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, const 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, const befs_data_stream *ds,
97 befs_btree_super * sup);
99 static int befs_bt_read_node(struct super_block *sb, const 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, const 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);
161 if (sup->magic != BEFS_BTREE_MAGIC) {
162 befs_error(sb, "Index header has bad magic.");
166 befs_debug(sb, "<--- %s", __func__);
170 befs_debug(sb, "<--- %s ERROR", __func__);
175 * befs_bt_read_node - read in btree node and convert to cpu byteorder
176 * @sb: Filesystem superblock
177 * @ds: Datastream to read from
178 * @node: Buffer in which to place the btree node
179 * @node_off: Starting offset (in bytes) of the node in @ds
181 * Calls befs_read_datastream to read in the indicated btree node and
182 * makes sure its header fields are in cpu byteorder, byteswapping if
184 * Note: node->bh must be NULL when this function is called the first time.
185 * Don't forget brelse(node->bh) after last call.
187 * On success, returns BEFS_OK and *@node contains the btree node that
188 * starts at @node_off, with the node->head fields in cpu byte order.
190 * On failure, BEFS_ERR is returned.
194 befs_bt_read_node(struct super_block *sb, const befs_data_stream *ds,
195 struct befs_btree_node *node, befs_off_t node_off)
199 befs_debug(sb, "---> %s", __func__);
204 node->bh = befs_read_datastream(sb, ds, node_off, &off);
206 befs_error(sb, "%s failed to read "
207 "node at %llu", __func__, node_off);
208 befs_debug(sb, "<--- %s ERROR", __func__);
213 (befs_btree_nodehead *) ((void *) node->bh->b_data + off);
215 befs_dump_index_node(sb, node->od_node);
217 node->head.left = fs64_to_cpu(sb, node->od_node->left);
218 node->head.right = fs64_to_cpu(sb, node->od_node->right);
219 node->head.overflow = fs64_to_cpu(sb, node->od_node->overflow);
220 node->head.all_key_count =
221 fs16_to_cpu(sb, node->od_node->all_key_count);
222 node->head.all_key_length =
223 fs16_to_cpu(sb, node->od_node->all_key_length);
225 befs_debug(sb, "<--- %s", __func__);
230 * befs_btree_find - Find a key in a befs B+tree
231 * @sb: Filesystem superblock
232 * @ds: Datastream containing btree
233 * @key: Key string to lookup in btree
234 * @value: Value stored with @key
236 * On success, returns BEFS_OK and sets *@value to the value stored
237 * with @key (usually the disk block number of an inode).
239 * On failure, returns BEFS_ERR or BEFS_BT_NOT_FOUND.
242 * Read the superblock and rootnode of the b+tree.
243 * Drill down through the interior nodes using befs_find_key().
244 * Once at the correct leaf node, use befs_find_key() again to get the
245 * actual value stored with the key.
248 befs_btree_find(struct super_block *sb, const befs_data_stream *ds,
249 const char *key, befs_off_t * value)
251 struct befs_btree_node *this_node;
252 befs_btree_super bt_super;
256 befs_debug(sb, "---> %s Key: %s", __func__, key);
258 if (befs_bt_read_super(sb, ds, &bt_super) != BEFS_OK) {
260 "befs_btree_find() failed to read index superblock");
264 this_node = kmalloc(sizeof(struct befs_btree_node),
267 befs_error(sb, "befs_btree_find() failed to allocate %zu "
268 "bytes of memory", sizeof(struct befs_btree_node));
272 this_node->bh = NULL;
274 /* read in root node */
275 node_off = bt_super.root_node_ptr;
276 if (befs_bt_read_node(sb, ds, this_node, node_off) != BEFS_OK) {
277 befs_error(sb, "befs_btree_find() failed to read "
278 "node at %llu", node_off);
282 while (!befs_leafnode(this_node)) {
283 res = befs_find_key(sb, this_node, key, &node_off);
284 /* if no key set, try the overflow node */
285 if (res == BEFS_BT_OVERFLOW)
286 node_off = this_node->head.overflow;
287 if (befs_bt_read_node(sb, ds, this_node, node_off) != BEFS_OK) {
288 befs_error(sb, "befs_btree_find() failed to read "
289 "node at %llu", node_off);
294 /* at a leaf node now, check if it is correct */
295 res = befs_find_key(sb, this_node, key, value);
297 brelse(this_node->bh);
300 if (res != BEFS_BT_MATCH) {
301 befs_error(sb, "<--- %s Key %s not found", __func__, key);
302 befs_debug(sb, "<--- %s ERROR", __func__);
304 return BEFS_BT_NOT_FOUND;
306 befs_debug(sb, "<--- %s Found key %s, value %llu", __func__,
314 befs_debug(sb, "<--- %s ERROR", __func__);
319 * befs_find_key - Search for a key within a node
320 * @sb: Filesystem superblock
321 * @node: Node to find the key within
322 * @findkey: Keystring to search for
323 * @value: If key is found, the value stored with the key is put here
325 * Finds exact match if one exists, and returns BEFS_BT_MATCH.
326 * If there is no match and node's value array is too small for key, return
328 * If no match and node should countain this key, return BEFS_BT_NOT_FOUND.
330 * Uses binary search instead of a linear.
333 befs_find_key(struct super_block *sb, struct befs_btree_node *node,
334 const char *findkey, befs_off_t * value)
336 int first, last, mid;
343 befs_debug(sb, "---> %s %s", __func__, findkey);
345 findkey_len = strlen(findkey);
347 /* if node can not contain key, just skip this node */
348 last = node->head.all_key_count - 1;
349 thiskey = befs_bt_get_key(sb, node, last, &keylen);
351 eq = befs_compare_strings(thiskey, keylen, findkey, findkey_len);
353 befs_debug(sb, "<--- node can't contain %s", findkey);
354 return BEFS_BT_OVERFLOW;
357 valarray = befs_bt_valarray(node);
359 /* simple binary search */
362 while (last >= first) {
363 mid = (last + first) / 2;
364 befs_debug(sb, "first: %d, last: %d, mid: %d", first, last,
366 thiskey = befs_bt_get_key(sb, node, mid, &keylen);
367 eq = befs_compare_strings(thiskey, keylen, findkey,
371 befs_debug(sb, "<--- %s found %s at %d",
372 __func__, thiskey, mid);
374 *value = fs64_to_cpu(sb, valarray[mid]);
375 return BEFS_BT_MATCH;
383 /* return an existing value so caller can arrive to a leaf node */
385 *value = fs64_to_cpu(sb, valarray[mid + 1]);
387 *value = fs64_to_cpu(sb, valarray[mid]);
388 befs_error(sb, "<--- %s %s not found", __func__, findkey);
389 befs_debug(sb, "<--- %s ERROR", __func__);
390 return BEFS_BT_NOT_FOUND;
394 * befs_btree_read - Traverse leafnodes of a btree
395 * @sb: Filesystem superblock
396 * @ds: Datastream containing btree
397 * @key_no: Key number (alphabetical order) of key to read
398 * @bufsize: Size of the buffer to return key in
399 * @keybuf: Pointer to a buffer to put the key in
400 * @keysize: Length of the returned key
401 * @value: Value stored with the returned key
403 * Here's how it works: Key_no is the index of the key/value pair to
404 * return in keybuf/value.
405 * Bufsize is the size of keybuf (BEFS_NAME_LEN+1 is a good size). Keysize is
406 * the number of characters in the key (just a convenience).
409 * Get the first leafnode of the tree. See if the requested key is in that
410 * node. If not, follow the node->right link to the next leafnode. Repeat
411 * until the (key_no)th key is found or the tree is out of keys.
414 befs_btree_read(struct super_block *sb, const befs_data_stream *ds,
415 loff_t key_no, size_t bufsize, char *keybuf, size_t * keysize,
418 struct befs_btree_node *this_node;
419 befs_btree_super bt_super;
429 befs_debug(sb, "---> %s", __func__);
431 if (befs_bt_read_super(sb, ds, &bt_super) != BEFS_OK) {
433 "befs_btree_read() failed to read index superblock");
437 this_node = kmalloc(sizeof(struct befs_btree_node), GFP_NOFS);
438 if (this_node == NULL) {
439 befs_error(sb, "befs_btree_read() failed to allocate %zu "
440 "bytes of memory", sizeof(struct befs_btree_node));
444 node_off = bt_super.root_node_ptr;
445 this_node->bh = NULL;
447 /* seeks down to first leafnode, reads it into this_node */
448 res = befs_btree_seekleaf(sb, ds, &bt_super, this_node, &node_off);
449 if (res == BEFS_BT_EMPTY) {
450 brelse(this_node->bh);
454 befs_debug(sb, "<--- %s Tree is EMPTY", __func__);
455 return BEFS_BT_EMPTY;
456 } else if (res == BEFS_ERR) {
460 /* find the leaf node containing the key_no key */
462 while (key_sum + this_node->head.all_key_count <= key_no) {
464 /* no more nodes to look in: key_no is too large */
465 if (this_node->head.right == BEFS_BT_INVAL) {
469 "<--- %s END of keys at %llu", __func__,
471 key_sum + this_node->head.all_key_count);
472 brelse(this_node->bh);
477 key_sum += this_node->head.all_key_count;
478 node_off = this_node->head.right;
480 if (befs_bt_read_node(sb, ds, this_node, node_off) != BEFS_OK) {
481 befs_error(sb, "%s failed to read node at %llu",
482 __func__, (unsigned long long)node_off);
487 /* how many keys into this_node is key_no */
488 cur_key = key_no - key_sum;
490 /* get pointers to datastructures within the node body */
491 valarray = befs_bt_valarray(this_node);
493 keystart = befs_bt_get_key(sb, this_node, cur_key, &keylen);
495 befs_debug(sb, "Read [%llu,%d]: keysize %d",
496 (long long unsigned int)node_off, (int)cur_key,
499 if (bufsize < keylen + 1) {
500 befs_error(sb, "%s keybuf too small (%zu) "
501 "for key of size %d", __func__, bufsize, keylen);
502 brelse(this_node->bh);
506 strlcpy(keybuf, keystart, keylen + 1);
507 *value = fs64_to_cpu(sb, valarray[cur_key]);
510 befs_debug(sb, "Read [%llu,%d]: Key \"%.*s\", Value %llu", node_off,
511 cur_key, keylen, keybuf, *value);
513 brelse(this_node->bh);
516 befs_debug(sb, "<--- %s", __func__);
526 befs_debug(sb, "<--- %s ERROR", __func__);
531 * befs_btree_seekleaf - Find the first leafnode in the btree
532 * @sb: Filesystem superblock
533 * @ds: Datastream containing btree
534 * @bt_super: Pointer to the superblock of the btree
535 * @this_node: Buffer to return the leafnode in
536 * @node_off: Pointer to offset of current node within datastream. Modified
539 * Helper function for btree traverse. Moves the current position to the
540 * start of the first leaf node.
542 * Also checks for an empty tree. If there are no keys, returns BEFS_BT_EMPTY.
545 befs_btree_seekleaf(struct super_block *sb, const befs_data_stream *ds,
546 befs_btree_super *bt_super,
547 struct befs_btree_node *this_node,
548 befs_off_t * node_off)
551 befs_debug(sb, "---> %s", __func__);
553 if (befs_bt_read_node(sb, ds, this_node, *node_off) != BEFS_OK) {
554 befs_error(sb, "%s failed to read "
555 "node at %llu", __func__, *node_off);
558 befs_debug(sb, "Seekleaf to root node %llu", *node_off);
560 if (this_node->head.all_key_count == 0 && befs_leafnode(this_node)) {
561 befs_debug(sb, "<--- %s Tree is EMPTY", __func__);
562 return BEFS_BT_EMPTY;
565 while (!befs_leafnode(this_node)) {
567 if (this_node->head.all_key_count == 0) {
568 befs_debug(sb, "%s encountered "
569 "an empty interior node: %llu. Using Overflow "
570 "node: %llu", __func__, *node_off,
571 this_node->head.overflow);
572 *node_off = this_node->head.overflow;
574 fs64 *valarray = befs_bt_valarray(this_node);
575 *node_off = fs64_to_cpu(sb, valarray[0]);
577 if (befs_bt_read_node(sb, ds, this_node, *node_off) != BEFS_OK) {
578 befs_error(sb, "%s failed to read "
579 "node at %llu", __func__, *node_off);
583 befs_debug(sb, "Seekleaf to child node %llu", *node_off);
585 befs_debug(sb, "Node %llu is a leaf node", *node_off);
590 befs_debug(sb, "<--- %s ERROR", __func__);
595 * befs_leafnode - Determine if the btree node is a leaf node or an
597 * @node: Pointer to node structure to test
599 * Return 1 if leaf, 0 if interior
602 befs_leafnode(struct befs_btree_node *node)
604 /* all interior nodes (and only interior nodes) have an overflow node */
605 if (node->head.overflow == BEFS_BT_INVAL)
612 * befs_bt_keylen_index - Finds start of keylen index in a node
613 * @node: Pointer to the node structure to find the keylen index within
615 * Returns a pointer to the start of the key length index array
616 * of the B+tree node *@node
618 * "The length of all the keys in the node is added to the size of the
619 * header and then rounded up to a multiple of four to get the beginning
620 * of the key length index" (p.88, practical filesystem design).
622 * Except that rounding up to 8 works, and rounding up to 4 doesn't.
625 befs_bt_keylen_index(struct befs_btree_node *node)
627 const int keylen_align = 8;
628 unsigned long int off =
629 (sizeof (befs_btree_nodehead) + node->head.all_key_length);
630 ulong tmp = off % keylen_align;
633 off += keylen_align - tmp;
635 return (fs16 *) ((void *) node->od_node + off);
639 * befs_bt_valarray - Finds the start of value array in a node
640 * @node: Pointer to the node structure to find the value array within
642 * Returns a pointer to the start of the value array
643 * of the node pointed to by the node header
646 befs_bt_valarray(struct befs_btree_node *node)
648 void *keylen_index_start = (void *) befs_bt_keylen_index(node);
649 size_t keylen_index_size = node->head.all_key_count * sizeof (fs16);
651 return (fs64 *) (keylen_index_start + keylen_index_size);
655 * befs_bt_keydata - Finds start of keydata array in a node
656 * @node: Pointer to the node structure to find the keydata array within
658 * Returns a pointer to the start of the keydata array
659 * of the node pointed to by the node header
662 befs_bt_keydata(struct befs_btree_node *node)
664 return (char *) ((void *) node->od_node + sizeof (befs_btree_nodehead));
668 * befs_bt_get_key - returns a pointer to the start of a key
669 * @sb: filesystem superblock
670 * @node: node in which to look for the key
671 * @index: the index of the key to get
672 * @keylen: modified to be the length of the key at @index
674 * Returns a valid pointer into @node on success.
675 * Returns NULL on failure (bad input) and sets *@keylen = 0
678 befs_bt_get_key(struct super_block *sb, struct befs_btree_node *node,
679 int index, u16 * keylen)
685 if (index < 0 || index > node->head.all_key_count) {
690 keystart = befs_bt_keydata(node);
691 keylen_index = befs_bt_keylen_index(node);
696 prev_key_end = fs16_to_cpu(sb, keylen_index[index - 1]);
698 *keylen = fs16_to_cpu(sb, keylen_index[index]) - prev_key_end;
700 return keystart + prev_key_end;
704 * befs_compare_strings - compare two strings
705 * @key1: pointer to the first key to be compared
706 * @keylen1: length in bytes of key1
707 * @key2: pointer to the second key to be compared
708 * @keylen2: length in bytes of key2
710 * Returns 0 if @key1 and @key2 are equal.
711 * Returns >0 if @key1 is greater.
712 * Returns <0 if @key2 is greater.
715 befs_compare_strings(const void *key1, int keylen1,
716 const void *key2, int keylen2)
718 int len = min_t(int, keylen1, keylen2);
719 int result = strncmp(key1, key2, len);
721 result = keylen1 - keylen2;
725 /* These will be used for non-string keyed btrees */
728 btree_compare_int32(cont void *key1, int keylen1, const void *key2, int keylen2)
730 return *(int32_t *) key1 - *(int32_t *) key2;
734 btree_compare_uint32(cont void *key1, int keylen1,
735 const void *key2, int keylen2)
737 if (*(u_int32_t *) key1 == *(u_int32_t *) key2)
739 else if (*(u_int32_t *) key1 > *(u_int32_t *) key2)
745 btree_compare_int64(cont void *key1, int keylen1, const void *key2, int keylen2)
747 if (*(int64_t *) key1 == *(int64_t *) key2)
749 else if (*(int64_t *) key1 > *(int64_t *) key2)
756 btree_compare_uint64(cont void *key1, int keylen1,
757 const void *key2, int keylen2)
759 if (*(u_int64_t *) key1 == *(u_int64_t *) key2)
761 else if (*(u_int64_t *) key1 > *(u_int64_t *) key2)
768 btree_compare_float(cont void *key1, int keylen1, const void *key2, int keylen2)
770 float result = *(float *) key1 - *(float *) key2;
774 return (result < 0.0f) ? -1 : 1;
778 btree_compare_double(cont void *key1, int keylen1,
779 const void *key2, int keylen2)
781 double result = *(double *) key1 - *(double *) key2;
785 return (result < 0.0) ? -1 : 1;