2 * To speed up listener socket lookup, create an array to store all sockets
3 * listening on the same port. This allows a decision to be made after finding
7 #include <net/sock_reuseport.h>
8 #include <linux/rcupdate.h>
10 #define INIT_SOCKS 128
12 static DEFINE_SPINLOCK(reuseport_lock);
14 static struct sock_reuseport *__reuseport_alloc(u16 max_socks)
16 size_t size = sizeof(struct sock_reuseport) +
17 sizeof(struct sock *) * max_socks;
18 struct sock_reuseport *reuse = kzalloc(size, GFP_ATOMIC);
23 reuse->max_socks = max_socks;
28 int reuseport_alloc(struct sock *sk)
30 struct sock_reuseport *reuse;
32 /* bh lock used since this function call may precede hlist lock in
33 * soft irq of receive path or setsockopt from process context
35 spin_lock_bh(&reuseport_lock);
36 WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
37 lockdep_is_held(&reuseport_lock)),
38 "multiple allocations for the same socket");
39 reuse = __reuseport_alloc(INIT_SOCKS);
41 spin_unlock_bh(&reuseport_lock);
47 rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
49 spin_unlock_bh(&reuseport_lock);
53 EXPORT_SYMBOL(reuseport_alloc);
55 static struct sock_reuseport *reuseport_grow(struct sock_reuseport *reuse)
57 struct sock_reuseport *more_reuse;
58 u32 more_socks_size, i;
60 more_socks_size = reuse->max_socks * 2U;
61 if (more_socks_size > U16_MAX)
64 more_reuse = __reuseport_alloc(more_socks_size);
68 more_reuse->max_socks = more_socks_size;
69 more_reuse->num_socks = reuse->num_socks;
71 memcpy(more_reuse->socks, reuse->socks,
72 reuse->num_socks * sizeof(struct sock *));
74 for (i = 0; i < reuse->num_socks; ++i)
75 rcu_assign_pointer(reuse->socks[i]->sk_reuseport_cb,
78 kfree_rcu(reuse, rcu);
83 * reuseport_add_sock - Add a socket to the reuseport group of another.
84 * @sk: New socket to add to the group.
85 * @sk2: Socket belonging to the existing reuseport group.
86 * May return ENOMEM and not add socket to group under memory pressure.
88 int reuseport_add_sock(struct sock *sk, const struct sock *sk2)
90 struct sock_reuseport *reuse;
92 spin_lock_bh(&reuseport_lock);
93 reuse = rcu_dereference_protected(sk2->sk_reuseport_cb,
94 lockdep_is_held(&reuseport_lock)),
95 WARN_ONCE(rcu_dereference_protected(sk->sk_reuseport_cb,
96 lockdep_is_held(&reuseport_lock)),
97 "socket already in reuseport group");
99 if (reuse->num_socks == reuse->max_socks) {
100 reuse = reuseport_grow(reuse);
102 spin_unlock_bh(&reuseport_lock);
107 reuse->socks[reuse->num_socks] = sk;
108 /* paired with smp_rmb() in reuseport_select_sock() */
111 rcu_assign_pointer(sk->sk_reuseport_cb, reuse);
113 spin_unlock_bh(&reuseport_lock);
117 EXPORT_SYMBOL(reuseport_add_sock);
119 void reuseport_detach_sock(struct sock *sk)
121 struct sock_reuseport *reuse;
124 spin_lock_bh(&reuseport_lock);
125 reuse = rcu_dereference_protected(sk->sk_reuseport_cb,
126 lockdep_is_held(&reuseport_lock));
127 rcu_assign_pointer(sk->sk_reuseport_cb, NULL);
129 for (i = 0; i < reuse->num_socks; i++) {
130 if (reuse->socks[i] == sk) {
131 reuse->socks[i] = reuse->socks[reuse->num_socks - 1];
133 if (reuse->num_socks == 0)
134 kfree_rcu(reuse, rcu);
138 spin_unlock_bh(&reuseport_lock);
140 EXPORT_SYMBOL(reuseport_detach_sock);
143 * reuseport_select_sock - Select a socket from an SO_REUSEPORT group.
144 * @sk: First socket in the group.
145 * @hash: Use this hash to select.
146 * Returns a socket that should receive the packet (or NULL on error).
148 struct sock *reuseport_select_sock(struct sock *sk, u32 hash)
150 struct sock_reuseport *reuse;
151 struct sock *sk2 = NULL;
155 reuse = rcu_dereference(sk->sk_reuseport_cb);
157 /* if memory allocation failed or add call is not yet complete */
161 socks = READ_ONCE(reuse->num_socks);
163 /* paired with smp_wmb() in reuseport_add_sock() */
166 sk2 = reuse->socks[reciprocal_scale(hash, socks)];
173 EXPORT_SYMBOL(reuseport_select_sock);