(right
->items
, left
->items
+ left_nritems
- push_items
,
875 push_items
* sizeof(struct btrfs_item
));
877 /* update the item pointers */
878 right_nritems
+= push_items
;
879 btrfs_set_header_nritems(&right
->header
, right_nritems
);
880 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
881 for (i
= 0; i
< right_nritems
; i
++) {
882 btrfs_set_item_offset(right
->items
+ i
, push_space
-
883 btrfs_item_size(right
->items
+ i
));
884 push_space
= btrfs_item_offset(right
->items
+ i
);
886 left_nritems
-= push_items
;
887 btrfs_set_header_nritems(&left
->header
, left_nritems
);
889 BUG_ON(list_empty(&left_buf
->dirty
));
890 BUG_ON(list_empty(&right_buf
->dirty
));
891 memcpy(&upper
->node
.ptrs
[slot
+ 1].key
,
892 &right
->items
[0].key
, sizeof(struct btrfs_disk_key
));
893 BUG_ON(list_empty(&upper
->dirty
));
895 /* then fixup the leaf pointer in the path */
896 if (path
->slots
[0] >= left_nritems
) {
897 path
->slots
[0] -= left_nritems
;
898 btrfs_block_release(root
, path
->nodes
[0]);
899 path
->nodes
[0] = right_buf
;
902 btrfs_block_release(root
, right_buf
);
907 * push some data in the path leaf to the left, trying to free up at
908 * least data_size bytes. returns zero if the push worked, nonzero otherwise
910 static int push_leaf_left(struct btrfs_trans_handle
*trans
, struct btrfs_root
911 *root
, struct btrfs_path
*path
, int data_size
)
913 struct btrfs_buffer
*right_buf
= path
->nodes
[0];
914 struct btrfs_leaf
*right
= &right_buf
->leaf
;
915 struct btrfs_buffer
*t
;
916 struct btrfs_leaf
*left
;
922 struct btrfs_item
*item
;
923 u32 old_left_nritems
;
927 slot
= path
->slots
[1];
931 if (!path
->nodes
[1]) {
934 t
= read_tree_block(root
, btrfs_node_blockptr(&path
->nodes
[1]->node
,
937 free_space
= btrfs_leaf_free_space(root
, left
);
938 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
939 btrfs_block_release(root
, t
);
943 /* cow and double check */
944 btrfs_cow_block(trans
, root
, t
, path
->nodes
[1], slot
- 1, &t
);
946 free_space
= btrfs_leaf_free_space(root
, left
);
947 if (free_space
< data_size
+ sizeof(struct btrfs_item
)) {
948 btrfs_block_release(root
, t
);
952 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
953 item
= right
->items
+ i
;
954 if (path
->slots
[0] == i
)
955 push_space
+= data_size
+ sizeof(*item
);
956 if (btrfs_item_size(item
) + sizeof(*item
) + push_space
>
960 push_space
+= btrfs_item_size(item
) + sizeof(*item
);
962 if (push_items
== 0) {
963 btrfs_block_release(root
, t
);
966 /* push data from right to left */
967 memcpy(left
->items
+ btrfs_header_nritems(&left
->header
),
968 right
->items
, push_items
* sizeof(struct btrfs_item
));
969 push_space
= BTRFS_LEAF_DATA_SIZE(root
) -
970 btrfs_item_offset(right
->items
+ push_items
-1);
971 memcpy(btrfs_leaf_data(left
) + leaf_data_end(root
, left
) - push_space
,
972 btrfs_leaf_data(right
) +
973 btrfs_item_offset(right
->items
+ push_items
- 1),
975 old_left_nritems
= btrfs_header_nritems(&left
->header
);
976 BUG_ON(old_left_nritems
< 0);
978 for (i
= old_left_nritems
; i
< old_left_nritems
+ push_items
; i
++) {
979 u32 ioff
= btrfs_item_offset(left
->items
+ i
);
980 btrfs_set_item_offset(left
->items
+ i
, ioff
-
981 (BTRFS_LEAF_DATA_SIZE(root
) -
982 btrfs_item_offset(left
->items
+
983 old_left_nritems
- 1)));
985 btrfs_set_header_nritems(&left
->header
, old_left_nritems
+ push_items
);
987 /* fixup right node */
988 push_space
= btrfs_item_offset(right
->items
+ push_items
- 1) -
989 leaf_data_end(root
, right
);
990 memmove(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
991 push_space
, btrfs_leaf_data(right
) +
992 leaf_data_end(root
, right
), push_space
);
993 memmove(right
->items
, right
->items
+ push_items
,
994 (btrfs_header_nritems(&right
->header
) - push_items
) *
995 sizeof(struct btrfs_item
));
996 btrfs_set_header_nritems(&right
->header
,
997 btrfs_header_nritems(&right
->header
) -
999 push_space
= BTRFS_LEAF_DATA_SIZE(root
);
1001 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
1002 btrfs_set_item_offset(right
->items
+ i
, push_space
-
1003 btrfs_item_size(right
->items
+ i
));
1004 push_space
= btrfs_item_offset(right
->items
+ i
);
1007 BUG_ON(list_empty(&t
->dirty
));
1008 BUG_ON(list_empty(&right_buf
->dirty
));
1010 wret
= fixup_low_keys(trans
, root
, path
, &right
->items
[0].key
, 1);
1014 /* then fixup the leaf pointer in the path */
1015 if (path
->slots
[0] < push_items
) {
1016 path
->slots
[0] += old_left_nritems
;
1017 btrfs_block_release(root
, path
->nodes
[0]);
1019 path
->slots
[1] -= 1;
1021 btrfs_block_release(root
, t
);
1022 path
->slots
[0] -= push_items
;
1024 BUG_ON(path
->slots
[0] < 0);
1029 * split the path's leaf in two, making sure there is at least data_size
1030 * available for the resulting leaf level of the path.
1032 * returns 0 if all went well and < 0 on failure.
1034 static int split_leaf(struct btrfs_trans_handle
*trans
, struct btrfs_root
1035 *root
, struct btrfs_path
*path
, int data_size
)
1037 struct btrfs_buffer
*l_buf
;
1038 struct btrfs_leaf
*l
;
1042 struct btrfs_leaf
*right
;
1043 struct btrfs_buffer
*right_buffer
;
1044 int space_needed
= data_size
+ sizeof(struct btrfs_item
);
1051 /* first try to make some room by pushing left and right */
1052 wret
= push_leaf_left(trans
, root
, path
, data_size
);
1056 wret
= push_leaf_right(trans
, root
, path
, data_size
);
1060 l_buf
= path
->nodes
[0];
1063 /* did the pushes work? */
1064 if (btrfs_leaf_free_space(root
, l
) >=
1065 sizeof(struct btrfs_item
) + data_size
)
1068 if (!path
->nodes
[1]) {
1069 ret
= insert_new_root(trans
, root
, path
, 1);
1073 slot
= path
->slots
[0];
1074 nritems
= btrfs_header_nritems(&l
->header
);
1075 mid
= (nritems
+ 1)/ 2;
1076 right_buffer
= btrfs_alloc_free_block(trans
, root
);
1077 BUG_ON(!right_buffer
);
1078 BUG_ON(mid
== nritems
);
1079 right
= &right_buffer
->leaf
;
1080 memset(&right
->header
, 0, sizeof(right
->header
));
1082 /* FIXME, just alloc a new leaf here */
1083 if (leaf_space_used(l
, mid
, nritems
- mid
) + space_needed
>
1084 BTRFS_LEAF_DATA_SIZE(root
))
1087 /* FIXME, just alloc a new leaf here */
1088 if (leaf_space_used(l
, 0, mid
+ 1) + space_needed
>
1089 BTRFS_LEAF_DATA_SIZE(root
))
1092 btrfs_set_header_nritems(&right
->header
, nritems
- mid
);
1093 btrfs_set_header_blocknr(&right
->header
, right_buffer
->blocknr
);
1094 btrfs_set_header_level(&right
->header
, 0);
1095 btrfs_set_header_parentid(&right
->header
,
1096 btrfs_header_parentid(&root
->node
->node
.header
));
1097 data_copy_size
= btrfs_item_end(l
->items
+ mid
) -
1098 leaf_data_end(root
, l
);
1099 memcpy(right
->items
, l
->items
+ mid
,
1100 (nritems
- mid
) * sizeof(struct btrfs_item
));
1101 memcpy(btrfs_leaf_data(right
) + BTRFS_LEAF_DATA_SIZE(root
) -
1102 data_copy_size
, btrfs_leaf_data(l
) +
1103 leaf_data_end(root
, l
), data_copy_size
);
1104 rt_data_off
= BTRFS_LEAF_DATA_SIZE(root
) -
1105 btrfs_item_end(l
->items
+ mid
);
1107 for (i
= 0; i
< btrfs_header_nritems(&right
->header
); i
++) {
1108 u32 ioff
= btrfs_item_offset(right
->items
+ i
);
1109 btrfs_set_item_offset(right
->items
+ i
, ioff
+ rt_data_off
);
1112 btrfs_set_header_nritems(&l
->header
, mid
);
1114 wret
= insert_ptr(trans
, root
, path
, &right
->items
[0].key
,
1115 right_buffer
->blocknr
, path
->slots
[1] + 1, 1);
1118 BUG_ON(list_empty(&right_buffer
->dirty
));
1119 BUG_ON(list_empty(&l_buf
->dirty
));
1120 BUG_ON(path
->slots
[0] != slot
);
1122 btrfs_block_release(root
, path
->nodes
[0]);
1123 path
->nodes
[0] = right_buffer
;
1124 path
->slots
[0] -= mid
;
1125 path
->slots
[1] += 1;
1127 btrfs_block_release(root
, right_buffer
);
1128 BUG_ON(path
->slots
[0] < 0);
1133 * Given a key and some data, insert an item into the tree.
1134 * This does all the path init required, making room in the tree if needed.
1136 int btrfs_insert_empty_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
1137 *root
, struct btrfs_path
*path
, struct btrfs_key
1138 *cpu_key
, u32 data_size
)
1143 struct btrfs_leaf
*leaf
;
1144 struct btrfs_buffer
*leaf_buf
;
1146 unsigned int data_end
;
1147 struct btrfs_disk_key disk_key
;
1149 btrfs_cpu_key_to_disk(&disk_key
, cpu_key
);
1151 /* create a root if there isn't one */
1154 ret
= btrfs_search_slot(trans
, root
, cpu_key
, path
, data_size
, 1);
1156 btrfs_release_path(root
, path
);
1162 slot_orig
= path
->slots
[0];
1163 leaf_buf
= path
->nodes
[0];
1164 leaf
= &leaf_buf
->leaf
;
1166 nritems
= btrfs_header_nritems(&leaf
->header
);
1167 data_end
= leaf_data_end(root
, leaf
);
1169 if (btrfs_leaf_free_space(root
, leaf
) <
1170 sizeof(struct btrfs_item
) + data_size
)
1173 slot
= path
->slots
[0];
1175 if (slot
!= nritems
) {
1177 unsigned int old_data
= btrfs_item_end(leaf
->items
+ slot
);
1180 * item0..itemN ... dataN.offset..dataN.size .. data0.size
1182 /* first correct the data pointers */
1183 for (i
= slot
; i
< nritems
; i
++) {
1184 u32 ioff
= btrfs_item_offset(leaf
->items
+ i
);
1185 btrfs_set_item_offset(leaf
->items
+ i
,
1189 /* shift the items */
1190 memmove(leaf
->items
+ slot
+ 1, leaf
->items
+ slot
,
1191 (nritems
- slot
) * sizeof(struct btrfs_item
));
1193 /* shift the data */
1194 memmove(btrfs_leaf_data(leaf
) + data_end
- data_size
,
1195 btrfs_leaf_data(leaf
) +
1196 data_end
, old_data
- data_end
);
1197 data_end
= old_data
;
1199 /* setup the item for the new data */
1200 memcpy(&leaf
->items
[slot
].key
, &disk_key
,
1201 sizeof(struct btrfs_disk_key
));
1202 btrfs_set_item_offset(leaf
->items
+ slot
, data_end
- data_size
);
1203 btrfs_set_item_size(leaf
->items
+ slot
, data_size
);
1204 btrfs_set_header_nritems(&leaf
->header
, nritems
+ 1);
1208 ret
= fixup_low_keys(trans
, root
, path
, &disk_key
, 1);
1210 BUG_ON(list_empty(&leaf_buf
->dirty
));
1211 if (btrfs_leaf_free_space(root
, leaf
) < 0)
1213 check_leaf(root
, path
, 0);
1219 * Given a key and some data, insert an item into the tree.
1220 * This does all the path init required, making room in the tree if needed.
1222 int btrfs_insert_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
1223 *root
, struct btrfs_key
*cpu_key
, void *data
, u32
1227 struct btrfs_path path
;
1230 btrfs_init_path(&path
);
1231 ret
= btrfs_insert_empty_item(trans
, root
, &path
, cpu_key
, data_size
);
1233 ptr
= btrfs_item_ptr(&path
.nodes
[0]->leaf
, path
.slots
[0], u8
);
1234 memcpy(ptr
, data
, data_size
);
1236 btrfs_release_path(root
, &path
);
1241 * delete the pointer from a given node.
1243 * If the delete empties a node, the node is removed from the tree,
1244 * continuing all the way the root if required. The root is converted into
1245 * a leaf if all the nodes are emptied.
1247 static int del_ptr(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1248 struct btrfs_path
*path
, int level
, int slot
)
1250 struct btrfs_node
*node
;
1251 struct btrfs_buffer
*parent
= path
->nodes
[level
];
1256 node
= &parent
->node
;
1257 nritems
= btrfs_header_nritems(&node
->header
);
1258 if (slot
!= nritems
-1) {
1259 memmove(node
->ptrs
+ slot
, node
->ptrs
+ slot
+ 1,
1260 sizeof(struct btrfs_key_ptr
) * (nritems
- slot
- 1));
1263 btrfs_set_header_nritems(&node
->header
, nritems
);
1264 if (nritems
== 0 && parent
== root
->node
) {
1265 BUG_ON(btrfs_header_level(&root
->node
->node
.header
) != 1);
1266 /* just turn the root into a leaf and break */
1267 btrfs_set_header_level(&root
->node
->node
.header
, 0);
1268 } else if (slot
== 0) {
1269 wret
= fixup_low_keys(trans
, root
, path
, &node
->ptrs
[0].key
,
1274 BUG_ON(list_empty(&parent
->dirty
));
1279 * delete the item at the leaf level in path. If that empties
1280 * the leaf, remove it from the tree
1282 int btrfs_del_item(struct btrfs_trans_handle
*trans
, struct btrfs_root
*root
,
1283 struct btrfs_path
*path
)
1286 struct btrfs_leaf
*leaf
;
1287 struct btrfs_buffer
*leaf_buf
;
1294 leaf_buf
= path
->nodes
[0];
1295 leaf
= &leaf_buf
->leaf
;
1296 slot
= path
->slots
[0];
1297 doff
= btrfs_item_offset(leaf
->items
+ slot
);
1298 dsize
= btrfs_item_size(leaf
->items
+ slot
);
1299 nritems
= btrfs_header_nritems(&leaf
->header
);
1301 if (slot
!= nritems
- 1) {
1303 int data_end
= leaf_data_end(root
, leaf
);
1304 memmove(btrfs_leaf_data(leaf
) + data_end
+ dsize
,
1305 btrfs_leaf_data(leaf
) + data_end
,
1307 for (i
= slot
+ 1; i
< nritems
; i
++) {
1308 u32 ioff
= btrfs_item_offset(leaf
->items
+ i
);
1309 btrfs_set_item_offset(leaf
->items
+ i
, ioff
+ dsize
);
1311 memmove(leaf
->items
+ slot
, leaf
->items
+ slot
+ 1,
1312 sizeof(struct btrfs_item
) *
1313 (nritems
- slot
- 1));
1315 btrfs_set_header_nritems(&leaf
->header
, nritems
- 1);
1317 /* delete the leaf if we've emptied it */
1319 if (leaf_buf
== root
->node
) {
1320 btrfs_set_header_level(&leaf
->header
, 0);
1321 BUG_ON(list_empty(&leaf_buf
->dirty
));
1323 clean_tree_block(trans
, root
, leaf_buf
);
1324 wret
= del_ptr(trans
, root
, path
, 1, path
->slots
[1]);
1327 wret
= btrfs_free_extent(trans
, root
,
1328 leaf_buf
->blocknr
, 1, 1);
1333 int used
= leaf_space_used(leaf
, 0, nritems
);
1335 wret
= fixup_low_keys(trans
, root
, path
,
1336 &leaf
->items
[0].key
, 1);
1340 BUG_ON(list_empty(&leaf_buf
->dirty
));
1342 /* delete the leaf if it is mostly empty */
1343 if (used
< BTRFS_LEAF_DATA_SIZE(root
) / 3) {
1344 /* push_leaf_left fixes the path.
1345 * make sure the path still points to our leaf
1346 * for possible call to del_ptr below
1348 slot
= path
->slots
[1];
1350 wret
= push_leaf_left(trans
, root
, path
, 1);
1353 if (path
->nodes
[0] == leaf_buf
&&
1354 btrfs_header_nritems(&leaf
->header
)) {
1355 wret
= push_leaf_right(trans
, root
, path
, 1);
1359 if (btrfs_header_nritems(&leaf
->header
) == 0) {
1360 u64 blocknr
= leaf_buf
->blocknr
;
1361 clean_tree_block(trans
, root
, leaf_buf
);
1362 wret
= del_ptr(trans
, root
, path
, 1, slot
);
1365 btrfs_block_release(root
, leaf_buf
);
1366 wret
= btrfs_free_extent(trans
, root
, blocknr
,
1371 btrfs_block_release(root
, leaf_buf
);
1379 * walk up the tree as far as required to find the next leaf.
1380 * returns 0 if it found something or 1 if there are no greater leaves.
1381 * returns < 0 on io errors.
1383 int btrfs_next_leaf(struct btrfs_root
*root
, struct btrfs_path
*path
)
1388 struct btrfs_buffer
*c
;
1389 struct btrfs_buffer
*next
= NULL
;
1391 while(level
< BTRFS_MAX_LEVEL
) {
1392 if (!path
->nodes
[level
])
1394 slot
= path
->slots
[level
] + 1;
1395 c
= path
->nodes
[level
];
1396 if (slot
>= btrfs_header_nritems(&c
->node
.header
)) {
1400 blocknr
= btrfs_node_blockptr(&c
->node
, slot
);
1402 btrfs_block_release(root
, next
);
1403 next
= read_tree_block(root
, blocknr
);
1406 path
->slots
[level
] = slot
;
1409 c
= path
->nodes
[level
];
1410 btrfs_block_release(root
, c
);
1411 path
->nodes
[level
] = next
;
1412 path
->slots
[level
] = 0;
1415 next
= read_tree_block(root
,
1416 btrfs_node_blockptr(&next
->node
, 0));