Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
1/* SPDX-License-Identifier: GPL-2.0-or-later */
2/*
3 Red Black Trees
4 (C) 1999 Andrea Arcangeli <andrea@suse.de>
5
6
7 linux/include/linux/rbtree.h
8
9 To use rbtrees you'll have to implement your own insert and search cores.
10 This will avoid us to use callbacks and to drop drammatically performances.
11 I know it's not the cleaner way, but in C (not in C++) to get
12 performances and genericity...
13
14 See Documentation/rbtree.txt for documentation and samples.
15*/
16
17#ifndef _LINUX_RBTREE_H
18#define _LINUX_RBTREE_H
19
20#include <linux/kernel.h>
21#include <linux/stddef.h>
22#include <linux/rcupdate.h>
23
24struct rb_node {
25 unsigned long __rb_parent_color;
26 struct rb_node *rb_right;
27 struct rb_node *rb_left;
28} __attribute__((aligned(sizeof(long))));
29 /* The alignment might seem pointless, but allegedly CRIS needs it */
30
31struct rb_root {
32 struct rb_node *rb_node;
33};
34
35#define rb_parent(r) ((struct rb_node *)((r)->__rb_parent_color & ~3))
36
37#define RB_ROOT (struct rb_root) { NULL, }
38#define rb_entry(ptr, type, member) container_of(ptr, type, member)
39
40#define RB_EMPTY_ROOT(root) (READ_ONCE((root)->rb_node) == NULL)
41
42/* 'empty' nodes are nodes that are known not to be inserted in an rbtree */
43#define RB_EMPTY_NODE(node) \
44 ((node)->__rb_parent_color == (unsigned long)(node))
45#define RB_CLEAR_NODE(node) \
46 ((node)->__rb_parent_color = (unsigned long)(node))
47
48
49extern void rb_insert_color(struct rb_node *, struct rb_root *);
50extern void rb_erase(struct rb_node *, struct rb_root *);
51
52
53/* Find logical next and previous nodes in a tree */
54extern struct rb_node *rb_next(const struct rb_node *);
55extern struct rb_node *rb_prev(const struct rb_node *);
56extern struct rb_node *rb_first(const struct rb_root *);
57extern struct rb_node *rb_last(const struct rb_root *);
58
59/* Postorder iteration - always visit the parent after its children */
60extern struct rb_node *rb_first_postorder(const struct rb_root *);
61extern struct rb_node *rb_next_postorder(const struct rb_node *);
62
63/* Fast replacement of a single node without remove/rebalance/add/rebalance */
64extern void rb_replace_node(struct rb_node *victim, struct rb_node *new,
65 struct rb_root *root);
66extern void rb_replace_node_rcu(struct rb_node *victim, struct rb_node *new,
67 struct rb_root *root);
68
69static inline void rb_link_node(struct rb_node *node, struct rb_node *parent,
70 struct rb_node **rb_link)
71{
72 node->__rb_parent_color = (unsigned long)parent;
73 node->rb_left = node->rb_right = NULL;
74
75 *rb_link = node;
76}
77
78static inline void rb_link_node_rcu(struct rb_node *node, struct rb_node *parent,
79 struct rb_node **rb_link)
80{
81 node->__rb_parent_color = (unsigned long)parent;
82 node->rb_left = node->rb_right = NULL;
83
84 rcu_assign_pointer(*rb_link, node);
85}
86
87#define rb_entry_safe(ptr, type, member) \
88 ({ typeof(ptr) ____ptr = (ptr); \
89 ____ptr ? rb_entry(____ptr, type, member) : NULL; \
90 })
91
92/**
93 * rbtree_postorder_for_each_entry_safe - iterate in post-order over rb_root of
94 * given type allowing the backing memory of @pos to be invalidated
95 *
96 * @pos: the 'type *' to use as a loop cursor.
97 * @n: another 'type *' to use as temporary storage
98 * @root: 'rb_root *' of the rbtree.
99 * @field: the name of the rb_node field within 'type'.
100 *
101 * rbtree_postorder_for_each_entry_safe() provides a similar guarantee as
102 * list_for_each_entry_safe() and allows the iteration to continue independent
103 * of changes to @pos by the body of the loop.
104 *
105 * Note, however, that it cannot handle other modifications that re-order the
106 * rbtree it is iterating over. This includes calling rb_erase() on @pos, as
107 * rb_erase() may rebalance the tree, causing us to miss some nodes.
108 */
109#define rbtree_postorder_for_each_entry_safe(pos, n, root, field) \
110 for (pos = rb_entry_safe(rb_first_postorder(root), typeof(*pos), field); \
111 pos && ({ n = rb_entry_safe(rb_next_postorder(&pos->field), \
112 typeof(*pos), field); 1; }); \
113 pos = n)
114
115/*
116 * Leftmost-cached rbtrees.
117 *
118 * We do not cache the rightmost node based on footprint
119 * size vs number of potential users that could benefit
120 * from O(1) rb_last(). Just not worth it, users that want
121 * this feature can always implement the logic explicitly.
122 * Furthermore, users that want to cache both pointers may
123 * find it a bit asymmetric, but that's ok.
124 */
125struct rb_root_cached {
126 struct rb_root rb_root;
127 struct rb_node *rb_leftmost;
128};
129
130#define RB_ROOT_CACHED (struct rb_root_cached) { {NULL, }, NULL }
131
132/* Same as rb_first(), but O(1) */
133#define rb_first_cached(root) (root)->rb_leftmost
134
135static inline void rb_insert_color_cached(struct rb_node *node,
136 struct rb_root_cached *root,
137 bool leftmost)
138{
139 if (leftmost)
140 root->rb_leftmost = node;
141 rb_insert_color(node, &root->rb_root);
142}
143
144static inline void rb_erase_cached(struct rb_node *node,
145 struct rb_root_cached *root)
146{
147 if (root->rb_leftmost == node)
148 root->rb_leftmost = rb_next(node);
149 rb_erase(node, &root->rb_root);
150}
151
152static inline void rb_replace_node_cached(struct rb_node *victim,
153 struct rb_node *new,
154 struct rb_root_cached *root)
155{
156 if (root->rb_leftmost == victim)
157 root->rb_leftmost = new;
158 rb_replace_node(victim, new, &root->rb_root);
159}
160
161#endif /* _LINUX_RBTREE_H */