xref: /linux/net/ipv4/inet_hashtables.c (revision 13abf8130139c2ccd4962a7e5a8902be5e6cb5a7)
1 /*
2  * INET		An implementation of the TCP/IP protocol suite for the LINUX
3  *		operating system.  INET is implemented using the BSD Socket
4  *		interface as the means of communication with the user level.
5  *
6  *		Generic INET transport hashtables
7  *
8  * Authors:	Lotsa people, from code originally in tcp
9  *
10  *	This program is free software; you can redistribute it and/or
11  *      modify it under the terms of the GNU General Public License
12  *      as published by the Free Software Foundation; either version
13  *      2 of the License, or (at your option) any later version.
14  */
15 
16 #include <linux/config.h>
17 #include <linux/module.h>
18 #include <linux/sched.h>
19 #include <linux/slab.h>
20 #include <linux/wait.h>
21 
22 #include <net/inet_connection_sock.h>
23 #include <net/inet_hashtables.h>
24 
25 /*
26  * Allocate and initialize a new local port bind bucket.
27  * The bindhash mutex for snum's hash chain must be held here.
28  */
29 struct inet_bind_bucket *inet_bind_bucket_create(kmem_cache_t *cachep,
30 						 struct inet_bind_hashbucket *head,
31 						 const unsigned short snum)
32 {
33 	struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, SLAB_ATOMIC);
34 
35 	if (tb != NULL) {
36 		tb->port      = snum;
37 		tb->fastreuse = 0;
38 		INIT_HLIST_HEAD(&tb->owners);
39 		hlist_add_head(&tb->node, &head->chain);
40 	}
41 	return tb;
42 }
43 
44 EXPORT_SYMBOL(inet_bind_bucket_create);
45 
46 /*
47  * Caller must hold hashbucket lock for this tb with local BH disabled
48  */
49 void inet_bind_bucket_destroy(kmem_cache_t *cachep, struct inet_bind_bucket *tb)
50 {
51 	if (hlist_empty(&tb->owners)) {
52 		__hlist_del(&tb->node);
53 		kmem_cache_free(cachep, tb);
54 	}
55 }
56 
57 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
58 		    const unsigned short snum)
59 {
60 	inet_sk(sk)->num = snum;
61 	sk_add_bind_node(sk, &tb->owners);
62 	inet_csk(sk)->icsk_bind_hash = tb;
63 }
64 
65 EXPORT_SYMBOL(inet_bind_hash);
66 
67 /*
68  * Get rid of any references to a local port held by the given sock.
69  */
70 static void __inet_put_port(struct inet_hashinfo *hashinfo, struct sock *sk)
71 {
72 	const int bhash = inet_bhashfn(inet_sk(sk)->num, hashinfo->bhash_size);
73 	struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
74 	struct inet_bind_bucket *tb;
75 
76 	spin_lock(&head->lock);
77 	tb = inet_csk(sk)->icsk_bind_hash;
78 	__sk_del_bind_node(sk);
79 	inet_csk(sk)->icsk_bind_hash = NULL;
80 	inet_sk(sk)->num = 0;
81 	inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
82 	spin_unlock(&head->lock);
83 }
84 
85 void inet_put_port(struct inet_hashinfo *hashinfo, struct sock *sk)
86 {
87 	local_bh_disable();
88 	__inet_put_port(hashinfo, sk);
89 	local_bh_enable();
90 }
91 
92 EXPORT_SYMBOL(inet_put_port);
93 
94 /*
95  * This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
96  * Look, when several writers sleep and reader wakes them up, all but one
97  * immediately hit write lock and grab all the cpus. Exclusive sleep solves
98  * this, _but_ remember, it adds useless work on UP machines (wake up each
99  * exclusive lock release). It should be ifdefed really.
100  */
101 void inet_listen_wlock(struct inet_hashinfo *hashinfo)
102 {
103 	write_lock(&hashinfo->lhash_lock);
104 
105 	if (atomic_read(&hashinfo->lhash_users)) {
106 		DEFINE_WAIT(wait);
107 
108 		for (;;) {
109 			prepare_to_wait_exclusive(&hashinfo->lhash_wait,
110 						  &wait, TASK_UNINTERRUPTIBLE);
111 			if (!atomic_read(&hashinfo->lhash_users))
112 				break;
113 			write_unlock_bh(&hashinfo->lhash_lock);
114 			schedule();
115 			write_lock_bh(&hashinfo->lhash_lock);
116 		}
117 
118 		finish_wait(&hashinfo->lhash_wait, &wait);
119 	}
120 }
121 
122 EXPORT_SYMBOL(inet_listen_wlock);
123 
124 /*
125  * Don't inline this cruft. Here are some nice properties to exploit here. The
126  * BSD API does not allow a listening sock to specify the remote port nor the
127  * remote address for the connection. So always assume those are both
128  * wildcarded during the search since they can never be otherwise.
129  */
130 struct sock *__inet_lookup_listener(const struct hlist_head *head, const u32 daddr,
131 				    const unsigned short hnum, const int dif)
132 {
133 	struct sock *result = NULL, *sk;
134 	const struct hlist_node *node;
135 	int hiscore = -1;
136 
137 	sk_for_each(sk, node, head) {
138 		const struct inet_sock *inet = inet_sk(sk);
139 
140 		if (inet->num == hnum && !ipv6_only_sock(sk)) {
141 			const __u32 rcv_saddr = inet->rcv_saddr;
142 			int score = sk->sk_family == PF_INET ? 1 : 0;
143 
144 			if (rcv_saddr) {
145 				if (rcv_saddr != daddr)
146 					continue;
147 				score += 2;
148 			}
149 			if (sk->sk_bound_dev_if) {
150 				if (sk->sk_bound_dev_if != dif)
151 					continue;
152 				score += 2;
153 			}
154 			if (score == 5)
155 				return sk;
156 			if (score > hiscore) {
157 				hiscore	= score;
158 				result	= sk;
159 			}
160 		}
161 	}
162 	return result;
163 }
164 
165 EXPORT_SYMBOL_GPL(__inet_lookup_listener);
166