xref: /linux/net/rds/rdma_transport.c (revision 088e88be5a380cc4e81963a9a02815da465d144f)
1 /*
2  * Copyright (c) 2009, 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/module.h>
34 #include <rdma/rdma_cm.h>
35 
36 #include "rds_single_path.h"
37 #include "rdma_transport.h"
38 #include "ib.h"
39 
40 /* Global IPv4 and IPv6 RDS RDMA listener cm_id */
41 static struct rdma_cm_id *rds_rdma_listen_id;
42 #if IS_ENABLED(CONFIG_IPV6)
43 static struct rdma_cm_id *rds6_rdma_listen_id;
44 #endif
45 
46 static int rds_rdma_cm_event_handler_cmn(struct rdma_cm_id *cm_id,
47 					 struct rdma_cm_event *event,
48 					 bool isv6)
49 {
50 	/* this can be null in the listening path */
51 	struct rds_connection *conn = cm_id->context;
52 	struct rds_transport *trans;
53 	int ret = 0;
54 	int *err;
55 	u8 len;
56 
57 	rdsdebug("conn %p id %p handling event %u (%s)\n", conn, cm_id,
58 		 event->event, rdma_event_msg(event->event));
59 
60 	if (cm_id->device->node_type == RDMA_NODE_IB_CA)
61 		trans = &rds_ib_transport;
62 
63 	/* Prevent shutdown from tearing down the connection
64 	 * while we're executing. */
65 	if (conn) {
66 		mutex_lock(&conn->c_cm_lock);
67 
68 		/* If the connection is being shut down, bail out
69 		 * right away. We return 0 so cm_id doesn't get
70 		 * destroyed prematurely */
71 		if (rds_conn_state(conn) == RDS_CONN_DISCONNECTING) {
72 			/* Reject incoming connections while we're tearing
73 			 * down an existing one. */
74 			if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST)
75 				ret = 1;
76 			goto out;
77 		}
78 	}
79 
80 	switch (event->event) {
81 	case RDMA_CM_EVENT_CONNECT_REQUEST:
82 		ret = trans->cm_handle_connect(cm_id, event, isv6);
83 		break;
84 
85 	case RDMA_CM_EVENT_ADDR_RESOLVED:
86 		rdma_set_service_type(cm_id, conn->c_tos);
87 		/* XXX do we need to clean up if this fails? */
88 		ret = rdma_resolve_route(cm_id,
89 					 RDS_RDMA_RESOLVE_TIMEOUT_MS);
90 		break;
91 
92 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
93 		/* Connection could have been dropped so make sure the
94 		 * cm_id is valid before proceeding
95 		 */
96 		if (conn) {
97 			struct rds_ib_connection *ibic;
98 
99 			ibic = conn->c_transport_data;
100 			if (ibic && ibic->i_cm_id == cm_id)
101 				ret = trans->cm_initiate_connect(cm_id, isv6);
102 			else
103 				rds_conn_drop(conn);
104 		}
105 		break;
106 
107 	case RDMA_CM_EVENT_ESTABLISHED:
108 		trans->cm_connect_complete(conn, event);
109 		break;
110 
111 	case RDMA_CM_EVENT_REJECTED:
112 		if (!conn)
113 			break;
114 		err = (int *)rdma_consumer_reject_data(cm_id, event, &len);
115 		if (!err ||
116 		    (err && len >= sizeof(*err) &&
117 		     ((*err) <= RDS_RDMA_REJ_INCOMPAT))) {
118 			pr_warn("RDS/RDMA: conn <%pI6c, %pI6c> rejected, dropping connection\n",
119 				&conn->c_laddr, &conn->c_faddr);
120 
121 			if (!conn->c_tos)
122 				conn->c_proposed_version = RDS_PROTOCOL_COMPAT_VERSION;
123 
124 			rds_conn_drop(conn);
125 		}
126 		rdsdebug("Connection rejected: %s\n",
127 			 rdma_reject_msg(cm_id, event->status));
128 		break;
129 	case RDMA_CM_EVENT_ADDR_ERROR:
130 	case RDMA_CM_EVENT_ROUTE_ERROR:
131 	case RDMA_CM_EVENT_CONNECT_ERROR:
132 	case RDMA_CM_EVENT_UNREACHABLE:
133 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
134 	case RDMA_CM_EVENT_ADDR_CHANGE:
135 		if (conn)
136 			rds_conn_drop(conn);
137 		break;
138 
139 	case RDMA_CM_EVENT_DISCONNECTED:
140 		rdsdebug("DISCONNECT event - dropping connection "
141 			 "%pI6c->%pI6c\n", &conn->c_laddr,
142 			 &conn->c_faddr);
143 		rds_conn_drop(conn);
144 		break;
145 
146 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
147 		if (conn) {
148 			pr_info("RDS: RDMA_CM_EVENT_TIMEWAIT_EXIT event: dropping connection %pI6c->%pI6c\n",
149 				&conn->c_laddr, &conn->c_faddr);
150 			rds_conn_drop(conn);
151 		}
152 		break;
153 
154 	default:
155 		/* things like device disconnect? */
156 		printk(KERN_ERR "RDS: unknown event %u (%s)!\n",
157 		       event->event, rdma_event_msg(event->event));
158 		break;
159 	}
160 
161 out:
162 	if (conn)
163 		mutex_unlock(&conn->c_cm_lock);
164 
165 	rdsdebug("id %p event %u (%s) handling ret %d\n", cm_id, event->event,
166 		 rdma_event_msg(event->event), ret);
167 
168 	return ret;
169 }
170 
171 int rds_rdma_cm_event_handler(struct rdma_cm_id *cm_id,
172 			      struct rdma_cm_event *event)
173 {
174 	return rds_rdma_cm_event_handler_cmn(cm_id, event, false);
175 }
176 
177 #if IS_ENABLED(CONFIG_IPV6)
178 int rds6_rdma_cm_event_handler(struct rdma_cm_id *cm_id,
179 			       struct rdma_cm_event *event)
180 {
181 	return rds_rdma_cm_event_handler_cmn(cm_id, event, true);
182 }
183 #endif
184 
185 static int rds_rdma_listen_init_common(rdma_cm_event_handler handler,
186 				       struct sockaddr *sa,
187 				       struct rdma_cm_id **ret_cm_id)
188 {
189 	struct rdma_cm_id *cm_id;
190 	int ret;
191 
192 	cm_id = rdma_create_id(&init_net, handler, NULL,
193 			       RDMA_PS_TCP, IB_QPT_RC);
194 	if (IS_ERR(cm_id)) {
195 		ret = PTR_ERR(cm_id);
196 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
197 		       "rdma_create_id() returned %d\n", ret);
198 		return ret;
199 	}
200 
201 	/*
202 	 * XXX I bet this binds the cm_id to a device.  If we want to support
203 	 * fail-over we'll have to take this into consideration.
204 	 */
205 	ret = rdma_bind_addr(cm_id, sa);
206 	if (ret) {
207 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
208 		       "rdma_bind_addr() returned %d\n", ret);
209 		goto out;
210 	}
211 
212 	ret = rdma_listen(cm_id, 128);
213 	if (ret) {
214 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
215 		       "rdma_listen() returned %d\n", ret);
216 		goto out;
217 	}
218 
219 	rdsdebug("cm %p listening on port %u\n", cm_id, RDS_PORT);
220 
221 	*ret_cm_id = cm_id;
222 	cm_id = NULL;
223 out:
224 	if (cm_id)
225 		rdma_destroy_id(cm_id);
226 	return ret;
227 }
228 
229 /* Initialize the RDS RDMA listeners.  We create two listeners for
230  * compatibility reason.  The one on RDS_PORT is used for IPv4
231  * requests only.  The one on RDS_CM_PORT is used for IPv6 requests
232  * only.  So only IPv6 enabled RDS module will communicate using this
233  * port.
234  */
235 static int rds_rdma_listen_init(void)
236 {
237 	int ret;
238 #if IS_ENABLED(CONFIG_IPV6)
239 	struct sockaddr_in6 sin6;
240 #endif
241 	struct sockaddr_in sin;
242 
243 	sin.sin_family = PF_INET;
244 	sin.sin_addr.s_addr = htonl(INADDR_ANY);
245 	sin.sin_port = htons(RDS_PORT);
246 	ret = rds_rdma_listen_init_common(rds_rdma_cm_event_handler,
247 					  (struct sockaddr *)&sin,
248 					  &rds_rdma_listen_id);
249 	if (ret != 0)
250 		return ret;
251 
252 #if IS_ENABLED(CONFIG_IPV6)
253 	sin6.sin6_family = PF_INET6;
254 	sin6.sin6_addr = in6addr_any;
255 	sin6.sin6_port = htons(RDS_CM_PORT);
256 	sin6.sin6_scope_id = 0;
257 	sin6.sin6_flowinfo = 0;
258 	ret = rds_rdma_listen_init_common(rds6_rdma_cm_event_handler,
259 					  (struct sockaddr *)&sin6,
260 					  &rds6_rdma_listen_id);
261 	/* Keep going even when IPv6 is not enabled in the system. */
262 	if (ret != 0)
263 		rdsdebug("Cannot set up IPv6 RDMA listener\n");
264 #endif
265 	return 0;
266 }
267 
268 static void rds_rdma_listen_stop(void)
269 {
270 	if (rds_rdma_listen_id) {
271 		rdsdebug("cm %p\n", rds_rdma_listen_id);
272 		rdma_destroy_id(rds_rdma_listen_id);
273 		rds_rdma_listen_id = NULL;
274 	}
275 #if IS_ENABLED(CONFIG_IPV6)
276 	if (rds6_rdma_listen_id) {
277 		rdsdebug("cm %p\n", rds6_rdma_listen_id);
278 		rdma_destroy_id(rds6_rdma_listen_id);
279 		rds6_rdma_listen_id = NULL;
280 	}
281 #endif
282 }
283 
284 static int rds_rdma_init(void)
285 {
286 	int ret;
287 
288 	ret = rds_ib_init();
289 	if (ret)
290 		goto out;
291 
292 	ret = rds_rdma_listen_init();
293 	if (ret)
294 		rds_ib_exit();
295 out:
296 	return ret;
297 }
298 module_init(rds_rdma_init);
299 
300 static void rds_rdma_exit(void)
301 {
302 	/* stop listening first to ensure no new connections are attempted */
303 	rds_rdma_listen_stop();
304 	rds_ib_exit();
305 }
306 module_exit(rds_rdma_exit);
307 
308 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
309 MODULE_DESCRIPTION("RDS: IB transport");
310 MODULE_LICENSE("Dual BSD/GPL");
311