xref: /linux/net/rds/rdma_transport.c (revision 23c48a124b469cee2eb0c75e6d22d366d1caa118)
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 /* Per IB specification 7.7.3, service level is a 4-bit field. */
47 #define TOS_TO_SL(tos)		((tos) & 0xF)
48 
49 static int rds_rdma_cm_event_handler_cmn(struct rdma_cm_id *cm_id,
50 					 struct rdma_cm_event *event,
51 					 bool isv6)
52 {
53 	/* this can be null in the listening path */
54 	struct rds_connection *conn = cm_id->context;
55 	struct rds_transport *trans;
56 	int ret = 0;
57 	int *err;
58 	u8 len;
59 
60 	rdsdebug("conn %p id %p handling event %u (%s)\n", conn, cm_id,
61 		 event->event, rdma_event_msg(event->event));
62 
63 	if (cm_id->device->node_type == RDMA_NODE_IB_CA)
64 		trans = &rds_ib_transport;
65 
66 	/* Prevent shutdown from tearing down the connection
67 	 * while we're executing. */
68 	if (conn) {
69 		mutex_lock(&conn->c_cm_lock);
70 
71 		/* If the connection is being shut down, bail out
72 		 * right away. We return 0 so cm_id doesn't get
73 		 * destroyed prematurely */
74 		if (rds_conn_state(conn) == RDS_CONN_DISCONNECTING) {
75 			/* Reject incoming connections while we're tearing
76 			 * down an existing one. */
77 			if (event->event == RDMA_CM_EVENT_CONNECT_REQUEST)
78 				ret = 1;
79 			goto out;
80 		}
81 	}
82 
83 	switch (event->event) {
84 	case RDMA_CM_EVENT_CONNECT_REQUEST:
85 		ret = trans->cm_handle_connect(cm_id, event, isv6);
86 		break;
87 
88 	case RDMA_CM_EVENT_ADDR_RESOLVED:
89 		rdma_set_service_type(cm_id, conn->c_tos);
90 		rdma_set_min_rnr_timer(cm_id, IB_RNR_TIMER_000_32);
91 		/* XXX do we need to clean up if this fails? */
92 		ret = rdma_resolve_route(cm_id,
93 					 RDS_RDMA_RESOLVE_TIMEOUT_MS);
94 		break;
95 
96 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
97 		/* Connection could have been dropped so make sure the
98 		 * cm_id is valid before proceeding
99 		 */
100 		if (conn) {
101 			struct rds_ib_connection *ibic;
102 
103 			ibic = conn->c_transport_data;
104 			if (ibic && ibic->i_cm_id == cm_id) {
105 				cm_id->route.path_rec[0].sl =
106 					TOS_TO_SL(conn->c_tos);
107 				ret = trans->cm_initiate_connect(cm_id, isv6);
108 			} else {
109 				rds_conn_drop(conn);
110 			}
111 		}
112 		break;
113 
114 	case RDMA_CM_EVENT_ESTABLISHED:
115 		if (conn)
116 			trans->cm_connect_complete(conn, event);
117 		break;
118 
119 	case RDMA_CM_EVENT_REJECTED:
120 		if (!conn)
121 			break;
122 		err = (int *)rdma_consumer_reject_data(cm_id, event, &len);
123 		if (!err ||
124 		    (err && len >= sizeof(*err) &&
125 		     ((*err) <= RDS_RDMA_REJ_INCOMPAT))) {
126 			pr_warn("RDS/RDMA: conn <%pI6c, %pI6c> rejected, dropping connection\n",
127 				&conn->c_laddr, &conn->c_faddr);
128 
129 			if (!conn->c_tos)
130 				conn->c_proposed_version = RDS_PROTOCOL_COMPAT_VERSION;
131 
132 			rds_conn_drop(conn);
133 		}
134 		rdsdebug("Connection rejected: %s\n",
135 			 rdma_reject_msg(cm_id, event->status));
136 		break;
137 	case RDMA_CM_EVENT_ADDR_ERROR:
138 	case RDMA_CM_EVENT_ROUTE_ERROR:
139 	case RDMA_CM_EVENT_CONNECT_ERROR:
140 	case RDMA_CM_EVENT_UNREACHABLE:
141 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
142 	case RDMA_CM_EVENT_ADDR_CHANGE:
143 		if (conn)
144 			rds_conn_drop(conn);
145 		break;
146 
147 	case RDMA_CM_EVENT_DISCONNECTED:
148 		if (!conn)
149 			break;
150 		rdsdebug("DISCONNECT event - dropping connection "
151 			 "%pI6c->%pI6c\n", &conn->c_laddr,
152 			 &conn->c_faddr);
153 		rds_conn_drop(conn);
154 		break;
155 
156 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
157 		if (conn) {
158 			pr_info("RDS: RDMA_CM_EVENT_TIMEWAIT_EXIT event: dropping connection %pI6c->%pI6c\n",
159 				&conn->c_laddr, &conn->c_faddr);
160 			rds_conn_drop(conn);
161 		}
162 		break;
163 
164 	default:
165 		/* things like device disconnect? */
166 		printk(KERN_ERR "RDS: unknown event %u (%s)!\n",
167 		       event->event, rdma_event_msg(event->event));
168 		break;
169 	}
170 
171 out:
172 	if (conn)
173 		mutex_unlock(&conn->c_cm_lock);
174 
175 	rdsdebug("id %p event %u (%s) handling ret %d\n", cm_id, event->event,
176 		 rdma_event_msg(event->event), ret);
177 
178 	return ret;
179 }
180 
181 int rds_rdma_cm_event_handler(struct rdma_cm_id *cm_id,
182 			      struct rdma_cm_event *event)
183 {
184 	return rds_rdma_cm_event_handler_cmn(cm_id, event, false);
185 }
186 
187 #if IS_ENABLED(CONFIG_IPV6)
188 int rds6_rdma_cm_event_handler(struct rdma_cm_id *cm_id,
189 			       struct rdma_cm_event *event)
190 {
191 	return rds_rdma_cm_event_handler_cmn(cm_id, event, true);
192 }
193 #endif
194 
195 static int rds_rdma_listen_init_common(rdma_cm_event_handler handler,
196 				       struct sockaddr *sa,
197 				       struct rdma_cm_id **ret_cm_id)
198 {
199 	struct rdma_cm_id *cm_id;
200 	int ret;
201 
202 	cm_id = rdma_create_id(&init_net, handler, NULL,
203 			       RDMA_PS_TCP, IB_QPT_RC);
204 	if (IS_ERR(cm_id)) {
205 		ret = PTR_ERR(cm_id);
206 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
207 		       "rdma_create_id() returned %d\n", ret);
208 		return ret;
209 	}
210 
211 	/*
212 	 * XXX I bet this binds the cm_id to a device.  If we want to support
213 	 * fail-over we'll have to take this into consideration.
214 	 */
215 	ret = rdma_bind_addr(cm_id, sa);
216 	if (ret) {
217 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
218 		       "rdma_bind_addr() returned %d\n", ret);
219 		goto out;
220 	}
221 
222 	ret = rdma_listen(cm_id, 128);
223 	if (ret) {
224 		printk(KERN_ERR "RDS/RDMA: failed to setup listener, "
225 		       "rdma_listen() returned %d\n", ret);
226 		goto out;
227 	}
228 
229 	rdsdebug("cm %p listening on port %u\n", cm_id, RDS_PORT);
230 
231 	*ret_cm_id = cm_id;
232 	cm_id = NULL;
233 out:
234 	if (cm_id)
235 		rdma_destroy_id(cm_id);
236 	return ret;
237 }
238 
239 /* Initialize the RDS RDMA listeners.  We create two listeners for
240  * compatibility reason.  The one on RDS_PORT is used for IPv4
241  * requests only.  The one on RDS_CM_PORT is used for IPv6 requests
242  * only.  So only IPv6 enabled RDS module will communicate using this
243  * port.
244  */
245 static int rds_rdma_listen_init(void)
246 {
247 	int ret;
248 #if IS_ENABLED(CONFIG_IPV6)
249 	struct sockaddr_in6 sin6;
250 #endif
251 	struct sockaddr_in sin;
252 
253 	sin.sin_family = PF_INET;
254 	sin.sin_addr.s_addr = htonl(INADDR_ANY);
255 	sin.sin_port = htons(RDS_PORT);
256 	ret = rds_rdma_listen_init_common(rds_rdma_cm_event_handler,
257 					  (struct sockaddr *)&sin,
258 					  &rds_rdma_listen_id);
259 	if (ret != 0)
260 		return ret;
261 
262 #if IS_ENABLED(CONFIG_IPV6)
263 	sin6.sin6_family = PF_INET6;
264 	sin6.sin6_addr = in6addr_any;
265 	sin6.sin6_port = htons(RDS_CM_PORT);
266 	sin6.sin6_scope_id = 0;
267 	sin6.sin6_flowinfo = 0;
268 	ret = rds_rdma_listen_init_common(rds6_rdma_cm_event_handler,
269 					  (struct sockaddr *)&sin6,
270 					  &rds6_rdma_listen_id);
271 	/* Keep going even when IPv6 is not enabled in the system. */
272 	if (ret != 0)
273 		rdsdebug("Cannot set up IPv6 RDMA listener\n");
274 #endif
275 	return 0;
276 }
277 
278 static void rds_rdma_listen_stop(void)
279 {
280 	if (rds_rdma_listen_id) {
281 		rdsdebug("cm %p\n", rds_rdma_listen_id);
282 		rdma_destroy_id(rds_rdma_listen_id);
283 		rds_rdma_listen_id = NULL;
284 	}
285 #if IS_ENABLED(CONFIG_IPV6)
286 	if (rds6_rdma_listen_id) {
287 		rdsdebug("cm %p\n", rds6_rdma_listen_id);
288 		rdma_destroy_id(rds6_rdma_listen_id);
289 		rds6_rdma_listen_id = NULL;
290 	}
291 #endif
292 }
293 
294 static int rds_rdma_init(void)
295 {
296 	int ret;
297 
298 	ret = rds_ib_init();
299 	if (ret)
300 		goto out;
301 
302 	ret = rds_rdma_listen_init();
303 	if (ret)
304 		rds_ib_exit();
305 out:
306 	return ret;
307 }
308 module_init(rds_rdma_init);
309 
310 static void rds_rdma_exit(void)
311 {
312 	/* stop listening first to ensure no new connections are attempted */
313 	rds_rdma_listen_stop();
314 	rds_ib_exit();
315 }
316 module_exit(rds_rdma_exit);
317 
318 MODULE_AUTHOR("Oracle Corporation <rds-devel@oss.oracle.com>");
319 MODULE_DESCRIPTION("RDS: IB transport");
320 MODULE_LICENSE("Dual BSD/GPL");
321