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 withing
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"
36 * The btree functions in this file are built on top of the
37 * datastream.c interface, which is in turn built on top of the
41 /* Befs B+tree structure:
43 * The first thing in the tree is the tree superblock. It tells you
44 * all kinds of useful things about the tree, like where the rootnode
45 * is located, and the size of the nodes (always 1024 with current version
48 * The rest of the tree consists of a series of nodes. Nodes contain a header
49 * (struct befs_btree_nodehead), the packed key data, an array of shorts
50 * containing the ending offsets for each of the keys, and an array of
51 * befs_off_t values. In interior nodes, the keys are the ending keys for
52 * the childnode they point to, and the values are offsets into the
53 * datastream containing the tree.
58 * The book states 2 confusing things about befs b+trees. First,
59 * it states that the overflow field of node headers is used by internal nodes
60 * to point to another node that "effectively continues this one". Here is what
61 * I believe that means. Each key in internal nodes points to another node that
62 * contains key values less than itself. Inspection reveals that the last key
63 * in the internal node is not the last key in the index. Keys that are
64 * greater than the last key in the internal node go into the overflow node.
65 * I imagine there is a performance reason for this.
67 * Second, it states that the header of a btree node is sufficient to
68 * distinguish internal nodes from leaf nodes. Without saying exactly how.
69 * After figuring out the first, it becomes obvious that internal nodes have
70 * overflow nodes and leafnodes do not.
74 * Currently, this code is only good for directory B+trees.
75 * In order to be used for other BFS indexes, it needs to be extended to handle
76 * duplicate keys and non-string keytypes (int32, int64, float, double).
80 * In memory structure of each btree node
83 befs_btree_nodehead head; /* head of node converted to cpu byteorder */
84 struct buffer_head *bh;
85 befs_btree_nodehead *od_node; /* on disk node */
89 static const befs_off_t befs_bt_inval = 0xffffffffffffffffULL;
92 static int befs_btree_seekleaf(struct super_block *sb, befs_data_stream * ds,
93 befs_btree_super * bt_super,
94 befs_btree_node * this_node,
95 befs_off_t * node_off);
97 static int befs_bt_read_super(struct super_block *sb, befs_data_stream * ds,
98 befs_btree_super * sup);
100 static int befs_bt_read_node(struct super_block *sb, befs_data_stream * ds,
101 befs_btree_node * node, befs_off_t node_off);
103 static int befs_leafnode(befs_btree_node * node);
105 static u16 *befs_bt_keylen_index(befs_btree_node * node);
107 static befs_off_t *befs_bt_valarray(befs_btree_node * node);
109 static char *befs_bt_keydata(befs_btree_node * node);
111 static int befs_find_key(struct super_block *sb, befs_btree_node * node,
112 const char *findkey, befs_off_t * value);
114 static char *befs_bt_get_key(struct super_block *sb, befs_btree_node * node,
115 int index, u16 * keylen);
117 static int befs_compare_strings(const void *key1, int keylen1,
118 const void *key2, int keylen2);
121 * befs_bt_read_super - read in btree superblock convert to cpu byteorder
122 * @sb: Filesystem superblock
123 * @ds: Datastream to read from
124 * @sup: Buffer in which to place the btree superblock
126 * Calls befs_read_datastream to read in the btree superblock and
127 * makes sure it is in cpu byteorder, byteswapping if necessary.
129 * On success, returns BEFS_OK and *@sup contains the btree superblock,
132 * On failure, BEFS_ERR is returned.
135 befs_bt_read_super(struct super_block *sb, befs_data_stream * ds,
136 befs_btree_super * sup)
138 struct buffer_head *bh = NULL;
139 befs_btree_super *od_sup = NULL;
141 befs_debug(sb, "---> befs_btree_read_super()");
143 bh = befs_read_datastream(sb, ds, 0, NULL);
146 befs_error(sb, "Couldn't read index header.");
149 od_sup = (befs_btree_super *) bh->b_data;
150 befs_dump_index_entry(sb, od_sup);
152 sup->magic = fs32_to_cpu(sb, od_sup->magic);
153 sup->node_size = fs32_to_cpu(sb, od_sup->node_size);
154 sup->max_depth = fs32_to_cpu(sb, od_sup->max_depth);
155 sup->data_type = fs32_to_cpu(sb, od_sup->data_type);
156 sup->root_node_ptr = fs64_to_cpu(sb, od_sup->root_node_ptr);
157 sup->free_node_ptr = fs64_to_cpu(sb, od_sup->free_node_ptr);
158 sup->max_size = fs64_to_cpu(sb, od_sup->max_size);
161 if (sup->magic != BEFS_BTREE_MAGIC) {
162 befs_error(sb, "Index header has bad magic.");
166 befs_debug(sb, "<--- befs_btree_read_super()");
170 befs_debug(sb, "<--- befs_btree_read_super() ERROR");
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 called first
185 * time. 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, befs_data_stream * ds,
195 befs_btree_node * node, befs_off_t node_off)
199 befs_debug(sb, "---> befs_bt_read_node()");
204 node->bh = befs_read_datastream(sb, ds, node_off, &off);
206 befs_error(sb, "befs_bt_read_node() failed to read "
207 "node at %Lu", node_off);
208 befs_debug(sb, "<--- befs_bt_read_node() ERROR");
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, "<--- befs_btree_read_node()");
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 sucess, 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 * actuall value stored with the key.
248 befs_btree_find(struct super_block *sb, befs_data_stream * ds,
249 const char *key, befs_off_t * value)
251 befs_btree_node *this_node = NULL;
252 befs_btree_super bt_super;
256 befs_debug(sb, "---> befs_btree_find() Key: %s", 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 = (befs_btree_node *) kmalloc(sizeof (befs_btree_node),
267 befs_error(sb, "befs_btree_find() failed to allocate %u "
268 "bytes of memory", sizeof (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 %Lu", node_off);
282 while (!befs_leafnode(this_node)) {
283 res = befs_find_key(sb, this_node, key, &node_off);
284 if (res == BEFS_BT_NOT_FOUND)
285 node_off = this_node->head.overflow;
286 /* if no match, go to overflow node */
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 %Lu", node_off);
294 /* at the correct leaf node now */
296 res = befs_find_key(sb, this_node, key, value);
298 brelse(this_node->bh);
301 if (res != BEFS_BT_MATCH) {
302 befs_debug(sb, "<--- befs_btree_find() Key %s not found", key);
304 return BEFS_BT_NOT_FOUND;
306 befs_debug(sb, "<--- befs_btree_find() Found key %s, value %Lu",
314 befs_debug(sb, "<--- befs_btree_find() ERROR");
319 * befs_find_key - Search for a key within a node
320 * @sb: Filesystem superblock
321 * @node: Node to find the key within
322 * @key: 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 no exact match, finds first key in node that is greater
327 * (alphabetically) than the search key and returns BEFS_BT_PARMATCH
328 * (for partial match, I guess). Can you think of something better to
331 * If no key was a match or greater than the search key, return
334 * Use binary search instead of a linear.
337 befs_find_key(struct super_block *sb, befs_btree_node * node,
338 const char *findkey, befs_off_t * value)
340 int first, last, mid;
345 befs_off_t *valarray;
347 befs_debug(sb, "---> befs_find_key() %s", findkey);
351 findkey_len = strlen(findkey);
353 /* if node can not contain key, just skeep this node */
354 last = node->head.all_key_count - 1;
355 thiskey = befs_bt_get_key(sb, node, last, &keylen);
357 eq = befs_compare_strings(thiskey, keylen, findkey, findkey_len);
359 befs_debug(sb, "<--- befs_find_key() %s not found", findkey);
360 return BEFS_BT_NOT_FOUND;
363 valarray = befs_bt_valarray(node);
365 /* simple binary search */
368 while (last >= first) {
369 mid = (last + first) / 2;
370 befs_debug(sb, "first: %d, last: %d, mid: %d", first, last,
372 thiskey = befs_bt_get_key(sb, node, mid, &keylen);
373 eq = befs_compare_strings(thiskey, keylen, findkey,
377 befs_debug(sb, "<--- befs_find_key() found %s at %d",
380 *value = fs64_to_cpu(sb, valarray[mid]);
381 return BEFS_BT_MATCH;
389 *value = fs64_to_cpu(sb, valarray[mid + 1]);
391 *value = fs64_to_cpu(sb, valarray[mid]);
392 befs_debug(sb, "<--- befs_find_key() found %s at %d", thiskey, mid);
393 return BEFS_BT_PARMATCH;
397 * befs_btree_read - Traverse leafnodes of a btree
398 * @sb: Filesystem superblock
399 * @ds: Datastream containing btree
400 * @key_no: Key number (alphabetical order) of key to read
401 * @bufsize: Size of the buffer to return key in
402 * @keybuf: Pointer to a buffer to put the key in
403 * @keysize: Length of the returned key
404 * @value: Value stored with the returned key
406 * Heres how it works: Key_no is the index of the key/value pair to
407 * return in keybuf/value.
408 * Bufsize is the size of keybuf (BEFS_NAME_LEN+1 is a good size). Keysize is
409 * the number of charecters in the key (just a convenience).
412 * Get the first leafnode of the tree. See if the requested key is in that
413 * node. If not, follow the node->right link to the next leafnode. Repeat
414 * until the (key_no)th key is found or the tree is out of keys.
417 befs_btree_read(struct super_block *sb, befs_data_stream * ds,
418 loff_t key_no, size_t bufsize, char *keybuf, size_t * keysize,
421 befs_btree_node *this_node;
422 befs_btree_super bt_super;
423 befs_off_t node_off = 0;
425 befs_off_t *valarray;
432 befs_debug(sb, "---> befs_btree_read()");
434 if (befs_bt_read_super(sb, ds, &bt_super) != BEFS_OK) {
436 "befs_btree_read() failed to read index superblock");
440 if ((this_node = (befs_btree_node *)
441 kmalloc(sizeof (befs_btree_node), GFP_NOFS)) == NULL) {
442 befs_error(sb, "befs_btree_read() failed to allocate %u "
443 "bytes of memory", sizeof (befs_btree_node));
447 node_off = bt_super.root_node_ptr;
448 this_node->bh = NULL;
450 /* seeks down to first leafnode, reads it into this_node */
451 res = befs_btree_seekleaf(sb, ds, &bt_super, this_node, &node_off);
452 if (res == BEFS_BT_EMPTY) {
453 brelse(this_node->bh);
457 befs_debug(sb, "<--- befs_btree_read() Tree is EMPTY");
458 return BEFS_BT_EMPTY;
459 } else if (res == BEFS_ERR) {
463 /* find the leaf node containing the key_no key */
465 while (key_sum + this_node->head.all_key_count <= key_no) {
467 /* no more nodes to look in: key_no is too large */
468 if (this_node->head.right == befs_bt_inval) {
472 "<--- befs_btree_read() END of keys at %Lu",
473 key_sum + this_node->head.all_key_count);
474 brelse(this_node->bh);
479 key_sum += this_node->head.all_key_count;
480 node_off = this_node->head.right;
482 if (befs_bt_read_node(sb, ds, this_node, node_off) != BEFS_OK) {
483 befs_error(sb, "befs_btree_read() failed to read "
484 "node at %Lu", node_off);
489 /* how many keys into this_node is key_no */
490 cur_key = key_no - key_sum;
492 /* get pointers to datastructures within the node body */
493 valarray = befs_bt_valarray(this_node);
495 keystart = befs_bt_get_key(sb, this_node, cur_key, &keylen);
497 befs_debug(sb, "Read [%Lu,%d]: keysize %d", node_off, cur_key, keylen);
499 if (bufsize < keylen + 1) {
500 befs_error(sb, "befs_btree_read() keybuf too small (%u) "
501 "for key of size %d", bufsize, keylen);
502 brelse(this_node->bh);
506 strncpy(keybuf, keystart, keylen);
507 *value = fs64_to_cpu(sb, valarray[cur_key]);
509 keybuf[keylen] = '\0';
511 befs_debug(sb, "Read [%Lu,%d]: Key \"%.*s\", Value %Lu", node_off,
512 cur_key, keylen, keybuf, *value);
514 brelse(this_node->bh);
517 befs_debug(sb, "<--- befs_btree_read()");
527 befs_debug(sb, "<--- befs_btree_read() ERROR");
532 * befs_btree_seekleaf - Find the first leafnode in the btree
533 * @sb: Filesystem superblock
534 * @ds: Datastream containing btree
535 * @bt_super: Pointer to the superblock of the btree
536 * @this_node: Buffer to return the leafnode in
537 * @node_off: Pointer to offset of current node within datastream. Modified
541 * Helper function for btree traverse. Moves the current position to the
542 * start of the first leaf node.
544 * Also checks for an empty tree. If there are no keys, returns BEFS_BT_EMPTY.
547 befs_btree_seekleaf(struct super_block *sb, befs_data_stream * ds,
548 befs_btree_super * bt_super, befs_btree_node * this_node,
549 befs_off_t * node_off)
552 befs_debug(sb, "---> befs_btree_seekleaf()");
554 if (befs_bt_read_node(sb, ds, this_node, *node_off) != BEFS_OK) {
555 befs_error(sb, "befs_btree_seekleaf() failed to read "
556 "node at %Lu", *node_off);
559 befs_debug(sb, "Seekleaf to root node %Lu", *node_off);
561 if (this_node->head.all_key_count == 0 && befs_leafnode(this_node)) {
562 befs_debug(sb, "<--- befs_btree_seekleaf() Tree is EMPTY");
563 return BEFS_BT_EMPTY;
566 while (!befs_leafnode(this_node)) {
568 if (this_node->head.all_key_count == 0) {
569 befs_debug(sb, "befs_btree_seekleaf() encountered "
570 "an empty interior node: %Lu. Using Overflow "
571 "node: %Lu", *node_off,
572 this_node->head.overflow);
573 *node_off = this_node->head.overflow;
575 befs_off_t *valarray = befs_bt_valarray(this_node);
576 *node_off = fs64_to_cpu(sb, valarray[0]);
578 if (befs_bt_read_node(sb, ds, this_node, *node_off) != BEFS_OK) {
579 befs_error(sb, "befs_btree_seekleaf() failed to read "
580 "node at %Lu", *node_off);
584 befs_debug(sb, "Seekleaf to child node %Lu", *node_off);
586 befs_debug(sb, "Node %Lu is a leaf node", *node_off);
591 befs_debug(sb, "<--- befs_btree_seekleaf() ERROR");
596 * befs_leafnode - Determine if the btree node is a leaf node or an
598 * @node: Pointer to node structure to test
600 * Return 1 if leaf, 0 if interior
603 befs_leafnode(befs_btree_node * node)
605 /* all interior nodes (and only interior nodes) have an overflow node */
606 if (node->head.overflow == befs_bt_inval)
613 * befs_bt_keylen_index - Finds start of keylen index in a node
614 * @node: Pointer to the node structure to find the keylen index within
616 * Returns a pointer to the start of the key length index array
617 * of the B+tree node *@node
619 * "The length of all the keys in the node is added to the size of the
620 * header and then rounded up to a multiple of four to get the beginning
621 * of the key length index" (p.88, practical filesystem design).
623 * Except that rounding up to 8 works, and rounding up to 4 doesn't.
626 befs_bt_keylen_index(befs_btree_node * node)
628 const int keylen_align = 8;
629 unsigned long int off =
630 (sizeof (befs_btree_nodehead) + node->head.all_key_length);
631 ulong tmp = off % keylen_align;
634 off += keylen_align - tmp;
636 return (u16 *) ((void *) node->od_node + off);
640 * befs_bt_valarray - Finds the start of value array in a node
641 * @node: Pointer to the node structure to find the value array within
643 * Returns a pointer to the start of the value array
644 * of the node pointed to by the node header
647 befs_bt_valarray(befs_btree_node * node)
649 void *keylen_index_start = (void *) befs_bt_keylen_index(node);
650 size_t keylen_index_size = node->head.all_key_count * sizeof (u16);
652 return (befs_off_t *) (keylen_index_start + keylen_index_size);
656 * befs_bt_keydata - Finds start of keydata array in a node
657 * @node: Pointer to the node structure to find the keydata array within
659 * Returns a pointer to the start of the keydata array
660 * of the node pointed to by the node header
663 befs_bt_keydata(befs_btree_node * node)
665 return (char *) ((void *) node->od_node + sizeof (befs_btree_nodehead));
669 * befs_bt_get_key - returns a pointer to the start of a key
670 * @sb: filesystem superblock
671 * @node: node in which to look for the key
672 * @index: the index of the key to get
673 * @keylen: modified to be the length of the key at @index
675 * Returns a valid pointer into @node on success.
676 * Returns NULL on failure (bad input) and sets *@keylen = 0
679 befs_bt_get_key(struct super_block *sb, befs_btree_node * node,
680 int index, u16 * keylen)
686 if (index < 0 || index > node->head.all_key_count) {
691 keystart = befs_bt_keydata(node);
692 keylen_index = befs_bt_keylen_index(node);
697 prev_key_end = fs16_to_cpu(sb, keylen_index[index - 1]);
699 *keylen = fs16_to_cpu(sb, keylen_index[index]) - prev_key_end;
701 return keystart + prev_key_end;
705 * befs_compare_strings - compare two strings
706 * @key1: pointer to the first key to be compared
707 * @keylen1: length in bytes of key1
708 * @key2: pointer to the second key to be compared
709 * @kelen2: length in bytes of key2
711 * Returns 0 if @key1 and @key2 are equal.
712 * Returns >0 if @key1 is greater.
713 * Returns <0 if @key2 is greater..
716 befs_compare_strings(const void *key1, int keylen1,
717 const void *key2, int keylen2)
719 int len = min_t(int, keylen1, keylen2);
720 int result = strncmp(key1, key2, len);
722 result = keylen1 - keylen2;
726 /* These will be used for non-string keyed btrees */
729 btree_compare_int32(cont void *key1, int keylen1, const void *key2, int keylen2)
731 return *(int32_t *) key1 - *(int32_t *) key2;
735 btree_compare_uint32(cont void *key1, int keylen1,
736 const void *key2, int keylen2)
738 if (*(u_int32_t *) key1 == *(u_int32_t *) key2)
740 else if (*(u_int32_t *) key1 > *(u_int32_t *) key2)
746 btree_compare_int64(cont void *key1, int keylen1, const void *key2, int keylen2)
748 if (*(int64_t *) key1 == *(int64_t *) key2)
750 else if (*(int64_t *) key1 > *(int64_t *) key2)
757 btree_compare_uint64(cont void *key1, int keylen1,
758 const void *key2, int keylen2)
760 if (*(u_int64_t *) key1 == *(u_int64_t *) key2)
762 else if (*(u_int64_t *) key1 > *(u_int64_t *) key2)
769 btree_compare_float(cont void *key1, int keylen1, const void *key2, int keylen2)
771 float result = *(float *) key1 - *(float *) key2;
775 return (result < 0.0f) ? -1 : 1;
779 btree_compare_double(cont void *key1, int keylen1,
780 const void *key2, int keylen2)
782 double result = *(double *) key1 - *(double *) key2;
786 return (result < 0.0) ? -1 : 1;