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
rds_tcp_keepalive(struct socket * sock)42 void 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 * Since reconnects are only initiated from the node with the numerically
63 * smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side
64 * by moving them to CONNECTING in this function.
65 */
66 static
rds_tcp_accept_one_path(struct rds_connection * conn)67 struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn)
68 {
69 int i;
70 int npaths = max_t(int, 1, conn->c_npaths);
71
72 /* for mprds, all paths MUST be initiated by the peer
73 * with the smaller address.
74 */
75 if (rds_addr_cmp(&conn->c_faddr, &conn->c_laddr) >= 0) {
76 /* Make sure we initiate at least one path if this
77 * has not already been done; rds_start_mprds() will
78 * take care of additional paths, if necessary.
79 */
80 if (npaths == 1)
81 rds_conn_path_connect_if_down(&conn->c_path[0]);
82 return NULL;
83 }
84
85 for (i = 0; i < npaths; i++) {
86 struct rds_conn_path *cp = &conn->c_path[i];
87
88 if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
89 RDS_CONN_CONNECTING) ||
90 rds_conn_path_transition(cp, RDS_CONN_ERROR,
91 RDS_CONN_CONNECTING)) {
92 return cp->cp_transport_data;
93 }
94 }
95 return NULL;
96 }
97
rds_tcp_accept_one(struct socket * sock)98 int rds_tcp_accept_one(struct socket *sock)
99 {
100 struct socket *new_sock = NULL;
101 struct rds_connection *conn;
102 int ret;
103 struct inet_sock *inet;
104 struct rds_tcp_connection *rs_tcp = NULL;
105 int conn_state;
106 struct rds_conn_path *cp;
107 struct in6_addr *my_addr, *peer_addr;
108 #if !IS_ENABLED(CONFIG_IPV6)
109 struct in6_addr saddr, daddr;
110 #endif
111 int dev_if = 0;
112
113 if (!sock) /* module unload or netns delete in progress */
114 return -ENETUNREACH;
115
116 ret = kernel_accept(sock, &new_sock, O_NONBLOCK);
117 if (ret)
118 return ret;
119
120 rds_tcp_keepalive(new_sock);
121 if (!rds_tcp_tune(new_sock)) {
122 ret = -EINVAL;
123 goto out;
124 }
125
126 inet = inet_sk(new_sock->sk);
127
128 #if IS_ENABLED(CONFIG_IPV6)
129 my_addr = &new_sock->sk->sk_v6_rcv_saddr;
130 peer_addr = &new_sock->sk->sk_v6_daddr;
131 #else
132 ipv6_addr_set_v4mapped(inet->inet_saddr, &saddr);
133 ipv6_addr_set_v4mapped(inet->inet_daddr, &daddr);
134 my_addr = &saddr;
135 peer_addr = &daddr;
136 #endif
137 rdsdebug("accepted family %d tcp %pI6c:%u -> %pI6c:%u\n",
138 sock->sk->sk_family,
139 my_addr, ntohs(inet->inet_sport),
140 peer_addr, ntohs(inet->inet_dport));
141
142 #if IS_ENABLED(CONFIG_IPV6)
143 /* sk_bound_dev_if is not set if the peer address is not link local
144 * address. In this case, it happens that mcast_oif is set. So
145 * just use it.
146 */
147 if ((ipv6_addr_type(my_addr) & IPV6_ADDR_LINKLOCAL) &&
148 !(ipv6_addr_type(peer_addr) & IPV6_ADDR_LINKLOCAL)) {
149 struct ipv6_pinfo *inet6;
150
151 inet6 = inet6_sk(new_sock->sk);
152 dev_if = READ_ONCE(inet6->mcast_oif);
153 } else {
154 dev_if = new_sock->sk->sk_bound_dev_if;
155 }
156 #endif
157
158 if (!rds_tcp_laddr_check(sock_net(sock->sk), peer_addr, dev_if)) {
159 /* local address connection is only allowed via loopback */
160 ret = -EOPNOTSUPP;
161 goto out;
162 }
163
164 conn = rds_conn_create(sock_net(sock->sk),
165 my_addr, peer_addr,
166 &rds_tcp_transport, 0, GFP_KERNEL, dev_if);
167
168 if (IS_ERR(conn)) {
169 ret = PTR_ERR(conn);
170 goto out;
171 }
172 /* An incoming SYN request came in, and TCP just accepted it.
173 *
174 * If the client reboots, this conn will need to be cleaned up.
175 * rds_tcp_state_change() will do that cleanup
176 */
177 rs_tcp = rds_tcp_accept_one_path(conn);
178 if (!rs_tcp)
179 goto rst_nsk;
180 mutex_lock(&rs_tcp->t_conn_path_lock);
181 cp = rs_tcp->t_cpath;
182 conn_state = rds_conn_path_state(cp);
183 WARN_ON(conn_state == RDS_CONN_UP);
184 if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_ERROR)
185 goto rst_nsk;
186 if (rs_tcp->t_sock) {
187 /* Duelling SYN has been handled in rds_tcp_accept_one() */
188 rds_tcp_reset_callbacks(new_sock, cp);
189 /* rds_connect_path_complete() marks RDS_CONN_UP */
190 rds_connect_path_complete(cp, RDS_CONN_RESETTING);
191 } else {
192 rds_tcp_set_callbacks(new_sock, cp);
193 rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
194 }
195 new_sock = NULL;
196 ret = 0;
197 if (conn->c_npaths == 0)
198 rds_send_ping(cp->cp_conn, cp->cp_index);
199 goto out;
200 rst_nsk:
201 /* reset the newly returned accept sock and bail.
202 * It is safe to set linger on new_sock because the RDS connection
203 * has not been brought up on new_sock, so no RDS-level data could
204 * be pending on it. By setting linger, we achieve the side-effect
205 * of avoiding TIME_WAIT state on new_sock.
206 */
207 sock_no_linger(new_sock->sk);
208 kernel_sock_shutdown(new_sock, SHUT_RDWR);
209 ret = 0;
210 out:
211 if (rs_tcp)
212 mutex_unlock(&rs_tcp->t_conn_path_lock);
213 if (new_sock)
214 sock_release(new_sock);
215 return ret;
216 }
217
rds_tcp_listen_data_ready(struct sock * sk)218 void rds_tcp_listen_data_ready(struct sock *sk)
219 {
220 void (*ready)(struct sock *sk);
221
222 trace_sk_data_ready(sk);
223 rdsdebug("listen data ready sk %p\n", sk);
224
225 read_lock_bh(&sk->sk_callback_lock);
226 ready = sk->sk_user_data;
227 if (!ready) { /* check for teardown race */
228 ready = sk->sk_data_ready;
229 goto out;
230 }
231
232 /*
233 * ->sk_data_ready is also called for a newly established child socket
234 * before it has been accepted and the accepter has set up their
235 * data_ready.. we only want to queue listen work for our listening
236 * socket
237 *
238 * (*ready)() may be null if we are racing with netns delete, and
239 * the listen socket is being torn down.
240 */
241 if (sk->sk_state == TCP_LISTEN)
242 rds_tcp_accept_work(sk);
243 else
244 ready = rds_tcp_listen_sock_def_readable(sock_net(sk));
245
246 out:
247 read_unlock_bh(&sk->sk_callback_lock);
248 if (ready)
249 ready(sk);
250 }
251
rds_tcp_listen_init(struct net * net,bool isv6)252 struct socket *rds_tcp_listen_init(struct net *net, bool isv6)
253 {
254 struct socket *sock = NULL;
255 struct sockaddr_storage ss;
256 struct sockaddr_in6 *sin6;
257 struct sockaddr_in *sin;
258 int addr_len;
259 int ret;
260
261 ret = sock_create_kern(net, isv6 ? PF_INET6 : PF_INET, SOCK_STREAM,
262 IPPROTO_TCP, &sock);
263 if (ret < 0) {
264 rdsdebug("could not create %s listener socket: %d\n",
265 isv6 ? "IPv6" : "IPv4", ret);
266 goto out;
267 }
268
269 sock->sk->sk_reuse = SK_CAN_REUSE;
270 tcp_sock_set_nodelay(sock->sk);
271
272 write_lock_bh(&sock->sk->sk_callback_lock);
273 sock->sk->sk_user_data = sock->sk->sk_data_ready;
274 sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
275 write_unlock_bh(&sock->sk->sk_callback_lock);
276
277 if (isv6) {
278 sin6 = (struct sockaddr_in6 *)&ss;
279 sin6->sin6_family = PF_INET6;
280 sin6->sin6_addr = in6addr_any;
281 sin6->sin6_port = htons(RDS_TCP_PORT);
282 sin6->sin6_scope_id = 0;
283 sin6->sin6_flowinfo = 0;
284 addr_len = sizeof(*sin6);
285 } else {
286 sin = (struct sockaddr_in *)&ss;
287 sin->sin_family = PF_INET;
288 sin->sin_addr.s_addr = htonl(INADDR_ANY);
289 sin->sin_port = htons(RDS_TCP_PORT);
290 addr_len = sizeof(*sin);
291 }
292
293 ret = kernel_bind(sock, (struct sockaddr *)&ss, addr_len);
294 if (ret < 0) {
295 rdsdebug("could not bind %s listener socket: %d\n",
296 isv6 ? "IPv6" : "IPv4", ret);
297 goto out;
298 }
299
300 ret = sock->ops->listen(sock, 64);
301 if (ret < 0)
302 goto out;
303
304 return sock;
305 out:
306 if (sock)
307 sock_release(sock);
308 return NULL;
309 }
310
rds_tcp_listen_stop(struct socket * sock,struct work_struct * acceptor)311 void rds_tcp_listen_stop(struct socket *sock, struct work_struct *acceptor)
312 {
313 struct sock *sk;
314
315 if (!sock)
316 return;
317
318 sk = sock->sk;
319
320 /* serialize with and prevent further callbacks */
321 lock_sock(sk);
322 write_lock_bh(&sk->sk_callback_lock);
323 if (sk->sk_user_data) {
324 sk->sk_data_ready = sk->sk_user_data;
325 sk->sk_user_data = NULL;
326 }
327 write_unlock_bh(&sk->sk_callback_lock);
328 release_sock(sk);
329
330 /* wait for accepts to stop and close the socket */
331 flush_workqueue(rds_wq);
332 flush_work(acceptor);
333 sock_release(sock);
334 }
335