]>
git.proxmox.com Git - mirror_zfs.git/blob - include/sys/btree.h
883abb5181c93eeedf695b338ce613c4ac056e4f
4 * This file and its contents are supplied under the terms of the
5 * Common Development and Distribution License ("CDDL"), version 1.0.
6 * You may only use this file in accordance with the terms of version
9 * A full copy of the text of the CDDL should have accompanied this
10 * source. A copy of the CDDL is also available via the Internet at
11 * http://www.illumos.org/license/CDDL.
16 * Copyright (c) 2019 by Delphix. All rights reserved.
26 #include <sys/zfs_context.h>
29 * This file defines the interface for a B-Tree implementation for ZFS. The
30 * tree can be used to store arbitrary sortable data types with low overhead
31 * and good operation performance. In addition the tree intelligently
32 * optimizes bulk in-order insertions to improve memory use and performance.
34 * Note that for all B-Tree functions, the values returned are pointers to the
35 * internal copies of the data in the tree. The internal data can only be
36 * safely mutated if the changes cannot change the ordering of the element
37 * with respect to any other elements in the tree.
39 * The major drawback of the B-Tree is that any returned elements or indexes
40 * are only valid until a side-effectful operation occurs, since these can
41 * result in reallocation or relocation of data. Side effectful operations are
42 * defined as insertion, removal, and zfs_btree_destroy_nodes.
44 * The B-Tree has two types of nodes: core nodes, and leaf nodes. Core
45 * nodes have an array of children pointing to other nodes, and an array of
46 * elements that act as separators between the elements of the subtrees rooted
47 * at its children. Leaf nodes only contain data elements, and form the bottom
48 * layer of the tree. Unlike B+ Trees, in this B-Tree implementation the
49 * elements in the core nodes are not copies of or references to leaf node
50 * elements. Each element occurs only once in the tree, no matter what kind
53 * The tree's height is the same throughout, unlike many other forms of search
54 * tree. Each node (except for the root) must be between half minus one and
55 * completely full of elements (and children) at all times. Any operation that
56 * would put the node outside of that range results in a rebalancing operation
57 * (taking, merging, or splitting).
59 * This tree was implemented using descriptions from Wikipedia's articles on
60 * B-Trees and B+ Trees.
64 * Decreasing these values results in smaller memmove operations, but more of
65 * them, and increased memory overhead. Increasing these values results in
66 * higher variance in operation time, and reduces memory overhead.
68 #define BTREE_CORE_ELEMS 126
69 #define BTREE_LEAF_SIZE 4096
71 extern kmem_cache_t
*zfs_btree_leaf_cache
;
73 typedef struct zfs_btree_hdr
{
74 struct zfs_btree_core
*bth_parent
;
76 * Set to -1 to indicate core nodes. Other values represent first
77 * valid element offset for leaf nodes.
81 * For both leaf and core nodes, represents the number of elements in
82 * the node. For core nodes, they will have bth_count + 1 children.
87 typedef struct zfs_btree_core
{
88 zfs_btree_hdr_t btc_hdr
;
89 zfs_btree_hdr_t
*btc_children
[BTREE_CORE_ELEMS
+ 1];
93 typedef struct zfs_btree_leaf
{
94 zfs_btree_hdr_t btl_hdr
;
98 typedef struct zfs_btree_index
{
99 zfs_btree_hdr_t
*bti_node
;
102 * True if the location is before the list offset, false if it's at
105 boolean_t bti_before
;
108 typedef struct btree
{
109 int (*bt_compar
) (const void *, const void *);
112 uint32_t bt_leaf_cap
;
114 uint64_t bt_num_elems
;
115 uint64_t bt_num_nodes
;
116 zfs_btree_hdr_t
*bt_root
;
117 zfs_btree_leaf_t
*bt_bulk
; // non-null if bulk loading
121 * Allocate and deallocate caches for btree nodes.
123 void zfs_btree_init(void);
124 void zfs_btree_fini(void);
127 * Initialize an B-Tree. Arguments are:
129 * tree - the tree to be initialized
130 * compar - function to compare two nodes, it must return exactly: -1, 0, or +1
131 * -1 for <, 0 for ==, and +1 for >
132 * size - the value of sizeof(struct my_type)
133 * lsize - custom leaf size
135 void zfs_btree_create(zfs_btree_t
*, int (*) (const void *, const void *),
137 void zfs_btree_create_custom(zfs_btree_t
*, int (*)(const void *, const void *),
141 * Find a node with a matching value in the tree. Returns the matching node
142 * found. If not found, it returns NULL and then if "where" is not NULL it sets
143 * "where" for use with zfs_btree_add_idx() or zfs_btree_nearest().
145 * node - node that has the value being looked for
146 * where - position for use with zfs_btree_nearest() or zfs_btree_add_idx(),
149 void *zfs_btree_find(zfs_btree_t
*, const void *, zfs_btree_index_t
*);
152 * Insert a node into the tree.
154 * node - the node to insert
155 * where - position as returned from zfs_btree_find()
157 void zfs_btree_add_idx(zfs_btree_t
*, const void *, const zfs_btree_index_t
*);
160 * Return the first or last valued node in the tree. Will return NULL if the
161 * tree is empty. The index can be NULL if the location of the first or last
162 * element isn't required.
164 void *zfs_btree_first(zfs_btree_t
*, zfs_btree_index_t
*);
165 void *zfs_btree_last(zfs_btree_t
*, zfs_btree_index_t
*);
168 * Return the next or previous valued node in the tree. The second index can
169 * safely be NULL, if the location of the next or previous value isn't
172 void *zfs_btree_next(zfs_btree_t
*, const zfs_btree_index_t
*,
173 zfs_btree_index_t
*);
174 void *zfs_btree_prev(zfs_btree_t
*, const zfs_btree_index_t
*,
175 zfs_btree_index_t
*);
178 * Get a value from a tree and an index.
180 void *zfs_btree_get(zfs_btree_t
*, zfs_btree_index_t
*);
183 * Add a single value to the tree. The value must not compare equal to any
184 * other node already in the tree. Note that the value will be copied out, not
185 * inserted directly. It is safe to free or destroy the value once this
188 void zfs_btree_add(zfs_btree_t
*, const void *);
191 * Remove a single value from the tree. The value must be in the tree. The
192 * pointer passed in may be a pointer into a tree-controlled buffer, but it
195 void zfs_btree_remove(zfs_btree_t
*, const void *);
198 * Remove the value at the given location from the tree.
200 void zfs_btree_remove_idx(zfs_btree_t
*, zfs_btree_index_t
*);
203 * Return the number of nodes in the tree
205 ulong_t
zfs_btree_numnodes(zfs_btree_t
*);
208 * Used to destroy any remaining nodes in a tree. The cookie argument should
209 * be initialized to NULL before the first call. Returns a node that has been
210 * removed from the tree and may be free()'d. Returns NULL when the tree is
213 * Once you call zfs_btree_destroy_nodes(), you can only continuing calling it
214 * and finally zfs_btree_destroy(). No other B-Tree routines will be valid.
216 * cookie - an index used to save state between calls to
217 * zfs_btree_destroy_nodes()
221 * struct my_data *node;
222 * zfs_btree_index_t *cookie;
225 * while ((node = zfs_btree_destroy_nodes(tree, &cookie)) != NULL)
226 * data_destroy(node);
227 * zfs_btree_destroy(tree);
229 void *zfs_btree_destroy_nodes(zfs_btree_t
*, zfs_btree_index_t
**);
232 * Destroys all nodes in the tree quickly. This doesn't give the caller an
233 * opportunity to iterate over each node and do its own cleanup; for that, use
234 * zfs_btree_destroy_nodes().
236 void zfs_btree_clear(zfs_btree_t
*);
239 * Final destroy of an B-Tree. Arguments are:
241 * tree - the empty tree to destroy
243 void zfs_btree_destroy(zfs_btree_t
*tree
);
245 /* Runs a variety of self-checks on the btree to verify integrity. */
246 void zfs_btree_verify(zfs_btree_t
*tree
);
252 #endif /* _BTREE_H */