xref: /openbmc/linux/net/rds/ib_cm.c (revision 7587eb18)
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 #include <linux/ratelimit.h>
38 
39 #include "rds.h"
40 #include "ib.h"
41 
42 /*
43  * Set the selected protocol version
44  */
45 static void rds_ib_set_protocol(struct rds_connection *conn, unsigned int version)
46 {
47 	conn->c_version = version;
48 }
49 
50 /*
51  * Set up flow control
52  */
53 static void rds_ib_set_flow_control(struct rds_connection *conn, u32 credits)
54 {
55 	struct rds_ib_connection *ic = conn->c_transport_data;
56 
57 	if (rds_ib_sysctl_flow_control && credits != 0) {
58 		/* We're doing flow control */
59 		ic->i_flowctl = 1;
60 		rds_ib_send_add_credits(conn, credits);
61 	} else {
62 		ic->i_flowctl = 0;
63 	}
64 }
65 
66 /*
67  * Tune RNR behavior. Without flow control, we use a rather
68  * low timeout, but not the absolute minimum - this should
69  * be tunable.
70  *
71  * We already set the RNR retry count to 7 (which is the
72  * smallest infinite number :-) above.
73  * If flow control is off, we want to change this back to 0
74  * so that we learn quickly when our credit accounting is
75  * buggy.
76  *
77  * Caller passes in a qp_attr pointer - don't waste stack spacv
78  * by allocation this twice.
79  */
80 static void
81 rds_ib_tune_rnr(struct rds_ib_connection *ic, struct ib_qp_attr *attr)
82 {
83 	int ret;
84 
85 	attr->min_rnr_timer = IB_RNR_TIMER_000_32;
86 	ret = ib_modify_qp(ic->i_cm_id->qp, attr, IB_QP_MIN_RNR_TIMER);
87 	if (ret)
88 		printk(KERN_NOTICE "ib_modify_qp(IB_QP_MIN_RNR_TIMER): err=%d\n", -ret);
89 }
90 
91 /*
92  * Connection established.
93  * We get here for both outgoing and incoming connection.
94  */
95 void rds_ib_cm_connect_complete(struct rds_connection *conn, struct rdma_cm_event *event)
96 {
97 	const struct rds_ib_connect_private *dp = NULL;
98 	struct rds_ib_connection *ic = conn->c_transport_data;
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 	if (conn->c_version < RDS_PROTOCOL(3,1)) {
115 		printk(KERN_NOTICE "RDS/IB: Connection to %pI4 version %u.%u failed,"
116 		       " no longer supported\n",
117 		       &conn->c_faddr,
118 		       RDS_PROTOCOL_MAJOR(conn->c_version),
119 		       RDS_PROTOCOL_MINOR(conn->c_version));
120 		rds_conn_destroy(conn);
121 		return;
122 	} else {
123 		printk(KERN_NOTICE "RDS/IB: connected to %pI4 version %u.%u%s\n",
124 		       &conn->c_faddr,
125 		       RDS_PROTOCOL_MAJOR(conn->c_version),
126 		       RDS_PROTOCOL_MINOR(conn->c_version),
127 		       ic->i_flowctl ? ", flow control" : "");
128 	}
129 
130 	/*
131 	 * Init rings and fill recv. this needs to wait until protocol negotiation
132 	 * is complete, since ring layout is different from 3.0 to 3.1.
133 	 */
134 	rds_ib_send_init_ring(ic);
135 	rds_ib_recv_init_ring(ic);
136 	/* Post receive buffers - as a side effect, this will update
137 	 * the posted credit count. */
138 	rds_ib_recv_refill(conn, 1, GFP_KERNEL);
139 
140 	/* Tune RNR behavior */
141 	rds_ib_tune_rnr(ic, &qp_attr);
142 
143 	qp_attr.qp_state = IB_QPS_RTS;
144 	err = ib_modify_qp(ic->i_cm_id->qp, &qp_attr, IB_QP_STATE);
145 	if (err)
146 		printk(KERN_NOTICE "ib_modify_qp(IB_QP_STATE, RTS): err=%d\n", err);
147 
148 	/* update ib_device with this local ipaddr */
149 	err = rds_ib_update_ipaddr(ic->rds_ibdev, conn->c_laddr);
150 	if (err)
151 		printk(KERN_ERR "rds_ib_update_ipaddr failed (%d)\n",
152 			err);
153 
154 	/* If the peer gave us the last packet it saw, process this as if
155 	 * we had received a regular ACK. */
156 	if (dp) {
157 		/* dp structure start is not guaranteed to be 8 bytes aligned.
158 		 * Since dp_ack_seq is 64-bit extended load operations can be
159 		 * used so go through get_unaligned to avoid unaligned errors.
160 		 */
161 		__be64 dp_ack_seq = get_unaligned(&dp->dp_ack_seq);
162 
163 		if (dp_ack_seq)
164 			rds_send_drop_acked(conn, be64_to_cpu(dp_ack_seq),
165 					    NULL);
166 	}
167 
168 	rds_connect_complete(conn);
169 }
170 
171 static void rds_ib_cm_fill_conn_param(struct rds_connection *conn,
172 			struct rdma_conn_param *conn_param,
173 			struct rds_ib_connect_private *dp,
174 			u32 protocol_version,
175 			u32 max_responder_resources,
176 			u32 max_initiator_depth)
177 {
178 	struct rds_ib_connection *ic = conn->c_transport_data;
179 	struct rds_ib_device *rds_ibdev = ic->rds_ibdev;
180 
181 	memset(conn_param, 0, sizeof(struct rdma_conn_param));
182 
183 	conn_param->responder_resources =
184 		min_t(u32, rds_ibdev->max_responder_resources, max_responder_resources);
185 	conn_param->initiator_depth =
186 		min_t(u32, rds_ibdev->max_initiator_depth, max_initiator_depth);
187 	conn_param->retry_count = min_t(unsigned int, rds_ib_retry_count, 7);
188 	conn_param->rnr_retry_count = 7;
189 
190 	if (dp) {
191 		memset(dp, 0, sizeof(*dp));
192 		dp->dp_saddr = conn->c_laddr;
193 		dp->dp_daddr = conn->c_faddr;
194 		dp->dp_protocol_major = RDS_PROTOCOL_MAJOR(protocol_version);
195 		dp->dp_protocol_minor = RDS_PROTOCOL_MINOR(protocol_version);
196 		dp->dp_protocol_minor_mask = cpu_to_be16(RDS_IB_SUPPORTED_PROTOCOLS);
197 		dp->dp_ack_seq = cpu_to_be64(rds_ib_piggyb_ack(ic));
198 
199 		/* Advertise flow control */
200 		if (ic->i_flowctl) {
201 			unsigned int credits;
202 
203 			credits = IB_GET_POST_CREDITS(atomic_read(&ic->i_credits));
204 			dp->dp_credit = cpu_to_be32(credits);
205 			atomic_sub(IB_SET_POST_CREDITS(credits), &ic->i_credits);
206 		}
207 
208 		conn_param->private_data = dp;
209 		conn_param->private_data_len = sizeof(*dp);
210 	}
211 }
212 
213 static void rds_ib_cq_event_handler(struct ib_event *event, void *data)
214 {
215 	rdsdebug("event %u (%s) data %p\n",
216 		 event->event, ib_event_msg(event->event), data);
217 }
218 
219 /* Plucking the oldest entry from the ring can be done concurrently with
220  * the thread refilling the ring.  Each ring operation is protected by
221  * spinlocks and the transient state of refilling doesn't change the
222  * recording of which entry is oldest.
223  *
224  * This relies on IB only calling one cq comp_handler for each cq so that
225  * there will only be one caller of rds_recv_incoming() per RDS connection.
226  */
227 static void rds_ib_cq_comp_handler_recv(struct ib_cq *cq, void *context)
228 {
229 	struct rds_connection *conn = context;
230 	struct rds_ib_connection *ic = conn->c_transport_data;
231 
232 	rdsdebug("conn %p cq %p\n", conn, cq);
233 
234 	rds_ib_stats_inc(s_ib_evt_handler_call);
235 
236 	tasklet_schedule(&ic->i_recv_tasklet);
237 }
238 
239 static void poll_scq(struct rds_ib_connection *ic, struct ib_cq *cq,
240 		     struct ib_wc *wcs)
241 {
242 	int nr, i;
243 	struct ib_wc *wc;
244 
245 	while ((nr = ib_poll_cq(cq, RDS_IB_WC_MAX, wcs)) > 0) {
246 		for (i = 0; i < nr; i++) {
247 			wc = wcs + i;
248 			rdsdebug("wc wr_id 0x%llx status %u byte_len %u imm_data %u\n",
249 				 (unsigned long long)wc->wr_id, wc->status,
250 				 wc->byte_len, be32_to_cpu(wc->ex.imm_data));
251 
252 			if (wc->wr_id <= ic->i_send_ring.w_nr ||
253 			    wc->wr_id == RDS_IB_ACK_WR_ID)
254 				rds_ib_send_cqe_handler(ic, wc);
255 			else
256 				rds_ib_mr_cqe_handler(ic, wc);
257 
258 		}
259 	}
260 }
261 
262 static void rds_ib_tasklet_fn_send(unsigned long data)
263 {
264 	struct rds_ib_connection *ic = (struct rds_ib_connection *)data;
265 	struct rds_connection *conn = ic->conn;
266 
267 	rds_ib_stats_inc(s_ib_tasklet_call);
268 
269 	poll_scq(ic, ic->i_send_cq, ic->i_send_wc);
270 	ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
271 	poll_scq(ic, ic->i_send_cq, ic->i_send_wc);
272 
273 	if (rds_conn_up(conn) &&
274 	    (!test_bit(RDS_LL_SEND_FULL, &conn->c_flags) ||
275 	    test_bit(0, &conn->c_map_queued)))
276 		rds_send_xmit(ic->conn);
277 }
278 
279 static void poll_rcq(struct rds_ib_connection *ic, struct ib_cq *cq,
280 		     struct ib_wc *wcs,
281 		     struct rds_ib_ack_state *ack_state)
282 {
283 	int nr, i;
284 	struct ib_wc *wc;
285 
286 	while ((nr = ib_poll_cq(cq, RDS_IB_WC_MAX, wcs)) > 0) {
287 		for (i = 0; i < nr; i++) {
288 			wc = wcs + i;
289 			rdsdebug("wc wr_id 0x%llx status %u byte_len %u imm_data %u\n",
290 				 (unsigned long long)wc->wr_id, wc->status,
291 				 wc->byte_len, be32_to_cpu(wc->ex.imm_data));
292 
293 			rds_ib_recv_cqe_handler(ic, wc, ack_state);
294 		}
295 	}
296 }
297 
298 static void rds_ib_tasklet_fn_recv(unsigned long data)
299 {
300 	struct rds_ib_connection *ic = (struct rds_ib_connection *)data;
301 	struct rds_connection *conn = ic->conn;
302 	struct rds_ib_device *rds_ibdev = ic->rds_ibdev;
303 	struct rds_ib_ack_state state;
304 
305 	if (!rds_ibdev)
306 		rds_conn_drop(conn);
307 
308 	rds_ib_stats_inc(s_ib_tasklet_call);
309 
310 	memset(&state, 0, sizeof(state));
311 	poll_rcq(ic, ic->i_recv_cq, ic->i_recv_wc, &state);
312 	ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
313 	poll_rcq(ic, ic->i_recv_cq, ic->i_recv_wc, &state);
314 
315 	if (state.ack_next_valid)
316 		rds_ib_set_ack(ic, state.ack_next, state.ack_required);
317 	if (state.ack_recv_valid && state.ack_recv > ic->i_ack_recv) {
318 		rds_send_drop_acked(conn, state.ack_recv, NULL);
319 		ic->i_ack_recv = state.ack_recv;
320 	}
321 
322 	if (rds_conn_up(conn))
323 		rds_ib_attempt_ack(ic);
324 }
325 
326 static void rds_ib_qp_event_handler(struct ib_event *event, void *data)
327 {
328 	struct rds_connection *conn = data;
329 	struct rds_ib_connection *ic = conn->c_transport_data;
330 
331 	rdsdebug("conn %p ic %p event %u (%s)\n", conn, ic, event->event,
332 		 ib_event_msg(event->event));
333 
334 	switch (event->event) {
335 	case IB_EVENT_COMM_EST:
336 		rdma_notify(ic->i_cm_id, IB_EVENT_COMM_EST);
337 		break;
338 	default:
339 		rdsdebug("Fatal QP Event %u (%s) "
340 			"- connection %pI4->%pI4, reconnecting\n",
341 			event->event, ib_event_msg(event->event),
342 			&conn->c_laddr, &conn->c_faddr);
343 		rds_conn_drop(conn);
344 		break;
345 	}
346 }
347 
348 static void rds_ib_cq_comp_handler_send(struct ib_cq *cq, void *context)
349 {
350 	struct rds_connection *conn = context;
351 	struct rds_ib_connection *ic = conn->c_transport_data;
352 
353 	rdsdebug("conn %p cq %p\n", conn, cq);
354 
355 	rds_ib_stats_inc(s_ib_evt_handler_call);
356 
357 	tasklet_schedule(&ic->i_send_tasklet);
358 }
359 
360 /*
361  * This needs to be very careful to not leave IS_ERR pointers around for
362  * cleanup to trip over.
363  */
364 static int rds_ib_setup_qp(struct rds_connection *conn)
365 {
366 	struct rds_ib_connection *ic = conn->c_transport_data;
367 	struct ib_device *dev = ic->i_cm_id->device;
368 	struct ib_qp_init_attr attr;
369 	struct ib_cq_init_attr cq_attr = {};
370 	struct rds_ib_device *rds_ibdev;
371 	int ret, fr_queue_space;
372 
373 	/*
374 	 * It's normal to see a null device if an incoming connection races
375 	 * with device removal, so we don't print a warning.
376 	 */
377 	rds_ibdev = rds_ib_get_client_data(dev);
378 	if (!rds_ibdev)
379 		return -EOPNOTSUPP;
380 
381 	/* The fr_queue_space is currently set to 512, to add extra space on
382 	 * completion queue and send queue. This extra space is used for FRMR
383 	 * registration and invalidation work requests
384 	 */
385 	fr_queue_space = (rds_ibdev->use_fastreg ? RDS_IB_DEFAULT_FR_WR : 0);
386 
387 	/* add the conn now so that connection establishment has the dev */
388 	rds_ib_add_conn(rds_ibdev, conn);
389 
390 	if (rds_ibdev->max_wrs < ic->i_send_ring.w_nr + 1)
391 		rds_ib_ring_resize(&ic->i_send_ring, rds_ibdev->max_wrs - 1);
392 	if (rds_ibdev->max_wrs < ic->i_recv_ring.w_nr + 1)
393 		rds_ib_ring_resize(&ic->i_recv_ring, rds_ibdev->max_wrs - 1);
394 
395 	/* Protection domain and memory range */
396 	ic->i_pd = rds_ibdev->pd;
397 
398 	cq_attr.cqe = ic->i_send_ring.w_nr + fr_queue_space + 1;
399 
400 	ic->i_send_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_send,
401 				     rds_ib_cq_event_handler, conn,
402 				     &cq_attr);
403 	if (IS_ERR(ic->i_send_cq)) {
404 		ret = PTR_ERR(ic->i_send_cq);
405 		ic->i_send_cq = NULL;
406 		rdsdebug("ib_create_cq send failed: %d\n", ret);
407 		goto out;
408 	}
409 
410 	cq_attr.cqe = ic->i_recv_ring.w_nr;
411 	ic->i_recv_cq = ib_create_cq(dev, rds_ib_cq_comp_handler_recv,
412 				     rds_ib_cq_event_handler, conn,
413 				     &cq_attr);
414 	if (IS_ERR(ic->i_recv_cq)) {
415 		ret = PTR_ERR(ic->i_recv_cq);
416 		ic->i_recv_cq = NULL;
417 		rdsdebug("ib_create_cq recv failed: %d\n", ret);
418 		goto out;
419 	}
420 
421 	ret = ib_req_notify_cq(ic->i_send_cq, IB_CQ_NEXT_COMP);
422 	if (ret) {
423 		rdsdebug("ib_req_notify_cq send failed: %d\n", ret);
424 		goto out;
425 	}
426 
427 	ret = ib_req_notify_cq(ic->i_recv_cq, IB_CQ_SOLICITED);
428 	if (ret) {
429 		rdsdebug("ib_req_notify_cq recv failed: %d\n", ret);
430 		goto out;
431 	}
432 
433 	/* XXX negotiate max send/recv with remote? */
434 	memset(&attr, 0, sizeof(attr));
435 	attr.event_handler = rds_ib_qp_event_handler;
436 	attr.qp_context = conn;
437 	/* + 1 to allow for the single ack message */
438 	attr.cap.max_send_wr = ic->i_send_ring.w_nr + fr_queue_space + 1;
439 	attr.cap.max_recv_wr = ic->i_recv_ring.w_nr + 1;
440 	attr.cap.max_send_sge = rds_ibdev->max_sge;
441 	attr.cap.max_recv_sge = RDS_IB_RECV_SGE;
442 	attr.sq_sig_type = IB_SIGNAL_REQ_WR;
443 	attr.qp_type = IB_QPT_RC;
444 	attr.send_cq = ic->i_send_cq;
445 	attr.recv_cq = ic->i_recv_cq;
446 	atomic_set(&ic->i_fastreg_wrs, RDS_IB_DEFAULT_FR_WR);
447 
448 	/*
449 	 * XXX this can fail if max_*_wr is too large?  Are we supposed
450 	 * to back off until we get a value that the hardware can support?
451 	 */
452 	ret = rdma_create_qp(ic->i_cm_id, ic->i_pd, &attr);
453 	if (ret) {
454 		rdsdebug("rdma_create_qp failed: %d\n", ret);
455 		goto out;
456 	}
457 
458 	ic->i_send_hdrs = ib_dma_alloc_coherent(dev,
459 					   ic->i_send_ring.w_nr *
460 						sizeof(struct rds_header),
461 					   &ic->i_send_hdrs_dma, GFP_KERNEL);
462 	if (!ic->i_send_hdrs) {
463 		ret = -ENOMEM;
464 		rdsdebug("ib_dma_alloc_coherent send failed\n");
465 		goto out;
466 	}
467 
468 	ic->i_recv_hdrs = ib_dma_alloc_coherent(dev,
469 					   ic->i_recv_ring.w_nr *
470 						sizeof(struct rds_header),
471 					   &ic->i_recv_hdrs_dma, GFP_KERNEL);
472 	if (!ic->i_recv_hdrs) {
473 		ret = -ENOMEM;
474 		rdsdebug("ib_dma_alloc_coherent recv failed\n");
475 		goto out;
476 	}
477 
478 	ic->i_ack = ib_dma_alloc_coherent(dev, sizeof(struct rds_header),
479 				       &ic->i_ack_dma, GFP_KERNEL);
480 	if (!ic->i_ack) {
481 		ret = -ENOMEM;
482 		rdsdebug("ib_dma_alloc_coherent ack failed\n");
483 		goto out;
484 	}
485 
486 	ic->i_sends = vzalloc_node(ic->i_send_ring.w_nr * sizeof(struct rds_ib_send_work),
487 				   ibdev_to_node(dev));
488 	if (!ic->i_sends) {
489 		ret = -ENOMEM;
490 		rdsdebug("send allocation failed\n");
491 		goto out;
492 	}
493 
494 	ic->i_recvs = vzalloc_node(ic->i_recv_ring.w_nr * sizeof(struct rds_ib_recv_work),
495 				   ibdev_to_node(dev));
496 	if (!ic->i_recvs) {
497 		ret = -ENOMEM;
498 		rdsdebug("recv allocation failed\n");
499 		goto out;
500 	}
501 
502 	rds_ib_recv_init_ack(ic);
503 
504 	rdsdebug("conn %p pd %p cq %p %p\n", conn, ic->i_pd,
505 		 ic->i_send_cq, ic->i_recv_cq);
506 
507 out:
508 	rds_ib_dev_put(rds_ibdev);
509 	return ret;
510 }
511 
512 static u32 rds_ib_protocol_compatible(struct rdma_cm_event *event)
513 {
514 	const struct rds_ib_connect_private *dp = event->param.conn.private_data;
515 	u16 common;
516 	u32 version = 0;
517 
518 	/*
519 	 * rdma_cm private data is odd - when there is any private data in the
520 	 * request, we will be given a pretty large buffer without telling us the
521 	 * original size. The only way to tell the difference is by looking at
522 	 * the contents, which are initialized to zero.
523 	 * If the protocol version fields aren't set, this is a connection attempt
524 	 * from an older version. This could could be 3.0 or 2.0 - we can't tell.
525 	 * We really should have changed this for OFED 1.3 :-(
526 	 */
527 
528 	/* Be paranoid. RDS always has privdata */
529 	if (!event->param.conn.private_data_len) {
530 		printk(KERN_NOTICE "RDS incoming connection has no private data, "
531 			"rejecting\n");
532 		return 0;
533 	}
534 
535 	/* Even if len is crap *now* I still want to check it. -ASG */
536 	if (event->param.conn.private_data_len < sizeof (*dp) ||
537 	    dp->dp_protocol_major == 0)
538 		return RDS_PROTOCOL_3_0;
539 
540 	common = be16_to_cpu(dp->dp_protocol_minor_mask) & RDS_IB_SUPPORTED_PROTOCOLS;
541 	if (dp->dp_protocol_major == 3 && common) {
542 		version = RDS_PROTOCOL_3_0;
543 		while ((common >>= 1) != 0)
544 			version++;
545 	} else
546 		printk_ratelimited(KERN_NOTICE "RDS: Connection from %pI4 using incompatible protocol version %u.%u\n",
547 				&dp->dp_saddr,
548 				dp->dp_protocol_major,
549 				dp->dp_protocol_minor);
550 	return version;
551 }
552 
553 int rds_ib_cm_handle_connect(struct rdma_cm_id *cm_id,
554 				    struct rdma_cm_event *event)
555 {
556 	__be64 lguid = cm_id->route.path_rec->sgid.global.interface_id;
557 	__be64 fguid = cm_id->route.path_rec->dgid.global.interface_id;
558 	const struct rds_ib_connect_private *dp = event->param.conn.private_data;
559 	struct rds_ib_connect_private dp_rep;
560 	struct rds_connection *conn = NULL;
561 	struct rds_ib_connection *ic = NULL;
562 	struct rdma_conn_param conn_param;
563 	u32 version;
564 	int err = 1, destroy = 1;
565 
566 	/* Check whether the remote protocol version matches ours. */
567 	version = rds_ib_protocol_compatible(event);
568 	if (!version)
569 		goto out;
570 
571 	rdsdebug("saddr %pI4 daddr %pI4 RDSv%u.%u lguid 0x%llx fguid "
572 		 "0x%llx\n", &dp->dp_saddr, &dp->dp_daddr,
573 		 RDS_PROTOCOL_MAJOR(version), RDS_PROTOCOL_MINOR(version),
574 		 (unsigned long long)be64_to_cpu(lguid),
575 		 (unsigned long long)be64_to_cpu(fguid));
576 
577 	/* RDS/IB is not currently netns aware, thus init_net */
578 	conn = rds_conn_create(&init_net, dp->dp_daddr, dp->dp_saddr,
579 			       &rds_ib_transport, GFP_KERNEL);
580 	if (IS_ERR(conn)) {
581 		rdsdebug("rds_conn_create failed (%ld)\n", PTR_ERR(conn));
582 		conn = NULL;
583 		goto out;
584 	}
585 
586 	/*
587 	 * The connection request may occur while the
588 	 * previous connection exist, e.g. in case of failover.
589 	 * But as connections may be initiated simultaneously
590 	 * by both hosts, we have a random backoff mechanism -
591 	 * see the comment above rds_queue_reconnect()
592 	 */
593 	mutex_lock(&conn->c_cm_lock);
594 	if (!rds_conn_transition(conn, RDS_CONN_DOWN, RDS_CONN_CONNECTING)) {
595 		if (rds_conn_state(conn) == RDS_CONN_UP) {
596 			rdsdebug("incoming connect while connecting\n");
597 			rds_conn_drop(conn);
598 			rds_ib_stats_inc(s_ib_listen_closed_stale);
599 		} else
600 		if (rds_conn_state(conn) == RDS_CONN_CONNECTING) {
601 			/* Wait and see - our connect may still be succeeding */
602 			rds_ib_stats_inc(s_ib_connect_raced);
603 		}
604 		goto out;
605 	}
606 
607 	ic = conn->c_transport_data;
608 
609 	rds_ib_set_protocol(conn, version);
610 	rds_ib_set_flow_control(conn, be32_to_cpu(dp->dp_credit));
611 
612 	/* If the peer gave us the last packet it saw, process this as if
613 	 * we had received a regular ACK. */
614 	if (dp->dp_ack_seq)
615 		rds_send_drop_acked(conn, be64_to_cpu(dp->dp_ack_seq), NULL);
616 
617 	BUG_ON(cm_id->context);
618 	BUG_ON(ic->i_cm_id);
619 
620 	ic->i_cm_id = cm_id;
621 	cm_id->context = conn;
622 
623 	/* We got halfway through setting up the ib_connection, if we
624 	 * fail now, we have to take the long route out of this mess. */
625 	destroy = 0;
626 
627 	err = rds_ib_setup_qp(conn);
628 	if (err) {
629 		rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", err);
630 		goto out;
631 	}
632 
633 	rds_ib_cm_fill_conn_param(conn, &conn_param, &dp_rep, version,
634 		event->param.conn.responder_resources,
635 		event->param.conn.initiator_depth);
636 
637 	/* rdma_accept() calls rdma_reject() internally if it fails */
638 	err = rdma_accept(cm_id, &conn_param);
639 	if (err)
640 		rds_ib_conn_error(conn, "rdma_accept failed (%d)\n", err);
641 
642 out:
643 	if (conn)
644 		mutex_unlock(&conn->c_cm_lock);
645 	if (err)
646 		rdma_reject(cm_id, NULL, 0);
647 	return destroy;
648 }
649 
650 
651 int rds_ib_cm_initiate_connect(struct rdma_cm_id *cm_id)
652 {
653 	struct rds_connection *conn = cm_id->context;
654 	struct rds_ib_connection *ic = conn->c_transport_data;
655 	struct rdma_conn_param conn_param;
656 	struct rds_ib_connect_private dp;
657 	int ret;
658 
659 	/* If the peer doesn't do protocol negotiation, we must
660 	 * default to RDSv3.0 */
661 	rds_ib_set_protocol(conn, RDS_PROTOCOL_3_0);
662 	ic->i_flowctl = rds_ib_sysctl_flow_control;	/* advertise flow control */
663 
664 	ret = rds_ib_setup_qp(conn);
665 	if (ret) {
666 		rds_ib_conn_error(conn, "rds_ib_setup_qp failed (%d)\n", ret);
667 		goto out;
668 	}
669 
670 	rds_ib_cm_fill_conn_param(conn, &conn_param, &dp, RDS_PROTOCOL_VERSION,
671 		UINT_MAX, UINT_MAX);
672 	ret = rdma_connect(cm_id, &conn_param);
673 	if (ret)
674 		rds_ib_conn_error(conn, "rdma_connect failed (%d)\n", ret);
675 
676 out:
677 	/* Beware - returning non-zero tells the rdma_cm to destroy
678 	 * the cm_id. We should certainly not do it as long as we still
679 	 * "own" the cm_id. */
680 	if (ret) {
681 		if (ic->i_cm_id == cm_id)
682 			ret = 0;
683 	}
684 	return ret;
685 }
686 
687 int rds_ib_conn_connect(struct rds_connection *conn)
688 {
689 	struct rds_ib_connection *ic = conn->c_transport_data;
690 	struct sockaddr_in src, dest;
691 	int ret;
692 
693 	/* XXX I wonder what affect the port space has */
694 	/* delegate cm event handler to rdma_transport */
695 	ic->i_cm_id = rdma_create_id(&init_net, rds_rdma_cm_event_handler, conn,
696 				     RDMA_PS_TCP, IB_QPT_RC);
697 	if (IS_ERR(ic->i_cm_id)) {
698 		ret = PTR_ERR(ic->i_cm_id);
699 		ic->i_cm_id = NULL;
700 		rdsdebug("rdma_create_id() failed: %d\n", ret);
701 		goto out;
702 	}
703 
704 	rdsdebug("created cm id %p for conn %p\n", ic->i_cm_id, conn);
705 
706 	src.sin_family = AF_INET;
707 	src.sin_addr.s_addr = (__force u32)conn->c_laddr;
708 	src.sin_port = (__force u16)htons(0);
709 
710 	dest.sin_family = AF_INET;
711 	dest.sin_addr.s_addr = (__force u32)conn->c_faddr;
712 	dest.sin_port = (__force u16)htons(RDS_PORT);
713 
714 	ret = rdma_resolve_addr(ic->i_cm_id, (struct sockaddr *)&src,
715 				(struct sockaddr *)&dest,
716 				RDS_RDMA_RESOLVE_TIMEOUT_MS);
717 	if (ret) {
718 		rdsdebug("addr resolve failed for cm id %p: %d\n", ic->i_cm_id,
719 			 ret);
720 		rdma_destroy_id(ic->i_cm_id);
721 		ic->i_cm_id = NULL;
722 	}
723 
724 out:
725 	return ret;
726 }
727 
728 /*
729  * This is so careful about only cleaning up resources that were built up
730  * so that it can be called at any point during startup.  In fact it
731  * can be called multiple times for a given connection.
732  */
733 void rds_ib_conn_shutdown(struct rds_connection *conn)
734 {
735 	struct rds_ib_connection *ic = conn->c_transport_data;
736 	int err = 0;
737 
738 	rdsdebug("cm %p pd %p cq %p %p qp %p\n", ic->i_cm_id,
739 		 ic->i_pd, ic->i_send_cq, ic->i_recv_cq,
740 		 ic->i_cm_id ? ic->i_cm_id->qp : NULL);
741 
742 	if (ic->i_cm_id) {
743 		struct ib_device *dev = ic->i_cm_id->device;
744 
745 		rdsdebug("disconnecting cm %p\n", ic->i_cm_id);
746 		err = rdma_disconnect(ic->i_cm_id);
747 		if (err) {
748 			/* Actually this may happen quite frequently, when
749 			 * an outgoing connect raced with an incoming connect.
750 			 */
751 			rdsdebug("failed to disconnect, cm: %p err %d\n",
752 				ic->i_cm_id, err);
753 		}
754 
755 		/*
756 		 * We want to wait for tx and rx completion to finish
757 		 * before we tear down the connection, but we have to be
758 		 * careful not to get stuck waiting on a send ring that
759 		 * only has unsignaled sends in it.  We've shutdown new
760 		 * sends before getting here so by waiting for signaled
761 		 * sends to complete we're ensured that there will be no
762 		 * more tx processing.
763 		 */
764 		wait_event(rds_ib_ring_empty_wait,
765 			   rds_ib_ring_empty(&ic->i_recv_ring) &&
766 			   (atomic_read(&ic->i_signaled_sends) == 0) &&
767 			   (atomic_read(&ic->i_fastreg_wrs) == RDS_IB_DEFAULT_FR_WR));
768 		tasklet_kill(&ic->i_send_tasklet);
769 		tasklet_kill(&ic->i_recv_tasklet);
770 
771 		/* first destroy the ib state that generates callbacks */
772 		if (ic->i_cm_id->qp)
773 			rdma_destroy_qp(ic->i_cm_id);
774 		if (ic->i_send_cq)
775 			ib_destroy_cq(ic->i_send_cq);
776 		if (ic->i_recv_cq)
777 			ib_destroy_cq(ic->i_recv_cq);
778 
779 		/* then free the resources that ib callbacks use */
780 		if (ic->i_send_hdrs)
781 			ib_dma_free_coherent(dev,
782 					   ic->i_send_ring.w_nr *
783 						sizeof(struct rds_header),
784 					   ic->i_send_hdrs,
785 					   ic->i_send_hdrs_dma);
786 
787 		if (ic->i_recv_hdrs)
788 			ib_dma_free_coherent(dev,
789 					   ic->i_recv_ring.w_nr *
790 						sizeof(struct rds_header),
791 					   ic->i_recv_hdrs,
792 					   ic->i_recv_hdrs_dma);
793 
794 		if (ic->i_ack)
795 			ib_dma_free_coherent(dev, sizeof(struct rds_header),
796 					     ic->i_ack, ic->i_ack_dma);
797 
798 		if (ic->i_sends)
799 			rds_ib_send_clear_ring(ic);
800 		if (ic->i_recvs)
801 			rds_ib_recv_clear_ring(ic);
802 
803 		rdma_destroy_id(ic->i_cm_id);
804 
805 		/*
806 		 * Move connection back to the nodev list.
807 		 */
808 		if (ic->rds_ibdev)
809 			rds_ib_remove_conn(ic->rds_ibdev, conn);
810 
811 		ic->i_cm_id = NULL;
812 		ic->i_pd = NULL;
813 		ic->i_send_cq = NULL;
814 		ic->i_recv_cq = NULL;
815 		ic->i_send_hdrs = NULL;
816 		ic->i_recv_hdrs = NULL;
817 		ic->i_ack = NULL;
818 	}
819 	BUG_ON(ic->rds_ibdev);
820 
821 	/* Clear pending transmit */
822 	if (ic->i_data_op) {
823 		struct rds_message *rm;
824 
825 		rm = container_of(ic->i_data_op, struct rds_message, data);
826 		rds_message_put(rm);
827 		ic->i_data_op = NULL;
828 	}
829 
830 	/* Clear the ACK state */
831 	clear_bit(IB_ACK_IN_FLIGHT, &ic->i_ack_flags);
832 #ifdef KERNEL_HAS_ATOMIC64
833 	atomic64_set(&ic->i_ack_next, 0);
834 #else
835 	ic->i_ack_next = 0;
836 #endif
837 	ic->i_ack_recv = 0;
838 
839 	/* Clear flow control state */
840 	ic->i_flowctl = 0;
841 	atomic_set(&ic->i_credits, 0);
842 
843 	rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
844 	rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
845 
846 	if (ic->i_ibinc) {
847 		rds_inc_put(&ic->i_ibinc->ii_inc);
848 		ic->i_ibinc = NULL;
849 	}
850 
851 	vfree(ic->i_sends);
852 	ic->i_sends = NULL;
853 	vfree(ic->i_recvs);
854 	ic->i_recvs = NULL;
855 }
856 
857 int rds_ib_conn_alloc(struct rds_connection *conn, gfp_t gfp)
858 {
859 	struct rds_ib_connection *ic;
860 	unsigned long flags;
861 	int ret;
862 
863 	/* XXX too lazy? */
864 	ic = kzalloc(sizeof(struct rds_ib_connection), gfp);
865 	if (!ic)
866 		return -ENOMEM;
867 
868 	ret = rds_ib_recv_alloc_caches(ic);
869 	if (ret) {
870 		kfree(ic);
871 		return ret;
872 	}
873 
874 	INIT_LIST_HEAD(&ic->ib_node);
875 	tasklet_init(&ic->i_send_tasklet, rds_ib_tasklet_fn_send,
876 		     (unsigned long)ic);
877 	tasklet_init(&ic->i_recv_tasklet, rds_ib_tasklet_fn_recv,
878 		     (unsigned long)ic);
879 	mutex_init(&ic->i_recv_mutex);
880 #ifndef KERNEL_HAS_ATOMIC64
881 	spin_lock_init(&ic->i_ack_lock);
882 #endif
883 	atomic_set(&ic->i_signaled_sends, 0);
884 
885 	/*
886 	 * rds_ib_conn_shutdown() waits for these to be emptied so they
887 	 * must be initialized before it can be called.
888 	 */
889 	rds_ib_ring_init(&ic->i_send_ring, rds_ib_sysctl_max_send_wr);
890 	rds_ib_ring_init(&ic->i_recv_ring, rds_ib_sysctl_max_recv_wr);
891 
892 	ic->conn = conn;
893 	conn->c_transport_data = ic;
894 
895 	spin_lock_irqsave(&ib_nodev_conns_lock, flags);
896 	list_add_tail(&ic->ib_node, &ib_nodev_conns);
897 	spin_unlock_irqrestore(&ib_nodev_conns_lock, flags);
898 
899 
900 	rdsdebug("conn %p conn ic %p\n", conn, conn->c_transport_data);
901 	return 0;
902 }
903 
904 /*
905  * Free a connection. Connection must be shut down and not set for reconnect.
906  */
907 void rds_ib_conn_free(void *arg)
908 {
909 	struct rds_ib_connection *ic = arg;
910 	spinlock_t	*lock_ptr;
911 
912 	rdsdebug("ic %p\n", ic);
913 
914 	/*
915 	 * Conn is either on a dev's list or on the nodev list.
916 	 * A race with shutdown() or connect() would cause problems
917 	 * (since rds_ibdev would change) but that should never happen.
918 	 */
919 	lock_ptr = ic->rds_ibdev ? &ic->rds_ibdev->spinlock : &ib_nodev_conns_lock;
920 
921 	spin_lock_irq(lock_ptr);
922 	list_del(&ic->ib_node);
923 	spin_unlock_irq(lock_ptr);
924 
925 	rds_ib_recv_free_caches(ic);
926 
927 	kfree(ic);
928 }
929 
930 
931 /*
932  * An error occurred on the connection
933  */
934 void
935 __rds_ib_conn_error(struct rds_connection *conn, const char *fmt, ...)
936 {
937 	va_list ap;
938 
939 	rds_conn_drop(conn);
940 
941 	va_start(ap, fmt);
942 	vprintk(fmt, ap);
943 	va_end(ap);
944 }
945