bpf, sockmap: Reschedule is now done through backlog
[platform/kernel/linux-starfive.git] / net / core / stream.c
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  *     SUCS NET3:
4  *
5  *     Generic stream handling routines. These are generic for most
6  *     protocols. Even IP. Tonight 8-).
7  *     This is used because TCP, LLC (others too) layer all have mostly
8  *     identical sendmsg() and recvmsg() code.
9  *     So we (will) share it here.
10  *
11  *     Authors:        Arnaldo Carvalho de Melo <acme@conectiva.com.br>
12  *                     (from old tcp.c code)
13  *                     Alan Cox <alan@lxorguk.ukuu.org.uk> (Borrowed comments 8-))
14  */
15
16 #include <linux/module.h>
17 #include <linux/sched/signal.h>
18 #include <linux/net.h>
19 #include <linux/signal.h>
20 #include <linux/tcp.h>
21 #include <linux/wait.h>
22 #include <net/sock.h>
23
24 /**
25  * sk_stream_write_space - stream socket write_space callback.
26  * @sk: socket
27  *
28  * FIXME: write proper description
29  */
30 void sk_stream_write_space(struct sock *sk)
31 {
32         struct socket *sock = sk->sk_socket;
33         struct socket_wq *wq;
34
35         if (__sk_stream_is_writeable(sk, 1) && sock) {
36                 clear_bit(SOCK_NOSPACE, &sock->flags);
37
38                 rcu_read_lock();
39                 wq = rcu_dereference(sk->sk_wq);
40                 if (skwq_has_sleeper(wq))
41                         wake_up_interruptible_poll(&wq->wait, EPOLLOUT |
42                                                 EPOLLWRNORM | EPOLLWRBAND);
43                 if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
44                         sock_wake_async(wq, SOCK_WAKE_SPACE, POLL_OUT);
45                 rcu_read_unlock();
46         }
47 }
48
49 /**
50  * sk_stream_wait_connect - Wait for a socket to get into the connected state
51  * @sk: sock to wait on
52  * @timeo_p: for how long to wait
53  *
54  * Must be called with the socket locked.
55  */
56 int sk_stream_wait_connect(struct sock *sk, long *timeo_p)
57 {
58         DEFINE_WAIT_FUNC(wait, woken_wake_function);
59         struct task_struct *tsk = current;
60         int done;
61
62         do {
63                 int err = sock_error(sk);
64                 if (err)
65                         return err;
66                 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV))
67                         return -EPIPE;
68                 if (!*timeo_p)
69                         return -EAGAIN;
70                 if (signal_pending(tsk))
71                         return sock_intr_errno(*timeo_p);
72
73                 add_wait_queue(sk_sleep(sk), &wait);
74                 sk->sk_write_pending++;
75                 done = sk_wait_event(sk, timeo_p,
76                                      !READ_ONCE(sk->sk_err) &&
77                                      !((1 << READ_ONCE(sk->sk_state)) &
78                                        ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)), &wait);
79                 remove_wait_queue(sk_sleep(sk), &wait);
80                 sk->sk_write_pending--;
81         } while (!done);
82         return 0;
83 }
84 EXPORT_SYMBOL(sk_stream_wait_connect);
85
86 /**
87  * sk_stream_closing - Return 1 if we still have things to send in our buffers.
88  * @sk: socket to verify
89  */
90 static int sk_stream_closing(const struct sock *sk)
91 {
92         return (1 << READ_ONCE(sk->sk_state)) &
93                (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK);
94 }
95
96 void sk_stream_wait_close(struct sock *sk, long timeout)
97 {
98         if (timeout) {
99                 DEFINE_WAIT_FUNC(wait, woken_wake_function);
100
101                 add_wait_queue(sk_sleep(sk), &wait);
102
103                 do {
104                         if (sk_wait_event(sk, &timeout, !sk_stream_closing(sk), &wait))
105                                 break;
106                 } while (!signal_pending(current) && timeout);
107
108                 remove_wait_queue(sk_sleep(sk), &wait);
109         }
110 }
111 EXPORT_SYMBOL(sk_stream_wait_close);
112
113 /**
114  * sk_stream_wait_memory - Wait for more memory for a socket
115  * @sk: socket to wait for memory
116  * @timeo_p: for how long
117  */
118 int sk_stream_wait_memory(struct sock *sk, long *timeo_p)
119 {
120         int err = 0;
121         long vm_wait = 0;
122         long current_timeo = *timeo_p;
123         DEFINE_WAIT_FUNC(wait, woken_wake_function);
124
125         if (sk_stream_memory_free(sk))
126                 current_timeo = vm_wait = prandom_u32_max(HZ / 5) + 2;
127
128         add_wait_queue(sk_sleep(sk), &wait);
129
130         while (1) {
131                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
132
133                 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
134                         goto do_error;
135                 if (!*timeo_p)
136                         goto do_eagain;
137                 if (signal_pending(current))
138                         goto do_interrupted;
139                 sk_clear_bit(SOCKWQ_ASYNC_NOSPACE, sk);
140                 if (sk_stream_memory_free(sk) && !vm_wait)
141                         break;
142
143                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
144                 sk->sk_write_pending++;
145                 sk_wait_event(sk, &current_timeo, READ_ONCE(sk->sk_err) ||
146                                                   (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) ||
147                                                   (sk_stream_memory_free(sk) &&
148                                                   !vm_wait), &wait);
149                 sk->sk_write_pending--;
150
151                 if (vm_wait) {
152                         vm_wait -= current_timeo;
153                         current_timeo = *timeo_p;
154                         if (current_timeo != MAX_SCHEDULE_TIMEOUT &&
155                             (current_timeo -= vm_wait) < 0)
156                                 current_timeo = 0;
157                         vm_wait = 0;
158                 }
159                 *timeo_p = current_timeo;
160         }
161 out:
162         if (!sock_flag(sk, SOCK_DEAD))
163                 remove_wait_queue(sk_sleep(sk), &wait);
164         return err;
165
166 do_error:
167         err = -EPIPE;
168         goto out;
169 do_eagain:
170         /* Make sure that whenever EAGAIN is returned, EPOLLOUT event can
171          * be generated later.
172          * When TCP receives ACK packets that make room, tcp_check_space()
173          * only calls tcp_new_space() if SOCK_NOSPACE is set.
174          */
175         set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
176         err = -EAGAIN;
177         goto out;
178 do_interrupted:
179         err = sock_intr_errno(*timeo_p);
180         goto out;
181 }
182 EXPORT_SYMBOL(sk_stream_wait_memory);
183
184 int sk_stream_error(struct sock *sk, int flags, int err)
185 {
186         if (err == -EPIPE)
187                 err = sock_error(sk) ? : -EPIPE;
188         if (err == -EPIPE && !(flags & MSG_NOSIGNAL))
189                 send_sig(SIGPIPE, current, 0);
190         return err;
191 }
192 EXPORT_SYMBOL(sk_stream_error);
193
194 void sk_stream_kill_queues(struct sock *sk)
195 {
196         /* First the read buffer. */
197         __skb_queue_purge(&sk->sk_receive_queue);
198
199         /* Next, the error queue.
200          * We need to use queue lock, because other threads might
201          * add packets to the queue without socket lock being held.
202          */
203         skb_queue_purge(&sk->sk_error_queue);
204
205         /* Next, the write queue. */
206         WARN_ON_ONCE(!skb_queue_empty(&sk->sk_write_queue));
207
208         /* Account for returned memory. */
209         sk_mem_reclaim_final(sk);
210
211         WARN_ON_ONCE(sk->sk_wmem_queued);
212
213         /* It is _impossible_ for the backlog to contain anything
214          * when we get here.  All user references to this socket
215          * have gone away, only the net layer knows can touch it.
216          */
217 }
218 EXPORT_SYMBOL(sk_stream_kill_queues);