Loading...
1/*
2 * Copyright (c) 2006 Oracle. 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
38#include "rds.h"
39#include "tcp.h"
40
41/*
42 * cheesy, but simple..
43 */
44static void rds_tcp_accept_worker(struct work_struct *work);
45static DECLARE_WORK(rds_tcp_listen_work, rds_tcp_accept_worker);
46static struct socket *rds_tcp_listen_sock;
47
48static int rds_tcp_accept_one(struct socket *sock)
49{
50 struct socket *new_sock = NULL;
51 struct rds_connection *conn;
52 int ret;
53 struct inet_sock *inet;
54
55 ret = sock_create_lite(sock->sk->sk_family, sock->sk->sk_type,
56 sock->sk->sk_protocol, &new_sock);
57 if (ret)
58 goto out;
59
60 new_sock->type = sock->type;
61 new_sock->ops = sock->ops;
62 ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
63 if (ret < 0)
64 goto out;
65
66 rds_tcp_tune(new_sock);
67
68 inet = inet_sk(new_sock->sk);
69
70 rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
71 &inet->inet_saddr, ntohs(inet->inet_sport),
72 &inet->inet_daddr, ntohs(inet->inet_dport));
73
74 conn = rds_conn_create(inet->inet_saddr, inet->inet_daddr,
75 &rds_tcp_transport, GFP_KERNEL);
76 if (IS_ERR(conn)) {
77 ret = PTR_ERR(conn);
78 goto out;
79 }
80
81 /*
82 * see the comment above rds_queue_delayed_reconnect()
83 */
84 if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
85 if (rds_conn_state(conn) == RDS_CONN_UP)
86 rds_tcp_stats_inc(s_tcp_listen_closed_stale);
87 else
88 rds_tcp_stats_inc(s_tcp_connect_raced);
89 rds_conn_drop(conn);
90 ret = 0;
91 goto out;
92 }
93
94 rds_tcp_set_callbacks(new_sock, conn);
95 rds_connect_complete(conn);
96 new_sock = NULL;
97 ret = 0;
98
99out:
100 if (new_sock)
101 sock_release(new_sock);
102 return ret;
103}
104
105static void rds_tcp_accept_worker(struct work_struct *work)
106{
107 while (rds_tcp_accept_one(rds_tcp_listen_sock) == 0)
108 cond_resched();
109}
110
111void rds_tcp_listen_data_ready(struct sock *sk, int bytes)
112{
113 void (*ready)(struct sock *sk, int bytes);
114
115 rdsdebug("listen data ready sk %p\n", sk);
116
117 read_lock_bh(&sk->sk_callback_lock);
118 ready = sk->sk_user_data;
119 if (!ready) { /* check for teardown race */
120 ready = sk->sk_data_ready;
121 goto out;
122 }
123
124 /*
125 * ->sk_data_ready is also called for a newly established child socket
126 * before it has been accepted and the accepter has set up their
127 * data_ready.. we only want to queue listen work for our listening
128 * socket
129 */
130 if (sk->sk_state == TCP_LISTEN)
131 queue_work(rds_wq, &rds_tcp_listen_work);
132
133out:
134 read_unlock_bh(&sk->sk_callback_lock);
135 ready(sk, bytes);
136}
137
138int rds_tcp_listen_init(void)
139{
140 struct sockaddr_in sin;
141 struct socket *sock = NULL;
142 int ret;
143
144 ret = sock_create(PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
145 if (ret < 0)
146 goto out;
147
148 sock->sk->sk_reuse = 1;
149 rds_tcp_nonagle(sock);
150
151 write_lock_bh(&sock->sk->sk_callback_lock);
152 sock->sk->sk_user_data = sock->sk->sk_data_ready;
153 sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
154 write_unlock_bh(&sock->sk->sk_callback_lock);
155
156 sin.sin_family = PF_INET,
157 sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
158 sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
159
160 ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
161 if (ret < 0)
162 goto out;
163
164 ret = sock->ops->listen(sock, 64);
165 if (ret < 0)
166 goto out;
167
168 rds_tcp_listen_sock = sock;
169 sock = NULL;
170out:
171 if (sock)
172 sock_release(sock);
173 return ret;
174}
175
176void rds_tcp_listen_stop(void)
177{
178 struct socket *sock = rds_tcp_listen_sock;
179 struct sock *sk;
180
181 if (!sock)
182 return;
183
184 sk = sock->sk;
185
186 /* serialize with and prevent further callbacks */
187 lock_sock(sk);
188 write_lock_bh(&sk->sk_callback_lock);
189 if (sk->sk_user_data) {
190 sk->sk_data_ready = sk->sk_user_data;
191 sk->sk_user_data = NULL;
192 }
193 write_unlock_bh(&sk->sk_callback_lock);
194 release_sock(sk);
195
196 /* wait for accepts to stop and close the socket */
197 flush_workqueue(rds_wq);
198 sock_release(sock);
199 rds_tcp_listen_sock = NULL;
200}
1/*
2 * Copyright (c) 2006 Oracle. 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
38#include "rds.h"
39#include "tcp.h"
40
41int rds_tcp_keepalive(struct socket *sock)
42{
43 /* values below based on xs_udp_default_timeout */
44 int keepidle = 5; /* send a probe 'keepidle' secs after last data */
45 int keepcnt = 5; /* number of unack'ed probes before declaring dead */
46 int keepalive = 1;
47 int ret = 0;
48
49 ret = kernel_setsockopt(sock, SOL_SOCKET, SO_KEEPALIVE,
50 (char *)&keepalive, sizeof(keepalive));
51 if (ret < 0)
52 goto bail;
53
54 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPCNT,
55 (char *)&keepcnt, sizeof(keepcnt));
56 if (ret < 0)
57 goto bail;
58
59 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPIDLE,
60 (char *)&keepidle, sizeof(keepidle));
61 if (ret < 0)
62 goto bail;
63
64 /* KEEPINTVL is the interval between successive probes. We follow
65 * the model in xs_tcp_finish_connecting() and re-use keepidle.
66 */
67 ret = kernel_setsockopt(sock, IPPROTO_TCP, TCP_KEEPINTVL,
68 (char *)&keepidle, sizeof(keepidle));
69bail:
70 return ret;
71}
72
73/* rds_tcp_accept_one_path(): if accepting on cp_index > 0, make sure the
74 * client's ipaddr < server's ipaddr. Otherwise, close the accepted
75 * socket and force a reconneect from smaller -> larger ip addr. The reason
76 * we special case cp_index 0 is to allow the rds probe ping itself to itself
77 * get through efficiently.
78 * Since reconnects are only initiated from the node with the numerically
79 * smaller ip address, we recycle conns in RDS_CONN_ERROR on the passive side
80 * by moving them to CONNECTING in this function.
81 */
82struct rds_tcp_connection *rds_tcp_accept_one_path(struct rds_connection *conn)
83{
84 int i;
85 bool peer_is_smaller = (conn->c_faddr < conn->c_laddr);
86 int npaths = max_t(int, 1, conn->c_npaths);
87
88 /* for mprds, all paths MUST be initiated by the peer
89 * with the smaller address.
90 */
91 if (!peer_is_smaller) {
92 /* Make sure we initiate at least one path if this
93 * has not already been done; rds_start_mprds() will
94 * take care of additional paths, if necessary.
95 */
96 if (npaths == 1)
97 rds_conn_path_connect_if_down(&conn->c_path[0]);
98 return NULL;
99 }
100
101 for (i = 0; i < npaths; i++) {
102 struct rds_conn_path *cp = &conn->c_path[i];
103
104 if (rds_conn_path_transition(cp, RDS_CONN_DOWN,
105 RDS_CONN_CONNECTING) ||
106 rds_conn_path_transition(cp, RDS_CONN_ERROR,
107 RDS_CONN_CONNECTING)) {
108 return cp->cp_transport_data;
109 }
110 }
111 return NULL;
112}
113
114int rds_tcp_accept_one(struct socket *sock)
115{
116 struct socket *new_sock = NULL;
117 struct rds_connection *conn;
118 int ret;
119 struct inet_sock *inet;
120 struct rds_tcp_connection *rs_tcp = NULL;
121 int conn_state;
122 struct rds_conn_path *cp;
123
124 if (!sock) /* module unload or netns delete in progress */
125 return -ENETUNREACH;
126
127 ret = sock_create_kern(sock_net(sock->sk), sock->sk->sk_family,
128 sock->sk->sk_type, sock->sk->sk_protocol,
129 &new_sock);
130 if (ret)
131 goto out;
132
133 new_sock->type = sock->type;
134 new_sock->ops = sock->ops;
135 ret = sock->ops->accept(sock, new_sock, O_NONBLOCK);
136 if (ret < 0)
137 goto out;
138
139 ret = rds_tcp_keepalive(new_sock);
140 if (ret < 0)
141 goto out;
142
143 rds_tcp_tune(new_sock);
144
145 inet = inet_sk(new_sock->sk);
146
147 rdsdebug("accepted tcp %pI4:%u -> %pI4:%u\n",
148 &inet->inet_saddr, ntohs(inet->inet_sport),
149 &inet->inet_daddr, ntohs(inet->inet_dport));
150
151 conn = rds_conn_create(sock_net(sock->sk),
152 inet->inet_saddr, inet->inet_daddr,
153 &rds_tcp_transport, GFP_KERNEL);
154 if (IS_ERR(conn)) {
155 ret = PTR_ERR(conn);
156 goto out;
157 }
158 /* An incoming SYN request came in, and TCP just accepted it.
159 *
160 * If the client reboots, this conn will need to be cleaned up.
161 * rds_tcp_state_change() will do that cleanup
162 */
163 rs_tcp = rds_tcp_accept_one_path(conn);
164 if (!rs_tcp)
165 goto rst_nsk;
166 mutex_lock(&rs_tcp->t_conn_path_lock);
167 cp = rs_tcp->t_cpath;
168 conn_state = rds_conn_path_state(cp);
169 WARN_ON(conn_state == RDS_CONN_UP);
170 if (conn_state != RDS_CONN_CONNECTING && conn_state != RDS_CONN_ERROR)
171 goto rst_nsk;
172 if (rs_tcp->t_sock) {
173 /* Need to resolve a duelling SYN between peers.
174 * We have an outstanding SYN to this peer, which may
175 * potentially have transitioned to the RDS_CONN_UP state,
176 * so we must quiesce any send threads before resetting
177 * c_transport_data.
178 */
179 if (ntohl(inet->inet_saddr) < ntohl(inet->inet_daddr) ||
180 !cp->cp_outgoing) {
181 goto rst_nsk;
182 } else {
183 rds_tcp_reset_callbacks(new_sock, cp);
184 cp->cp_outgoing = 0;
185 /* rds_connect_path_complete() marks RDS_CONN_UP */
186 rds_connect_path_complete(cp, RDS_CONN_RESETTING);
187 }
188 } else {
189 rds_tcp_set_callbacks(new_sock, cp);
190 rds_connect_path_complete(cp, RDS_CONN_CONNECTING);
191 }
192 new_sock = NULL;
193 ret = 0;
194 goto out;
195rst_nsk:
196 /* reset the newly returned accept sock and bail */
197 kernel_sock_shutdown(new_sock, SHUT_RDWR);
198 ret = 0;
199out:
200 if (rs_tcp)
201 mutex_unlock(&rs_tcp->t_conn_path_lock);
202 if (new_sock)
203 sock_release(new_sock);
204 return ret;
205}
206
207void rds_tcp_listen_data_ready(struct sock *sk)
208{
209 void (*ready)(struct sock *sk);
210
211 rdsdebug("listen data ready sk %p\n", sk);
212
213 read_lock_bh(&sk->sk_callback_lock);
214 ready = sk->sk_user_data;
215 if (!ready) { /* check for teardown race */
216 ready = sk->sk_data_ready;
217 goto out;
218 }
219
220 /*
221 * ->sk_data_ready is also called for a newly established child socket
222 * before it has been accepted and the accepter has set up their
223 * data_ready.. we only want to queue listen work for our listening
224 * socket
225 */
226 if (sk->sk_state == TCP_LISTEN)
227 rds_tcp_accept_work(sk);
228 else
229 ready = rds_tcp_listen_sock_def_readable(sock_net(sk));
230
231out:
232 read_unlock_bh(&sk->sk_callback_lock);
233 ready(sk);
234}
235
236struct socket *rds_tcp_listen_init(struct net *net)
237{
238 struct sockaddr_in sin;
239 struct socket *sock = NULL;
240 int ret;
241
242 ret = sock_create_kern(net, PF_INET, SOCK_STREAM, IPPROTO_TCP, &sock);
243 if (ret < 0)
244 goto out;
245
246 sock->sk->sk_reuse = SK_CAN_REUSE;
247 rds_tcp_nonagle(sock);
248
249 write_lock_bh(&sock->sk->sk_callback_lock);
250 sock->sk->sk_user_data = sock->sk->sk_data_ready;
251 sock->sk->sk_data_ready = rds_tcp_listen_data_ready;
252 write_unlock_bh(&sock->sk->sk_callback_lock);
253
254 sin.sin_family = PF_INET;
255 sin.sin_addr.s_addr = (__force u32)htonl(INADDR_ANY);
256 sin.sin_port = (__force u16)htons(RDS_TCP_PORT);
257
258 ret = sock->ops->bind(sock, (struct sockaddr *)&sin, sizeof(sin));
259 if (ret < 0)
260 goto out;
261
262 ret = sock->ops->listen(sock, 64);
263 if (ret < 0)
264 goto out;
265
266 return sock;
267out:
268 if (sock)
269 sock_release(sock);
270 return NULL;
271}
272
273void rds_tcp_listen_stop(struct socket *sock)
274{
275 struct sock *sk;
276
277 if (!sock)
278 return;
279
280 sk = sock->sk;
281
282 /* serialize with and prevent further callbacks */
283 lock_sock(sk);
284 write_lock_bh(&sk->sk_callback_lock);
285 if (sk->sk_user_data) {
286 sk->sk_data_ready = sk->sk_user_data;
287 sk->sk_user_data = NULL;
288 }
289 write_unlock_bh(&sk->sk_callback_lock);
290 release_sock(sk);
291
292 /* wait for accepts to stop and close the socket */
293 flush_workqueue(rds_wq);
294 sock_release(sock);
295}