summaryrefslogtreecommitdiffstats
path: root/net/9p/trans_rdma.c
blob: 3533d2a53ab649e634487edfa63dcc965a3ffa07 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
/*
 * linux/fs/9p/trans_rdma.c
 *
 * RDMA transport layer based on the trans_fd.c implementation.
 *
 *  Copyright (C) 2008 by Tom Tucker <tom@opengridcomputing.com>
 *  Copyright (C) 2006 by Russ Cox <rsc@swtch.com>
 *  Copyright (C) 2004-2005 by Latchesar Ionkov <lucho@ionkov.net>
 *  Copyright (C) 2004-2008 by Eric Van Hensbergen <ericvh@gmail.com>
 *  Copyright (C) 1997-2002 by Ron Minnich <rminnich@sarnoff.com>
 *
 *  This program is free software; you can redistribute it and/or modify
 *  it under the terms of the GNU General Public License version 2
 *  as published by the Free Software Foundation.
 *
 *  This program is distributed in the hope that it will be useful,
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
 *  GNU General Public License for more details.
 *
 *  You should have received a copy of the GNU General Public License
 *  along with this program; if not, write to:
 *  Free Software Foundation
 *  51 Franklin Street, Fifth Floor
 *  Boston, MA  02111-1301  USA
 *
 */

#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt

#include <linux/in.h>
#include <linux/module.h>
#include <linux/net.h>
#include <linux/ipv6.h>
#include <linux/kthread.h>
#include <linux/errno.h>
#include <linux/kernel.h>
#include <linux/un.h>
#include <linux/uaccess.h>
#include <linux/inet.h>
#include <linux/idr.h>
#include <linux/file.h>
#include <linux/parser.h>
#include <linux/semaphore.h>
#include <linux/slab.h>
#include <net/9p/9p.h>
#include <net/9p/client.h>
#include <net/9p/transport.h>
#include <rdma/ib_verbs.h>
#include <rdma/rdma_cm.h>

#define P9_PORT			5640
#define P9_RDMA_SQ_DEPTH	32
#define P9_RDMA_RQ_DEPTH	32
#define P9_RDMA_SEND_SGE	4
#define P9_RDMA_RECV_SGE	4
#define P9_RDMA_IRD		0
#define P9_RDMA_ORD		0
#define P9_RDMA_TIMEOUT		30000		/* 30 seconds */
#define P9_RDMA_MAXSIZE		(1024*1024)	/* 1MB */

/**
 * struct p9_trans_rdma - RDMA transport instance
 *
 * @state: tracks the transport state machine for connection setup and tear down
 * @cm_id: The RDMA CM ID
 * @pd: Protection Domain pointer
 * @qp: Queue Pair pointer
 * @cq: Completion Queue pointer
 * @dm_mr: DMA Memory Region pointer
 * @lkey: The local access only memory region key
 * @timeout: Number of uSecs to wait for connection management events
 * @sq_depth: The depth of the Send Queue
 * @sq_sem: Semaphore for the SQ
 * @rq_depth: The depth of the Receive Queue.
 * @rq_sem: Semaphore for the RQ
 * @excess_rc : Amount of posted Receive Contexts without a pending request.
 *		See rdma_request()
 * @addr: The remote peer's address
 * @req_lock: Protects the active request list
 * @cm_done: Completion event for connection management tracking
 */
struct p9_trans_rdma {
	enum {
		P9_RDMA_INIT,
		P9_RDMA_ADDR_RESOLVED,
		P9_RDMA_ROUTE_RESOLVED,
		P9_RDMA_CONNECTED,
		P9_RDMA_FLUSHING,
		P9_RDMA_CLOSING,
		P9_RDMA_CLOSED,
	} state;
	struct rdma_cm_id *cm_id;
	struct ib_pd *pd;
	struct ib_qp *qp;
	struct ib_cq *cq;
	struct ib_mr *dma_mr;
	u32 lkey;
	long timeout;
	int sq_depth;
	struct semaphore sq_sem;
	int rq_depth;
	struct semaphore rq_sem;
	atomic_t excess_rc;
	struct sockaddr_in addr;
	spinlock_t req_lock;

	struct completion cm_done;
};

/**
 * p9_rdma_context - Keeps track of in-process WR
 *
 * @wc_op: The original WR op for when the CQE completes in error.
 * @busa: Bus address to unmap when the WR completes
 * @req: Keeps track of requests (send)
 * @rc: Keepts track of replies (receive)
 */
struct p9_rdma_req;
struct p9_rdma_context {
	enum ib_wc_opcode wc_op;
	dma_addr_t busa;
	union {
		struct p9_req_t *req;
		struct p9_fcall *rc;
	};
};

/**
 * p9_rdma_opts - Collection of mount options
 * @port: port of connection
 * @sq_depth: The requested depth of the SQ. This really doesn't need
 * to be any deeper than the number of threads used in the client
 * @rq_depth: The depth of the RQ. Should be greater than or equal to SQ depth
 * @timeout: Time to wait in msecs for CM events
 */
struct p9_rdma_opts {
	short port;
	int sq_depth;
	int rq_depth;
	long timeout;
	int privport;
};

/*
 * Option Parsing (code inspired by NFS code)
 */
enum {
	/* Options that take integer arguments */
	Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout,
	/* Options that take no argument */
	Opt_privport,
	Opt_err,
};

static match_table_t tokens = {
	{Opt_port, "port=%u"},
	{Opt_sq_depth, "sq=%u"},
	{Opt_rq_depth, "rq=%u"},
	{Opt_timeout, "timeout=%u"},
	{Opt_privport, "privport"},
	{Opt_err, NULL},
};

/**
 * parse_opts - parse mount options into rdma options structure
 * @params: options string passed from mount
 * @opts: rdma transport-specific structure to parse options into
 *
 * Returns 0 upon success, -ERRNO upon failure
 */
static int parse_opts(char *params, struct p9_rdma_opts *opts)
{
	char *p;
	substring_t args[MAX_OPT_ARGS];
	int option;
	char *options, *tmp_options;

	opts->port = P9_PORT;
	opts->sq_depth = P9_RDMA_SQ_DEPTH;
	opts->rq_depth = P9_RDMA_RQ_DEPTH;
	opts->timeout = P9_RDMA_TIMEOUT;
	opts->privport = 0;

	if (!params)
		return 0;

	tmp_options = kstrdup(params, GFP_KERNEL);
	if (!tmp_options) {
		p9_debug(P9_DEBUG_ERROR,
			 "failed to allocate copy of option string\n");
		return -ENOMEM;
	}
	options = tmp_options;

	while ((p = strsep(&options, ",")) != NULL) {
		int token;
		int r;
		if (!*p)
			continue;
		token = match_token(p, tokens, args);
		if ((token != Opt_err) && (token != Opt_privport)) {
			r = match_int(&args[0], &option);
			if (r < 0) {
				p9_debug(P9_DEBUG_ERROR,
					 "integer field, but no integer?\n");
				continue;
			}
		}
		switch (token) {
		case Opt_port:
			opts->port = option;
			break;
		case Opt_sq_depth:
			opts->sq_depth = option;
			break;
		case Opt_rq_depth:
			opts->rq_depth = option;
			break;
		case Opt_timeout:
			opts->timeout = option;
			break;
		case Opt_privport:
			opts->privport = 1;
			break;
		default:
			continue;
		}
	}
	/* RQ must be at least as large as the SQ */
	opts->rq_depth = max(opts->rq_depth, opts->sq_depth);
	kfree(tmp_options);
	return 0;
}

static int
p9_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event)
{
	struct p9_client *c = id->context;
	struct p9_trans_rdma *rdma = c->trans;
	switch (event->event) {
	case RDMA_CM_EVENT_ADDR_RESOLVED:
		BUG_ON(rdma->state != P9_RDMA_INIT);
		rdma->state = P9_RDMA_ADDR_RESOLVED;
		break;

	case RDMA_CM_EVENT_ROUTE_RESOLVED:
		BUG_ON(rdma->state != P9_RDMA_ADDR_RESOLVED);
		rdma->state = P9_RDMA_ROUTE_RESOLVED;
		break;

	case RDMA_CM_EVENT_ESTABLISHED:
		BUG_ON(rdma->state != P9_RDMA_ROUTE_RESOLVED);
		rdma->state = P9_RDMA_CONNECTED;
		break;

	case RDMA_CM_EVENT_DISCONNECTED:
		if (rdma)
			rdma->state = P9_RDMA_CLOSED;
		if (c)
			c->status = Disconnected;
		break;

	case RDMA_CM_EVENT_TIMEWAIT_EXIT:
		break;

	case RDMA_CM_EVENT_ADDR_CHANGE:
	case RDMA_CM_EVENT_ROUTE_ERROR:
	case RDMA_CM_EVENT_DEVICE_REMOVAL:
	case RDMA_CM_EVENT_MULTICAST_JOIN:
	case RDMA_CM_EVENT_MULTICAST_ERROR:
	case RDMA_CM_EVENT_REJECTED:
	case RDMA_CM_EVENT_CONNECT_REQUEST:
	case RDMA_CM_EVENT_CONNECT_RESPONSE:
	case RDMA_CM_EVENT_CONNECT_ERROR:
	case RDMA_CM_EVENT_ADDR_ERROR:
	case RDMA_CM_EVENT_UNREACHABLE:
		c->status = Disconnected;
		rdma_disconnect(rdma->cm_id);
		break;
	default:
		BUG();
	}
	complete(&rdma->cm_done);
	return 0;
}

static void
handle_recv(struct p9_client *client, struct p9_trans_rdma *rdma,
	    struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
{
	struct p9_req_t *req;
	int err = 0;
	int16_t tag;

	req = NULL;
	ib_dma_unmap_single(rdma->cm_id->device, c->busa, client->msize,
							 DMA_FROM_DEVICE);

	if (status != IB_WC_SUCCESS)
		goto err_out;

	err = p9_parse_header(c->rc, NULL, NULL, &tag, 1);
	if (err)
		goto err_out;

	req = p9_tag_lookup(client, tag);
	if (!req)
		goto err_out;

	/* Check that we have not yet received a reply for this request.
	 */
	if (unlikely(req->rc)) {
		pr_err("Duplicate reply for request %d", tag);
		goto err_out;
	}

	req->rc = c->rc;
	p9_client_cb(client, req, REQ_STATUS_RCVD);

	return;

 err_out:
	p9_debug(P9_DEBUG_ERROR, "req %p err %d status %d\n", req, err, status);
	rdma->state = P9_RDMA_FLUSHING;
	client->status = Disconnected;
}

static void
handle_send(struct p9_client *client, struct p9_trans_rdma *rdma,
	    struct p9_rdma_context *c, enum ib_wc_status status, u32 byte_len)
{
	ib_dma_unmap_single(rdma->cm_id->device,
			    c->busa, c->req->tc->size,
			    DMA_TO_DEVICE);
}

static void qp_event_handler(struct ib_event *event, void *context)
{
	p9_debug(P9_DEBUG_ERROR, "QP event %d context %p\n",
		 event->event, context);
}

static void cq_comp_handler(struct ib_cq *cq, void *cq_context)
{
	struct p9_client *client = cq_context;
	struct p9_trans_rdma *rdma = client->trans;
	int ret;
	struct ib_wc wc;

	ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);
	while ((ret = ib_poll_cq(cq, 1, &wc)) > 0) {
		struct p9_rdma_context *c = (void *) (unsigned long) wc.wr_id;

		switch (c->wc_op) {
		case IB_WC_RECV:
			handle_recv(client, rdma, c, wc.status, wc.byte_len);
			up(&rdma->rq_sem);
			break;

		case IB_WC_SEND:
			handle_send(client, rdma, c, wc.status, wc.byte_len);
			up(&rdma->sq_sem);
			break;

		default:
			pr_err("unexpected completion type, c->wc_op=%d, wc.opcode=%d, status=%d\n",
			       c->wc_op, wc.opcode, wc.status);
			break;
		}
		kfree(c);
	}
}

static void cq_event_handler(struct ib_event *e, void *v)
{
	p9_debug(P9_DEBUG_ERROR, "CQ event %d context %p\n", e->event, v);
}

static void rdma_destroy_trans(struct p9_trans_rdma *rdma)
{
	if (!rdma)
		return;

	if (rdma->dma_mr && !IS_ERR(rdma->dma_mr))
		ib_dereg_mr(rdma->dma_mr);

	if (rdma->qp && !IS_ERR(rdma->qp))
		ib_destroy_qp(rdma->qp);

	if (rdma->pd && !IS_ERR(rdma->pd))
		ib_dealloc_pd(rdma->pd);

	if (rdma->cq && !IS_ERR(rdma->cq))
		ib_destroy_cq(rdma->cq);

	if (rdma->cm_id && !IS_ERR(rdma->cm_id))
		rdma_destroy_id(rdma->cm_id);

	kfree(rdma);
}

static int
post_recv(struct p9_client *client, struct p9_rdma_context *c)
{
	struct p9_trans_rdma *rdma = client->trans;
	struct ib_recv_wr wr, *bad_wr;
	struct ib_sge sge;

	c->busa = ib_dma_map_single(rdma->cm_id->device,
				    c->rc->sdata, client->msize,
				    DMA_FROM_DEVICE);
	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa))
		goto error;

	sge.addr = c->busa;
	sge.length = client->msize;
	sge.lkey = rdma->lkey;

	wr.next = NULL;
	c->wc_op = IB_WC_RECV;
	wr.wr_id = (unsigned long) c;
	wr.sg_list = &sge;
	wr.num_sge = 1;
	return ib_post_recv(rdma->qp, &wr, &bad_wr);

 error:
	p9_debug(P9_DEBUG_ERROR, "EIO\n");
	return -EIO;
}

static int rdma_request(struct p9_client *client, struct p9_req_t *req)
{
	struct p9_trans_rdma *rdma = client->trans;
	struct ib_send_wr wr, *bad_wr;
	struct ib_sge sge;
	int err = 0;
	unsigned long flags;
	struct p9_rdma_context *c = NULL;
	struct p9_rdma_context *rpl_context = NULL;

	/* When an error occurs between posting the recv and the send,
	 * there will be a receive context posted without a pending request.
	 * Since there is no way to "un-post" it, we remember it and skip
	 * post_recv() for the next request.
	 * So here,
	 * see if we are this `next request' and need to absorb an excess rc.
	 * If yes, then drop and free our own, and do not recv_post().
	 **/
	if (unlikely(atomic_read(&rdma->excess_rc) > 0)) {
		if ((atomic_sub_return(1, &rdma->excess_rc) >= 0)) {
			/* Got one ! */
			kfree(req->rc);
			req->rc = NULL;
			goto dont_need_post_recv;
		} else {
			/* We raced and lost. */
			atomic_inc(&rdma->excess_rc);
		}
	}

	/* Allocate an fcall for the reply */
	rpl_context = kmalloc(sizeof *rpl_context, GFP_NOFS);
	if (!rpl_context) {
		err = -ENOMEM;
		goto recv_error;
	}
	rpl_context->rc = req->rc;

	/*
	 * Post a receive buffer for this request. We need to ensure
	 * there is a reply buffer available for every outstanding
	 * request. A flushed request can result in no reply for an
	 * outstanding request, so we must keep a count to avoid
	 * overflowing the RQ.
	 */
	if (down_interruptible(&rdma->rq_sem)) {
		err = -EINTR;
		goto recv_error;
	}

	err = post_recv(client, rpl_context);
	if (err) {
		p9_debug(P9_DEBUG_FCALL, "POST RECV failed\n");
		goto recv_error;
	}
	/* remove posted receive buffer from request structure */
	req->rc = NULL;

dont_need_post_recv:
	/* Post the request */
	c = kmalloc(sizeof *c, GFP_NOFS);
	if (!c) {
		err = -ENOMEM;
		goto send_error;
	}
	c->req = req;

	c->busa = ib_dma_map_single(rdma->cm_id->device,
				    c->req->tc->sdata, c->req->tc->size,
				    DMA_TO_DEVICE);
	if (ib_dma_mapping_error(rdma->cm_id->device, c->busa)) {
		err = -EIO;
		goto send_error;
	}

	sge.addr = c->busa;
	sge.length = c->req->tc->size;
	sge.lkey = rdma->lkey;

	wr.next = NULL;
	c->wc_op = IB_WC_SEND;
	wr.wr_id = (unsigned long) c;
	wr.opcode = IB_WR_SEND;
	wr.send_flags = IB_SEND_SIGNALED;
	wr.sg_list = &sge;
	wr.num_sge = 1;

	if (down_interruptible(&rdma->sq_sem)) {
		err = -EINTR;
		goto send_error;
	}

	/* Mark request as `sent' *before* we actually send it,
	 * because doing if after could erase the REQ_STATUS_RCVD
	 * status in case of a very fast reply.
	 */
	req->status = REQ_STATUS_SENT;
	err = ib_post_send(rdma->qp, &wr, &bad_wr);
	if (err)
		goto send_error;

	/* Success */
	return 0;

 /* Handle errors that happened during or while preparing the send: */
 send_error:
	req->status = REQ_STATUS_ERROR;
	kfree(c);
	p9_debug(P9_DEBUG_ERROR, "Error %d in rdma_request()\n", err);

	/* Ach.
	 *  We did recv_post(), but not send. We have one recv_post in excess.
	 */
	atomic_inc(&rdma->excess_rc);
	return err;

 /* Handle errors that happened during or while preparing post_recv(): */
 recv_error:
	kfree(rpl_context);
	spin_lock_irqsave(&rdma->req_lock, flags);
	if (rdma->state < P9_RDMA_CLOSING) {
		rdma->state = P9_RDMA_CLOSING;
		spin_unlock_irqrestore(&rdma->req_lock, flags);
		rdma_disconnect(rdma->cm_id);
	} else
		spin_unlock_irqrestore(&rdma->req_lock, flags);
	return err;
}

static void rdma_close(struct p9_client *client)
{
	struct p9_trans_rdma *rdma;

	if (!client)
		return;

	rdma = client->trans;
	if (!rdma)
		return;

	client->status = Disconnected;
	rdma_disconnect(rdma->cm_id);
	rdma_destroy_trans(rdma);
}

/**
 * alloc_rdma - Allocate and initialize the rdma transport structure
 * @opts: Mount options structure
 */
static struct p9_trans_rdma *alloc_rdma(struct p9_rdma_opts *opts)
{
	struct p9_trans_rdma *rdma;

	rdma = kzalloc(sizeof(struct p9_trans_rdma), GFP_KERNEL);
	if (!rdma)
		return NULL;

	rdma->sq_depth = opts->sq_depth;
	rdma->rq_depth = opts->rq_depth;
	rdma->timeout = opts->timeout;
	spin_lock_init(&rdma->req_lock);
	init_completion(&rdma->cm_done);
	sema_init(&rdma->sq_sem, rdma->sq_depth);
	sema_init(&rdma->rq_sem, rdma->rq_depth);
	atomic_set(&rdma->excess_rc, 0);

	return rdma;
}

static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
{
	/* Nothing to do here.
	 * We will take care of it (if we have to) in rdma_cancelled()
	 */
	return 1;
}

/* A request has been fully flushed without a reply.
 * That means we have posted one buffer in excess.
 */
static int rdma_cancelled(struct p9_client *client, struct p9_req_t *req)
{
	struct p9_trans_rdma *rdma = client->trans;
	atomic_inc(&rdma->excess_rc);
	return 0;
}

static int p9_rdma_bind_privport(struct p9_trans_rdma *rdma)
{
	struct sockaddr_in cl = {
		.sin_family = AF_INET,
		.sin_addr.s_addr = htonl(INADDR_ANY),
	};
	int port, err = -EINVAL;

	for (port = P9_DEF_MAX_RESVPORT; port >= P9_DEF_MIN_RESVPORT; port--) {
		cl.sin_port = htons((ushort)port);
		err = rdma_bind_addr(rdma->cm_id, (struct sockaddr *)&cl);
		if (err != -EADDRINUSE)
			break;
	}
	return err;
}

/**
 * trans_create_rdma - Transport method for creating atransport instance
 * @client: client instance
 * @addr: IP address string
 * @args: Mount options string
 */
static int
rdma_create_trans(struct p9_client *client, const char *addr, char *args)
{
	int err;
	struct p9_rdma_opts opts;
	struct p9_trans_rdma *rdma;
	struct rdma_conn_param conn_param;
	struct ib_qp_init_attr qp_attr;
	struct ib_device_attr devattr;

	/* Parse the transport specific mount options */
	err = parse_opts(args, &opts);
	if (err < 0)
		return err;

	/* Create and initialize the RDMA transport structure */
	rdma = alloc_rdma(&opts);
	if (!rdma)
		return -ENOMEM;

	/* Create the RDMA CM ID */
	rdma->cm_id = rdma_create_id(p9_cm_event_handler, client, RDMA_PS_TCP,
				     IB_QPT_RC);
	if (IS_ERR(rdma->cm_id))
		goto error;

	/* Associate the client with the transport */
	client->trans = rdma;

	/* Bind to a privileged port if we need to */
	if (opts.privport) {
		err = p9_rdma_bind_privport(rdma);
		if (err < 0) {
			pr_err("%s (%d): problem binding to privport: %d\n",
			       __func__, task_pid_nr(current), -err);
			goto error;
		}
	}

	/* Resolve the server's address */
	rdma->addr.sin_family = AF_INET;
	rdma->addr.sin_addr.s_addr = in_aton(addr);
	rdma->addr.sin_port = htons(opts.port);
	err = rdma_resolve_addr(rdma->cm_id, NULL,
				(struct sockaddr *)&rdma->addr,
				rdma->timeout);
	if (err)
		goto error;
	err = wait_for_completion_interruptible(&rdma->cm_done);
	if (err || (rdma->state != P9_RDMA_ADDR_RESOLVED))
		goto error;

	/* Resolve the route to the server */
	err = rdma_resolve_route(rdma->cm_id, rdma->timeout);
	if (err)
		goto error;
	err = wait_for_completion_interruptible(&rdma->cm_done);
	if (err || (rdma->state != P9_RDMA_ROUTE_RESOLVED))
		goto error;

	/* Query the device attributes */
	err = ib_query_device(rdma->cm_id->device, &devattr);
	if (err)
		goto error;

	/* Create the Completion Queue */
	rdma->cq = ib_create_cq(rdma->cm_id->device, cq_comp_handler,
				cq_event_handler, client,
				opts.sq_depth + opts.rq_depth + 1, 0);
	if (IS_ERR(rdma->cq))
		goto error;
	ib_req_notify_cq(rdma->cq, IB_CQ_NEXT_COMP);

	/* Create the Protection Domain */
	rdma->pd = ib_alloc_pd(rdma->cm_id->device);
	if (IS_ERR(rdma->pd))
		goto error;

	/* Cache the DMA lkey in the transport */
	rdma->dma_mr = NULL;
	if (devattr.device_cap_flags & IB_DEVICE_LOCAL_DMA_LKEY)
		rdma->lkey = rdma->cm_id->device->local_dma_lkey;
	else {
		rdma->dma_mr = ib_get_dma_mr(rdma->pd, IB_ACCESS_LOCAL_WRITE);
		if (IS_ERR(rdma->dma_mr))
			goto error;
		rdma->lkey = rdma->dma_mr->lkey;
	}

	/* Create the Queue Pair */
	memset(&qp_attr, 0, sizeof qp_attr);
	qp_attr.event_handler = qp_event_handler;
	qp_attr.qp_context = client;
	qp_attr.cap.max_send_wr = opts.sq_depth;
	qp_attr.cap.max_recv_wr = opts.rq_depth;
	qp_attr.cap.max_send_sge = P9_RDMA_SEND_SGE;
	qp_attr.cap.max_recv_sge = P9_RDMA_RECV_SGE;
	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
	qp_attr.qp_type = IB_QPT_RC;
	qp_attr.send_cq = rdma->cq;
	qp_attr.recv_cq = rdma->cq;
	err = rdma_create_qp(rdma->cm_id, rdma->pd, &qp_attr);
	if (err)
		goto error;
	rdma->qp = rdma->cm_id->qp;

	/* Request a connection */
	memset(&conn_param, 0, sizeof(conn_param));
	conn_param.private_data = NULL;
	conn_param.private_data_len = 0;
	conn_param.responder_resources = P9_RDMA_IRD;
	conn_param.initiator_depth = P9_RDMA_ORD;
	err = rdma_connect(rdma->cm_id, &conn_param);
	if (err)
		goto error;
	err = wait_for_completion_interruptible(&rdma->cm_done);
	if (err || (rdma->state != P9_RDMA_CONNECTED))
		goto error;

	client->status = Connected;

	return 0;

error:
	rdma_destroy_trans(rdma);
	return -ENOTCONN;
}

static struct p9_trans_module p9_rdma_trans = {
	.name = "rdma",
	.maxsize = P9_RDMA_MAXSIZE,
	.def = 0,
	.owner = THIS_MODULE,
	.create = rdma_create_trans,
	.close = rdma_close,
	.request = rdma_request,
	.cancel = rdma_cancel,
	.cancelled = rdma_cancelled,
};

/**
 * p9_trans_rdma_init - Register the 9P RDMA transport driver
 */
static int __init p9_trans_rdma_init(void)
{
	v9fs_register_trans(&p9_rdma_trans);
	return 0;
}

static void __exit p9_trans_rdma_exit(void)
{
	v9fs_unregister_trans(&p9_rdma_trans);
}

module_init(p9_trans_rdma_init);
module_exit(p9_trans_rdma_exit);

MODULE_AUTHOR("Tom Tucker <tom@opengridcomputing.com>");
MODULE_DESCRIPTION("RDMA Transport for 9P");
MODULE_LICENSE("Dual BSD/GPL");