xref: /openbmc/linux/net/rds/ib_cm.c (revision d5532ee7)
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/in.h>
35 #include <linux/slab.h>
36 #include <linux/vmalloc.h>
37 
38 #include "rds.h"
39 #include "ib.h"
40 
41 /*
42  * Set the selected protocol version
43  */
44 static void rds_ib_set_protocol(struct rds_connection *conn, unsigned int version)
45 {
46 	conn->c_version = version;
47 }
48 
49 /*
50  * Set up flow control
51  */
52 static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits)
53 {
54 	struct rds_ib_connection *ic = conn->c_transport_data;
55 
56 	if (rds_ib_sysctl_flow_control && credits != 0) {
57 		/* We're doing flow control */
58 		ic->i_flowctl = 1;
59 		rds_ib_send_add_credits(conn, credits);
60 	} else {
61 		ic->i_flowctl = 0;
62 	}
63 }
64 
65 /*
66  * Tune RNR behavior. Without flow control, we use a rather
67  * low timeout, but not the absolute minimum - this should
68  * be tunable.
69  *
70  * We already set the RNR retry count to 7 (which is the
71  * smallest infinite number :-) above.
72  * If flow control is off, we want to change this back to 0
73  * so that we learn quickly when our credit accounting is
74  * buggy.
75  *
76  * Caller passes in a qp_attr pointer - don't waste stack spacv
77  * by allocation this twice.
78  */
79 static void
80 rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr)
81 {
82 	int ret;
83 
84 	attr->min_rnr_timer = IB_RNR_TIMER_000_32;
85 	ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER);
86 	if (ret)
87 		printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret);
88 }
89 
90 /*
91  * Connection established.
92  * We get here for both outgoing and incoming connection.
93  */
94 void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
95 {
96 	const struct rds_ib_connect_private *dp = NULL;
97 	struct rds_ib_connection *ic = conn->c_transport_data;
98 	struct rds_ib_device *rds_ibdev;
99 	struct ib_qp_attr qp_attr;
100 	int err;
101 
102 	if (event->param.conn.private_data_len >= sizeof(*dp)) {
103 		dp = event->param.conn.private_data;
104 
105 		/* make sure it isn't empty data */
106 		if (dp->dp_protocol_major) {
107 			rds_ib_set_protocol(conn,
108 				RDS_PROTOCOL(dp->dp_protocol_major,
109 				dp->dp_protocol_minor));
110 			rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
111 		}
112 	}
113 
114 	printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n",
115 			&conn->c_faddr,
116 			RDS_PROTOCOL_MAJOR(conn->c_version),
117 			RDS_PROTOCOL_MINOR(conn->c_version),
118 			ic->i_flowctl ? ", flow control" : "");
119 
120 	/*
121 	 * Init rings and fill recv. this needs to wait until protocol negotiation
122 	 * is complete, since ring layout is different from 3.0 to 3.1.
123 	 */
124 	rds_ib_send_init_ring(ic);
125 	rds_ib_recv_init_ring(ic);
126 	/* Post receive buffers - as a side effect, this will update
127 	 * the posted credit count. */
128 	rds_ib_recv_refill(conn, GFP_KERNEL, GFP_HIGHUSER, 1);
129 
130 	/* Tune RNR behavior */
131 	rds_ib_tune_rnr(ic, &qp_attr);
132 
133 	qp_attr.qp_state = IB_QPS_RTS;
134 	err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
135 	if (err)
136 		printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
137 
138 	/* update ib_device with this local ipaddr & conn */
139 	rds_ibdev = ib_get_client_data(ic->i_cm_id->device, &rds_ib_client);
140 	err = rds_ib_update_ipaddr(rds_ibdev, conn->c_laddr);
141 	if (err)
142 		printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n", err);
143 	rds_ib_add_conn(rds_ibdev, conn);
144 
145 	/* If the peer gave us the last packet it saw, process this as if
146 	 * we had received a regular ACK. */
147 	if (dp && dp->dp_ack_seq)
148 		rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
149 
150 	rds_connect_complete(conn);
151 }
152 
153 static void rds_ib_cm_fill_conn_param(struct rds_connection *conn,
154 			struct rdma_conn_param *conn_param,
155 			struct rds_ib_connect_private *dp,
156 			u32 protocol_version)
157 {
158 	memset(conn_param, 0, sizeof(struct rdma_conn_param));
159 	/* XXX tune these? */
160 	conn_param->responder_resources = 1;
161 	conn_param->initiator_depth = 1;
162 	conn_param->retry_count = min_t(unsigned int, rds_ib_retry_count, 7);
163 	conn_param->rnr_retry_count = 7;
164 
165 	if (dp) {
166 		struct rds_ib_connection *ic = conn->c_transport_data;
167 
168 		memset(dp, 0, sizeof(*dp));
169 		dp->dp_saddr = conn->c_laddr;
170 		dp->dp_daddr = conn->c_faddr;
171 		dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version);
172 		dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version);
173 		dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IB_SUPPORTED_PROTOCOLS);
174 		dp->dp_ack_seq = rds_ib_piggyb_ack(ic);
175 
176 		/* Advertise flow control */
177 		if (ic->i_flowctl) {
178 			unsigned int credits;
179 
180 			credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits));
181 			dp->dp_credit = cpu_to_be32(credits);
182 			atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits);
183 		}
184 
185 		conn_param->private_data = dp;
186 		conn_param->private_data_len = sizeof(*dp);
187 	}
188 }
189 
190 static void rds_ib_cq_event_handler(struct ib_event *event, void *data)
191 {
192 	rdsdebug("event %u data %p\n", event->event, data);
193 }
194 
195 static void rds_ib_qp_event_handler(struct ib_event *event, void *data)
196 {
197 	struct rds_connection *conn = data;
198 	struct rds_ib_connection *ic = conn->c_transport_data;
199 
200 	rdsdebug("conn %p ic %p event %u\n", conn, ic, event->event);
201 
202 	switch (event->event) {
203 	case IB_EVENT_COMM_EST:
204 		rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST);
205 		break;
206 	default:
207 		rdsdebug("Fatal QP Event %u "
208 			"- connection %pI4->%pI4, reconnecting\n",
209 			event->event, &conn->c_laddr, &conn->c_faddr);
210 		rds_conn_drop(conn);
211 		break;
212 	}
213 }
214 
215 /*
216  * This needs to be very careful to not leave IS_ERR pointers around for
217  * cleanup to trip over.
218  */
219 static int rds_ib_setup_qp(struct rds_connection *conn)
220 {
221 	struct rds_ib_connection *ic = conn->c_transport_data;
222 	struct ib_device *dev = ic->i_cm_id->device;
223 	struct ib_qp_init_attr attr;
224 	struct rds_ib_device *rds_ibdev;
225 	int ret;
226 
227 	/* rds_ib_add_one creates a rds_ib_device object per IB device,
228 	 * and allocates a protection domain, memory range and FMR pool
229 	 * for each.  If that fails for any reason, it will not register
230 	 * the rds_ibdev at all.
231 	 */
232 	rds_ibdev = ib_get_client_data(dev, &rds_ib_client);
233 	if (rds_ibdev == NULL) {
234 		if (printk_ratelimit())
235 			printk(KERN_NOTICE "RDS/IB: No client_data for device %s\n",
236 					dev->name);
237 		return -EOPNOTSUPP;
238 	}
239 
240 	if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
241 		rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
242 	if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
243 		rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
244 
245 	/* Protection domain and memory range */
246 	ic->i_pd = rds_ibdev->pd;
247 	ic->i_mr = rds_ibdev->mr;
248 
249 	ic->i_send_cq = ib_create_cq(dev, rds_ib_send_cq_comp_handler,
250 				     rds_ib_cq_event_handler, conn,
251 				     ic->i_send_ring.w_nr + 1, 0);
252 	if (IS_ERR(ic->i_send_cq)) {
253 		ret = PTR_ERR(ic->i_send_cq);
254 		ic->i_send_cq = NULL;
255 		rdsdebug("ib_create_cq send failed: %d\n", ret);
256 		goto out;
257 	}
258 
259 	ic->i_recv_cq = ib_create_cq(dev, rds_ib_recv_cq_comp_handler,
260 				     rds_ib_cq_event_handler, conn,
261 				     ic->i_recv_ring.w_nr, 0);
262 	if (IS_ERR(ic->i_recv_cq)) {
263 		ret = PTR_ERR(ic->i_recv_cq);
264 		ic->i_recv_cq = NULL;
265 		rdsdebug("ib_create_cq recv failed: %d\n", ret);
266 		goto out;
267 	}
268 
269 	ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
270 	if (ret) {
271 		rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
272 		goto out;
273 	}
274 
275 	ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
276 	if (ret) {
277 		rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
278 		goto out;
279 	}
280 
281 	/* XXX negotiate max send/recv with remote? */
282 	memset(&attr, 0, sizeof(attr));
283 	attr.event_handler = rds_ib_qp_event_handler;
284 	attr.qp_context = conn;
285 	/* + 1 to allow for the single ack message */
286 	attr.cap.max_send_wr = ic->i_send_ring.w_nr + 1;
287 	attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
288 	attr.cap.max_send_sge = rds_ibdev->max_sge;
289 	attr.cap.max_recv_sge = RDS_IB_RECV_SGE;
290 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
291 	attr.qp_type = IB_QPT_RC;
292 	attr.send_cq = ic->i_send_cq;
293 	attr.recv_cq = ic->i_recv_cq;
294 
295 	/*
296 	 * XXX this can fail if max_*_wr is too large?  Are we supposed
297 	 * to back off until we get a value that the hardware can support?
298 	 */
299 	ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
300 	if (ret) {
301 		rdsdebug("rdma_create_qp failed: %d\n", ret);
302 		goto out;
303 	}
304 
305 	ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
306 					   ic->i_send_ring.w_nr *
307 						sizeof(struct rds_header),
308 					   &ic->i_send_hdrs_dma, GFP_KERNEL);
309 	if (ic->i_send_hdrs == NULL) {
310 		ret = -ENOMEM;
311 		rdsdebug("ib_dma_alloc_coherent send failed\n");
312 		goto out;
313 	}
314 
315 	ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
316 					   ic->i_recv_ring.w_nr *
317 						sizeof(struct rds_header),
318 					   &ic->i_recv_hdrs_dma, GFP_KERNEL);
319 	if (ic->i_recv_hdrs == NULL) {
320 		ret = -ENOMEM;
321 		rdsdebug("ib_dma_alloc_coherent recv failed\n");
322 		goto out;
323 	}
324 
325 	ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
326 				       &ic->i_ack_dma, GFP_KERNEL);
327 	if (ic->i_ack == NULL) {
328 		ret = -ENOMEM;
329 		rdsdebug("ib_dma_alloc_coherent ack failed\n");
330 		goto out;
331 	}
332 
333 	ic->i_sends = vmalloc(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work));
334 	if (ic->i_sends == NULL) {
335 		ret = -ENOMEM;
336 		rdsdebug("send allocation failed\n");
337 		goto out;
338 	}
339 	memset(ic->i_sends, 0, ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work));
340 
341 	ic->i_recvs = vmalloc(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work));
342 	if (ic->i_recvs == NULL) {
343 		ret = -ENOMEM;
344 		rdsdebug("recv allocation failed\n");
345 		goto out;
346 	}
347 	memset(ic->i_recvs, 0, ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work));
348 
349 	rds_ib_recv_init_ack(ic);
350 
351 	rdsdebug("conn %p pd %p mr %p cq %p %p\n", conn, ic->i_pd, ic->i_mr,
352 		 ic->i_send_cq, ic->i_recv_cq);
353 
354 out:
355 	return ret;
356 }
357 
358 static u32 rds_ib_protocol_compatible(struct rdma_cm_event *event)
359 {
360 	const struct rds_ib_connect_private *dp = event->param.conn.private_data;
361 	u16 common;
362 	u32 version = 0;
363 
364 	/*
365 	 * rdma_cm private data is odd - when there is any private data in the
366 	 * request, we will be given a pretty large buffer without telling us the
367 	 * original size. The only way to tell the difference is by looking at
368 	 * the contents, which are initialized to zero.
369 	 * If the protocol version fields aren't set, this is a connection attempt
370 	 * from an older version. This could could be 3.0 or 2.0 - we can't tell.
371 	 * We really should have changed this for OFED 1.3 :-(
372 	 */
373 
374 	/* Be paranoid. RDS always has privdata */
375 	if (!event->param.conn.private_data_len) {
376 		printk(KERN_NOTICE "RDS incoming connection has no private data, "
377 			"rejecting\n");
378 		return 0;
379 	}
380 
381 	/* Even if len is crap *now* I still want to check it. -ASG */
382 	if (event->param.conn.private_data_len < sizeof (*dp) ||
383 	    dp->dp_protocol_major == 0)
384 		return RDS_PROTOCOL_3_0;
385 
386 	common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IB_SUPPORTED_PROTOCOLS;
387 	if (dp->dp_protocol_major == 3 && common) {
388 		version = RDS_PROTOCOL_3_0;
389 		while ((common >>= 1) != 0)
390 			version++;
391 	} else if (printk_ratelimit()) {
392 		printk(KERN_NOTICE "RDS: Connection from %pI4 using "
393 			"incompatible protocol version %u.%u\n",
394 			&dp->dp_saddr,
395 			dp->dp_protocol_major,
396 			dp->dp_protocol_minor);
397 	}
398 	return version;
399 }
400 
401 int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
402 				    struct rdma_cm_event *event)
403 {
404 	__be64 lguid = cm_id->route.path_rec->sgid.global.interface_id;
405 	__be64 fguid = cm_id->route.path_rec->dgid.global.interface_id;
406 	const struct rds_ib_connect_private *dp = event->param.conn.private_data;
407 	struct rds_ib_connect_private dp_rep;
408 	struct rds_connection *conn = NULL;
409 	struct rds_ib_connection *ic = NULL;
410 	struct rdma_conn_param conn_param;
411 	u32 version;
412 	int err, destroy = 1;
413 
414 	/* Check whether the remote protocol version matches ours. */
415 	version = rds_ib_protocol_compatible(event);
416 	if (!version)
417 		goto out;
418 
419 	rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u lguid 0x%llx fguid "
420 		 "0x%llx\n", &dp->dp_saddr, &dp->dp_daddr,
421 		 RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version),
422 		 (unsigned long long)be64_to_cpu(lguid),
423 		 (unsigned long long)be64_to_cpu(fguid));
424 
425 	conn = rds_conn_create(dp->dp_daddr, dp->dp_saddr, &rds_ib_transport,
426 			       GFP_KERNEL);
427 	if (IS_ERR(conn)) {
428 		rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn));
429 		conn = NULL;
430 		goto out;
431 	}
432 
433 	/*
434 	 * The connection request may occur while the
435 	 * previous connection exist, e.g. in case of failover.
436 	 * But as connections may be initiated simultaneously
437 	 * by both hosts, we have a random backoff mechanism -
438 	 * see the comment above rds_queue_reconnect()
439 	 */
440 	mutex_lock(&conn->c_cm_lock);
441 	if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
442 		if (rds_conn_state(conn) == RDS_CONN_UP) {
443 			rdsdebug("incoming connect while connecting\n");
444 			rds_conn_drop(conn);
445 			rds_ib_stats_inc(s_ib_listen_closed_stale);
446 		} else
447 		if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
448 			/* Wait and see - our connect may still be succeeding */
449 			rds_ib_stats_inc(s_ib_connect_raced);
450 		}
451 		mutex_unlock(&conn->c_cm_lock);
452 		goto out;
453 	}
454 
455 	ic = conn->c_transport_data;
456 
457 	rds_ib_set_protocol(conn, version);
458 	rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
459 
460 	/* If the peer gave us the last packet it saw, process this as if
461 	 * we had received a regular ACK. */
462 	if (dp->dp_ack_seq)
463 		rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
464 
465 	BUG_ON(cm_id->context);
466 	BUG_ON(ic->i_cm_id);
467 
468 	ic->i_cm_id = cm_id;
469 	cm_id->context = conn;
470 
471 	/* We got halfway through setting up the ib_connection, if we
472 	 * fail now, we have to take the long route out of this mess. */
473 	destroy = 0;
474 
475 	err = rds_ib_setup_qp(conn);
476 	if (err) {
477 		rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", err);
478 		mutex_unlock(&conn->c_cm_lock);
479 		goto out;
480 	}
481 
482 	rds_ib_cm_fill_conn_param(conn, &conn_param, &dp_rep, version);
483 
484 	/* rdma_accept() calls rdma_reject() internally if it fails */
485 	err = rdma_accept(cm_id, &conn_param);
486 	mutex_unlock(&conn->c_cm_lock);
487 	if (err) {
488 		rds_ib_conn_error(conn, "rdma_accept failed (%d)\n", err);
489 		goto out;
490 	}
491 
492 	return 0;
493 
494 out:
495 	rdma_reject(cm_id, NULL, 0);
496 	return destroy;
497 }
498 
499 
500 int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id)
501 {
502 	struct rds_connection *conn = cm_id->context;
503 	struct rds_ib_connection *ic = conn->c_transport_data;
504 	struct rdma_conn_param conn_param;
505 	struct rds_ib_connect_private dp;
506 	int ret;
507 
508 	/* If the peer doesn't do protocol negotiation, we must
509 	 * default to RDSv3.0 */
510 	rds_ib_set_protocol(conn, RDS_PROTOCOL_3_0);
511 	ic->i_flowctl = rds_ib_sysctl_flow_control;	/* advertise flow control */
512 
513 	ret = rds_ib_setup_qp(conn);
514 	if (ret) {
515 		rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", ret);
516 		goto out;
517 	}
518 
519 	rds_ib_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION);
520 
521 	ret = rdma_connect(cm_id, &conn_param);
522 	if (ret)
523 		rds_ib_conn_error(conn, "rdma_connect failed (%d)\n", ret);
524 
525 out:
526 	/* Beware - returning non-zero tells the rdma_cm to destroy
527 	 * the cm_id. We should certainly not do it as long as we still
528 	 * "own" the cm_id. */
529 	if (ret) {
530 		if (ic->i_cm_id == cm_id)
531 			ret = 0;
532 	}
533 	return ret;
534 }
535 
536 int rds_ib_conn_connect(struct rds_connection *conn)
537 {
538 	struct rds_ib_connection *ic = conn->c_transport_data;
539 	struct sockaddr_in src, dest;
540 	int ret;
541 
542 	/* XXX I wonder what affect the port space has */
543 	/* delegate cm event handler to rdma_transport */
544 	ic->i_cm_id = rdma_create_id(rds_rdma_cm_event_handler, conn,
545 				     RDMA_PS_TCP);
546 	if (IS_ERR(ic->i_cm_id)) {
547 		ret = PTR_ERR(ic->i_cm_id);
548 		ic->i_cm_id = NULL;
549 		rdsdebug("rdma_create_id() failed: %d\n", ret);
550 		goto out;
551 	}
552 
553 	rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn);
554 
555 	src.sin_family = AF_INET;
556 	src.sin_addr.s_addr = (__force u32)conn->c_laddr;
557 	src.sin_port = (__force u16)htons(0);
558 
559 	dest.sin_family = AF_INET;
560 	dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
561 	dest.sin_port = (__force u16)htons(RDS_PORT);
562 
563 	ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src,
564 				(struct sockaddr *)&dest,
565 				RDS_RDMA_RESOLVE_TIMEOUT_MS);
566 	if (ret) {
567 		rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id,
568 			 ret);
569 		rdma_destroy_id(ic->i_cm_id);
570 		ic->i_cm_id = NULL;
571 	}
572 
573 out:
574 	return ret;
575 }
576 
577 /*
578  * This is so careful about only cleaning up resources that were built up
579  * so that it can be called at any point during startup.  In fact it
580  * can be called multiple times for a given connection.
581  */
582 void rds_ib_conn_shutdown(struct rds_connection *conn)
583 {
584 	struct rds_ib_connection *ic = conn->c_transport_data;
585 	int err = 0;
586 
587 	rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id,
588 		 ic->i_pd, ic->i_send_cq, ic->i_recv_cq,
589 		 ic->i_cm_id ? ic->i_cm_id->qp : NULL);
590 
591 	if (ic->i_cm_id) {
592 		struct ib_device *dev = ic->i_cm_id->device;
593 
594 		rdsdebug("disconnecting cm %p\n", ic->i_cm_id);
595 		err = rdma_disconnect(ic->i_cm_id);
596 		if (err) {
597 			/* Actually this may happen quite frequently, when
598 			 * an outgoing connect raced with an incoming connect.
599 			 */
600 			rdsdebug("failed to disconnect, cm: %p err %d\n",
601 				ic->i_cm_id, err);
602 		}
603 
604 		wait_event(rds_ib_ring_empty_wait,
605 			rds_ib_ring_empty(&ic->i_send_ring) &&
606 			rds_ib_ring_empty(&ic->i_recv_ring));
607 
608 		if (ic->i_send_hdrs)
609 			ib_dma_free_coherent(dev,
610 					   ic->i_send_ring.w_nr *
611 						sizeof(struct rds_header),
612 					   ic->i_send_hdrs,
613 					   ic->i_send_hdrs_dma);
614 
615 		if (ic->i_recv_hdrs)
616 			ib_dma_free_coherent(dev,
617 					   ic->i_recv_ring.w_nr *
618 						sizeof(struct rds_header),
619 					   ic->i_recv_hdrs,
620 					   ic->i_recv_hdrs_dma);
621 
622 		if (ic->i_ack)
623 			ib_dma_free_coherent(dev, sizeof(struct rds_header),
624 					     ic->i_ack, ic->i_ack_dma);
625 
626 		if (ic->i_sends)
627 			rds_ib_send_clear_ring(ic);
628 		if (ic->i_recvs)
629 			rds_ib_recv_clear_ring(ic);
630 
631 		if (ic->i_cm_id->qp)
632 			rdma_destroy_qp(ic->i_cm_id);
633 		if (ic->i_send_cq)
634 			ib_destroy_cq(ic->i_send_cq);
635 		if (ic->i_recv_cq)
636 			ib_destroy_cq(ic->i_recv_cq);
637 		rdma_destroy_id(ic->i_cm_id);
638 
639 		/*
640 		 * Move connection back to the nodev list.
641 		 */
642 		if (ic->rds_ibdev)
643 			rds_ib_remove_conn(ic->rds_ibdev, conn);
644 
645 		ic->i_cm_id = NULL;
646 		ic->i_pd = NULL;
647 		ic->i_mr = NULL;
648 		ic->i_send_cq = NULL;
649 		ic->i_recv_cq = NULL;
650 		ic->i_send_hdrs = NULL;
651 		ic->i_recv_hdrs = NULL;
652 		ic->i_ack = NULL;
653 	}
654 	BUG_ON(ic->rds_ibdev);
655 
656 	/* Clear pending transmit */
657 	if (ic->i_rm) {
658 		rds_message_put(ic->i_rm);
659 		ic->i_rm = NULL;
660 	}
661 
662 	/* Clear the ACK state */
663 	clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
664 #ifdef KERNEL_HAS_ATOMIC64
665 	atomic64_set(&ic->i_ack_next, 0);
666 #else
667 	ic->i_ack_next = 0;
668 #endif
669 	ic->i_ack_recv = 0;
670 
671 	/* Clear flow control state */
672 	ic->i_flowctl = 0;
673 	atomic_set(&ic->i_credits, 0);
674 
675 	rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
676 	rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
677 
678 	if (ic->i_ibinc) {
679 		rds_inc_put(&ic->i_ibinc->ii_inc);
680 		ic->i_ibinc = NULL;
681 	}
682 
683 	vfree(ic->i_sends);
684 	ic->i_sends = NULL;
685 	vfree(ic->i_recvs);
686 	ic->i_recvs = NULL;
687 }
688 
689 int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp)
690 {
691 	struct rds_ib_connection *ic;
692 	unsigned long flags;
693 
694 	/* XXX too lazy? */
695 	ic = kzalloc(sizeof(struct rds_ib_connection), GFP_KERNEL);
696 	if (ic == NULL)
697 		return -ENOMEM;
698 
699 	INIT_LIST_HEAD(&ic->ib_node);
700 	tasklet_init(&ic->i_recv_tasklet, rds_ib_recv_tasklet_fn,
701 		     (unsigned long) ic);
702 	mutex_init(&ic->i_recv_mutex);
703 #ifndef KERNEL_HAS_ATOMIC64
704 	spin_lock_init(&ic->i_ack_lock);
705 #endif
706 
707 	/*
708 	 * rds_ib_conn_shutdown() waits for these to be emptied so they
709 	 * must be initialized before it can be called.
710 	 */
711 	rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
712 	rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
713 
714 	ic->conn = conn;
715 	conn->c_transport_data = ic;
716 
717 	spin_lock_irqsave(&ib_nodev_conns_lock, flags);
718 	list_add_tail(&ic->ib_node, &ib_nodev_conns);
719 	spin_unlock_irqrestore(&ib_nodev_conns_lock, flags);
720 
721 
722 	rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data);
723 	return 0;
724 }
725 
726 /*
727  * Free a connection. Connection must be shut down and not set for reconnect.
728  */
729 void rds_ib_conn_free(void *arg)
730 {
731 	struct rds_ib_connection *ic = arg;
732 	spinlock_t	*lock_ptr;
733 
734 	rdsdebug("ic %p\n", ic);
735 
736 	/*
737 	 * Conn is either on a dev's list or on the nodev list.
738 	 * A race with shutdown() or connect() would cause problems
739 	 * (since rds_ibdev would change) but that should never happen.
740 	 */
741 	lock_ptr = ic->rds_ibdev ? &ic->rds_ibdev->spinlock : &ib_nodev_conns_lock;
742 
743 	spin_lock_irq(lock_ptr);
744 	list_del(&ic->ib_node);
745 	spin_unlock_irq(lock_ptr);
746 
747 	kfree(ic);
748 }
749 
750 
751 /*
752  * An error occurred on the connection
753  */
754 void
755 __rds_ib_conn_error(struct rds_connection *conn, const char *fmt, ...)
756 {
757 	va_list ap;
758 
759 	rds_conn_drop(conn);
760 
761 	va_start(ap, fmt);
762 	vprintk(fmt, ap);
763 	va_end(ap);
764 }
765