SAFE public projects git trees. - safe/jmp/linux-2.6/blob - net/sunrpc/xprtrdma/transport.c

   1 /*
   2  * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the BSD-type
   8  * license below:
   9  *
  10  * Redistribution and use in source and binary forms, with or without
  11  * modification, are permitted provided that the following conditions
  12  * are met:
  13  *
  14  *      Redistributions of source code must retain the above copyright
  15  *      notice, this list of conditions and the following disclaimer.
  16  *
  17  *      Redistributions in binary form must reproduce the above
  18  *      copyright notice, this list of conditions and the following
  19  *      disclaimer in the documentation and/or other materials provided
  20  *      with the distribution.
  21  *
  22  *      Neither the name of the Network Appliance, Inc. nor the names of
  23  *      its contributors may be used to endorse or promote products
  24  *      derived from this software without specific prior written
  25  *      permission.
  26  *
  27  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  28  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  29  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  30  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  31  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  32  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  33  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  34  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  35  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  36  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  37  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  38  */
  39
  40 /*
  41  * transport.c
  42  *
  43  * This file contains the top-level implementation of an RPC RDMA
  44  * transport.
  45  *
  46  * Naming convention: functions beginning with xprt_ are part of the
  47  * transport switch. All others are RPC RDMA internal.
  48  */
  49
  50 #include <linux/module.h>
  51 #include <linux/init.h>
  52 #include <linux/slab.h>
  53 #include <linux/seq_file.h>
  54
  55 #include "xprt_rdma.h"
  56
  57 #ifdef RPC_DEBUG
  58 # define RPCDBG_FACILITY        RPCDBG_TRANS
  59 #endif
  60
  61 MODULE_LICENSE("Dual BSD/GPL");
  62
  63 MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
  64 MODULE_AUTHOR("Network Appliance, Inc.");
  65
  66 /*
  67  * tunables
  68  */
  69
  70 static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
  71 static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
  72 static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
  73 static unsigned int xprt_rdma_inline_write_padding;
  74 static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_FRMR;
  75                 int xprt_rdma_pad_optimize = 0;
  76
  77 #ifdef RPC_DEBUG
  78
  79 static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
  80 static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
  81 static unsigned int zero;
  82 static unsigned int max_padding = PAGE_SIZE;
  83 static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
  84 static unsigned int max_memreg = RPCRDMA_LAST - 1;
  85
  86 static struct ctl_table_header *sunrpc_table_header;
  87
  88 static ctl_table xr_tunables_table[] = {
  89         {
  90                 .procname       = "rdma_slot_table_entries",
  91                 .data           = &xprt_rdma_slot_table_entries,
  92                 .maxlen         = sizeof(unsigned int),
  93                 .mode           = 0644,
  94                 .proc_handler   = proc_dointvec_minmax,
  95                 .extra1         = &min_slot_table_size,
  96                 .extra2         = &max_slot_table_size
  97         },
  98         {
  99                 .procname       = "rdma_max_inline_read",
 100                 .data           = &xprt_rdma_max_inline_read,
 101                 .maxlen         = sizeof(unsigned int),
 102                 .mode           = 0644,
 103                 .proc_handler   = proc_dointvec,
 104         },
 105         {
 106                 .procname       = "rdma_max_inline_write",
 107                 .data           = &xprt_rdma_max_inline_write,
 108                 .maxlen         = sizeof(unsigned int),
 109                 .mode           = 0644,
 110                 .proc_handler   = proc_dointvec,
 111         },
 112         {
 113                 .procname       = "rdma_inline_write_padding",
 114                 .data           = &xprt_rdma_inline_write_padding,
 115                 .maxlen         = sizeof(unsigned int),
 116                 .mode           = 0644,
 117                 .proc_handler   = proc_dointvec_minmax,
 118                 .extra1         = &zero,
 119                 .extra2         = &max_padding,
 120         },
 121         {
 122                 .procname       = "rdma_memreg_strategy",
 123                 .data           = &xprt_rdma_memreg_strategy,
 124                 .maxlen         = sizeof(unsigned int),
 125                 .mode           = 0644,
 126                 .proc_handler   = proc_dointvec_minmax,
 127                 .extra1         = &min_memreg,
 128                 .extra2         = &max_memreg,
 129         },
 130         {
 131                 .procname       = "rdma_pad_optimize",
 132                 .data           = &xprt_rdma_pad_optimize,
 133                 .maxlen         = sizeof(unsigned int),
 134                 .mode           = 0644,
 135                 .proc_handler   = proc_dointvec,
 136         },
 137         { },
 138 };
 139
 140 static ctl_table sunrpc_table[] = {
 141         {
 142                 .procname       = "sunrpc",
 143                 .mode           = 0555,
 144                 .child          = xr_tunables_table
 145         },
 146         { },
 147 };
 148
 149 #endif
 150
 151 static struct rpc_xprt_ops xprt_rdma_procs;     /* forward reference */
 152
 153 static void
 154 xprt_rdma_format_addresses(struct rpc_xprt *xprt)
 155 {
 156         struct sockaddr *sap = (struct sockaddr *)
 157                                         &rpcx_to_rdmad(xprt).addr;
 158         struct sockaddr_in *sin = (struct sockaddr_in *)sap;
 159         char buf[64];
 160
 161         (void)rpc_ntop(sap, buf, sizeof(buf));
 162         xprt->address_strings[RPC_DISPLAY_ADDR] = kstrdup(buf, GFP_KERNEL);
 163
 164         snprintf(buf, sizeof(buf), "%u", rpc_get_port(sap));
 165         xprt->address_strings[RPC_DISPLAY_PORT] = kstrdup(buf, GFP_KERNEL);
 166
 167         xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
 168
 169         snprintf(buf, sizeof(buf), "%08x", ntohl(sin->sin_addr.s_addr));
 170         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = kstrdup(buf, GFP_KERNEL);
 171
 172         snprintf(buf, sizeof(buf), "%4hx", rpc_get_port(sap));
 173         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = kstrdup(buf, GFP_KERNEL);
 174
 175         /* netid */
 176         xprt->address_strings[RPC_DISPLAY_NETID] = "rdma";
 177 }
 178
 179 static void
 180 xprt_rdma_free_addresses(struct rpc_xprt *xprt)
 181 {
 182         unsigned int i;
 183
 184         for (i = 0; i < RPC_DISPLAY_MAX; i++)
 185                 switch (i) {
 186                 case RPC_DISPLAY_PROTO:
 187                 case RPC_DISPLAY_NETID:
 188                         continue;
 189                 default:
 190                         kfree(xprt->address_strings[i]);
 191                 }
 192 }
 193
 194 static void
 195 xprt_rdma_connect_worker(struct work_struct *work)
 196 {
 197         struct rpcrdma_xprt *r_xprt =
 198                 container_of(work, struct rpcrdma_xprt, rdma_connect.work);
 199         struct rpc_xprt *xprt = &r_xprt->xprt;
 200         int rc = 0;
 201
 202         if (!xprt->shutdown) {
 203                 xprt_clear_connected(xprt);
 204
 205                 dprintk("RPC:       %s: %sconnect\n", __func__,
 206                                 r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
 207                 rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
 208                 if (rc)
 209                         goto out;
 210         }
 211         goto out_clear;
 212
 213 out:
 214         xprt_wake_pending_tasks(xprt, rc);
 215
 216 out_clear:
 217         dprintk("RPC:       %s: exit\n", __func__);
 218         xprt_clear_connecting(xprt);
 219 }
 220
 221 /*
 222  * xprt_rdma_destroy
 223  *
 224  * Destroy the xprt.
 225  * Free all memory associated with the object, including its own.
 226  * NOTE: none of the *destroy methods free memory for their top-level
 227  * objects, even though they may have allocated it (they do free
 228  * private memory). It's up to the caller to handle it. In this
 229  * case (RDMA transport), all structure memory is inlined with the
 230  * struct rpcrdma_xprt.
 231  */
 232 static void
 233 xprt_rdma_destroy(struct rpc_xprt *xprt)
 234 {
 235         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 236         int rc;
 237
 238         dprintk("RPC:       %s: called\n", __func__);
 239
 240         cancel_delayed_work(&r_xprt->rdma_connect);
 241         flush_scheduled_work();
 242
 243         xprt_clear_connected(xprt);
 244
 245         rpcrdma_buffer_destroy(&r_xprt->rx_buf);
 246         rc = rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
 247         if (rc)
 248                 dprintk("RPC:       %s: rpcrdma_ep_destroy returned %i\n",
 249                         __func__, rc);
 250         rpcrdma_ia_close(&r_xprt->rx_ia);
 251
 252         xprt_rdma_free_addresses(xprt);
 253
 254         kfree(xprt->slot);
 255         xprt->slot = NULL;
 256         kfree(xprt);
 257
 258         dprintk("RPC:       %s: returning\n", __func__);
 259
 260         module_put(THIS_MODULE);
 261 }
 262
 263 static const struct rpc_timeout xprt_rdma_default_timeout = {
 264         .to_initval = 60 * HZ,
 265         .to_maxval = 60 * HZ,
 266 };
 267
 268 /**
 269  * xprt_setup_rdma - Set up transport to use RDMA
 270  *
 271  * @args: rpc transport arguments
 272  */
 273 static struct rpc_xprt *
 274 xprt_setup_rdma(struct xprt_create *args)
 275 {
 276         struct rpcrdma_create_data_internal cdata;
 277         struct rpc_xprt *xprt;
 278         struct rpcrdma_xprt *new_xprt;
 279         struct rpcrdma_ep *new_ep;
 280         struct sockaddr_in *sin;
 281         int rc;
 282
 283         if (args->addrlen > sizeof(xprt->addr)) {
 284                 dprintk("RPC:       %s: address too large\n", __func__);
 285                 return ERR_PTR(-EBADF);
 286         }
 287
 288         xprt = kzalloc(sizeof(struct rpcrdma_xprt), GFP_KERNEL);
 289         if (xprt == NULL) {
 290                 dprintk("RPC:       %s: couldn't allocate rpcrdma_xprt\n",
 291                         __func__);
 292                 return ERR_PTR(-ENOMEM);
 293         }
 294
 295         xprt->max_reqs = xprt_rdma_slot_table_entries;
 296         xprt->slot = kcalloc(xprt->max_reqs,
 297                                 sizeof(struct rpc_rqst), GFP_KERNEL);
 298         if (xprt->slot == NULL) {
 299                 dprintk("RPC:       %s: couldn't allocate %d slots\n",
 300                         __func__, xprt->max_reqs);
 301                 kfree(xprt);
 302                 return ERR_PTR(-ENOMEM);
 303         }
 304
 305         /* 60 second timeout, no retries */
 306         xprt->timeout = &xprt_rdma_default_timeout;
 307         xprt->bind_timeout = (60U * HZ);
 308         xprt->connect_timeout = (60U * HZ);
 309         xprt->reestablish_timeout = (5U * HZ);
 310         xprt->idle_timeout = (5U * 60 * HZ);
 311
 312         xprt->resvport = 0;             /* privileged port not needed */
 313         xprt->tsh_size = 0;             /* RPC-RDMA handles framing */
 314         xprt->max_payload = RPCRDMA_MAX_DATA_SEGS * PAGE_SIZE;
 315         xprt->ops = &xprt_rdma_procs;
 316
 317         /*
 318          * Set up RDMA-specific connect data.
 319          */
 320
 321         /* Put server RDMA address in local cdata */
 322         memcpy(&cdata.addr, args->dstaddr, args->addrlen);
 323
 324         /* Ensure xprt->addr holds valid server TCP (not RDMA)
 325          * address, for any side protocols which peek at it */
 326         xprt->prot = IPPROTO_TCP;
 327         xprt->addrlen = args->addrlen;
 328         memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
 329
 330         sin = (struct sockaddr_in *)&cdata.addr;
 331         if (ntohs(sin->sin_port) != 0)
 332                 xprt_set_bound(xprt);
 333
 334         dprintk("RPC:       %s: %pI4:%u\n",
 335                 __func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port));
 336
 337         /* Set max requests */
 338         cdata.max_requests = xprt->max_reqs;
 339
 340         /* Set some length limits */
 341         cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
 342         cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
 343
 344         cdata.inline_wsize = xprt_rdma_max_inline_write;
 345         if (cdata.inline_wsize > cdata.wsize)
 346                 cdata.inline_wsize = cdata.wsize;
 347
 348         cdata.inline_rsize = xprt_rdma_max_inline_read;
 349         if (cdata.inline_rsize > cdata.rsize)
 350                 cdata.inline_rsize = cdata.rsize;
 351
 352         cdata.padding = xprt_rdma_inline_write_padding;
 353
 354         /*
 355          * Create new transport instance, which includes initialized
 356          *  o ia
 357          *  o endpoint
 358          *  o buffers
 359          */
 360
 361         new_xprt = rpcx_to_rdmax(xprt);
 362
 363         rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
 364                                 xprt_rdma_memreg_strategy);
 365         if (rc)
 366                 goto out1;
 367
 368         /*
 369          * initialize and create ep
 370          */
 371         new_xprt->rx_data = cdata;
 372         new_ep = &new_xprt->rx_ep;
 373         new_ep->rep_remote_addr = cdata.addr;
 374
 375         rc = rpcrdma_ep_create(&new_xprt->rx_ep,
 376                                 &new_xprt->rx_ia, &new_xprt->rx_data);
 377         if (rc)
 378                 goto out2;
 379
 380         /*
 381          * Allocate pre-registered send and receive buffers for headers and
 382          * any inline data. Also specify any padding which will be provided
 383          * from a preregistered zero buffer.
 384          */
 385         rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
 386                                 &new_xprt->rx_data);
 387         if (rc)
 388                 goto out3;
 389
 390         /*
 391          * Register a callback for connection events. This is necessary because
 392          * connection loss notification is async. We also catch connection loss
 393          * when reaping receives.
 394          */
 395         INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
 396         new_ep->rep_func = rpcrdma_conn_func;
 397         new_ep->rep_xprt = xprt;
 398
 399         xprt_rdma_format_addresses(xprt);
 400
 401         if (!try_module_get(THIS_MODULE))
 402                 goto out4;
 403
 404         return xprt;
 405
 406 out4:
 407         xprt_rdma_free_addresses(xprt);
 408         rc = -EINVAL;
 409 out3:
 410         (void) rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
 411 out2:
 412         rpcrdma_ia_close(&new_xprt->rx_ia);
 413 out1:
 414         kfree(xprt->slot);
 415         kfree(xprt);
 416         return ERR_PTR(rc);
 417 }
 418
 419 /*
 420  * Close a connection, during shutdown or timeout/reconnect
 421  */
 422 static void
 423 xprt_rdma_close(struct rpc_xprt *xprt)
 424 {
 425         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 426
 427         dprintk("RPC:       %s: closing\n", __func__);
 428         if (r_xprt->rx_ep.rep_connected > 0)
 429                 xprt->reestablish_timeout = 0;
 430         xprt_disconnect_done(xprt);
 431         (void) rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
 432 }
 433
 434 static void
 435 xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
 436 {
 437         struct sockaddr_in *sap;
 438
 439         sap = (struct sockaddr_in *)&xprt->addr;
 440         sap->sin_port = htons(port);
 441         sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
 442         sap->sin_port = htons(port);
 443         dprintk("RPC:       %s: %u\n", __func__, port);
 444 }
 445
 446 static void
 447 xprt_rdma_connect(struct rpc_task *task)
 448 {
 449         struct rpc_xprt *xprt = (struct rpc_xprt *)task->tk_xprt;
 450         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 451
 452         if (!xprt_test_and_set_connecting(xprt)) {
 453                 if (r_xprt->rx_ep.rep_connected != 0) {
 454                         /* Reconnect */
 455                         schedule_delayed_work(&r_xprt->rdma_connect,
 456                                 xprt->reestablish_timeout);
 457                         xprt->reestablish_timeout <<= 1;
 458                         if (xprt->reestablish_timeout > (30 * HZ))
 459                                 xprt->reestablish_timeout = (30 * HZ);
 460                         else if (xprt->reestablish_timeout < (5 * HZ))
 461                                 xprt->reestablish_timeout = (5 * HZ);
 462                 } else {
 463                         schedule_delayed_work(&r_xprt->rdma_connect, 0);
 464                         if (!RPC_IS_ASYNC(task))
 465                                 flush_scheduled_work();
 466                 }
 467         }
 468 }
 469
 470 static int
 471 xprt_rdma_reserve_xprt(struct rpc_task *task)
 472 {
 473         struct rpc_xprt *xprt = task->tk_xprt;
 474         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 475         int credits = atomic_read(&r_xprt->rx_buf.rb_credits);
 476
 477         /* == RPC_CWNDSCALE @ init, but *after* setup */
 478         if (r_xprt->rx_buf.rb_cwndscale == 0UL) {
 479                 r_xprt->rx_buf.rb_cwndscale = xprt->cwnd;
 480                 dprintk("RPC:       %s: cwndscale %lu\n", __func__,
 481                         r_xprt->rx_buf.rb_cwndscale);
 482                 BUG_ON(r_xprt->rx_buf.rb_cwndscale <= 0);
 483         }
 484         xprt->cwnd = credits * r_xprt->rx_buf.rb_cwndscale;
 485         return xprt_reserve_xprt_cong(task);
 486 }
 487
 488 /*
 489  * The RDMA allocate/free functions need the task structure as a place
 490  * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
 491  * sequence. For this reason, the recv buffers are attached to send
 492  * buffers for portions of the RPC. Note that the RPC layer allocates
 493  * both send and receive buffers in the same call. We may register
 494  * the receive buffer portion when using reply chunks.
 495  */
 496 static void *
 497 xprt_rdma_allocate(struct rpc_task *task, size_t size)
 498 {
 499         struct rpc_xprt *xprt = task->tk_xprt;
 500         struct rpcrdma_req *req, *nreq;
 501
 502         req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
 503         BUG_ON(NULL == req);
 504
 505         if (size > req->rl_size) {
 506                 dprintk("RPC:       %s: size %zd too large for buffer[%zd]: "
 507                         "prog %d vers %d proc %d\n",
 508                         __func__, size, req->rl_size,
 509                         task->tk_client->cl_prog, task->tk_client->cl_vers,
 510                         task->tk_msg.rpc_proc->p_proc);
 511                 /*
 512                  * Outgoing length shortage. Our inline write max must have
 513                  * been configured to perform direct i/o.
 514                  *
 515                  * This is therefore a large metadata operation, and the
 516                  * allocate call was made on the maximum possible message,
 517                  * e.g. containing long filename(s) or symlink data. In
 518                  * fact, while these metadata operations *might* carry
 519                  * large outgoing payloads, they rarely *do*. However, we
 520                  * have to commit to the request here, so reallocate and
 521                  * register it now. The data path will never require this
 522                  * reallocation.
 523                  *
 524                  * If the allocation or registration fails, the RPC framework
 525                  * will (doggedly) retry.
 526                  */
 527                 if (rpcx_to_rdmax(xprt)->rx_ia.ri_memreg_strategy ==
 528                                 RPCRDMA_BOUNCEBUFFERS) {
 529                         /* forced to "pure inline" */
 530                         dprintk("RPC:       %s: too much data (%zd) for inline "
 531                                         "(r/w max %d/%d)\n", __func__, size,
 532                                         rpcx_to_rdmad(xprt).inline_rsize,
 533                                         rpcx_to_rdmad(xprt).inline_wsize);
 534                         size = req->rl_size;
 535                         rpc_exit(task, -EIO);           /* fail the operation */
 536                         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 537                         goto out;
 538                 }
 539                 if (task->tk_flags & RPC_TASK_SWAPPER)
 540                         nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
 541                 else
 542                         nreq = kmalloc(sizeof *req + size, GFP_NOFS);
 543                 if (nreq == NULL)
 544                         goto outfail;
 545
 546                 if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
 547                                 nreq->rl_base, size + sizeof(struct rpcrdma_req)
 548                                 - offsetof(struct rpcrdma_req, rl_base),
 549                                 &nreq->rl_handle, &nreq->rl_iov)) {
 550                         kfree(nreq);
 551                         goto outfail;
 552                 }
 553                 rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
 554                 nreq->rl_size = size;
 555                 nreq->rl_niovs = 0;
 556                 nreq->rl_nchunks = 0;
 557                 nreq->rl_buffer = (struct rpcrdma_buffer *)req;
 558                 nreq->rl_reply = req->rl_reply;
 559                 memcpy(nreq->rl_segments,
 560                         req->rl_segments, sizeof nreq->rl_segments);
 561                 /* flag the swap with an unused field */
 562                 nreq->rl_iov.length = 0;
 563                 req->rl_reply = NULL;
 564                 req = nreq;
 565         }
 566         dprintk("RPC:       %s: size %zd, request 0x%p\n", __func__, size, req);
 567 out:
 568         req->rl_connect_cookie = 0;     /* our reserved value */
 569         return req->rl_xdr_buf;
 570
 571 outfail:
 572         rpcrdma_buffer_put(req);
 573         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 574         return NULL;
 575 }
 576
 577 /*
 578  * This function returns all RDMA resources to the pool.
 579  */
 580 static void
 581 xprt_rdma_free(void *buffer)
 582 {
 583         struct rpcrdma_req *req;
 584         struct rpcrdma_xprt *r_xprt;
 585         struct rpcrdma_rep *rep;
 586         int i;
 587
 588         if (buffer == NULL)
 589                 return;
 590
 591         req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
 592         if (req->rl_iov.length == 0) {  /* see allocate above */
 593                 r_xprt = container_of(((struct rpcrdma_req *) req->rl_buffer)->rl_buffer,
 594                                       struct rpcrdma_xprt, rx_buf);
 595         } else
 596                 r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
 597         rep = req->rl_reply;
 598
 599         dprintk("RPC:       %s: called on 0x%p%s\n",
 600                 __func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
 601
 602         /*
 603          * Finish the deregistration. When using mw bind, this was
 604          * begun in rpcrdma_reply_handler(). In all other modes, we
 605          * do it here, in thread context. The process is considered
 606          * complete when the rr_func vector becomes NULL - this
 607          * was put in place during rpcrdma_reply_handler() - the wait
 608          * call below will not block if the dereg is "done". If
 609          * interrupted, our framework will clean up.
 610          */
 611         for (i = 0; req->rl_nchunks;) {
 612                 --req->rl_nchunks;
 613                 i += rpcrdma_deregister_external(
 614                         &req->rl_segments[i], r_xprt, NULL);
 615         }
 616
 617         if (rep && wait_event_interruptible(rep->rr_unbind, !rep->rr_func)) {
 618                 rep->rr_func = NULL;    /* abandon the callback */
 619                 req->rl_reply = NULL;
 620         }
 621
 622         if (req->rl_iov.length == 0) {  /* see allocate above */
 623                 struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
 624                 oreq->rl_reply = req->rl_reply;
 625                 (void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
 626                                                    req->rl_handle,
 627                                                    &req->rl_iov);
 628                 kfree(req);
 629                 req = oreq;
 630         }
 631
 632         /* Put back request+reply buffers */
 633         rpcrdma_buffer_put(req);
 634 }
 635
 636 /*
 637  * send_request invokes the meat of RPC RDMA. It must do the following:
 638  *  1.  Marshal the RPC request into an RPC RDMA request, which means
 639  *      putting a header in front of data, and creating IOVs for RDMA
 640  *      from those in the request.
 641  *  2.  In marshaling, detect opportunities for RDMA, and use them.
 642  *  3.  Post a recv message to set up asynch completion, then send
 643  *      the request (rpcrdma_ep_post).
 644  *  4.  No partial sends are possible in the RPC-RDMA protocol (as in UDP).
 645  */
 646
 647 static int
 648 xprt_rdma_send_request(struct rpc_task *task)
 649 {
 650         struct rpc_rqst *rqst = task->tk_rqstp;
 651         struct rpc_xprt *xprt = task->tk_xprt;
 652         struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 653         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 654
 655         /* marshal the send itself */
 656         if (req->rl_niovs == 0 && rpcrdma_marshal_req(rqst) != 0) {
 657                 r_xprt->rx_stats.failed_marshal_count++;
 658                 dprintk("RPC:       %s: rpcrdma_marshal_req failed\n",
 659                         __func__);
 660                 return -EIO;
 661         }
 662
 663         if (req->rl_reply == NULL)              /* e.g. reconnection */
 664                 rpcrdma_recv_buffer_get(req);
 665
 666         if (req->rl_reply) {
 667                 req->rl_reply->rr_func = rpcrdma_reply_handler;
 668                 /* this need only be done once, but... */
 669                 req->rl_reply->rr_xprt = xprt;
 670         }
 671
 672         /* Must suppress retransmit to maintain credits */
 673         if (req->rl_connect_cookie == xprt->connect_cookie)
 674                 goto drop_connection;
 675         req->rl_connect_cookie = xprt->connect_cookie;
 676
 677         if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
 678                 goto drop_connection;
 679
 680         task->tk_bytes_sent += rqst->rq_snd_buf.len;
 681         rqst->rq_bytes_sent = 0;
 682         return 0;
 683
 684 drop_connection:
 685         xprt_disconnect_done(xprt);
 686         return -ENOTCONN;       /* implies disconnect */
 687 }
 688
 689 static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
 690 {
 691         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 692         long idle_time = 0;
 693
 694         if (xprt_connected(xprt))
 695                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
 696
 697         seq_printf(seq,
 698           "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
 699           "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
 700
 701            0,   /* need a local port? */
 702            xprt->stat.bind_count,
 703            xprt->stat.connect_count,
 704            xprt->stat.connect_time,
 705            idle_time,
 706            xprt->stat.sends,
 707            xprt->stat.recvs,
 708            xprt->stat.bad_xids,
 709            xprt->stat.req_u,
 710            xprt->stat.bklog_u,
 711
 712            r_xprt->rx_stats.read_chunk_count,
 713            r_xprt->rx_stats.write_chunk_count,
 714            r_xprt->rx_stats.reply_chunk_count,
 715            r_xprt->rx_stats.total_rdma_request,
 716            r_xprt->rx_stats.total_rdma_reply,
 717            r_xprt->rx_stats.pullup_copy_count,
 718            r_xprt->rx_stats.fixup_copy_count,
 719            r_xprt->rx_stats.hardway_register_count,
 720            r_xprt->rx_stats.failed_marshal_count,
 721            r_xprt->rx_stats.bad_reply_count);
 722 }
 723
 724 /*
 725  * Plumbing for rpc transport switch and kernel module
 726  */
 727
 728 static struct rpc_xprt_ops xprt_rdma_procs = {
 729         .reserve_xprt           = xprt_rdma_reserve_xprt,
 730         .release_xprt           = xprt_release_xprt_cong, /* sunrpc/xprt.c */
 731         .release_request        = xprt_release_rqst_cong,       /* ditto */
 732         .set_retrans_timeout    = xprt_set_retrans_timeout_def, /* ditto */
 733         .rpcbind                = rpcb_getport_async,   /* sunrpc/rpcb_clnt.c */
 734         .set_port               = xprt_rdma_set_port,
 735         .connect                = xprt_rdma_connect,
 736         .buf_alloc              = xprt_rdma_allocate,
 737         .buf_free               = xprt_rdma_free,
 738         .send_request           = xprt_rdma_send_request,
 739         .close                  = xprt_rdma_close,
 740         .destroy                = xprt_rdma_destroy,
 741         .print_stats            = xprt_rdma_print_stats
 742 };
 743
 744 static struct xprt_class xprt_rdma = {
 745         .list                   = LIST_HEAD_INIT(xprt_rdma.list),
 746         .name                   = "rdma",
 747         .owner                  = THIS_MODULE,
 748         .ident                  = XPRT_TRANSPORT_RDMA,
 749         .setup                  = xprt_setup_rdma,
 750 };
 751
 752 static void __exit xprt_rdma_cleanup(void)
 753 {
 754         int rc;
 755
 756         dprintk(KERN_INFO "RPCRDMA Module Removed, deregister RPC RDMA transport\n");
 757 #ifdef RPC_DEBUG
 758         if (sunrpc_table_header) {
 759                 unregister_sysctl_table(sunrpc_table_header);
 760                 sunrpc_table_header = NULL;
 761         }
 762 #endif
 763         rc = xprt_unregister_transport(&xprt_rdma);
 764         if (rc)
 765                 dprintk("RPC:       %s: xprt_unregister returned %i\n",
 766                         __func__, rc);
 767 }
 768
 769 static int __init xprt_rdma_init(void)
 770 {
 771         int rc;
 772
 773         rc = xprt_register_transport(&xprt_rdma);
 774
 775         if (rc)
 776                 return rc;
 777
 778         dprintk(KERN_INFO "RPCRDMA Module Init, register RPC RDMA transport\n");
 779
 780         dprintk(KERN_INFO "Defaults:\n");
 781         dprintk(KERN_INFO "\tSlots %d\n"
 782                 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
 783                 xprt_rdma_slot_table_entries,
 784                 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
 785         dprintk(KERN_INFO "\tPadding %d\n\tMemreg %d\n",
 786                 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
 787
 788 #ifdef RPC_DEBUG
 789         if (!sunrpc_table_header)
 790                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
 791 #endif
 792         return 0;
 793 }
 794
 795 module_init(xprt_rdma_init);
 796 module_exit(xprt_rdma_cleanup);