root/net/sunrpc/xprtrdma/svc_rdma_transport.c

DEFINITIONS

1. qp_event_handler
2. svc_rdma_create_xprt
3. svc_rdma_parse_connect_private
4. handle_connect_req
5. rdma_listen_handler
6. rdma_cma_handler
7. svc_rdma_create
8. svc_rdma_accept
9. svc_rdma_detach
10. __svc_rdma_free
11. svc_rdma_free
12. svc_rdma_has_wspace
13. svc_rdma_secure_port
14. svc_rdma_kill_temp_xprt

   1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause
   2 /*
   3  * Copyright (c) 2015-2018 Oracle. All rights reserved.
   4  * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
   5  * Copyright (c) 2005-2007 Network Appliance, Inc. All rights reserved.
   6  *
   7  * This software is available to you under a choice of one of two
   8  * licenses.  You may choose to be licensed under the terms of the GNU
   9  * General Public License (GPL) Version 2, available from the file
  10  * COPYING in the main directory of this source tree, or the BSD-type
  11  * license below:
  12  *
  13  * Redistribution and use in source and binary forms, with or without
  14  * modification, are permitted provided that the following conditions
  15  * are met:
  16  *
  17  *      Redistributions of source code must retain the above copyright
  18  *      notice, this list of conditions and the following disclaimer.
  19  *
  20  *      Redistributions in binary form must reproduce the above
  21  *      copyright notice, this list of conditions and the following
  22  *      disclaimer in the documentation and/or other materials provided
  23  *      with the distribution.
  24  *
  25  *      Neither the name of the Network Appliance, Inc. nor the names of
  26  *      its contributors may be used to endorse or promote products
  27  *      derived from this software without specific prior written
  28  *      permission.
  29  *
  30  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  31  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  32  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  33  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  34  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  35  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  36  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  37  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  38  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  39  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  40  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  41  *
  42  * Author: Tom Tucker <tom@opengridcomputing.com>
  43  */
  44 
  45 #include <linux/interrupt.h>
  46 #include <linux/sched.h>
  47 #include <linux/slab.h>
  48 #include <linux/spinlock.h>
  49 #include <linux/workqueue.h>
  50 #include <linux/export.h>
  51 
  52 #include <rdma/ib_verbs.h>
  53 #include <rdma/rdma_cm.h>
  54 #include <rdma/rw.h>
  55 
  56 #include <linux/sunrpc/addr.h>
  57 #include <linux/sunrpc/debug.h>
  58 #include <linux/sunrpc/rpc_rdma.h>
  59 #include <linux/sunrpc/svc_xprt.h>
  60 #include <linux/sunrpc/svc_rdma.h>
  61 
  62 #include "xprt_rdma.h"
  63 #include <trace/events/rpcrdma.h>
  64 
  65 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
  66 
  67 static struct svcxprt_rdma *svc_rdma_create_xprt(struct svc_serv *serv,
  68                                                  struct net *net);
  69 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
  70                                         struct net *net,
  71                                         struct sockaddr *sa, int salen,
  72                                         int flags);
  73 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
  74 static void svc_rdma_detach(struct svc_xprt *xprt);
  75 static void svc_rdma_free(struct svc_xprt *xprt);
  76 static int svc_rdma_has_wspace(struct svc_xprt *xprt);
  77 static void svc_rdma_secure_port(struct svc_rqst *);
  78 static void svc_rdma_kill_temp_xprt(struct svc_xprt *);
  79 
  80 static const struct svc_xprt_ops svc_rdma_ops = {
  81         .xpo_create = svc_rdma_create,
  82         .xpo_recvfrom = svc_rdma_recvfrom,
  83         .xpo_sendto = svc_rdma_sendto,
  84         .xpo_release_rqst = svc_rdma_release_rqst,
  85         .xpo_detach = svc_rdma_detach,
  86         .xpo_free = svc_rdma_free,
  87         .xpo_has_wspace = svc_rdma_has_wspace,
  88         .xpo_accept = svc_rdma_accept,
  89         .xpo_secure_port = svc_rdma_secure_port,
  90         .xpo_kill_temp_xprt = svc_rdma_kill_temp_xprt,
  91 };
  92 
  93 struct svc_xprt_class svc_rdma_class = {
  94         .xcl_name = "rdma",
  95         .xcl_owner = THIS_MODULE,
  96         .xcl_ops = &svc_rdma_ops,
  97         .xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
  98         .xcl_ident = XPRT_TRANSPORT_RDMA,
  99 };
 100 
 101 /* QP event handler */
 102 static void qp_event_handler(struct ib_event *event, void *context)
 103 {
 104         struct svc_xprt *xprt = context;
 105 
 106         trace_svcrdma_qp_error(event, (struct sockaddr *)&xprt->xpt_remote);
 107         switch (event->event) {
 108         /* These are considered benign events */
 109         case IB_EVENT_PATH_MIG:
 110         case IB_EVENT_COMM_EST:
 111         case IB_EVENT_SQ_DRAINED:
 112         case IB_EVENT_QP_LAST_WQE_REACHED:
 113                 break;
 114 
 115         /* These are considered fatal events */
 116         case IB_EVENT_PATH_MIG_ERR:
 117         case IB_EVENT_QP_FATAL:
 118         case IB_EVENT_QP_REQ_ERR:
 119         case IB_EVENT_QP_ACCESS_ERR:
 120         case IB_EVENT_DEVICE_FATAL:
 121         default:
 122                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
 123                 svc_xprt_enqueue(xprt);
 124                 break;
 125         }
 126 }
 127 
 128 static struct svcxprt_rdma *svc_rdma_create_xprt(struct svc_serv *serv,
 129                                                  struct net *net)
 130 {
 131         struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
 132 
 133         if (!cma_xprt) {
 134                 dprintk("svcrdma: failed to create new transport\n");
 135                 return NULL;
 136         }
 137         svc_xprt_init(net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
 138         INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
 139         INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
 140         INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
 141         INIT_LIST_HEAD(&cma_xprt->sc_send_ctxts);
 142         init_llist_head(&cma_xprt->sc_recv_ctxts);
 143         INIT_LIST_HEAD(&cma_xprt->sc_rw_ctxts);
 144         init_waitqueue_head(&cma_xprt->sc_send_wait);
 145 
 146         spin_lock_init(&cma_xprt->sc_lock);
 147         spin_lock_init(&cma_xprt->sc_rq_dto_lock);
 148         spin_lock_init(&cma_xprt->sc_send_lock);
 149         spin_lock_init(&cma_xprt->sc_rw_ctxt_lock);
 150 
 151         /*
 152          * Note that this implies that the underlying transport support
 153          * has some form of congestion control (see RFC 7530 section 3.1
 154          * paragraph 2). For now, we assume that all supported RDMA
 155          * transports are suitable here.
 156          */
 157         set_bit(XPT_CONG_CTRL, &cma_xprt->sc_xprt.xpt_flags);
 158 
 159         return cma_xprt;
 160 }
 161 
 162 static void
 163 svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
 164                                struct rdma_conn_param *param)
 165 {
 166         const struct rpcrdma_connect_private *pmsg = param->private_data;
 167 
 168         if (pmsg &&
 169             pmsg->cp_magic == rpcrdma_cmp_magic &&
 170             pmsg->cp_version == RPCRDMA_CMP_VERSION) {
 171                 newxprt->sc_snd_w_inv = pmsg->cp_flags &
 172                                         RPCRDMA_CMP_F_SND_W_INV_OK;
 173 
 174                 dprintk("svcrdma: client send_size %u, recv_size %u "
 175                         "remote inv %ssupported\n",
 176                         rpcrdma_decode_buffer_size(pmsg->cp_send_size),
 177                         rpcrdma_decode_buffer_size(pmsg->cp_recv_size),
 178                         newxprt->sc_snd_w_inv ? "" : "un");
 179         }
 180 }
 181 
 182 /*
 183  * This function handles the CONNECT_REQUEST event on a listening
 184  * endpoint. It is passed the cma_id for the _new_ connection. The context in
 185  * this cma_id is inherited from the listening cma_id and is the svc_xprt
 186  * structure for the listening endpoint.
 187  *
 188  * This function creates a new xprt for the new connection and enqueues it on
 189  * the accept queue for the listent xprt. When the listen thread is kicked, it
 190  * will call the recvfrom method on the listen xprt which will accept the new
 191  * connection.
 192  */
 193 static void handle_connect_req(struct rdma_cm_id *new_cma_id,
 194                                struct rdma_conn_param *param)
 195 {
 196         struct svcxprt_rdma *listen_xprt = new_cma_id->context;
 197         struct svcxprt_rdma *newxprt;
 198         struct sockaddr *sa;
 199 
 200         /* Create a new transport */
 201         newxprt = svc_rdma_create_xprt(listen_xprt->sc_xprt.xpt_server,
 202                                        listen_xprt->sc_xprt.xpt_net);
 203         if (!newxprt)
 204                 return;
 205         newxprt->sc_cm_id = new_cma_id;
 206         new_cma_id->context = newxprt;
 207         svc_rdma_parse_connect_private(newxprt, param);
 208 
 209         /* Save client advertised inbound read limit for use later in accept. */
 210         newxprt->sc_ord = param->initiator_depth;
 211 
 212         sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
 213         svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
 214         /* The remote port is arbitrary and not under the control of the
 215          * client ULP. Set it to a fixed value so that the DRC continues
 216          * to be effective after a reconnect.
 217          */
 218         rpc_set_port((struct sockaddr *)&newxprt->sc_xprt.xpt_remote, 0);
 219 
 220         sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
 221         svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
 222 
 223         /*
 224          * Enqueue the new transport on the accept queue of the listening
 225          * transport
 226          */
 227         spin_lock(&listen_xprt->sc_lock);
 228         list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
 229         spin_unlock(&listen_xprt->sc_lock);
 230 
 231         set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
 232         svc_xprt_enqueue(&listen_xprt->sc_xprt);
 233 }
 234 
 235 /*
 236  * Handles events generated on the listening endpoint. These events will be
 237  * either be incoming connect requests or adapter removal  events.
 238  */
 239 static int rdma_listen_handler(struct rdma_cm_id *cma_id,
 240                                struct rdma_cm_event *event)
 241 {
 242         struct sockaddr *sap = (struct sockaddr *)&cma_id->route.addr.src_addr;
 243 
 244         trace_svcrdma_cm_event(event, sap);
 245 
 246         switch (event->event) {
 247         case RDMA_CM_EVENT_CONNECT_REQUEST:
 248                 dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
 249                         "event = %s (%d)\n", cma_id, cma_id->context,
 250                         rdma_event_msg(event->event), event->event);
 251                 handle_connect_req(cma_id, &event->param.conn);
 252                 break;
 253         default:
 254                 /* NB: No device removal upcall for INADDR_ANY listeners */
 255                 dprintk("svcrdma: Unexpected event on listening endpoint %p, "
 256                         "event = %s (%d)\n", cma_id,
 257                         rdma_event_msg(event->event), event->event);
 258                 break;
 259         }
 260 
 261         return 0;
 262 }
 263 
 264 static int rdma_cma_handler(struct rdma_cm_id *cma_id,
 265                             struct rdma_cm_event *event)
 266 {
 267         struct sockaddr *sap = (struct sockaddr *)&cma_id->route.addr.dst_addr;
 268         struct svcxprt_rdma *rdma = cma_id->context;
 269         struct svc_xprt *xprt = &rdma->sc_xprt;
 270 
 271         trace_svcrdma_cm_event(event, sap);
 272 
 273         switch (event->event) {
 274         case RDMA_CM_EVENT_ESTABLISHED:
 275                 /* Accept complete */
 276                 svc_xprt_get(xprt);
 277                 dprintk("svcrdma: Connection completed on DTO xprt=%p, "
 278                         "cm_id=%p\n", xprt, cma_id);
 279                 clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
 280                 svc_xprt_enqueue(xprt);
 281                 break;
 282         case RDMA_CM_EVENT_DISCONNECTED:
 283                 dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
 284                         xprt, cma_id);
 285                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
 286                 svc_xprt_enqueue(xprt);
 287                 svc_xprt_put(xprt);
 288                 break;
 289         case RDMA_CM_EVENT_DEVICE_REMOVAL:
 290                 dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
 291                         "event = %s (%d)\n", cma_id, xprt,
 292                         rdma_event_msg(event->event), event->event);
 293                 set_bit(XPT_CLOSE, &xprt->xpt_flags);
 294                 svc_xprt_enqueue(xprt);
 295                 svc_xprt_put(xprt);
 296                 break;
 297         default:
 298                 dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
 299                         "event = %s (%d)\n", cma_id,
 300                         rdma_event_msg(event->event), event->event);
 301                 break;
 302         }
 303         return 0;
 304 }
 305 
 306 /*
 307  * Create a listening RDMA service endpoint.
 308  */
 309 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
 310                                         struct net *net,
 311                                         struct sockaddr *sa, int salen,
 312                                         int flags)
 313 {
 314         struct rdma_cm_id *listen_id;
 315         struct svcxprt_rdma *cma_xprt;
 316         int ret;
 317 
 318         dprintk("svcrdma: Creating RDMA listener\n");
 319         if ((sa->sa_family != AF_INET) && (sa->sa_family != AF_INET6)) {
 320                 dprintk("svcrdma: Address family %d is not supported.\n", sa->sa_family);
 321                 return ERR_PTR(-EAFNOSUPPORT);
 322         }
 323         cma_xprt = svc_rdma_create_xprt(serv, net);
 324         if (!cma_xprt)
 325                 return ERR_PTR(-ENOMEM);
 326         set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
 327         strcpy(cma_xprt->sc_xprt.xpt_remotebuf, "listener");
 328 
 329         listen_id = rdma_create_id(net, rdma_listen_handler, cma_xprt,
 330                                    RDMA_PS_TCP, IB_QPT_RC);
 331         if (IS_ERR(listen_id)) {
 332                 ret = PTR_ERR(listen_id);
 333                 dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
 334                 goto err0;
 335         }
 336 
 337         /* Allow both IPv4 and IPv6 sockets to bind a single port
 338          * at the same time.
 339          */
 340 #if IS_ENABLED(CONFIG_IPV6)
 341         ret = rdma_set_afonly(listen_id, 1);
 342         if (ret) {
 343                 dprintk("svcrdma: rdma_set_afonly failed = %d\n", ret);
 344                 goto err1;
 345         }
 346 #endif
 347         ret = rdma_bind_addr(listen_id, sa);
 348         if (ret) {
 349                 dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
 350                 goto err1;
 351         }
 352         cma_xprt->sc_cm_id = listen_id;
 353 
 354         ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
 355         if (ret) {
 356                 dprintk("svcrdma: rdma_listen failed = %d\n", ret);
 357                 goto err1;
 358         }
 359 
 360         /*
 361          * We need to use the address from the cm_id in case the
 362          * caller specified 0 for the port number.
 363          */
 364         sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
 365         svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
 366 
 367         return &cma_xprt->sc_xprt;
 368 
 369  err1:
 370         rdma_destroy_id(listen_id);
 371  err0:
 372         kfree(cma_xprt);
 373         return ERR_PTR(ret);
 374 }
 375 
 376 /*
 377  * This is the xpo_recvfrom function for listening endpoints. Its
 378  * purpose is to accept incoming connections. The CMA callback handler
 379  * has already created a new transport and attached it to the new CMA
 380  * ID.
 381  *
 382  * There is a queue of pending connections hung on the listening
 383  * transport. This queue contains the new svc_xprt structure. This
 384  * function takes svc_xprt structures off the accept_q and completes
 385  * the connection.
 386  */
 387 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
 388 {
 389         struct svcxprt_rdma *listen_rdma;
 390         struct svcxprt_rdma *newxprt = NULL;
 391         struct rdma_conn_param conn_param;
 392         struct rpcrdma_connect_private pmsg;
 393         struct ib_qp_init_attr qp_attr;
 394         unsigned int ctxts, rq_depth;
 395         struct ib_device *dev;
 396         int ret = 0;
 397         RPC_IFDEBUG(struct sockaddr *sap);
 398 
 399         listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
 400         clear_bit(XPT_CONN, &xprt->xpt_flags);
 401         /* Get the next entry off the accept list */
 402         spin_lock(&listen_rdma->sc_lock);
 403         if (!list_empty(&listen_rdma->sc_accept_q)) {
 404                 newxprt = list_entry(listen_rdma->sc_accept_q.next,
 405                                      struct svcxprt_rdma, sc_accept_q);
 406                 list_del_init(&newxprt->sc_accept_q);
 407         }
 408         if (!list_empty(&listen_rdma->sc_accept_q))
 409                 set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
 410         spin_unlock(&listen_rdma->sc_lock);
 411         if (!newxprt)
 412                 return NULL;
 413 
 414         dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
 415                 newxprt, newxprt->sc_cm_id);
 416 
 417         dev = newxprt->sc_cm_id->device;
 418         newxprt->sc_port_num = newxprt->sc_cm_id->port_num;
 419 
 420         /* Qualify the transport resource defaults with the
 421          * capabilities of this particular device */
 422         /* Transport header, head iovec, tail iovec */
 423         newxprt->sc_max_send_sges = 3;
 424         /* Add one SGE per page list entry */
 425         newxprt->sc_max_send_sges += (svcrdma_max_req_size / PAGE_SIZE) + 1;
 426         if (newxprt->sc_max_send_sges > dev->attrs.max_send_sge)
 427                 newxprt->sc_max_send_sges = dev->attrs.max_send_sge;
 428         newxprt->sc_max_req_size = svcrdma_max_req_size;
 429         newxprt->sc_max_requests = svcrdma_max_requests;
 430         newxprt->sc_max_bc_requests = svcrdma_max_bc_requests;
 431         rq_depth = newxprt->sc_max_requests + newxprt->sc_max_bc_requests;
 432         if (rq_depth > dev->attrs.max_qp_wr) {
 433                 pr_warn("svcrdma: reducing receive depth to %d\n",
 434                         dev->attrs.max_qp_wr);
 435                 rq_depth = dev->attrs.max_qp_wr;
 436                 newxprt->sc_max_requests = rq_depth - 2;
 437                 newxprt->sc_max_bc_requests = 2;
 438         }
 439         newxprt->sc_fc_credits = cpu_to_be32(newxprt->sc_max_requests);
 440         ctxts = rdma_rw_mr_factor(dev, newxprt->sc_port_num, RPCSVC_MAXPAGES);
 441         ctxts *= newxprt->sc_max_requests;
 442         newxprt->sc_sq_depth = rq_depth + ctxts;
 443         if (newxprt->sc_sq_depth > dev->attrs.max_qp_wr) {
 444                 pr_warn("svcrdma: reducing send depth to %d\n",
 445                         dev->attrs.max_qp_wr);
 446                 newxprt->sc_sq_depth = dev->attrs.max_qp_wr;
 447         }
 448         atomic_set(&newxprt->sc_sq_avail, newxprt->sc_sq_depth);
 449 
 450         newxprt->sc_pd = ib_alloc_pd(dev, 0);
 451         if (IS_ERR(newxprt->sc_pd)) {
 452                 dprintk("svcrdma: error creating PD for connect request\n");
 453                 goto errout;
 454         }
 455         newxprt->sc_sq_cq = ib_alloc_cq_any(dev, newxprt, newxprt->sc_sq_depth,
 456                                             IB_POLL_WORKQUEUE);
 457         if (IS_ERR(newxprt->sc_sq_cq)) {
 458                 dprintk("svcrdma: error creating SQ CQ for connect request\n");
 459                 goto errout;
 460         }
 461         newxprt->sc_rq_cq =
 462                 ib_alloc_cq_any(dev, newxprt, rq_depth, IB_POLL_WORKQUEUE);
 463         if (IS_ERR(newxprt->sc_rq_cq)) {
 464                 dprintk("svcrdma: error creating RQ CQ for connect request\n");
 465                 goto errout;
 466         }
 467 
 468         memset(&qp_attr, 0, sizeof qp_attr);
 469         qp_attr.event_handler = qp_event_handler;
 470         qp_attr.qp_context = &newxprt->sc_xprt;
 471         qp_attr.port_num = newxprt->sc_port_num;
 472         qp_attr.cap.max_rdma_ctxs = ctxts;
 473         qp_attr.cap.max_send_wr = newxprt->sc_sq_depth - ctxts;
 474         qp_attr.cap.max_recv_wr = rq_depth;
 475         qp_attr.cap.max_send_sge = newxprt->sc_max_send_sges;
 476         qp_attr.cap.max_recv_sge = 1;
 477         qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
 478         qp_attr.qp_type = IB_QPT_RC;
 479         qp_attr.send_cq = newxprt->sc_sq_cq;
 480         qp_attr.recv_cq = newxprt->sc_rq_cq;
 481         dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n",
 482                 newxprt->sc_cm_id, newxprt->sc_pd);
 483         dprintk("    cap.max_send_wr = %d, cap.max_recv_wr = %d\n",
 484                 qp_attr.cap.max_send_wr, qp_attr.cap.max_recv_wr);
 485         dprintk("    cap.max_send_sge = %d, cap.max_recv_sge = %d\n",
 486                 qp_attr.cap.max_send_sge, qp_attr.cap.max_recv_sge);
 487 
 488         ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
 489         if (ret) {
 490                 dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
 491                 goto errout;
 492         }
 493         newxprt->sc_qp = newxprt->sc_cm_id->qp;
 494 
 495         if (!(dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
 496                 newxprt->sc_snd_w_inv = false;
 497         if (!rdma_protocol_iwarp(dev, newxprt->sc_port_num) &&
 498             !rdma_ib_or_roce(dev, newxprt->sc_port_num))
 499                 goto errout;
 500 
 501         if (!svc_rdma_post_recvs(newxprt))
 502                 goto errout;
 503 
 504         /* Swap out the handler */
 505         newxprt->sc_cm_id->event_handler = rdma_cma_handler;
 506 
 507         /* Construct RDMA-CM private message */
 508         pmsg.cp_magic = rpcrdma_cmp_magic;
 509         pmsg.cp_version = RPCRDMA_CMP_VERSION;
 510         pmsg.cp_flags = 0;
 511         pmsg.cp_send_size = pmsg.cp_recv_size =
 512                 rpcrdma_encode_buffer_size(newxprt->sc_max_req_size);
 513 
 514         /* Accept Connection */
 515         set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
 516         memset(&conn_param, 0, sizeof conn_param);
 517         conn_param.responder_resources = 0;
 518         conn_param.initiator_depth = min_t(int, newxprt->sc_ord,
 519                                            dev->attrs.max_qp_init_rd_atom);
 520         if (!conn_param.initiator_depth) {
 521                 dprintk("svcrdma: invalid ORD setting\n");
 522                 ret = -EINVAL;
 523                 goto errout;
 524         }
 525         conn_param.private_data = &pmsg;
 526         conn_param.private_data_len = sizeof(pmsg);
 527         ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
 528         if (ret)
 529                 goto errout;
 530 
 531 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG)
 532         dprintk("svcrdma: new connection %p accepted:\n", newxprt);
 533         sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
 534         dprintk("    local address   : %pIS:%u\n", sap, rpc_get_port(sap));
 535         sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
 536         dprintk("    remote address  : %pIS:%u\n", sap, rpc_get_port(sap));
 537         dprintk("    max_sge         : %d\n", newxprt->sc_max_send_sges);
 538         dprintk("    sq_depth        : %d\n", newxprt->sc_sq_depth);
 539         dprintk("    rdma_rw_ctxs    : %d\n", ctxts);
 540         dprintk("    max_requests    : %d\n", newxprt->sc_max_requests);
 541         dprintk("    ord             : %d\n", conn_param.initiator_depth);
 542 #endif
 543 
 544         trace_svcrdma_xprt_accept(&newxprt->sc_xprt);
 545         return &newxprt->sc_xprt;
 546 
 547  errout:
 548         dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
 549         trace_svcrdma_xprt_fail(&newxprt->sc_xprt);
 550         /* Take a reference in case the DTO handler runs */
 551         svc_xprt_get(&newxprt->sc_xprt);
 552         if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
 553                 ib_destroy_qp(newxprt->sc_qp);
 554         rdma_destroy_id(newxprt->sc_cm_id);
 555         /* This call to put will destroy the transport */
 556         svc_xprt_put(&newxprt->sc_xprt);
 557         return NULL;
 558 }
 559 
 560 /*
 561  * When connected, an svc_xprt has at least two references:
 562  *
 563  * - A reference held by the cm_id between the ESTABLISHED and
 564  *   DISCONNECTED events. If the remote peer disconnected first, this
 565  *   reference could be gone.
 566  *
 567  * - A reference held by the svc_recv code that called this function
 568  *   as part of close processing.
 569  *
 570  * At a minimum one references should still be held.
 571  */
 572 static void svc_rdma_detach(struct svc_xprt *xprt)
 573 {
 574         struct svcxprt_rdma *rdma =
 575                 container_of(xprt, struct svcxprt_rdma, sc_xprt);
 576 
 577         /* Disconnect and flush posted WQE */
 578         rdma_disconnect(rdma->sc_cm_id);
 579 }
 580 
 581 static void __svc_rdma_free(struct work_struct *work)
 582 {
 583         struct svcxprt_rdma *rdma =
 584                 container_of(work, struct svcxprt_rdma, sc_work);
 585         struct svc_xprt *xprt = &rdma->sc_xprt;
 586 
 587         trace_svcrdma_xprt_free(xprt);
 588 
 589         if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
 590                 ib_drain_qp(rdma->sc_qp);
 591 
 592         svc_rdma_flush_recv_queues(rdma);
 593 
 594         /* Final put of backchannel client transport */
 595         if (xprt->xpt_bc_xprt) {
 596                 xprt_put(xprt->xpt_bc_xprt);
 597                 xprt->xpt_bc_xprt = NULL;
 598         }
 599 
 600         svc_rdma_destroy_rw_ctxts(rdma);
 601         svc_rdma_send_ctxts_destroy(rdma);
 602         svc_rdma_recv_ctxts_destroy(rdma);
 603 
 604         /* Destroy the QP if present (not a listener) */
 605         if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
 606                 ib_destroy_qp(rdma->sc_qp);
 607 
 608         if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
 609                 ib_free_cq(rdma->sc_sq_cq);
 610 
 611         if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
 612                 ib_free_cq(rdma->sc_rq_cq);
 613 
 614         if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
 615                 ib_dealloc_pd(rdma->sc_pd);
 616 
 617         /* Destroy the CM ID */
 618         rdma_destroy_id(rdma->sc_cm_id);
 619 
 620         kfree(rdma);
 621 }
 622 
 623 static void svc_rdma_free(struct svc_xprt *xprt)
 624 {
 625         struct svcxprt_rdma *rdma =
 626                 container_of(xprt, struct svcxprt_rdma, sc_xprt);
 627 
 628         INIT_WORK(&rdma->sc_work, __svc_rdma_free);
 629         schedule_work(&rdma->sc_work);
 630 }
 631 
 632 static int svc_rdma_has_wspace(struct svc_xprt *xprt)
 633 {
 634         struct svcxprt_rdma *rdma =
 635                 container_of(xprt, struct svcxprt_rdma, sc_xprt);
 636 
 637         /*
 638          * If there are already waiters on the SQ,
 639          * return false.
 640          */
 641         if (waitqueue_active(&rdma->sc_send_wait))
 642                 return 0;
 643 
 644         /* Otherwise return true. */
 645         return 1;
 646 }
 647 
 648 static void svc_rdma_secure_port(struct svc_rqst *rqstp)
 649 {
 650         set_bit(RQ_SECURE, &rqstp->rq_flags);
 651 }
 652 
 653 static void svc_rdma_kill_temp_xprt(struct svc_xprt *xprt)
 654 {
 655 }