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 nonblock,int flags,int * addr_len)50 static int unix_bpf_recvmsg(struct sock *sk, struct msghdr *msg,
51 size_t len, int nonblock, int flags,
52 int *addr_len)
53 {
54 struct unix_sock *u = unix_sk(sk);
55 struct sk_psock *psock;
56 int copied;
57
58 if (!len)
59 return 0;
60
61 psock = sk_psock_get(sk);
62 if (unlikely(!psock))
63 return __unix_recvmsg(sk, msg, len, flags);
64
65 mutex_lock(&u->iolock);
66 if (!skb_queue_empty(&sk->sk_receive_queue) &&
67 sk_psock_queue_empty(psock)) {
68 mutex_unlock(&u->iolock);
69 sk_psock_put(sk, psock);
70 return __unix_recvmsg(sk, msg, len, flags);
71 }
72
73 msg_bytes_ready:
74 copied = sk_msg_recvmsg(sk, psock, msg, len, flags);
75 if (!copied) {
76 long timeo;
77 int data;
78
79 timeo = sock_rcvtimeo(sk, nonblock);
80 data = unix_msg_wait_data(sk, psock, timeo);
81 if (data) {
82 if (!sk_psock_queue_empty(psock))
83 goto msg_bytes_ready;
84 mutex_unlock(&u->iolock);
85 sk_psock_put(sk, psock);
86 return __unix_recvmsg(sk, msg, len, flags);
87 }
88 copied = -EAGAIN;
89 }
90 mutex_unlock(&u->iolock);
91 sk_psock_put(sk, psock);
92 return copied;
93 }
94
95 static struct proto *unix_dgram_prot_saved __read_mostly;
96 static DEFINE_SPINLOCK(unix_dgram_prot_lock);
97 static struct proto unix_dgram_bpf_prot;
98
99 static struct proto *unix_stream_prot_saved __read_mostly;
100 static DEFINE_SPINLOCK(unix_stream_prot_lock);
101 static struct proto unix_stream_bpf_prot;
102
unix_dgram_bpf_rebuild_protos(struct proto * prot,const struct proto * base)103 static void unix_dgram_bpf_rebuild_protos(struct proto *prot, const struct proto *base)
104 {
105 *prot = *base;
106 prot->close = sock_map_close;
107 prot->recvmsg = unix_bpf_recvmsg;
108 prot->sock_is_readable = sk_msg_is_readable;
109 }
110
unix_stream_bpf_rebuild_protos(struct proto * prot,const struct proto * base)111 static void unix_stream_bpf_rebuild_protos(struct proto *prot,
112 const struct proto *base)
113 {
114 *prot = *base;
115 prot->close = sock_map_close;
116 prot->recvmsg = unix_bpf_recvmsg;
117 prot->sock_is_readable = sk_msg_is_readable;
118 prot->unhash = sock_map_unhash;
119 }
120
unix_dgram_bpf_check_needs_rebuild(struct proto * ops)121 static void unix_dgram_bpf_check_needs_rebuild(struct proto *ops)
122 {
123 if (unlikely(ops != smp_load_acquire(&unix_dgram_prot_saved))) {
124 spin_lock_bh(&unix_dgram_prot_lock);
125 if (likely(ops != unix_dgram_prot_saved)) {
126 unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, ops);
127 smp_store_release(&unix_dgram_prot_saved, ops);
128 }
129 spin_unlock_bh(&unix_dgram_prot_lock);
130 }
131 }
132
unix_stream_bpf_check_needs_rebuild(struct proto * ops)133 static void unix_stream_bpf_check_needs_rebuild(struct proto *ops)
134 {
135 if (unlikely(ops != smp_load_acquire(&unix_stream_prot_saved))) {
136 spin_lock_bh(&unix_stream_prot_lock);
137 if (likely(ops != unix_stream_prot_saved)) {
138 unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, ops);
139 smp_store_release(&unix_stream_prot_saved, ops);
140 }
141 spin_unlock_bh(&unix_stream_prot_lock);
142 }
143 }
144
unix_dgram_bpf_update_proto(struct sock * sk,struct sk_psock * psock,bool restore)145 int unix_dgram_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
146 {
147 if (sk->sk_type != SOCK_DGRAM)
148 return -EOPNOTSUPP;
149
150 if (restore) {
151 sk->sk_write_space = psock->saved_write_space;
152 WRITE_ONCE(sk->sk_prot, psock->sk_proto);
153 return 0;
154 }
155
156 unix_dgram_bpf_check_needs_rebuild(psock->sk_proto);
157 WRITE_ONCE(sk->sk_prot, &unix_dgram_bpf_prot);
158 return 0;
159 }
160
unix_stream_bpf_update_proto(struct sock * sk,struct sk_psock * psock,bool restore)161 int unix_stream_bpf_update_proto(struct sock *sk, struct sk_psock *psock, bool restore)
162 {
163 if (restore) {
164 sk->sk_write_space = psock->saved_write_space;
165 WRITE_ONCE(sk->sk_prot, psock->sk_proto);
166 return 0;
167 }
168
169 unix_stream_bpf_check_needs_rebuild(psock->sk_proto);
170 WRITE_ONCE(sk->sk_prot, &unix_stream_bpf_prot);
171 return 0;
172 }
173
unix_bpf_build_proto(void)174 void __init unix_bpf_build_proto(void)
175 {
176 unix_dgram_bpf_rebuild_protos(&unix_dgram_bpf_prot, &unix_dgram_proto);
177 unix_stream_bpf_rebuild_protos(&unix_stream_bpf_prot, &unix_stream_proto);
178
179 }
180