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, wake: 1) && sock) { |
36 | clear_bit(SOCK_NOSPACE, addr: &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(sk_wq: wq, how: 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(p: tsk)) |
71 | return sock_intr_errno(timeo: *timeo_p); |
72 | |
73 | add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &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(wq_head: sk_sleep(sk), wq_entry: &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(wq_head: sk_sleep(sk), wq_entry: &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(wq_head: sk_sleep(sk), wq_entry: &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 ret, 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 = get_random_u32_below(HZ / 5) + 2; |
127 | |
128 | add_wait_queue(wq_head: sk_sleep(sk), wq_entry: &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, addr: &sk->sk_socket->flags); |
144 | sk->sk_write_pending++; |
145 | ret = sk_wait_event(sk, ¤t_timeo, READ_ONCE(sk->sk_err) || |
146 | (READ_ONCE(sk->sk_shutdown) & SEND_SHUTDOWN) || |
147 | (sk_stream_memory_free(sk) && !vm_wait), |
148 | &wait); |
149 | sk->sk_write_pending--; |
150 | if (ret < 0) |
151 | goto do_error; |
152 | |
153 | if (vm_wait) { |
154 | vm_wait -= current_timeo; |
155 | current_timeo = *timeo_p; |
156 | if (current_timeo != MAX_SCHEDULE_TIMEOUT && |
157 | (current_timeo -= vm_wait) < 0) |
158 | current_timeo = 0; |
159 | vm_wait = 0; |
160 | } |
161 | *timeo_p = current_timeo; |
162 | } |
163 | out: |
164 | if (!sock_flag(sk, flag: SOCK_DEAD)) |
165 | remove_wait_queue(wq_head: sk_sleep(sk), wq_entry: &wait); |
166 | return err; |
167 | |
168 | do_error: |
169 | err = -EPIPE; |
170 | goto out; |
171 | do_eagain: |
172 | /* Make sure that whenever EAGAIN is returned, EPOLLOUT event can |
173 | * be generated later. |
174 | * When TCP receives ACK packets that make room, tcp_check_space() |
175 | * only calls tcp_new_space() if SOCK_NOSPACE is set. |
176 | */ |
177 | set_bit(SOCK_NOSPACE, addr: &sk->sk_socket->flags); |
178 | err = -EAGAIN; |
179 | goto out; |
180 | do_interrupted: |
181 | err = sock_intr_errno(timeo: *timeo_p); |
182 | goto out; |
183 | } |
184 | EXPORT_SYMBOL(sk_stream_wait_memory); |
185 | |
186 | int sk_stream_error(struct sock *sk, int flags, int err) |
187 | { |
188 | if (err == -EPIPE) |
189 | err = sock_error(sk) ? : -EPIPE; |
190 | if (err == -EPIPE && !(flags & MSG_NOSIGNAL)) |
191 | send_sig(SIGPIPE, current, 0); |
192 | return err; |
193 | } |
194 | EXPORT_SYMBOL(sk_stream_error); |
195 | |
196 | void sk_stream_kill_queues(struct sock *sk) |
197 | { |
198 | /* First the read buffer. */ |
199 | __skb_queue_purge(list: &sk->sk_receive_queue); |
200 | |
201 | /* Next, the error queue. |
202 | * We need to use queue lock, because other threads might |
203 | * add packets to the queue without socket lock being held. |
204 | */ |
205 | skb_queue_purge(list: &sk->sk_error_queue); |
206 | |
207 | /* Next, the write queue. */ |
208 | WARN_ON_ONCE(!skb_queue_empty(&sk->sk_write_queue)); |
209 | |
210 | /* Account for returned memory. */ |
211 | sk_mem_reclaim_final(sk); |
212 | |
213 | WARN_ON_ONCE(sk->sk_wmem_queued); |
214 | |
215 | /* It is _impossible_ for the backlog to contain anything |
216 | * when we get here. All user references to this socket |
217 | * have gone away, only the net layer knows can touch it. |
218 | */ |
219 | } |
220 | EXPORT_SYMBOL(sk_stream_kill_queues); |
221 | |