trans_rdma.c 17.4 KB
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
/*
 * 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
 *
 */

#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		(4*4096)	/* Min SGE is 4, so we can
						 * safely advertise a maxsize
						 * of 64k */

/**
 * 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_count: Count of requests in the Receive Queue.
 * @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;
	atomic_t rq_count;
	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;
};

/*
 * Option Parsing (code inspired by NFS code)
 */
enum {
	/* Options that take integer arguments */
	Opt_port, Opt_rq_depth, Opt_sq_depth, Opt_timeout, 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_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;

	if (!params)
		return 0;

	tmp_options = kstrdup(params, GFP_KERNEL);
	if (!tmp_options) {
		P9_DPRINTK(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);
		r = match_int(&args[0], &option);
		if (r < 0) {
			P9_DPRINTK(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;
		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;

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

	return;

 err_out:
	P9_DPRINTK(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_DPRINTK(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:
			atomic_dec(&rdma->rq_count);
			handle_recv(client, rdma, c, wc.status, wc.byte_len);
			break;

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

		default:
			printk(KERN_ERR "9prdma: 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_DPRINTK(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_DPRINTK(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;

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

	/*
	 * If the request has a buffer, steal it, otherwise
	 * allocate a new one.  Typically, requests should already
	 * have receive buffers allocated and just swap them around
	 */
	if (!req->rc) {
		req->rc = kmalloc(sizeof(struct p9_fcall)+client->msize,
				  GFP_NOFS);
		if (req->rc) {
			req->rc->sdata = (char *) req->rc +
						sizeof(struct p9_fcall);
			req->rc->capacity = client->msize;
		}
	}
	rpl_context->rc = req->rc;
	if (!rpl_context->rc) {
		err = -ENOMEM;
		goto err_free2;
	}

	/*
	 * 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 (atomic_inc_return(&rdma->rq_count) <= rdma->rq_depth) {
		err = post_recv(client, rpl_context);
		if (err)
			goto err_free1;
	} else
		atomic_dec(&rdma->rq_count);

	/* remove posted receive buffer from request structure */
	req->rc = NULL;

	/* Post the request */
	c = kmalloc(sizeof *c, GFP_NOFS);
	if (!c) {
		err = -ENOMEM;
		goto err_free1;
	}
	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))
		goto 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))
		goto error;

	return ib_post_send(rdma->qp, &wr, &bad_wr);

 error:
	kfree(c);
	kfree(rpl_context->rc);
	kfree(rpl_context);
	P9_DPRINTK(P9_DEBUG_ERROR, "EIO\n");
	return -EIO;
 err_free1:
	kfree(rpl_context->rc);
 err_free2:
	kfree(rpl_context);
 err_close:
	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);
	atomic_set(&rdma->rq_count, 0);

	return rdma;
}

/* its not clear to me we can do anything after send has been posted */
static int rdma_cancel(struct p9_client *client, struct p9_req_t *req)
{
	return 1;
}

/**
 * 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;

	/* 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,
};

/**
 * 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");