xref: /openbmc/linux/net/9p/trans_rdma.c (revision 9efac679) 
1 /*
2  * linux/fs/9p/trans_rdma.c
3  *
4  * RDMA transport layer based on the trans_fd.c implementation.
5  *
6  *  Copyright (C) 2008 by Tom Tucker <tom@opengridcomputing.com>
7  *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
8  *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
9  *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
10  *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
11  *
12  *  This program is free software; you can redistribute it and/or modify
13  *  it under the terms of the GNU General Public License version 2
14  *  as published by the Free Software Foundation.
15  *
16  *  This program is distributed in the hope that it will be useful,
17  *  but WITHOUT ANY WARRANTY; without even the implied warranty of
18  *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
19  *  GNU General Public License for more details.
20  *
21  *  You should have received a copy of the GNU General Public License
22  *  along with this program; if not, write to:
23  *  Free Software Foundation
24  *  51 Franklin Street, Fifth Floor
25  *  Boston, MA  02111-1301  USA
26  *
27  */
28 
29 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
30 
31 #include <linux/in.h>
32 #include <linux/module.h>
33 #include <linux/net.h>
34 #include <linux/ipv6.h>
35 #include <linux/kthread.h>
36 #include <linux/errno.h>
37 #include <linux/kernel.h>
38 #include <linux/un.h>
39 #include <linux/uaccess.h>
40 #include <linux/inet.h>
41 #include <linux/idr.h>
42 #include <linux/file.h>
43 #include <linux/parser.h>
44 #include <linux/semaphore.h>
45 #include <linux/slab.h>
46 #include <net/9p/9p.h>
47 #include <net/9p/client.h>
48 #include <net/9p/transport.h>
49 #include <rdma/ib_verbs.h>
50 #include <rdma/rdma_cm.h>
51 
52 #define P9_PORT			5640
53 #define P9_RDMA_SQ_DEPTH	32
54 #define P9_RDMA_RQ_DEPTH	32
55 #define P9_RDMA_SEND_SGE	4
56 #define P9_RDMA_RECV_SGE	4
57 #define P9_RDMA_IRD		0
58 #define P9_RDMA_ORD		0
59 #define P9_RDMA_TIMEOUT		30000		/* 30 seconds */
60 #define P9_RDMA_MAXSIZE		(1024*1024)	/* 1MB */
61 
62 /**
63  * struct p9_trans_rdma - RDMA transport instance
64  *
65  * @state: tracks the transport state machine for connection setup and tear down
66  * @cm_id: The RDMA CM ID
67  * @pd: Protection Domain pointer
68  * @qp: Queue Pair pointer
69  * @cq: Completion Queue pointer
70  * @dm_mr: DMA Memory Region pointer
71  * @lkey: The local access only memory region key
72  * @timeout: Number of uSecs to wait for connection management events
73  * @sq_depth: The depth of the Send Queue
74  * @sq_sem: Semaphore for the SQ
75  * @rq_depth: The depth of the Receive Queue.
76  * @rq_sem: Semaphore for the RQ
77  * @excess_rc : Amount of posted Receive Contexts without a pending request.
78  *		See rdma_request()
79  * @addr: The remote peer's address
80  * @req_lock: Protects the active request list
81  * @cm_done: Completion event for connection management tracking
82  */
83 struct p9_trans_rdma {
84 	enum {
85 		P9_RDMA_INIT,
86 		P9_RDMA_ADDR_RESOLVED,
87 		P9_RDMA_ROUTE_RESOLVED,
88 		P9_RDMA_CONNECTED,
89 		P9_RDMA_FLUSHING,
90 		P9_RDMA_CLOSING,
91 		P9_RDMA_CLOSED,
92 	} state;
93 	struct rdma_cm_id *cm_id;
94 	struct ib_pd *pd;
95 	struct ib_qp *qp;
96 	struct ib_cq *cq;
97 	long timeout;
98 	int sq_depth;
99 	struct semaphore sq_sem;
100 	int rq_depth;
101 	struct semaphore rq_sem;
102 	atomic_t excess_rc;
103 	struct sockaddr_in addr;
104 	spinlock_t req_lock;
105 
106 	struct completion cm_done;
107 };
108 
109 /**
110  * p9_rdma_context - Keeps track of in-process WR
111  *
112  * @busa: Bus address to unmap when the WR completes
113  * @req: Keeps track of requests (send)
114  * @rc: Keepts track of replies (receive)
115  */
116 struct p9_rdma_req;
117 struct p9_rdma_context {
118 	struct ib_cqe cqe;
119 	dma_addr_t busa;
120 	union {
121 		struct p9_req_t *req;
122 		struct p9_fcall *rc;
123 	};
124 };
125 
126 /**
127  * p9_rdma_opts - Collection of mount options
128  * @port: port of connection
129  * @sq_depth: The requested depth of the SQ. This really doesn't need
130  * to be any deeper than the number of threads used in the client
131  * @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
132  * @timeout: Time to wait in msecs for CM events
133  */
134 struct p9_rdma_opts {
135 	short port;
136 	int sq_depth;
137 	int rq_depth;
138 	long timeout;
139 	int privport;
140 };
141 
142 /*
143  * Option Parsing (code inspired by NFS code)
144  */
145 enum {
146 	/* Options that take integer arguments */
147 	Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout,
148 	/* Options that take no argument */
149 	Opt_privport,
150 	Opt_err,
151 };
152 
153 static match_table_t tokens = {
154 	{Opt_port, "port=%u"},
155 	{Opt_sq_depth, "sq=%u"},
156 	{Opt_rq_depth, "rq=%u"},
157 	{Opt_timeout, "timeout=%u"},
158 	{Opt_privport, "privport"},
159 	{Opt_err, NULL},
160 };
161 
162 /**
163  * parse_opts - parse mount options into rdma options structure
164  * @params: options string passed from mount
165  * @opts: rdma transport-specific structure to parse options into
166  *
167  * Returns 0 upon success, -ERRNO upon failure
168  */
169 static int parse_opts(char *params, struct p9_rdma_opts *opts)
170 {
171 	char *p;
172 	substring_t args[MAX_OPT_ARGS];
173 	int option;
174 	char *options, *tmp_options;
175 
176 	opts->port = P9_PORT;
177 	opts->sq_depth = P9_RDMA_SQ_DEPTH;
178 	opts->rq_depth = P9_RDMA_RQ_DEPTH;
179 	opts->timeout = P9_RDMA_TIMEOUT;
180 	opts->privport = 0;
181 
182 	if (!params)
183 		return 0;
184 
185 	tmp_options = kstrdup(params, GFP_KERNEL);
186 	if (!tmp_options) {
187 		p9_debug(P9_DEBUG_ERROR,
188 			 "failed to allocate copy of option string\n");
189 		return -ENOMEM;
190 	}
191 	options = tmp_options;
192 
193 	while ((p = strsep(&options, ",")) != NULL) {
194 		int token;
195 		int r;
196 		if (!*p)
197 			continue;
198 		token = match_token(p, tokens, args);
199 		if ((token != Opt_err) && (token != Opt_privport)) {
200 			r = match_int(&args[0], &option);
201 			if (r < 0) {
202 				p9_debug(P9_DEBUG_ERROR,
203 					 "integer field, but no integer?\n");
204 				continue;
205 			}
206 		}
207 		switch (token) {
208 		case Opt_port:
209 			opts->port = option;
210 			break;
211 		case Opt_sq_depth:
212 			opts->sq_depth = option;
213 			break;
214 		case Opt_rq_depth:
215 			opts->rq_depth = option;
216 			break;
217 		case Opt_timeout:
218 			opts->timeout = option;
219 			break;
220 		case Opt_privport:
221 			opts->privport = 1;
222 			break;
223 		default:
224 			continue;
225 		}
226 	}
227 	/* RQ must be at least as large as the SQ */
228 	opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
229 	kfree(tmp_options);
230 	return 0;
231 }
232 
233 static int
234 p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
235 {
236 	struct p9_client *c = id->context;
237 	struct p9_trans_rdma *rdma = c->trans;
238 	switch (event->event) {
239 	case RDMA_CM_EVENT_ADDR_RESOLVED:
240 		BUG_ON(rdma->state != P9_RDMA_INIT);
241 		rdma->state = P9_RDMA_ADDR_RESOLVED;
242 		break;
243 
244 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
245 		BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
246 		rdma->state = P9_RDMA_ROUTE_RESOLVED;
247 		break;
248 
249 	case RDMA_CM_EVENT_ESTABLISHED:
250 		BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
251 		rdma->state = P9_RDMA_CONNECTED;
252 		break;
253 
254 	case RDMA_CM_EVENT_DISCONNECTED:
255 		if (rdma)
256 			rdma->state = P9_RDMA_CLOSED;
257 		if (c)
258 			c->status = Disconnected;
259 		break;
260 
261 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
262 		break;
263 
264 	case RDMA_CM_EVENT_ADDR_CHANGE:
265 	case RDMA_CM_EVENT_ROUTE_ERROR:
266 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
267 	case RDMA_CM_EVENT_MULTICAST_JOIN:
268 	case RDMA_CM_EVENT_MULTICAST_ERROR:
269 	case RDMA_CM_EVENT_REJECTED:
270 	case RDMA_CM_EVENT_CONNECT_REQUEST:
271 	case RDMA_CM_EVENT_CONNECT_RESPONSE:
272 	case RDMA_CM_EVENT_CONNECT_ERROR:
273 	case RDMA_CM_EVENT_ADDR_ERROR:
274 	case RDMA_CM_EVENT_UNREACHABLE:
275 		c->status = Disconnected;
276 		rdma_disconnect(rdma->cm_id);
277 		break;
278 	default:
279 		BUG();
280 	}
281 	complete(&rdma->cm_done);
282 	return 0;
283 }
284 
285 static void
286 recv_done(struct ib_cq *cq, struct ib_wc *wc)
287 {
288 	struct p9_client *client = cq->cq_context;
289 	struct p9_trans_rdma *rdma = client->trans;
290 	struct p9_rdma_context *c =
291 		container_of(wc->wr_cqe, struct p9_rdma_context, cqe);
292 	struct p9_req_t *req;
293 	int err = 0;
294 	int16_t tag;
295 
296 	req = NULL;
297 	ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
298 							 DMA_FROM_DEVICE);
299 
300 	if (wc->status != IB_WC_SUCCESS)
301 		goto err_out;
302 
303 	err = p9_parse_header(c->rc, NULL, NULL, &tag, 1);
304 	if (err)
305 		goto err_out;
306 
307 	req = p9_tag_lookup(client, tag);
308 	if (!req)
309 		goto err_out;
310 
311 	/* Check that we have not yet received a reply for this request.
312 	 */
313 	if (unlikely(req->rc)) {
314 		pr_err("Duplicate reply for request %d", tag);
315 		goto err_out;
316 	}
317 
318 	req->rc = c->rc;
319 	p9_client_cb(client, req, REQ_STATUS_RCVD);
320 
321  out:
322 	up(&rdma->rq_sem);
323 	kfree(c);
324 	return;
325 
326  err_out:
327 	p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n",
328 			req, err, wc->status);
329 	rdma->state = P9_RDMA_FLUSHING;
330 	client->status = Disconnected;
331 	goto out;
332 }
333 
334 static void
335 send_done(struct ib_cq *cq, struct ib_wc *wc)
336 {
337 	struct p9_client *client = cq->cq_context;
338 	struct p9_trans_rdma *rdma = client->trans;
339 	struct p9_rdma_context *c =
340 		container_of(wc->wr_cqe, struct p9_rdma_context, cqe);
341 
342 	ib_dma_unmap_single(rdma->cm_id->device,
343 			    c->busa, c->req->tc->size,
344 			    DMA_TO_DEVICE);
345 	up(&rdma->sq_sem);
346 	kfree(c);
347 }
348 
349 static void qp_event_handler(struct ib_event *event, void *context)
350 {
351 	p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
352 		 event->event, context);
353 }
354 
355 static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
356 {
357 	if (!rdma)
358 		return;
359 
360 	if (rdma->qp && !IS_ERR(rdma->qp))
361 		ib_destroy_qp(rdma->qp);
362 
363 	if (rdma->pd && !IS_ERR(rdma->pd))
364 		ib_dealloc_pd(rdma->pd);
365 
366 	if (rdma->cq && !IS_ERR(rdma->cq))
367 		ib_free_cq(rdma->cq);
368 
369 	if (rdma->cm_id && !IS_ERR(rdma->cm_id))
370 		rdma_destroy_id(rdma->cm_id);
371 
372 	kfree(rdma);
373 }
374 
375 static int
376 post_recv(struct p9_client *client, struct p9_rdma_context *c)
377 {
378 	struct p9_trans_rdma *rdma = client->trans;
379 	struct ib_recv_wr wr, *bad_wr;
380 	struct ib_sge sge;
381 
382 	c->busa = ib_dma_map_single(rdma->cm_id->device,
383 				    c->rc->sdata, client->msize,
384 				    DMA_FROM_DEVICE);
385 	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
386 		goto error;
387 
388 	c->cqe.done = recv_done;
389 
390 	sge.addr = c->busa;
391 	sge.length = client->msize;
392 	sge.lkey = rdma->pd->local_dma_lkey;
393 
394 	wr.next = NULL;
395 	wr.wr_cqe = &c->cqe;
396 	wr.sg_list = &sge;
397 	wr.num_sge = 1;
398 	return ib_post_recv(rdma->qp, &wr, &bad_wr);
399 
400  error:
401 	p9_debug(P9_DEBUG_ERROR, "EIO\n");
402 	return -EIO;
403 }
404 
405 static int rdma_request(struct p9_client *client, struct p9_req_t *req)
406 {
407 	struct p9_trans_rdma *rdma = client->trans;
408 	struct ib_send_wr wr, *bad_wr;
409 	struct ib_sge sge;
410 	int err = 0;
411 	unsigned long flags;
412 	struct p9_rdma_context *c = NULL;
413 	struct p9_rdma_context *rpl_context = NULL;
414 
415 	/* When an error occurs between posting the recv and the send,
416 	 * there will be a receive context posted without a pending request.
417 	 * Since there is no way to "un-post" it, we remember it and skip
418 	 * post_recv() for the next request.
419 	 * So here,
420 	 * see if we are this `next request' and need to absorb an excess rc.
421 	 * If yes, then drop and free our own, and do not recv_post().
422 	 **/
423 	if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
424 		if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
425 			/* Got one ! */
426 			kfree(req->rc);
427 			req->rc = NULL;
428 			goto dont_need_post_recv;
429 		} else {
430 			/* We raced and lost. */
431 			atomic_inc(&rdma->excess_rc);
432 		}
433 	}
434 
435 	/* Allocate an fcall for the reply */
436 	rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
437 	if (!rpl_context) {
438 		err = -ENOMEM;
439 		goto recv_error;
440 	}
441 	rpl_context->rc = req->rc;
442 
443 	/*
444 	 * Post a receive buffer for this request. We need to ensure
445 	 * there is a reply buffer available for every outstanding
446 	 * request. A flushed request can result in no reply for an
447 	 * outstanding request, so we must keep a count to avoid
448 	 * overflowing the RQ.
449 	 */
450 	if (down_interruptible(&rdma->rq_sem)) {
451 		err = -EINTR;
452 		goto recv_error;
453 	}
454 
455 	err = post_recv(client, rpl_context);
456 	if (err) {
457 		p9_debug(P9_DEBUG_FCALL, "POST RECV failed\n");
458 		goto recv_error;
459 	}
460 	/* remove posted receive buffer from request structure */
461 	req->rc = NULL;
462 
463 dont_need_post_recv:
464 	/* Post the request */
465 	c = kmalloc(sizeof *c, GFP_NOFS);
466 	if (!c) {
467 		err = -ENOMEM;
468 		goto send_error;
469 	}
470 	c->req = req;
471 
472 	c->busa = ib_dma_map_single(rdma->cm_id->device,
473 				    c->req->tc->sdata, c->req->tc->size,
474 				    DMA_TO_DEVICE);
475 	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
476 		err = -EIO;
477 		goto send_error;
478 	}
479 
480 	c->cqe.done = send_done;
481 
482 	sge.addr = c->busa;
483 	sge.length = c->req->tc->size;
484 	sge.lkey = rdma->pd->local_dma_lkey;
485 
486 	wr.next = NULL;
487 	wr.wr_cqe = &c->cqe;
488 	wr.opcode = IB_WR_SEND;
489 	wr.send_flags = IB_SEND_SIGNALED;
490 	wr.sg_list = &sge;
491 	wr.num_sge = 1;
492 
493 	if (down_interruptible(&rdma->sq_sem)) {
494 		err = -EINTR;
495 		goto send_error;
496 	}
497 
498 	/* Mark request as `sent' *before* we actually send it,
499 	 * because doing if after could erase the REQ_STATUS_RCVD
500 	 * status in case of a very fast reply.
501 	 */
502 	req->status = REQ_STATUS_SENT;
503 	err = ib_post_send(rdma->qp, &wr, &bad_wr);
504 	if (err)
505 		goto send_error;
506 
507 	/* Success */
508 	return 0;
509 
510  /* Handle errors that happened during or while preparing the send: */
511  send_error:
512 	req->status = REQ_STATUS_ERROR;
513 	kfree(c);
514 	p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);
515 
516 	/* Ach.
517 	 *  We did recv_post(), but not send. We have one recv_post in excess.
518 	 */
519 	atomic_inc(&rdma->excess_rc);
520 	return err;
521 
522  /* Handle errors that happened during or while preparing post_recv(): */
523  recv_error:
524 	kfree(rpl_context);
525 	spin_lock_irqsave(&rdma->req_lock, flags);
526 	if (rdma->state < P9_RDMA_CLOSING) {
527 		rdma->state = P9_RDMA_CLOSING;
528 		spin_unlock_irqrestore(&rdma->req_lock, flags);
529 		rdma_disconnect(rdma->cm_id);
530 	} else
531 		spin_unlock_irqrestore(&rdma->req_lock, flags);
532 	return err;
533 }
534 
535 static void rdma_close(struct p9_client *client)
536 {
537 	struct p9_trans_rdma *rdma;
538 
539 	if (!client)
540 		return;
541 
542 	rdma = client->trans;
543 	if (!rdma)
544 		return;
545 
546 	client->status = Disconnected;
547 	rdma_disconnect(rdma->cm_id);
548 	rdma_destroy_trans(rdma);
549 }
550 
551 /**
552  * alloc_rdma - Allocate and initialize the rdma transport structure
553  * @opts: Mount options structure
554  */
555 static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
556 {
557 	struct p9_trans_rdma *rdma;
558 
559 	rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
560 	if (!rdma)
561 		return NULL;
562 
563 	rdma->sq_depth = opts->sq_depth;
564 	rdma->rq_depth = opts->rq_depth;
565 	rdma->timeout = opts->timeout;
566 	spin_lock_init(&rdma->req_lock);
567 	init_completion(&rdma->cm_done);
568 	sema_init(&rdma->sq_sem, rdma->sq_depth);
569 	sema_init(&rdma->rq_sem, rdma->rq_depth);
570 	atomic_set(&rdma->excess_rc, 0);
571 
572 	return rdma;
573 }
574 
575 static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
576 {
577 	/* Nothing to do here.
578 	 * We will take care of it (if we have to) in rdma_cancelled()
579 	 */
580 	return 1;
581 }
582 
583 /* A request has been fully flushed without a reply.
584  * That means we have posted one buffer in excess.
585  */
586 static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
587 {
588 	struct p9_trans_rdma *rdma = client->trans;
589 	atomic_inc(&rdma->excess_rc);
590 	return 0;
591 }
592 
593 static int p9_rdma_bind_privport(struct p9_trans_rdma *rdma)
594 {
595 	struct sockaddr_in cl = {
596 		.sin_family = AF_INET,
597 		.sin_addr.s_addr = htonl(INADDR_ANY),
598 	};
599 	int port, err = -EINVAL;
600 
601 	for (port = P9_DEF_MAX_RESVPORT; port >= P9_DEF_MIN_RESVPORT; port--) {
602 		cl.sin_port = htons((ushort)port);
603 		err = rdma_bind_addr(rdma->cm_id, (struct sockaddr *)&cl);
604 		if (err != -EADDRINUSE)
605 			break;
606 	}
607 	return err;
608 }
609 
610 /**
611  * trans_create_rdma - Transport method for creating atransport instance
612  * @client: client instance
613  * @addr: IP address string
614  * @args: Mount options string
615  */
616 static int
617 rdma_create_trans(struct p9_client *client, const char *addr, char *args)
618 {
619 	int err;
620 	struct p9_rdma_opts opts;
621 	struct p9_trans_rdma *rdma;
622 	struct rdma_conn_param conn_param;
623 	struct ib_qp_init_attr qp_attr;
624 
625 	/* Parse the transport specific mount options */
626 	err = parse_opts(args, &opts);
627 	if (err < 0)
628 		return err;
629 
630 	/* Create and initialize the RDMA transport structure */
631 	rdma = alloc_rdma(&opts);
632 	if (!rdma)
633 		return -ENOMEM;
634 
635 	/* Create the RDMA CM ID */
636 	rdma->cm_id = rdma_create_id(&init_net, p9_cm_event_handler, client,
637 				     RDMA_PS_TCP, IB_QPT_RC);
638 	if (IS_ERR(rdma->cm_id))
639 		goto error;
640 
641 	/* Associate the client with the transport */
642 	client->trans = rdma;
643 
644 	/* Bind to a privileged port if we need to */
645 	if (opts.privport) {
646 		err = p9_rdma_bind_privport(rdma);
647 		if (err < 0) {
648 			pr_err("%s (%d): problem binding to privport: %d\n",
649 			       __func__, task_pid_nr(current), -err);
650 			goto error;
651 		}
652 	}
653 
654 	/* Resolve the server's address */
655 	rdma->addr.sin_family = AF_INET;
656 	rdma->addr.sin_addr.s_addr = in_aton(addr);
657 	rdma->addr.sin_port = htons(opts.port);
658 	err = rdma_resolve_addr(rdma->cm_id, NULL,
659 				(struct sockaddr *)&rdma->addr,
660 				rdma->timeout);
661 	if (err)
662 		goto error;
663 	err = wait_for_completion_interruptible(&rdma->cm_done);
664 	if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
665 		goto error;
666 
667 	/* Resolve the route to the server */
668 	err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
669 	if (err)
670 		goto error;
671 	err = wait_for_completion_interruptible(&rdma->cm_done);
672 	if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
673 		goto error;
674 
675 	/* Create the Completion Queue */
676 	rdma->cq = ib_alloc_cq(rdma->cm_id->device, client,
677 			opts.sq_depth + opts.rq_depth + 1,
678 			0, IB_POLL_SOFTIRQ);
679 	if (IS_ERR(rdma->cq))
680 		goto error;
681 
682 	/* Create the Protection Domain */
683 	rdma->pd = ib_alloc_pd(rdma->cm_id->device);
684 	if (IS_ERR(rdma->pd))
685 		goto error;
686 
687 	/* Create the Queue Pair */
688 	memset(&qp_attr, 0, sizeof qp_attr);
689 	qp_attr.event_handler = qp_event_handler;
690 	qp_attr.qp_context = client;
691 	qp_attr.cap.max_send_wr = opts.sq_depth;
692 	qp_attr.cap.max_recv_wr = opts.rq_depth;
693 	qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
694 	qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
695 	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
696 	qp_attr.qp_type = IB_QPT_RC;
697 	qp_attr.send_cq = rdma->cq;
698 	qp_attr.recv_cq = rdma->cq;
699 	err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
700 	if (err)
701 		goto error;
702 	rdma->qp = rdma->cm_id->qp;
703 
704 	/* Request a connection */
705 	memset(&conn_param, 0, sizeof(conn_param));
706 	conn_param.private_data = NULL;
707 	conn_param.private_data_len = 0;
708 	conn_param.responder_resources = P9_RDMA_IRD;
709 	conn_param.initiator_depth = P9_RDMA_ORD;
710 	err = rdma_connect(rdma->cm_id, &conn_param);
711 	if (err)
712 		goto error;
713 	err = wait_for_completion_interruptible(&rdma->cm_done);
714 	if (err || (rdma->state != P9_RDMA_CONNECTED))
715 		goto error;
716 
717 	client->status = Connected;
718 
719 	return 0;
720 
721 error:
722 	rdma_destroy_trans(rdma);
723 	return -ENOTCONN;
724 }
725 
726 static struct p9_trans_module p9_rdma_trans = {
727 	.name = "rdma",
728 	.maxsize = P9_RDMA_MAXSIZE,
729 	.def = 0,
730 	.owner = THIS_MODULE,
731 	.create = rdma_create_trans,
732 	.close = rdma_close,
733 	.request = rdma_request,
734 	.cancel = rdma_cancel,
735 	.cancelled = rdma_cancelled,
736 };
737 
738 /**
739  * p9_trans_rdma_init - Register the 9P RDMA transport driver
740  */
741 static int __init p9_trans_rdma_init(void)
742 {
743 	v9fs_register_trans(&p9_rdma_trans);
744 	return 0;
745 }
746 
747 static void __exit p9_trans_rdma_exit(void)
748 {
749 	v9fs_unregister_trans(&p9_rdma_trans);
750 }
751 
752 module_init(p9_trans_rdma_init);
753 module_exit(p9_trans_rdma_exit);
754 
755 MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
756 MODULE_DESCRIPTION("RDMA Transport for 9P");
757 MODULE_LICENSE("Dual BSD/GPL");
758