2 * linux/fs/befs/btree.c
4 * Copyright (C) 2001-2002 Will Dyson <will_dyson@pobox.com>
6 * Licensed under the GNU GPL. See the file COPYING for details.
8 * 2002-02-05: Sergey S. Kostyliov added binary search within
13 * Dominic Giampaolo, author of "Practical File System
14 * Design with the Be File System", for such a helpful book.
16 * Marcus J. Ranum, author of the b+tree package in
17 * comp.sources.misc volume 10. This code is not copied from that
18 * work, but it is partially based on it.
20 * Makoto Kato, author of the original BeFS for linux filesystem
24 #include <linux/kernel.h>
25 #include <linux/string.h>
26 #include <linux/slab.h>
28 #include <linux/buffer_head.h>
32 #include "datastream.h"
35 * The btree functions in this file are built on top of the
36 * datastream.c interface, which is in turn built on top of the
40 /* Befs B+tree structure:
42 * The first thing in the tree is the tree superblock. It tells you
43 * all kinds of useful things about the tree, like where the rootnode
44 * is located, and the size of the nodes (always 1024 with current version
47 * The rest of the tree consists of a series of nodes. Nodes contain a header
48 * (struct befs_btree_nodehead), the packed key data, an array of shorts
49 * containing the ending offsets for each of the keys, and an array of
50 * befs_off_t values. In interior nodes, the keys are the ending keys for
51 * the childnode they point to, and the values are offsets into the
52 * datastream containing the tree.
57 * The book states 2 confusing things about befs b+trees. First,
58 * it states that the overflow field of node headers is used by internal nodes
59 * to point to another node that "effectively continues this one". Here is what
60 * I believe that means. Each key in internal nodes points to another node that
61 * contains key values less than itself. Inspection reveals that the last key
62 * in the internal node is not the last key in the index. Keys that are
63 * greater than the last key in the internal node go into the overflow node.
64 * I imagine there is a performance reason for this.
66 * Second, it states that the header of a btree node is sufficient to
67 * distinguish internal nodes from leaf nodes. Without saying exactly how.
68 * After figuring out the first, it becomes obvious that internal nodes have
69 * overflow nodes and leafnodes do not.
73 * Currently, this code is only good for directory B+trees.
74 * In order to be used for other BFS indexes, it needs to be extended to handle
75 * duplicate keys and non-string keytypes (int32, int64, float, double).
79 * In memory structure of each btree node
81 struct befs_btree_node
{
82 befs_host_btree_nodehead head
; /* head of node converted to cpu byteorder */
83 struct buffer_head
*bh
;
84 befs_btree_nodehead
*od_node
; /* on disk node */
88 static const befs_off_t befs_bt_inval
= 0xffffffffffffffffULL
;
91 static int befs_btree_seekleaf(struct super_block
*sb
, befs_data_stream
* ds
,
92 befs_btree_super
* bt_super
,
93 struct befs_btree_node
*this_node
,
94 befs_off_t
* node_off
);
96 static int befs_bt_read_super(struct super_block
*sb
, befs_data_stream
* ds
,
97 befs_btree_super
* sup
);
99 static int befs_bt_read_node(struct super_block
*sb
, befs_data_stream
* ds
,
100 struct befs_btree_node
*node
,
101 befs_off_t node_off
);
103 static int befs_leafnode(struct befs_btree_node
*node
);
105 static fs16
*befs_bt_keylen_index(struct befs_btree_node
*node
);
107 static fs64
*befs_bt_valarray(struct befs_btree_node
*node
);
109 static char *befs_bt_keydata(struct befs_btree_node
*node
);
111 static int befs_find_key(struct super_block
*sb
,
112 struct befs_btree_node
*node
,
113 const char *findkey
, befs_off_t
* value
);
115 static char *befs_bt_get_key(struct super_block
*sb
,
116 struct befs_btree_node
*node
,
117 int index
, u16
* keylen
);
119 static int befs_compare_strings(const void *key1
, int keylen1
,
120 const void *key2
, int keylen2
);
123 * befs_bt_read_super - read in btree superblock convert to cpu byteorder
124 * @sb: Filesystem superblock
125 * @ds: Datastream to read from
126 * @sup: Buffer in which to place the btree superblock
128 * Calls befs_read_datastream to read in the btree superblock and
129 * makes sure it is in cpu byteorder, byteswapping if necessary.
131 * On success, returns BEFS_OK and *@sup contains the btree superblock,
134 * On failure, BEFS_ERR is returned.
137 befs_bt_read_super(struct super_block
*sb
, befs_data_stream
* ds
,
138 befs_btree_super
* sup
)
140 struct buffer_head
*bh
;
141 befs_disk_btree_super
*od_sup
;
143 befs_debug(sb
, "---> %s", __func__
);
145 bh
= befs_read_datastream(sb
, ds
, 0, NULL
);
148 befs_error(sb
, "Couldn't read index header.");
151 od_sup
= (befs_disk_btree_super
*) bh
->b_data
;
152 befs_dump_index_entry(sb
, od_sup
);
154 sup
->magic
= fs32_to_cpu(sb
, od_sup
->magic
);
155 sup
->node_size
= fs32_to_cpu(sb
, od_sup
->node_size
);
156 sup
->max_depth
= fs32_to_cpu(sb
, od_sup
->max_depth
);
157 sup
->data_type
= fs32_to_cpu(sb
, od_sup
->data_type
);
158 sup
->root_node_ptr
= fs64_to_cpu(sb
, od_sup
->root_node_ptr
);
159 sup
->free_node_ptr
= fs64_to_cpu(sb
, od_sup
->free_node_ptr
);
160 sup
->max_size
= fs64_to_cpu(sb
, od_sup
->max_size
);
163 if (sup
->magic
!= BEFS_BTREE_MAGIC
) {
164 befs_error(sb
, "Index header has bad magic.");
168 befs_debug(sb
, "<--- %s", __func__
);
172 befs_debug(sb
, "<--- %s ERROR", __func__
);
177 * befs_bt_read_node - read in btree node and convert to cpu byteorder
178 * @sb: Filesystem superblock
179 * @ds: Datastream to read from
180 * @node: Buffer in which to place the btree node
181 * @node_off: Starting offset (in bytes) of the node in @ds
183 * Calls befs_read_datastream to read in the indicated btree node and
184 * makes sure its header fields are in cpu byteorder, byteswapping if
186 * Note: node->bh must be NULL when this function called first
187 * time. Don't forget brelse(node->bh) after last call.
189 * On success, returns BEFS_OK and *@node contains the btree node that
190 * starts at @node_off, with the node->head fields in cpu byte order.
192 * On failure, BEFS_ERR is returned.
196 befs_bt_read_node(struct super_block
*sb
, befs_data_stream
* ds
,
197 struct befs_btree_node
*node
, befs_off_t node_off
)
201 befs_debug(sb
, "---> %s", __func__
);
206 node
->bh
= befs_read_datastream(sb
, ds
, node_off
, &off
);
208 befs_error(sb
, "%s failed to read "
209 "node at %llu", __func__
, node_off
);
210 befs_debug(sb
, "<--- %s ERROR", __func__
);
215 (befs_btree_nodehead
*) ((void *) node
->bh
->b_data
+ off
);
217 befs_dump_index_node(sb
, node
->od_node
);
219 node
->head
.left
= fs64_to_cpu(sb
, node
->od_node
->left
);
220 node
->head
.right
= fs64_to_cpu(sb
, node
->od_node
->right
);
221 node
->head
.overflow
= fs64_to_cpu(sb
, node
->od_node
->overflow
);
222 node
->head
.all_key_count
=
223 fs16_to_cpu(sb
, node
->od_node
->all_key_count
);
224 node
->head
.all_key_length
=
225 fs16_to_cpu(sb
, node
->od_node
->all_key_length
);
227 befs_debug(sb
, "<--- %s", __func__
);
232 * befs_btree_find - Find a key in a befs B+tree
233 * @sb: Filesystem superblock
234 * @ds: Datastream containing btree
235 * @key: Key string to lookup in btree
236 * @value: Value stored with @key
238 * On success, returns BEFS_OK and sets *@value to the value stored
239 * with @key (usually the disk block number of an inode).
241 * On failure, returns BEFS_ERR or BEFS_BT_NOT_FOUND.
244 * Read the superblock and rootnode of the b+tree.
245 * Drill down through the interior nodes using befs_find_key().
246 * Once at the correct leaf node, use befs_find_key() again to get the
247 * actuall value stored with the key.
250 befs_btree_find(struct super_block
*sb
, befs_data_stream
* ds
,
251 const char *key
, befs_off_t
* value
)
253 struct befs_btree_node
*this_node
;
254 befs_btree_super bt_super
;
258 befs_debug(sb
, "---> %s Key: %s", __func__
, key
);
260 if (befs_bt_read_super(sb
, ds
, &bt_super
) != BEFS_OK
) {
262 "befs_btree_find() failed to read index superblock");
266 this_node
= kmalloc(sizeof(struct befs_btree_node
),
269 befs_error(sb
, "befs_btree_find() failed to allocate %zu "
270 "bytes of memory", sizeof(struct befs_btree_node
));
274 this_node
->bh
= NULL
;
276 /* read in root node */
277 node_off
= bt_super
.root_node_ptr
;
278 if (befs_bt_read_node(sb
, ds
, this_node
, node_off
) != BEFS_OK
) {
279 befs_error(sb
, "befs_btree_find() failed to read "
280 "node at %llu", node_off
);
284 while (!befs_leafnode(this_node
)) {
285 res
= befs_find_key(sb
, this_node
, key
, &node_off
);
286 if (res
== BEFS_BT_NOT_FOUND
)
287 node_off
= this_node
->head
.overflow
;
288 /* if no match, go to overflow node */
289 if (befs_bt_read_node(sb
, ds
, this_node
, node_off
) != BEFS_OK
) {
290 befs_error(sb
, "befs_btree_find() failed to read "
291 "node at %llu", node_off
);
296 /* at the correct leaf node now */
298 res
= befs_find_key(sb
, this_node
, key
, value
);
300 brelse(this_node
->bh
);
303 if (res
!= BEFS_BT_MATCH
) {
304 befs_debug(sb
, "<--- %s Key %s not found", __func__
, key
);
306 return BEFS_BT_NOT_FOUND
;
308 befs_debug(sb
, "<--- %s Found key %s, value %llu", __func__
,
316 befs_debug(sb
, "<--- %s ERROR", __func__
);
321 * befs_find_key - Search for a key within a node
322 * @sb: Filesystem superblock
323 * @node: Node to find the key within
324 * @findkey: Keystring to search for
325 * @value: If key is found, the value stored with the key is put here
327 * finds exact match if one exists, and returns BEFS_BT_MATCH
328 * If no exact match, finds first key in node that is greater
329 * (alphabetically) than the search key and returns BEFS_BT_PARMATCH
330 * (for partial match, I guess). Can you think of something better to
333 * If no key was a match or greater than the search key, return
336 * Use binary search instead of a linear.
339 befs_find_key(struct super_block
*sb
, struct befs_btree_node
*node
,
340 const char *findkey
, befs_off_t
* value
)
342 int first
, last
, mid
;
349 befs_debug(sb
, "---> %s %s", __func__
, findkey
);
353 findkey_len
= strlen(findkey
);
355 /* if node can not contain key, just skeep this node */
356 last
= node
->head
.all_key_count
- 1;
357 thiskey
= befs_bt_get_key(sb
, node
, last
, &keylen
);
359 eq
= befs_compare_strings(thiskey
, keylen
, findkey
, findkey_len
);
361 befs_debug(sb
, "<--- %s %s not found", __func__
, findkey
);
362 return BEFS_BT_NOT_FOUND
;
365 valarray
= befs_bt_valarray(node
);
367 /* simple binary search */
370 while (last
>= first
) {
371 mid
= (last
+ first
) / 2;
372 befs_debug(sb
, "first: %d, last: %d, mid: %d", first
, last
,
374 thiskey
= befs_bt_get_key(sb
, node
, mid
, &keylen
);
375 eq
= befs_compare_strings(thiskey
, keylen
, findkey
,
379 befs_debug(sb
, "<--- %s found %s at %d",
380 __func__
, thiskey
, mid
);
382 *value
= fs64_to_cpu(sb
, valarray
[mid
]);
383 return BEFS_BT_MATCH
;
391 *value
= fs64_to_cpu(sb
, valarray
[mid
+ 1]);
393 *value
= fs64_to_cpu(sb
, valarray
[mid
]);
394 befs_debug(sb
, "<--- %s found %s at %d", __func__
, thiskey
, mid
);
395 return BEFS_BT_PARMATCH
;
399 * befs_btree_read - Traverse leafnodes of a btree
400 * @sb: Filesystem superblock
401 * @ds: Datastream containing btree
402 * @key_no: Key number (alphabetical order) of key to read
403 * @bufsize: Size of the buffer to return key in
404 * @keybuf: Pointer to a buffer to put the key in
405 * @keysize: Length of the returned key
406 * @value: Value stored with the returned key
408 * Heres how it works: Key_no is the index of the key/value pair to
409 * return in keybuf/value.
410 * Bufsize is the size of keybuf (BEFS_NAME_LEN+1 is a good size). Keysize is
411 * the number of characters in the key (just a convenience).
414 * Get the first leafnode of the tree. See if the requested key is in that
415 * node. If not, follow the node->right link to the next leafnode. Repeat
416 * until the (key_no)th key is found or the tree is out of keys.
419 befs_btree_read(struct super_block
*sb
, befs_data_stream
* ds
,
420 loff_t key_no
, size_t bufsize
, char *keybuf
, size_t * keysize
,
423 struct befs_btree_node
*this_node
;
424 befs_btree_super bt_super
;
425 befs_off_t node_off
= 0;
434 befs_debug(sb
, "---> %s", __func__
);
436 if (befs_bt_read_super(sb
, ds
, &bt_super
) != BEFS_OK
) {
438 "befs_btree_read() failed to read index superblock");
442 this_node
= kmalloc(sizeof(struct befs_btree_node
), GFP_NOFS
);
443 if (this_node
== NULL
) {
444 befs_error(sb
, "befs_btree_read() failed to allocate %zu "
445 "bytes of memory", sizeof(struct befs_btree_node
));
449 node_off
= bt_super
.root_node_ptr
;
450 this_node
->bh
= NULL
;
452 /* seeks down to first leafnode, reads it into this_node */
453 res
= befs_btree_seekleaf(sb
, ds
, &bt_super
, this_node
, &node_off
);
454 if (res
== BEFS_BT_EMPTY
) {
455 brelse(this_node
->bh
);
459 befs_debug(sb
, "<--- %s Tree is EMPTY", __func__
);
460 return BEFS_BT_EMPTY
;
461 } else if (res
== BEFS_ERR
) {
465 /* find the leaf node containing the key_no key */
467 while (key_sum
+ this_node
->head
.all_key_count
<= key_no
) {
469 /* no more nodes to look in: key_no is too large */
470 if (this_node
->head
.right
== befs_bt_inval
) {
474 "<--- %s END of keys at %llu", __func__
,
476 key_sum
+ this_node
->head
.all_key_count
);
477 brelse(this_node
->bh
);
482 key_sum
+= this_node
->head
.all_key_count
;
483 node_off
= this_node
->head
.right
;
485 if (befs_bt_read_node(sb
, ds
, this_node
, node_off
) != BEFS_OK
) {
486 befs_error(sb
, "%s failed to read node at %llu",
487 __func__
, (unsigned long long)node_off
);
492 /* how many keys into this_node is key_no */
493 cur_key
= key_no
- key_sum
;
495 /* get pointers to datastructures within the node body */
496 valarray
= befs_bt_valarray(this_node
);
498 keystart
= befs_bt_get_key(sb
, this_node
, cur_key
, &keylen
);
500 befs_debug(sb
, "Read [%llu,%d]: keysize %d",
501 (long long unsigned int)node_off
, (int)cur_key
,
504 if (bufsize
< keylen
+ 1) {
505 befs_error(sb
, "%s keybuf too small (%zu) "
506 "for key of size %d", __func__
, bufsize
, keylen
);
507 brelse(this_node
->bh
);
511 strlcpy(keybuf
, keystart
, keylen
+ 1);
512 *value
= fs64_to_cpu(sb
, valarray
[cur_key
]);
515 befs_debug(sb
, "Read [%llu,%d]: Key \"%.*s\", Value %llu", node_off
,
516 cur_key
, keylen
, keybuf
, *value
);
518 brelse(this_node
->bh
);
521 befs_debug(sb
, "<--- %s", __func__
);
531 befs_debug(sb
, "<--- %s ERROR", __func__
);
536 * befs_btree_seekleaf - Find the first leafnode in the btree
537 * @sb: Filesystem superblock
538 * @ds: Datastream containing btree
539 * @bt_super: Pointer to the superblock of the btree
540 * @this_node: Buffer to return the leafnode in
541 * @node_off: Pointer to offset of current node within datastream. Modified
545 * Helper function for btree traverse. Moves the current position to the
546 * start of the first leaf node.
548 * Also checks for an empty tree. If there are no keys, returns BEFS_BT_EMPTY.
551 befs_btree_seekleaf(struct super_block
*sb
, befs_data_stream
* ds
,
552 befs_btree_super
*bt_super
,
553 struct befs_btree_node
*this_node
,
554 befs_off_t
* node_off
)
557 befs_debug(sb
, "---> %s", __func__
);
559 if (befs_bt_read_node(sb
, ds
, this_node
, *node_off
) != BEFS_OK
) {
560 befs_error(sb
, "%s failed to read "
561 "node at %llu", __func__
, *node_off
);
564 befs_debug(sb
, "Seekleaf to root node %llu", *node_off
);
566 if (this_node
->head
.all_key_count
== 0 && befs_leafnode(this_node
)) {
567 befs_debug(sb
, "<--- %s Tree is EMPTY", __func__
);
568 return BEFS_BT_EMPTY
;
571 while (!befs_leafnode(this_node
)) {
573 if (this_node
->head
.all_key_count
== 0) {
574 befs_debug(sb
, "%s encountered "
575 "an empty interior node: %llu. Using Overflow "
576 "node: %llu", __func__
, *node_off
,
577 this_node
->head
.overflow
);
578 *node_off
= this_node
->head
.overflow
;
580 fs64
*valarray
= befs_bt_valarray(this_node
);
581 *node_off
= fs64_to_cpu(sb
, valarray
[0]);
583 if (befs_bt_read_node(sb
, ds
, this_node
, *node_off
) != BEFS_OK
) {
584 befs_error(sb
, "%s failed to read "
585 "node at %llu", __func__
, *node_off
);
589 befs_debug(sb
, "Seekleaf to child node %llu", *node_off
);
591 befs_debug(sb
, "Node %llu is a leaf node", *node_off
);
596 befs_debug(sb
, "<--- %s ERROR", __func__
);
601 * befs_leafnode - Determine if the btree node is a leaf node or an
603 * @node: Pointer to node structure to test
605 * Return 1 if leaf, 0 if interior
608 befs_leafnode(struct befs_btree_node
*node
)
610 /* all interior nodes (and only interior nodes) have an overflow node */
611 if (node
->head
.overflow
== befs_bt_inval
)
618 * befs_bt_keylen_index - Finds start of keylen index in a node
619 * @node: Pointer to the node structure to find the keylen index within
621 * Returns a pointer to the start of the key length index array
622 * of the B+tree node *@node
624 * "The length of all the keys in the node is added to the size of the
625 * header and then rounded up to a multiple of four to get the beginning
626 * of the key length index" (p.88, practical filesystem design).
628 * Except that rounding up to 8 works, and rounding up to 4 doesn't.
631 befs_bt_keylen_index(struct befs_btree_node
*node
)
633 const int keylen_align
= 8;
634 unsigned long int off
=
635 (sizeof (befs_btree_nodehead
) + node
->head
.all_key_length
);
636 ulong tmp
= off
% keylen_align
;
639 off
+= keylen_align
- tmp
;
641 return (fs16
*) ((void *) node
->od_node
+ off
);
645 * befs_bt_valarray - Finds the start of value array in a node
646 * @node: Pointer to the node structure to find the value array within
648 * Returns a pointer to the start of the value array
649 * of the node pointed to by the node header
652 befs_bt_valarray(struct befs_btree_node
*node
)
654 void *keylen_index_start
= (void *) befs_bt_keylen_index(node
);
655 size_t keylen_index_size
= node
->head
.all_key_count
* sizeof (fs16
);
657 return (fs64
*) (keylen_index_start
+ keylen_index_size
);
661 * befs_bt_keydata - Finds start of keydata array in a node
662 * @node: Pointer to the node structure to find the keydata array within
664 * Returns a pointer to the start of the keydata array
665 * of the node pointed to by the node header
668 befs_bt_keydata(struct befs_btree_node
*node
)
670 return (char *) ((void *) node
->od_node
+ sizeof (befs_btree_nodehead
));
674 * befs_bt_get_key - returns a pointer to the start of a key
675 * @sb: filesystem superblock
676 * @node: node in which to look for the key
677 * @index: the index of the key to get
678 * @keylen: modified to be the length of the key at @index
680 * Returns a valid pointer into @node on success.
681 * Returns NULL on failure (bad input) and sets *@keylen = 0
684 befs_bt_get_key(struct super_block
*sb
, struct befs_btree_node
*node
,
685 int index
, u16
* keylen
)
691 if (index
< 0 || index
> node
->head
.all_key_count
) {
696 keystart
= befs_bt_keydata(node
);
697 keylen_index
= befs_bt_keylen_index(node
);
702 prev_key_end
= fs16_to_cpu(sb
, keylen_index
[index
- 1]);
704 *keylen
= fs16_to_cpu(sb
, keylen_index
[index
]) - prev_key_end
;
706 return keystart
+ prev_key_end
;
710 * befs_compare_strings - compare two strings
711 * @key1: pointer to the first key to be compared
712 * @keylen1: length in bytes of key1
713 * @key2: pointer to the second key to be compared
714 * @keylen2: length in bytes of key2
716 * Returns 0 if @key1 and @key2 are equal.
717 * Returns >0 if @key1 is greater.
718 * Returns <0 if @key2 is greater..
721 befs_compare_strings(const void *key1
, int keylen1
,
722 const void *key2
, int keylen2
)
724 int len
= min_t(int, keylen1
, keylen2
);
725 int result
= strncmp(key1
, key2
, len
);
727 result
= keylen1
- keylen2
;
731 /* These will be used for non-string keyed btrees */
734 btree_compare_int32(cont
void *key1
, int keylen1
, const void *key2
, int keylen2
)
736 return *(int32_t *) key1
- *(int32_t *) key2
;
740 btree_compare_uint32(cont
void *key1
, int keylen1
,
741 const void *key2
, int keylen2
)
743 if (*(u_int32_t
*) key1
== *(u_int32_t
*) key2
)
745 else if (*(u_int32_t
*) key1
> *(u_int32_t
*) key2
)
751 btree_compare_int64(cont
void *key1
, int keylen1
, const void *key2
, int keylen2
)
753 if (*(int64_t *) key1
== *(int64_t *) key2
)
755 else if (*(int64_t *) key1
> *(int64_t *) key2
)
762 btree_compare_uint64(cont
void *key1
, int keylen1
,
763 const void *key2
, int keylen2
)
765 if (*(u_int64_t
*) key1
== *(u_int64_t
*) key2
)
767 else if (*(u_int64_t
*) key1
> *(u_int64_t
*) key2
)
774 btree_compare_float(cont
void *key1
, int keylen1
, const void *key2
, int keylen2
)
776 float result
= *(float *) key1
- *(float *) key2
;
780 return (result
< 0.0f
) ? -1 : 1;
784 btree_compare_double(cont
void *key1
, int keylen1
,
785 const void *key2
, int keylen2
)
787 double result
= *(double *) key1
- *(double *) key2
;
791 return (result
< 0.0) ? -1 : 1;