]> git.proxmox.com Git - mirror_ubuntu-bionic-kernel.git/blob - net/ipv4/inet_hashtables.c
[INET]: Generalise tcp_v4_lookup_listener
[mirror_ubuntu-bionic-kernel.git] / net / ipv4 / inet_hashtables.c
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_hashtables.h>
23
24 /*
25 * Allocate and initialize a new local port bind bucket.
26 * The bindhash mutex for snum's hash chain must be held here.
27 */
28 struct inet_bind_bucket *inet_bind_bucket_create(kmem_cache_t *cachep,
29 struct inet_bind_hashbucket *head,
30 const unsigned short snum)
31 {
32 struct inet_bind_bucket *tb = kmem_cache_alloc(cachep, SLAB_ATOMIC);
33
34 if (tb != NULL) {
35 tb->port = snum;
36 tb->fastreuse = 0;
37 INIT_HLIST_HEAD(&tb->owners);
38 hlist_add_head(&tb->node, &head->chain);
39 }
40 return tb;
41 }
42
43 EXPORT_SYMBOL(inet_bind_bucket_create);
44
45 /*
46 * Caller must hold hashbucket lock for this tb with local BH disabled
47 */
48 void inet_bind_bucket_destroy(kmem_cache_t *cachep, struct inet_bind_bucket *tb)
49 {
50 if (hlist_empty(&tb->owners)) {
51 __hlist_del(&tb->node);
52 kmem_cache_free(cachep, tb);
53 }
54 }
55
56 void inet_bind_hash(struct sock *sk, struct inet_bind_bucket *tb,
57 const unsigned short snum)
58 {
59 struct inet_sock *inet = inet_sk(sk);
60 inet->num = snum;
61 sk_add_bind_node(sk, &tb->owners);
62 inet->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 struct inet_sock *inet = inet_sk(sk);
73 const int bhash = inet_bhashfn(inet->num, hashinfo->bhash_size);
74 struct inet_bind_hashbucket *head = &hashinfo->bhash[bhash];
75 struct inet_bind_bucket *tb;
76
77 spin_lock(&head->lock);
78 tb = inet->bind_hash;
79 __sk_del_bind_node(sk);
80 inet->bind_hash = NULL;
81 inet->num = 0;
82 inet_bind_bucket_destroy(hashinfo->bind_bucket_cachep, tb);
83 spin_unlock(&head->lock);
84 }
85
86 void inet_put_port(struct inet_hashinfo *hashinfo, struct sock *sk)
87 {
88 local_bh_disable();
89 __inet_put_port(hashinfo, sk);
90 local_bh_enable();
91 }
92
93 EXPORT_SYMBOL(inet_put_port);
94
95 /*
96 * This lock without WQ_FLAG_EXCLUSIVE is good on UP and it can be very bad on SMP.
97 * Look, when several writers sleep and reader wakes them up, all but one
98 * immediately hit write lock and grab all the cpus. Exclusive sleep solves
99 * this, _but_ remember, it adds useless work on UP machines (wake up each
100 * exclusive lock release). It should be ifdefed really.
101 */
102 void inet_listen_wlock(struct inet_hashinfo *hashinfo)
103 {
104 write_lock(&hashinfo->lhash_lock);
105
106 if (atomic_read(&hashinfo->lhash_users)) {
107 DEFINE_WAIT(wait);
108
109 for (;;) {
110 prepare_to_wait_exclusive(&hashinfo->lhash_wait,
111 &wait, TASK_UNINTERRUPTIBLE);
112 if (!atomic_read(&hashinfo->lhash_users))
113 break;
114 write_unlock_bh(&hashinfo->lhash_lock);
115 schedule();
116 write_lock_bh(&hashinfo->lhash_lock);
117 }
118
119 finish_wait(&hashinfo->lhash_wait, &wait);
120 }
121 }
122
123 EXPORT_SYMBOL(inet_listen_wlock);
124
125 /*
126 * Don't inline this cruft. Here are some nice properties to exploit here. The
127 * BSD API does not allow a listening sock to specify the remote port nor the
128 * remote address for the connection. So always assume those are both
129 * wildcarded during the search since they can never be otherwise.
130 */
131 struct sock *__inet_lookup_listener(const struct hlist_head *head, const u32 daddr,
132 const unsigned short hnum, const int dif)
133 {
134 struct sock *result = NULL, *sk;
135 const struct hlist_node *node;
136 int hiscore = -1;
137
138 sk_for_each(sk, node, head) {
139 const struct inet_sock *inet = inet_sk(sk);
140
141 if (inet->num == hnum && !ipv6_only_sock(sk)) {
142 const __u32 rcv_saddr = inet->rcv_saddr;
143 int score = sk->sk_family == PF_INET ? 1 : 0;
144
145 if (rcv_saddr) {
146 if (rcv_saddr != daddr)
147 continue;
148 score += 2;
149 }
150 if (sk->sk_bound_dev_if) {
151 if (sk->sk_bound_dev_if != dif)
152 continue;
153 score += 2;
154 }
155 if (score == 5)
156 return sk;
157 if (score > hiscore) {
158 hiscore = score;
159 result = sk;
160 }
161 }
162 }
163 return result;
164 }