xref: /openbmc/linux/net/unix/unix_bpf.c (revision 772f9c31)
1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2021 Cong Wang <cong.wang@bytedance.com> */
3 
4 #include <linux/skmsg.h>
5 #include <linux/bpf.h>
6 #include <net/sock.h>
7 #include <net/af_unix.h>
8 
9 #define unix_sk_has_data(__sk, __psock)					\
10 		({	!skb_queue_empty(&__sk->sk_receive_queue) ||	\
11 			!skb_queue_empty(&__psock->ingress_skb) ||	\
12 			!list_empty(&__psock->ingress_msg);		\
13 		})
14 
unix_msg_wait_data(struct sock * sk,struct sk_psock * psock,long timeo)15 static int unix_msg_wait_data(struct sock *sk, struct sk_psock *psock,
16 			      long timeo)
17 {
18 	DEFINE_WAIT_FUNC(wait, woken_wake_function);
19 	struct unix_sock *u = unix_sk(sk);
20 	int ret = 0;
21 
22 	if (sk->sk_shutdown & RCV_SHUTDOWN)
23 		return 1;
24 
25 	if (!timeo)
26 		return ret;
27 
28 	add_wait_queue(sk_sleep(sk), &wait);
29 	sk_set_bit(SOCKWQ_ASYNC_WAITDATA, sk);
30 	if (!unix_sk_has_data(sk, psock)) {
31 		mutex_unlock(&u->iolock);
32 		wait_woken(&wait, TASK_INTERRUPTIBLE, timeo);
33 		mutex_lock(&u->iolock);
34 		ret = unix_sk_has_data(sk, psock);
35 	}
36 	sk_clear_bit(SOCKWQ_ASYNC_WAITDATA, sk);
37 	remove_wait_queue(sk_sleep(sk), &wait);
38 	return ret;
39 }
40 
__unix_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags)41 static int __unix_recvmsg(struct sock *sk, struct msghdr *msg,
42 			  size_t len, int flags)
43 {
44 	if (sk->sk_type == SOCK_DGRAM)
45 		return __unix_dgram_recvmsg(sk, msg, len, flags);
46 	else
47 		return __unix_stream_recvmsg(sk, msg, len, flags);
48 }
49 
unix_bpf_recvmsg(struct sock * sk,struct msghdr * msg,size_t len,int flags,int * addr_len)50 static int unix_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
51 			    size_t len, int flags, int *addr_len)
52 {
53 	struct unix_sock *u = unix_sk(sk);
54 	struct sk_psock *psock;
55 	int copied;
56 
57 	if (flags & MSG_OOB)
58 		return -EOPNOTSUPP;
59 
60 	if (!len)
61 		return 0;
62 
63 	psock = sk_psock_get(sk);
64 	if (unlikely(!psock))
65 		return __unix_recvmsg(sk, msg, len, flags);
66 
67 	mutex_lock(&u->iolock);
68 	if (!skb_queue_empty(&sk->sk_receive_queue) &&
69 	    sk_psock_queue_empty(psock)) {
70 		mutex_unlock(&u->iolock);
71 		sk_psock_put(sk, psock);
72 		return __unix_recvmsg(sk, msg, len, flags);
73 	}
74 
75 msg_bytes_ready:
76 	copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
77 	if (!copied) {
78 		long timeo;
79 		int data;
80 
81 		timeo = sock_rcvtimeo(sk, flags & MSG_DONTWAIT);
82 		data = unix_msg_wait_data(sk, psock, timeo);
83 		if (data) {
84 			if (!sk_psock_queue_empty(psock))
85 				goto msg_bytes_ready;
86 			mutex_unlock(&u->iolock);
87 			sk_psock_put(sk, psock);
88 			return __unix_recvmsg(sk, msg, len, flags);
89 		}
90 		copied = -EAGAIN;
91 	}
92 	mutex_unlock(&u->iolock);
93 	sk_psock_put(sk, psock);
94 	return copied;
95 }
96 
97 static struct proto *unix_dgram_prot_saved __read_mostly;
98 static DEFINE_SPINLOCK(unix_dgram_prot_lock);
99 static struct proto unix_dgram_bpf_prot;
100 
101 static struct proto *unix_stream_prot_saved __read_mostly;
102 static DEFINE_SPINLOCK(unix_stream_prot_lock);
103 static struct proto unix_stream_bpf_prot;
104 
unix_dgram_bpf_rebuild_protos(struct proto * prot,const struct proto * base)105 static void unix_dgram_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
106 {
107 	*prot        = *base;
108 	prot->close  = sock_map_close;
109 	prot->recvmsg = unix_bpf_recvmsg;
110 	prot->sock_is_readable = sk_msg_is_readable;
111 }
112 
unix_stream_bpf_rebuild_protos(struct proto * prot,const struct proto * base)113 static void unix_stream_bpf_rebuild_protos(struct proto *prot,
114 					   const struct proto *base)
115 {
116 	*prot        = *base;
117 	prot->close  = sock_map_close;
118 	prot->recvmsg = unix_bpf_recvmsg;
119 	prot->sock_is_readable = sk_msg_is_readable;
120 	prot->unhash  = sock_map_unhash;
121 }
122 
unix_dgram_bpf_check_needs_rebuild(struct proto * ops)123 static void unix_dgram_bpf_check_needs_rebuild(struct proto *ops)
124 {
125 	if (unlikely(ops != smp_load_acquire(&unix_dgram_prot_saved))) {
126 		spin_lock_bh(&unix_dgram_prot_lock);
127 		if (likely(ops != unix_dgram_prot_saved)) {
128 			unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, ops);
129 			smp_store_release(&unix_dgram_prot_saved, ops);
130 		}
131 		spin_unlock_bh(&unix_dgram_prot_lock);
132 	}
133 }
134 
unix_stream_bpf_check_needs_rebuild(struct proto * ops)135 static void unix_stream_bpf_check_needs_rebuild(struct proto *ops)
136 {
137 	if (unlikely(ops != smp_load_acquire(&unix_stream_prot_saved))) {
138 		spin_lock_bh(&unix_stream_prot_lock);
139 		if (likely(ops != unix_stream_prot_saved)) {
140 			unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, ops);
141 			smp_store_release(&unix_stream_prot_saved, ops);
142 		}
143 		spin_unlock_bh(&unix_stream_prot_lock);
144 	}
145 }
146 
unix_dgram_bpf_update_proto(struct sock * sk,struct sk_psock * psock,bool restore)147 int unix_dgram_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
148 {
149 	if (sk->sk_type != SOCK_DGRAM)
150 		return -EOPNOTSUPP;
151 
152 	if (restore) {
153 		sk->sk_write_space = psock->saved_write_space;
154 		sock_replace_proto(sk, psock->sk_proto);
155 		return 0;
156 	}
157 
158 	unix_dgram_bpf_check_needs_rebuild(psock->sk_proto);
159 	sock_replace_proto(sk, &unix_dgram_bpf_prot);
160 	return 0;
161 }
162 
unix_stream_bpf_update_proto(struct sock * sk,struct sk_psock * psock,bool restore)163 int unix_stream_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
164 {
165 	struct sock *sk_pair;
166 
167 	/* Restore does not decrement the sk_pair reference yet because we must
168 	 * keep the a reference to the socket until after an RCU grace period
169 	 * and any pending sends have completed.
170 	 */
171 	if (restore) {
172 		sk->sk_write_space = psock->saved_write_space;
173 		sock_replace_proto(sk, psock->sk_proto);
174 		return 0;
175 	}
176 
177 	/* psock_update_sk_prot can be called multiple times if psock is
178 	 * added to multiple maps and/or slots in the same map. There is
179 	 * also an edge case where replacing a psock with itself can trigger
180 	 * an extra psock_update_sk_prot during the insert process. So it
181 	 * must be safe to do multiple calls. Here we need to ensure we don't
182 	 * increment the refcnt through sock_hold many times. There will only
183 	 * be a single matching destroy operation.
184 	 */
185 	if (!psock->sk_pair) {
186 		sk_pair = unix_peer(sk);
187 		sock_hold(sk_pair);
188 		psock->sk_pair = sk_pair;
189 	}
190 
191 	unix_stream_bpf_check_needs_rebuild(psock->sk_proto);
192 	sock_replace_proto(sk, &unix_stream_bpf_prot);
193 	return 0;
194 }
195 
unix_bpf_build_proto(void)196 void __init unix_bpf_build_proto(void)
197 {
198 	unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, &unix_dgram_proto);
199 	unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, &unix_stream_proto);
200 
201 }
202