xref: /linux/net/ipv4/udp_bpf.c (revision 929e30f9312514902133c45e51c79088421ab084)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2020 Cloudflare Ltd https://cloudflare.com */
3 
4 #include <linux/skmsg.h>
5 #include <net/sock.h>
6 #include <net/udp.h>
7 #include <net/inet_common.h>
8 #include <asm/ioctls.h>
9 
10 #include "udp_impl.h"
11 
12 static struct proto *udpv6_prot_saved __read_mostly;
13 
14 static int sk_udp_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
15 			  int flags, int *addr_len)
16 {
17 #if IS_ENABLED(CONFIG_IPV6)
18 	if (sk->sk_family == AF_INET6)
19 		return udpv6_prot_saved->recvmsg(sk, msg, len, flags, addr_len);
20 #endif
21 	return udp_prot.recvmsg(sk, msg, len, flags, addr_len);
22 }
23 
24 static bool udp_sk_has_data(struct sock *sk)
25 {
26 	return !skb_queue_empty(&udp_sk(sk)->reader_queue) ||
27 	       !skb_queue_empty(&sk->sk_receive_queue);
28 }
29 
30 static bool psock_has_data(struct sk_psock *psock)
31 {
32 	return !skb_queue_empty(&psock->ingress_skb) ||
33 	       !sk_psock_queue_empty(psock);
34 }
35 
36 #define udp_msg_has_data(__sk, __psock)	\
37 		({ udp_sk_has_data(__sk) || psock_has_data(__psock); })
38 
39 static int udp_msg_wait_data(struct sock *sk, struct sk_psock *psock,
40 			     long timeo)
41 {
42 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
43 	int ret = 0;
44 
45 	if (sk->sk_shutdown & RCV_SHUTDOWN)
46 		return 1;
47 
48 	if (!timeo)
49 		return ret;
50 
51 	add_wait_queue(sk_sleep(sk), &wait);
52 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
53 	ret = udp_msg_has_data(sk, psock);
54 	if (!ret) {
55 		wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
56 		ret = udp_msg_has_data(sk, psock);
57 	}
58 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
59 	remove_wait_queue(sk_sleep(sk), &wait);
60 	return ret;
61 }
62 
63 static int udp_bpf_recvmsg(struct sock *sk, struct msghdr *msg, size_t len,
64 			   int flags, int *addr_len)
65 {
66 	struct sk_psock *psock;
67 	int copied, ret;
68 
69 	if (unlikely(flags & MSG_ERRQUEUE))
70 		return inet_recv_error(sk, msg, len, addr_len);
71 
72 	if (!len)
73 		return 0;
74 
75 	psock = sk_psock_get(sk);
76 	if (unlikely(!psock))
77 		return sk_udp_recvmsg(sk, msg, len, flags, addr_len);
78 
79 	if (!psock_has_data(psock)) {
80 		ret = sk_udp_recvmsg(sk, msg, len, flags, addr_len);
81 		goto out;
82 	}
83 
84 msg_bytes_ready:
85 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
86 	if (!copied) {
87 		long timeo;
88 		int data;
89 
90 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
91 		data = udp_msg_wait_data(sk, psock, timeo);
92 		if (data) {
93 			if (psock_has_data(psock))
94 				goto msg_bytes_ready;
95 			ret = sk_udp_recvmsg(sk, msg, len, flags, addr_len);
96 			goto out;
97 		}
98 		copied = -EAGAIN;
99 	}
100 	ret = copied;
101 out:
102 	sk_psock_put(sk, psock);
103 	return ret;
104 }
105 
106 enum {
107 	UDP_BPF_IPV4,
108 	UDP_BPF_IPV6,
109 	UDP_BPF_NUM_PROTS,
110 };
111 
112 static DEFINE_SPINLOCK(udpv6_prot_lock);
113 static struct proto udp_bpf_prots[UDP_BPF_NUM_PROTS];
114 
115 static int udp_bpf_ioctl(struct sock *sk, int cmd, int *karg)
116 {
117 	if (cmd != SIOCINQ)
118 		return udp_ioctl(sk, cmd, karg);
119 
120 	/* Since we don't hold a lock, sk_receive_queue may contain data.
121 	 * BPF might only be processing this data at the moment. We only
122 	 * care about the data in the ingress_msg here.
123 	 */
124 	*karg = sk_msg_first_len(sk);
125 	return 0;
126 }
127 
128 static void udp_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
129 {
130 	*prot			= *base;
131 	prot->close		= sock_map_close;
132 	prot->recvmsg		= udp_bpf_recvmsg;
133 	prot->sock_is_readable	= sk_msg_is_readable;
134 	prot->ioctl		= udp_bpf_ioctl;
135 }
136 
137 static void udp_bpf_check_v6_needs_rebuild(struct proto *ops)
138 {
139 	if (unlikely(ops != smp_load_acquire(&udpv6_prot_saved))) {
140 		spin_lock_bh(&udpv6_prot_lock);
141 		if (likely(ops != udpv6_prot_saved)) {
142 			udp_bpf_rebuild_protos(&udp_bpf_prots[UDP_BPF_IPV6], ops);
143 			smp_store_release(&udpv6_prot_saved, ops);
144 		}
145 		spin_unlock_bh(&udpv6_prot_lock);
146 	}
147 }
148 
149 static int __init udp_bpf_v4_build_proto(void)
150 {
151 	udp_bpf_rebuild_protos(&udp_bpf_prots[UDP_BPF_IPV4], &udp_prot);
152 	return 0;
153 }
154 late_initcall(udp_bpf_v4_build_proto);
155 
156 int udp_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
157 {
158 	int family = sk->sk_family == AF_INET ? UDP_BPF_IPV4 : UDP_BPF_IPV6;
159 
160 	if (restore) {
161 		sk->sk_write_space = psock->saved_write_space;
162 		sock_replace_proto(sk, psock->sk_proto);
163 		return 0;
164 	}
165 
166 	if (sk->sk_family == AF_INET6)
167 		udp_bpf_check_v6_needs_rebuild(psock->sk_proto);
168 
169 	sock_replace_proto(sk, &udp_bpf_prots[family]);
170 	return 0;
171 }
172 EXPORT_SYMBOL_GPL(udp_bpf_update_proto);
173