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: pointer to the socket structure
27 *
28 * This function is invoked when there's space available in the socket's
29 * send buffer for writing. It first checks if the socket is writable,
30 * clears the SOCK_NOSPACE flag indicating that memory for writing
31 * is now available, wakes up any processes waiting for write operations
32 * and sends asynchronous notifications if needed.
33 */
34void sk_stream_write_space(struct sock *sk)
35{
36 struct socket *sock = sk->sk_socket;
37 struct socket_wq *wq;
38
39 if (__sk_stream_is_writeable(sk, wake: 1) && sock) {
40 clear_bit(nr: SOCK_NOSPACE, addr: &sock->flags);
41
42 rcu_read_lock();
43 wq = rcu_dereference(sk->sk_wq);
44 if (skwq_has_sleeper(wq))
45 wake_up_interruptible_poll(&wq->wait, EPOLLOUT |
46 EPOLLWRNORM | EPOLLWRBAND);
47 if (wq && wq->fasync_list && !(sk->sk_shutdown & SEND_SHUTDOWN))
48 sock_wake_async(sk_wq: wq, how: SOCK_WAKE_SPACE, POLL_OUT);
49 rcu_read_unlock();
50 }
51}
52
53/**
54 * sk_stream_wait_connect - Wait for a socket to get into the connected state
55 * @sk: sock to wait on
56 * @timeo_p: for how long to wait
57 *
58 * Must be called with the socket locked.
59 */
60int sk_stream_wait_connect(struct sock *sk, long *timeo_p)
61{
62 DEFINE_WAIT_FUNC(wait, woken_wake_function);
63 struct task_struct *tsk = current;
64 int done;
65
66 do {
67 int err = sock_error(sk);
68 if (err)
69 return err;
70 if ((1 << sk->sk_state) & ~(TCPF_SYN_SENT | TCPF_SYN_RECV))
71 return -EPIPE;
72 if (!*timeo_p)
73 return -EAGAIN;
74 if (signal_pending(p: tsk))
75 return sock_intr_errno(timeo: *timeo_p);
76
77 add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
78 sk->sk_write_pending++;
79 done = sk_wait_event(sk, timeo_p,
80 !READ_ONCE(sk->sk_err) &&
81 !((1 << READ_ONCE(sk->sk_state)) &
82 ~(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)), &wait);
83 remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
84 sk->sk_write_pending--;
85 } while (!done);
86 return done < 0 ? done : 0;
87}
88EXPORT_SYMBOL(sk_stream_wait_connect);
89
90/**
91 * sk_stream_closing - Return 1 if we still have things to send in our buffers.
92 * @sk: socket to verify
93 */
94static int sk_stream_closing(const struct sock *sk)
95{
96 return (1 << READ_ONCE(sk->sk_state)) &
97 (TCPF_FIN_WAIT1 | TCPF_CLOSING | TCPF_LAST_ACK);
98}
99
100void sk_stream_wait_close(struct sock *sk, long timeout)
101{
102 if (timeout) {
103 DEFINE_WAIT_FUNC(wait, woken_wake_function);
104
105 add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
106
107 do {
108 if (sk_wait_event(sk, &timeout, !sk_stream_closing(sk), &wait))
109 break;
110 } while (!signal_pending(current) && timeout);
111
112 remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
113 }
114}
115EXPORT_SYMBOL(sk_stream_wait_close);
116
117/**
118 * sk_stream_wait_memory - Wait for more memory for a socket
119 * @sk: socket to wait for memory
120 * @timeo_p: for how long
121 */
122int sk_stream_wait_memory(struct sock *sk, long *timeo_p)
123{
124 int ret, err = 0;
125 long vm_wait = 0;
126 long current_timeo = *timeo_p;
127 DEFINE_WAIT_FUNC(wait, woken_wake_function);
128
129 if (sk_stream_memory_free(sk))
130 current_timeo = vm_wait = get_random_u32_below(HZ / 5) + 2;
131
132 add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
133
134 while (1) {
135 sk_set_bit(nr: SOCKWQ_ASYNC_NOSPACE, sk);
136
137 if (sk->sk_err || (sk->sk_shutdown & SEND_SHUTDOWN))
138 goto do_error;
139 if (!*timeo_p)
140 goto do_eagain;
141 if (signal_pending(current))
142 goto do_interrupted;
143 sk_clear_bit(nr: SOCKWQ_ASYNC_NOSPACE, sk);
144 if (sk_stream_memory_free(sk) && !vm_wait)
145 break;
146
147 set_bit(nr: SOCK_NOSPACE, addr: &sk->sk_socket->flags);
148 sk->sk_write_pending++;
149 ret = sk_wait_event(sk, &current_timeo, READ_ONCE(sk->sk_err) ||
150 (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) ||
151 (sk_stream_memory_free(sk) && !vm_wait),
152 &wait);
153 sk->sk_write_pending--;
154 if (ret < 0)
155 goto do_error;
156
157 if (vm_wait) {
158 vm_wait -= current_timeo;
159 current_timeo = *timeo_p;
160 if (current_timeo != MAX_SCHEDULE_TIMEOUT &&
161 (current_timeo -= vm_wait) < 0)
162 current_timeo = 0;
163 vm_wait = 0;
164 }
165 *timeo_p = current_timeo;
166 }
167out:
168 if (!sock_flag(sk, flag: SOCK_DEAD))
169 remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait);
170 return err;
171
172do_error:
173 err = -EPIPE;
174 goto out;
175do_eagain:
176 /* Make sure that whenever EAGAIN is returned, EPOLLOUT event can
177 * be generated later.
178 * When TCP receives ACK packets that make room, tcp_check_space()
179 * only calls tcp_new_space() if SOCK_NOSPACE is set.
180 */
181 set_bit(nr: SOCK_NOSPACE, addr: &sk->sk_socket->flags);
182 err = -EAGAIN;
183 goto out;
184do_interrupted:
185 err = sock_intr_errno(timeo: *timeo_p);
186 goto out;
187}
188EXPORT_SYMBOL(sk_stream_wait_memory);
189
190int sk_stream_error(struct sock *sk, int flags, int err)
191{
192 if (err == -EPIPE)
193 err = sock_error(sk) ? : -EPIPE;
194 if (err == -EPIPE && !(flags & MSG_NOSIGNAL))
195 send_sig(SIGPIPE, current, 0);
196 return err;
197}
198EXPORT_SYMBOL(sk_stream_error);
199
200void sk_stream_kill_queues(struct sock *sk)
201{
202 /* First the read buffer. */
203 __skb_queue_purge(list: &sk->sk_receive_queue);
204
205 /* Next, the error queue.
206 * We need to use queue lock, because other threads might
207 * add packets to the queue without socket lock being held.
208 */
209 skb_queue_purge(list: &sk->sk_error_queue);
210
211 /* Next, the write queue. */
212 WARN_ON_ONCE(!skb_queue_empty(&sk->sk_write_queue));
213
214 /* Account for returned memory. */
215 sk_mem_reclaim_final(sk);
216
217 WARN_ON_ONCE(sk->sk_wmem_queued);
218
219 /* It is _impossible_ for the backlog to contain anything
220 * when we get here. All user references to this socket
221 * have gone away, only the net layer knows can touch it.
222 */
223}
224EXPORT_SYMBOL(sk_stream_kill_queues);
225