xref: /openbmc/linux/net/9p/trans_rdma.c (revision 275876e2)
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 	struct ib_mr *dma_mr;
98 	u32 lkey;
99 	long timeout;
100 	int sq_depth;
101 	struct semaphore sq_sem;
102 	int rq_depth;
103 	struct semaphore rq_sem;
104 	atomic_t excess_rc;
105 	struct sockaddr_in addr;
106 	spinlock_t req_lock;
107 
108 	struct completion cm_done;
109 };
110 
111 /**
112  * p9_rdma_context - Keeps track of in-process WR
113  *
114  * @wc_op: The original WR op for when the CQE completes in error.
115  * @busa: Bus address to unmap when the WR completes
116  * @req: Keeps track of requests (send)
117  * @rc: Keepts track of replies (receive)
118  */
119 struct p9_rdma_req;
120 struct p9_rdma_context {
121 	enum ib_wc_opcode wc_op;
122 	dma_addr_t busa;
123 	union {
124 		struct p9_req_t *req;
125 		struct p9_fcall *rc;
126 	};
127 };
128 
129 /**
130  * p9_rdma_opts - Collection of mount options
131  * @port: port of connection
132  * @sq_depth: The requested depth of the SQ. This really doesn't need
133  * to be any deeper than the number of threads used in the client
134  * @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
135  * @timeout: Time to wait in msecs for CM events
136  */
137 struct p9_rdma_opts {
138 	short port;
139 	int sq_depth;
140 	int rq_depth;
141 	long timeout;
142 };
143 
144 /*
145  * Option Parsing (code inspired by NFS code)
146  */
147 enum {
148 	/* Options that take integer arguments */
149 	Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout, Opt_err,
150 };
151 
152 static match_table_t tokens = {
153 	{Opt_port, "port=%u"},
154 	{Opt_sq_depth, "sq=%u"},
155 	{Opt_rq_depth, "rq=%u"},
156 	{Opt_timeout, "timeout=%u"},
157 	{Opt_err, NULL},
158 };
159 
160 /**
161  * parse_opts - parse mount options into rdma options structure
162  * @params: options string passed from mount
163  * @opts: rdma transport-specific structure to parse options into
164  *
165  * Returns 0 upon success, -ERRNO upon failure
166  */
167 static int parse_opts(char *params, struct p9_rdma_opts *opts)
168 {
169 	char *p;
170 	substring_t args[MAX_OPT_ARGS];
171 	int option;
172 	char *options, *tmp_options;
173 
174 	opts->port = P9_PORT;
175 	opts->sq_depth = P9_RDMA_SQ_DEPTH;
176 	opts->rq_depth = P9_RDMA_RQ_DEPTH;
177 	opts->timeout = P9_RDMA_TIMEOUT;
178 
179 	if (!params)
180 		return 0;
181 
182 	tmp_options = kstrdup(params, GFP_KERNEL);
183 	if (!tmp_options) {
184 		p9_debug(P9_DEBUG_ERROR,
185 			 "failed to allocate copy of option string\n");
186 		return -ENOMEM;
187 	}
188 	options = tmp_options;
189 
190 	while ((p = strsep(&options, ",")) != NULL) {
191 		int token;
192 		int r;
193 		if (!*p)
194 			continue;
195 		token = match_token(p, tokens, args);
196 		if (token == Opt_err)
197 			continue;
198 		r = match_int(&args[0], &option);
199 		if (r < 0) {
200 			p9_debug(P9_DEBUG_ERROR,
201 				 "integer field, but no integer?\n");
202 			continue;
203 		}
204 		switch (token) {
205 		case Opt_port:
206 			opts->port = option;
207 			break;
208 		case Opt_sq_depth:
209 			opts->sq_depth = option;
210 			break;
211 		case Opt_rq_depth:
212 			opts->rq_depth = option;
213 			break;
214 		case Opt_timeout:
215 			opts->timeout = option;
216 			break;
217 		default:
218 			continue;
219 		}
220 	}
221 	/* RQ must be at least as large as the SQ */
222 	opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
223 	kfree(tmp_options);
224 	return 0;
225 }
226 
227 static int
228 p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
229 {
230 	struct p9_client *c = id->context;
231 	struct p9_trans_rdma *rdma = c->trans;
232 	switch (event->event) {
233 	case RDMA_CM_EVENT_ADDR_RESOLVED:
234 		BUG_ON(rdma->state != P9_RDMA_INIT);
235 		rdma->state = P9_RDMA_ADDR_RESOLVED;
236 		break;
237 
238 	case RDMA_CM_EVENT_ROUTE_RESOLVED:
239 		BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
240 		rdma->state = P9_RDMA_ROUTE_RESOLVED;
241 		break;
242 
243 	case RDMA_CM_EVENT_ESTABLISHED:
244 		BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
245 		rdma->state = P9_RDMA_CONNECTED;
246 		break;
247 
248 	case RDMA_CM_EVENT_DISCONNECTED:
249 		if (rdma)
250 			rdma->state = P9_RDMA_CLOSED;
251 		if (c)
252 			c->status = Disconnected;
253 		break;
254 
255 	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
256 		break;
257 
258 	case RDMA_CM_EVENT_ADDR_CHANGE:
259 	case RDMA_CM_EVENT_ROUTE_ERROR:
260 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
261 	case RDMA_CM_EVENT_MULTICAST_JOIN:
262 	case RDMA_CM_EVENT_MULTICAST_ERROR:
263 	case RDMA_CM_EVENT_REJECTED:
264 	case RDMA_CM_EVENT_CONNECT_REQUEST:
265 	case RDMA_CM_EVENT_CONNECT_RESPONSE:
266 	case RDMA_CM_EVENT_CONNECT_ERROR:
267 	case RDMA_CM_EVENT_ADDR_ERROR:
268 	case RDMA_CM_EVENT_UNREACHABLE:
269 		c->status = Disconnected;
270 		rdma_disconnect(rdma->cm_id);
271 		break;
272 	default:
273 		BUG();
274 	}
275 	complete(&rdma->cm_done);
276 	return 0;
277 }
278 
279 static void
280 handle_recv(struct p9_client *client, struct p9_trans_rdma *rdma,
281 	    struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
282 {
283 	struct p9_req_t *req;
284 	int err = 0;
285 	int16_t tag;
286 
287 	req = NULL;
288 	ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
289 							 DMA_FROM_DEVICE);
290 
291 	if (status != IB_WC_SUCCESS)
292 		goto err_out;
293 
294 	err = p9_parse_header(c->rc, NULL, NULL, &tag, 1);
295 	if (err)
296 		goto err_out;
297 
298 	req = p9_tag_lookup(client, tag);
299 	if (!req)
300 		goto err_out;
301 
302 	/* Check that we have not yet received a reply for this request.
303 	 */
304 	if (unlikely(req->rc)) {
305 		pr_err("Duplicate reply for request %d", tag);
306 		goto err_out;
307 	}
308 
309 	req->rc = c->rc;
310 	p9_client_cb(client, req, REQ_STATUS_RCVD);
311 
312 	return;
313 
314  err_out:
315 	p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n", req, err, status);
316 	rdma->state = P9_RDMA_FLUSHING;
317 	client->status = Disconnected;
318 }
319 
320 static void
321 handle_send(struct p9_client *client, struct p9_trans_rdma *rdma,
322 	    struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
323 {
324 	ib_dma_unmap_single(rdma->cm_id->device,
325 			    c->busa, c->req->tc->size,
326 			    DMA_TO_DEVICE);
327 }
328 
329 static void qp_event_handler(struct ib_event *event, void *context)
330 {
331 	p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
332 		 event->event, context);
333 }
334 
335 static void cq_comp_handler(struct ib_cq *cq, void *cq_context)
336 {
337 	struct p9_client *client = cq_context;
338 	struct p9_trans_rdma *rdma = client->trans;
339 	int ret;
340 	struct ib_wc wc;
341 
342 	ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
343 	while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
344 		struct p9_rdma_context *c = (void *) (unsigned long) wc.wr_id;
345 
346 		switch (c->wc_op) {
347 		case IB_WC_RECV:
348 			handle_recv(client, rdma, c, wc.status, wc.byte_len);
349 			up(&rdma->rq_sem);
350 			break;
351 
352 		case IB_WC_SEND:
353 			handle_send(client, rdma, c, wc.status, wc.byte_len);
354 			up(&rdma->sq_sem);
355 			break;
356 
357 		default:
358 			pr_err("unexpected completion type, c->wc_op=%d, wc.opcode=%d, status=%d\n",
359 			       c->wc_op, wc.opcode, wc.status);
360 			break;
361 		}
362 		kfree(c);
363 	}
364 }
365 
366 static void cq_event_handler(struct ib_event *e, void *v)
367 {
368 	p9_debug(P9_DEBUG_ERROR, "CQ event %d context %p\n", e->event, v);
369 }
370 
371 static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
372 {
373 	if (!rdma)
374 		return;
375 
376 	if (rdma->dma_mr && !IS_ERR(rdma->dma_mr))
377 		ib_dereg_mr(rdma->dma_mr);
378 
379 	if (rdma->qp && !IS_ERR(rdma->qp))
380 		ib_destroy_qp(rdma->qp);
381 
382 	if (rdma->pd && !IS_ERR(rdma->pd))
383 		ib_dealloc_pd(rdma->pd);
384 
385 	if (rdma->cq && !IS_ERR(rdma->cq))
386 		ib_destroy_cq(rdma->cq);
387 
388 	if (rdma->cm_id && !IS_ERR(rdma->cm_id))
389 		rdma_destroy_id(rdma->cm_id);
390 
391 	kfree(rdma);
392 }
393 
394 static int
395 post_recv(struct p9_client *client, struct p9_rdma_context *c)
396 {
397 	struct p9_trans_rdma *rdma = client->trans;
398 	struct ib_recv_wr wr, *bad_wr;
399 	struct ib_sge sge;
400 
401 	c->busa = ib_dma_map_single(rdma->cm_id->device,
402 				    c->rc->sdata, client->msize,
403 				    DMA_FROM_DEVICE);
404 	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
405 		goto error;
406 
407 	sge.addr = c->busa;
408 	sge.length = client->msize;
409 	sge.lkey = rdma->lkey;
410 
411 	wr.next = NULL;
412 	c->wc_op = IB_WC_RECV;
413 	wr.wr_id = (unsigned long) c;
414 	wr.sg_list = &sge;
415 	wr.num_sge = 1;
416 	return ib_post_recv(rdma->qp, &wr, &bad_wr);
417 
418  error:
419 	p9_debug(P9_DEBUG_ERROR, "EIO\n");
420 	return -EIO;
421 }
422 
423 static int rdma_request(struct p9_client *client, struct p9_req_t *req)
424 {
425 	struct p9_trans_rdma *rdma = client->trans;
426 	struct ib_send_wr wr, *bad_wr;
427 	struct ib_sge sge;
428 	int err = 0;
429 	unsigned long flags;
430 	struct p9_rdma_context *c = NULL;
431 	struct p9_rdma_context *rpl_context = NULL;
432 
433 	/* When an error occurs between posting the recv and the send,
434 	 * there will be a receive context posted without a pending request.
435 	 * Since there is no way to "un-post" it, we remember it and skip
436 	 * post_recv() for the next request.
437 	 * So here,
438 	 * see if we are this `next request' and need to absorb an excess rc.
439 	 * If yes, then drop and free our own, and do not recv_post().
440 	 **/
441 	if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
442 		if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
443 			/* Got one ! */
444 			kfree(req->rc);
445 			req->rc = NULL;
446 			goto dont_need_post_recv;
447 		} else {
448 			/* We raced and lost. */
449 			atomic_inc(&rdma->excess_rc);
450 		}
451 	}
452 
453 	/* Allocate an fcall for the reply */
454 	rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
455 	if (!rpl_context) {
456 		err = -ENOMEM;
457 		goto recv_error;
458 	}
459 	rpl_context->rc = req->rc;
460 
461 	/*
462 	 * Post a receive buffer for this request. We need to ensure
463 	 * there is a reply buffer available for every outstanding
464 	 * request. A flushed request can result in no reply for an
465 	 * outstanding request, so we must keep a count to avoid
466 	 * overflowing the RQ.
467 	 */
468 	if (down_interruptible(&rdma->rq_sem)) {
469 		err = -EINTR;
470 		goto recv_error;
471 	}
472 
473 	err = post_recv(client, rpl_context);
474 	if (err) {
475 		p9_debug(P9_DEBUG_FCALL, "POST RECV failed\n");
476 		goto recv_error;
477 	}
478 	/* remove posted receive buffer from request structure */
479 	req->rc = NULL;
480 
481 dont_need_post_recv:
482 	/* Post the request */
483 	c = kmalloc(sizeof *c, GFP_NOFS);
484 	if (!c) {
485 		err = -ENOMEM;
486 		goto send_error;
487 	}
488 	c->req = req;
489 
490 	c->busa = ib_dma_map_single(rdma->cm_id->device,
491 				    c->req->tc->sdata, c->req->tc->size,
492 				    DMA_TO_DEVICE);
493 	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
494 		err = -EIO;
495 		goto send_error;
496 	}
497 
498 	sge.addr = c->busa;
499 	sge.length = c->req->tc->size;
500 	sge.lkey = rdma->lkey;
501 
502 	wr.next = NULL;
503 	c->wc_op = IB_WC_SEND;
504 	wr.wr_id = (unsigned long) c;
505 	wr.opcode = IB_WR_SEND;
506 	wr.send_flags = IB_SEND_SIGNALED;
507 	wr.sg_list = &sge;
508 	wr.num_sge = 1;
509 
510 	if (down_interruptible(&rdma->sq_sem)) {
511 		err = -EINTR;
512 		goto send_error;
513 	}
514 
515 	/* Mark request as `sent' *before* we actually send it,
516 	 * because doing if after could erase the REQ_STATUS_RCVD
517 	 * status in case of a very fast reply.
518 	 */
519 	req->status = REQ_STATUS_SENT;
520 	err = ib_post_send(rdma->qp, &wr, &bad_wr);
521 	if (err)
522 		goto send_error;
523 
524 	/* Success */
525 	return 0;
526 
527  /* Handle errors that happened during or while preparing the send: */
528  send_error:
529 	req->status = REQ_STATUS_ERROR;
530 	kfree(c);
531 	p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);
532 
533 	/* Ach.
534 	 *  We did recv_post(), but not send. We have one recv_post in excess.
535 	 */
536 	atomic_inc(&rdma->excess_rc);
537 	return err;
538 
539  /* Handle errors that happened during or while preparing post_recv(): */
540  recv_error:
541 	kfree(rpl_context);
542 	spin_lock_irqsave(&rdma->req_lock, flags);
543 	if (rdma->state < P9_RDMA_CLOSING) {
544 		rdma->state = P9_RDMA_CLOSING;
545 		spin_unlock_irqrestore(&rdma->req_lock, flags);
546 		rdma_disconnect(rdma->cm_id);
547 	} else
548 		spin_unlock_irqrestore(&rdma->req_lock, flags);
549 	return err;
550 }
551 
552 static void rdma_close(struct p9_client *client)
553 {
554 	struct p9_trans_rdma *rdma;
555 
556 	if (!client)
557 		return;
558 
559 	rdma = client->trans;
560 	if (!rdma)
561 		return;
562 
563 	client->status = Disconnected;
564 	rdma_disconnect(rdma->cm_id);
565 	rdma_destroy_trans(rdma);
566 }
567 
568 /**
569  * alloc_rdma - Allocate and initialize the rdma transport structure
570  * @opts: Mount options structure
571  */
572 static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
573 {
574 	struct p9_trans_rdma *rdma;
575 
576 	rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
577 	if (!rdma)
578 		return NULL;
579 
580 	rdma->sq_depth = opts->sq_depth;
581 	rdma->rq_depth = opts->rq_depth;
582 	rdma->timeout = opts->timeout;
583 	spin_lock_init(&rdma->req_lock);
584 	init_completion(&rdma->cm_done);
585 	sema_init(&rdma->sq_sem, rdma->sq_depth);
586 	sema_init(&rdma->rq_sem, rdma->rq_depth);
587 	atomic_set(&rdma->excess_rc, 0);
588 
589 	return rdma;
590 }
591 
592 static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
593 {
594 	/* Nothing to do here.
595 	 * We will take care of it (if we have to) in rdma_cancelled()
596 	 */
597 	return 1;
598 }
599 
600 /* A request has been fully flushed without a reply.
601  * That means we have posted one buffer in excess.
602  */
603 static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
604 {
605 	struct p9_trans_rdma *rdma = client->trans;
606 	atomic_inc(&rdma->excess_rc);
607 	return 0;
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 	struct ib_device_attr devattr;
625 
626 	/* Parse the transport specific mount options */
627 	err = parse_opts(args, &opts);
628 	if (err < 0)
629 		return err;
630 
631 	/* Create and initialize the RDMA transport structure */
632 	rdma = alloc_rdma(&opts);
633 	if (!rdma)
634 		return -ENOMEM;
635 
636 	/* Create the RDMA CM ID */
637 	rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP,
638 				     IB_QPT_RC);
639 	if (IS_ERR(rdma->cm_id))
640 		goto error;
641 
642 	/* Associate the client with the transport */
643 	client->trans = rdma;
644 
645 	/* Resolve the server's address */
646 	rdma->addr.sin_family = AF_INET;
647 	rdma->addr.sin_addr.s_addr = in_aton(addr);
648 	rdma->addr.sin_port = htons(opts.port);
649 	err = rdma_resolve_addr(rdma->cm_id, NULL,
650 				(struct sockaddr *)&rdma->addr,
651 				rdma->timeout);
652 	if (err)
653 		goto error;
654 	err = wait_for_completion_interruptible(&rdma->cm_done);
655 	if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
656 		goto error;
657 
658 	/* Resolve the route to the server */
659 	err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
660 	if (err)
661 		goto error;
662 	err = wait_for_completion_interruptible(&rdma->cm_done);
663 	if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
664 		goto error;
665 
666 	/* Query the device attributes */
667 	err = ib_query_device(rdma->cm_id->device, &devattr);
668 	if (err)
669 		goto error;
670 
671 	/* Create the Completion Queue */
672 	rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
673 				cq_event_handler, client,
674 				opts.sq_depth + opts.rq_depth + 1, 0);
675 	if (IS_ERR(rdma->cq))
676 		goto error;
677 	ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
678 
679 	/* Create the Protection Domain */
680 	rdma->pd = ib_alloc_pd(rdma->cm_id->device);
681 	if (IS_ERR(rdma->pd))
682 		goto error;
683 
684 	/* Cache the DMA lkey in the transport */
685 	rdma->dma_mr = NULL;
686 	if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
687 		rdma->lkey = rdma->cm_id->device->local_dma_lkey;
688 	else {
689 		rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
690 		if (IS_ERR(rdma->dma_mr))
691 			goto error;
692 		rdma->lkey = rdma->dma_mr->lkey;
693 	}
694 
695 	/* Create the Queue Pair */
696 	memset(&qp_attr, 0, sizeof qp_attr);
697 	qp_attr.event_handler = qp_event_handler;
698 	qp_attr.qp_context = client;
699 	qp_attr.cap.max_send_wr = opts.sq_depth;
700 	qp_attr.cap.max_recv_wr = opts.rq_depth;
701 	qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
702 	qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
703 	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
704 	qp_attr.qp_type = IB_QPT_RC;
705 	qp_attr.send_cq = rdma->cq;
706 	qp_attr.recv_cq = rdma->cq;
707 	err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
708 	if (err)
709 		goto error;
710 	rdma->qp = rdma->cm_id->qp;
711 
712 	/* Request a connection */
713 	memset(&conn_param, 0, sizeof(conn_param));
714 	conn_param.private_data = NULL;
715 	conn_param.private_data_len = 0;
716 	conn_param.responder_resources = P9_RDMA_IRD;
717 	conn_param.initiator_depth = P9_RDMA_ORD;
718 	err = rdma_connect(rdma->cm_id, &conn_param);
719 	if (err)
720 		goto error;
721 	err = wait_for_completion_interruptible(&rdma->cm_done);
722 	if (err || (rdma->state != P9_RDMA_CONNECTED))
723 		goto error;
724 
725 	client->status = Connected;
726 
727 	return 0;
728 
729 error:
730 	rdma_destroy_trans(rdma);
731 	return -ENOTCONN;
732 }
733 
734 static struct p9_trans_module p9_rdma_trans = {
735 	.name = "rdma",
736 	.maxsize = P9_RDMA_MAXSIZE,
737 	.def = 0,
738 	.owner = THIS_MODULE,
739 	.create = rdma_create_trans,
740 	.close = rdma_close,
741 	.request = rdma_request,
742 	.cancel = rdma_cancel,
743 	.cancelled = rdma_cancelled,
744 };
745 
746 /**
747  * p9_trans_rdma_init - Register the 9P RDMA transport driver
748  */
749 static int __init p9_trans_rdma_init(void)
750 {
751 	v9fs_register_trans(&p9_rdma_trans);
752 	return 0;
753 }
754 
755 static void __exit p9_trans_rdma_exit(void)
756 {
757 	v9fs_unregister_trans(&p9_rdma_trans);
758 }
759 
760 module_init(p9_trans_rdma_init);
761 module_exit(p9_trans_rdma_exit);
762 
763 MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
764 MODULE_DESCRIPTION("RDMA Transport for 9P");
765 MODULE_LICENSE("Dual BSD/GPL");
766