net/9p/trans_rdma.c

   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");