tcp_listen.c

  1/*
  2 * Copyright (c) 2006, 2018 Oracle and/or its affiliates. All rights reserved.
  3 *
  4 * This software is available to you under a choice of one of two
  5 * licenses.  You may choose to be licensed under the terms of the GNU
  6 * General Public License (GPL) Version 2, available from the file
  7 * COPYING in the main directory of this source tree, or the
  8 * OpenIB.org BSD license below:
  9 *
 10 *     Redistribution and use in source and binary forms, with or
 11 *     without modification, are permitted provided that the following
 12 *     conditions are met:
 13 *
 14 *      - Redistributions of source code must retain the above
 15 *        copyright notice, this list of conditions and the following
 16 *        disclaimer.
 17 *
 18 *      - Redistributions in binary form must reproduce the above
 19 *        copyright notice, this list of conditions and the following
 20 *        disclaimer in the documentation and/or other materials
 21 *        provided with the distribution.
 22 *
 23 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
 24 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
 25 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
 26 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
 27 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
 28 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
 29 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
 30 * SOFTWARE.
 31 *
 32 */
 33#include <linux/kernel.h>
 34#include <linux/gfp.h>
 35#include <linux/in.h>
 36#include <net/tcp.h>
 37#include <trace/events/sock.h>
 38
 39#include "rds.h"
 40#include "tcp.h"
 41
 42void rds_tcp_keepalive(struct socket *sock)
 43{
 44	/* values below based on xs_udp_default_timeout */
 45	int keepidle = 5; /* send a probe 'keepidle' secs after last data */
 46	int keepcnt = 5; /* number of unack'ed probes before declaring dead */
 47
 48	sock_set_keepalive(sock->sk);
 49	tcp_sock_set_keepcnt(sock->sk, keepcnt);
 50	tcp_sock_set_keepidle(sock->sk, keepidle);
 51	/* KEEPINTVL is the interval between successive probes. We follow
 52	 * the model in xs_tcp_finish_connecting() and re-use keepidle.
 53	 */
 54	tcp_sock_set_keepintvl(sock->sk, keepidle);
 55}
 56
 57/* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the
 58 * client's ipaddr < server's ipaddr. Otherwise, close the accepted
 59 * socket and force a reconneect from smaller -> larger ip addr. The reason
 60 * we special case cp_index 0 is to allow the rds probe ping itself to itself
 61 * get through efficiently.
 62 */
 63static
 64struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn)
 65{
 66	int i;
 67	int npaths = max_t(int, 1, conn->c_npaths);
 68
 69	/* for mprds, all paths MUST be initiated by the peer
 70	 * with the smaller address.
 71	 */
 72	if (rds_addr_cmp(&conn->c_faddr, &conn->c_laddr) >= 0) {
 73		/* Make sure we initiate at least one path if this
 74		 * has not already been done; rds_start_mprds() will
 75		 * take care of additional paths, if necessary.
 76		 */
 77		if (npaths == 1)
 78			rds_conn_path_connect_if_down(&conn->c_path[0]);
 79		return NULL;
 80	}
 81
 82	for (i = 0; i < npaths; i++) {
 83		struct rds_conn_path *cp = &conn->c_path[i];
 84
 85		if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
 86					     RDS_CONN_CONNECTING)) {
 87			return cp->cp_transport_data;
 88		}
 89	}
 90	return NULL;
 91}
 92
 93int rds_tcp_accept_one(struct socket *sock)
 94{
 95	struct socket *new_sock = NULL;
 96	struct rds_connection *conn;
 97	int ret;
 98	struct inet_sock *inet;
 99	struct rds_tcp_connection *rs_tcp = NULL;
100	int conn_state;
101	struct rds_conn_path *cp;
102	struct in6_addr *my_addr, *peer_addr;
103#if !IS_ENABLED(CONFIG_IPV6)
104	struct in6_addr saddr, daddr;
105#endif
106	int dev_if = 0;
107
108	if (!sock) /* module unload or netns delete in progress */
109		return -ENETUNREACH;
110
111	ret = kernel_accept(sock, &new_sock, O_NONBLOCK);
112	if (ret)
113		return ret;
114
115	rds_tcp_keepalive(new_sock);
116	if (!rds_tcp_tune(new_sock)) {
117		ret = -EINVAL;
118		goto out;
119	}
120
121	inet = inet_sk(new_sock->sk);
122
123#if IS_ENABLED(CONFIG_IPV6)
124	my_addr = &new_sock->sk->sk_v6_rcv_saddr;
125	peer_addr = &new_sock->sk->sk_v6_daddr;
126#else
127	ipv6_addr_set_v4mapped(inet->inet_saddr, &saddr);
128	ipv6_addr_set_v4mapped(inet->inet_daddr, &daddr);
129	my_addr = &saddr;
130	peer_addr = &daddr;
131#endif
132	rdsdebug("accepted family %d tcp %pI6c:%u -> %pI6c:%u\n",
133		 sock->sk->sk_family,
134		 my_addr, ntohs(inet->inet_sport),
135		 peer_addr, ntohs(inet->inet_dport));
136
137#if IS_ENABLED(CONFIG_IPV6)
138	/* sk_bound_dev_if is not set if the peer address is not link local
139	 * address.  In this case, it happens that mcast_oif is set.  So
140	 * just use it.
141	 */
142	if ((ipv6_addr_type(my_addr) & IPV6_ADDR_LINKLOCAL) &&
143	    !(ipv6_addr_type(peer_addr) & IPV6_ADDR_LINKLOCAL)) {
144		struct ipv6_pinfo *inet6;
145
146		inet6 = inet6_sk(new_sock->sk);
147		dev_if = READ_ONCE(inet6->mcast_oif);
148	} else {
149		dev_if = new_sock->sk->sk_bound_dev_if;
150	}
151#endif
152
153	if (!rds_tcp_laddr_check(sock_net(sock->sk), peer_addr, dev_if)) {
154		/* local address connection is only allowed via loopback */
155		ret = -EOPNOTSUPP;
156		goto out;
157	}
158
159	conn = rds_conn_create(sock_net(sock->sk),
160			       my_addr, peer_addr,
161			       &rds_tcp_transport, 0, GFP_KERNEL, dev_if);
162
163	if (IS_ERR(conn)) {
164		ret = PTR_ERR(conn);
165		goto out;
166	}
167	/* An incoming SYN request came in, and TCP just accepted it.
168	 *
169	 * If the client reboots, this conn will need to be cleaned up.
170	 * rds_tcp_state_change() will do that cleanup
171	 */
172	rs_tcp = rds_tcp_accept_one_path(conn);
173	if (!rs_tcp)
174		goto rst_nsk;
175	mutex_lock(&rs_tcp->t_conn_path_lock);
176	cp = rs_tcp->t_cpath;
177	conn_state = rds_conn_path_state(cp);
178	WARN_ON(conn_state == RDS_CONN_UP);
179	if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_ERROR)
180		goto rst_nsk;
181	if (rs_tcp->t_sock) {
182		/* Duelling SYN has been handled in rds_tcp_accept_one() */
183		rds_tcp_reset_callbacks(new_sock, cp);
184		/* rds_connect_path_complete() marks RDS_CONN_UP */
185		rds_connect_path_complete(cp, RDS_CONN_RESETTING);
186	} else {
187		rds_tcp_set_callbacks(new_sock, cp);
188		rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
189	}
190	new_sock = NULL;
191	ret = 0;
192	if (conn->c_npaths == 0)
193		rds_send_ping(cp->cp_conn, cp->cp_index);
194	goto out;
195rst_nsk:
196	/* reset the newly returned accept sock and bail.
197	 * It is safe to set linger on new_sock because the RDS connection
198	 * has not been brought up on new_sock, so no RDS-level data could
199	 * be pending on it. By setting linger, we achieve the side-effect
200	 * of avoiding TIME_WAIT state on new_sock.
201	 */
202	sock_no_linger(new_sock->sk);
203	kernel_sock_shutdown(new_sock, SHUT_RDWR);
204	ret = 0;
205out:
206	if (rs_tcp)
207		mutex_unlock(&rs_tcp->t_conn_path_lock);
208	if (new_sock)
209		sock_release(new_sock);
210	return ret;
211}
212
213void rds_tcp_listen_data_ready(struct sock *sk)
214{
215	void (*ready)(struct sock *sk);
216
217	trace_sk_data_ready(sk);
218	rdsdebug("listen data ready sk %p\n", sk);
219
220	read_lock_bh(&sk->sk_callback_lock);
221	ready = sk->sk_user_data;
222	if (!ready) { /* check for teardown race */
223		ready = sk->sk_data_ready;
224		goto out;
225	}
226
227	/*
228	 * ->sk_data_ready is also called for a newly established child socket
229	 * before it has been accepted and the accepter has set up their
230	 * data_ready.. we only want to queue listen work for our listening
231	 * socket
232	 *
233	 * (*ready)() may be null if we are racing with netns delete, and
234	 * the listen socket is being torn down.
235	 */
236	if (sk->sk_state == TCP_LISTEN)
237		rds_tcp_accept_work(sk);
238	else
239		ready = rds_tcp_listen_sock_def_readable(sock_net(sk));
240
241out:
242	read_unlock_bh(&sk->sk_callback_lock);
243	if (ready)
244		ready(sk);
245}
246
247struct socket *rds_tcp_listen_init(struct net *net, bool isv6)
248{
249	struct socket *sock = NULL;
250	struct sockaddr_storage ss;
251	struct sockaddr_in6 *sin6;
252	struct sockaddr_in *sin;
253	int addr_len;
254	int ret;
255
256	ret = sock_create_kern(net, isv6 ? PF_INET6 : PF_INET, SOCK_STREAM,
257			       IPPROTO_TCP, &sock);
258	if (ret < 0) {
259		rdsdebug("could not create %s listener socket: %d\n",
260			 isv6 ? "IPv6" : "IPv4", ret);
261		goto out;
262	}
263
264	sock->sk->sk_reuse = SK_CAN_REUSE;
265	tcp_sock_set_nodelay(sock->sk);
266
267	write_lock_bh(&sock->sk->sk_callback_lock);
268	sock->sk->sk_user_data = sock->sk->sk_data_ready;
269	sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
270	write_unlock_bh(&sock->sk->sk_callback_lock);
271
272	if (isv6) {
273		sin6 = (struct sockaddr_in6 *)&ss;
274		sin6->sin6_family = PF_INET6;
275		sin6->sin6_addr = in6addr_any;
276		sin6->sin6_port = htons(RDS_TCP_PORT);
277		sin6->sin6_scope_id = 0;
278		sin6->sin6_flowinfo = 0;
279		addr_len = sizeof(*sin6);
280	} else {
281		sin = (struct sockaddr_in *)&ss;
282		sin->sin_family = PF_INET;
283		sin->sin_addr.s_addr = htonl(INADDR_ANY);
284		sin->sin_port = htons(RDS_TCP_PORT);
285		addr_len = sizeof(*sin);
286	}
287
288	ret = kernel_bind(sock, (struct sockaddr_unsized *)&ss, addr_len);
289	if (ret < 0) {
290		rdsdebug("could not bind %s listener socket: %d\n",
291			 isv6 ? "IPv6" : "IPv4", ret);
292		goto out;
293	}
294
295	ret = sock->ops->listen(sock, 64);
296	if (ret < 0)
297		goto out;
298
299	return sock;
300out:
301	if (sock)
302		sock_release(sock);
303	return NULL;
304}
305
306void rds_tcp_listen_stop(struct socket *sock, struct work_struct *acceptor)
307{
308	struct sock *sk;
309
310	if (!sock)
311		return;
312
313	sk = sock->sk;
314
315	/* serialize with and prevent further callbacks */
316	lock_sock(sk);
317	write_lock_bh(&sk->sk_callback_lock);
318	if (sk->sk_user_data) {
319		sk->sk_data_ready = sk->sk_user_data;
320		sk->sk_user_data = NULL;
321	}
322	write_unlock_bh(&sk->sk_callback_lock);
323	release_sock(sk);
324
325	/* wait for accepts to stop and close the socket */
326	flush_workqueue(rds_wq);
327	flush_work(acceptor);
328	sock_release(sock);
329}