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_sync(&r_xprt->rdma_connect);
 241
 242         xprt_clear_connected(xprt);
 243
 244         rpcrdma_buffer_destroy(&r_xprt->rx_buf);
 245         rc = rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
 246         if (rc)
 247                 dprintk("RPC:       %s: rpcrdma_ep_destroy returned %i\n",
 248                         __func__, rc);
 249         rpcrdma_ia_close(&r_xprt->rx_ia);
 250
 251         xprt_rdma_free_addresses(xprt);
 252
 253         xprt_free(xprt);
 254
 255         dprintk("RPC:       %s: returning\n", __func__);
 256
 257         module_put(THIS_MODULE);
 258 }
 259
 260 static const struct rpc_timeout xprt_rdma_default_timeout = {
 261         .to_initval = 60 * HZ,
 262         .to_maxval = 60 * HZ,
 263 };
 264
 265 /**
 266  * xprt_setup_rdma - Set up transport to use RDMA
 267  *
 268  * @args: rpc transport arguments
 269  */
 270 static struct rpc_xprt *
 271 xprt_setup_rdma(struct xprt_create *args)
 272 {
 273         struct rpcrdma_create_data_internal cdata;
 274         struct rpc_xprt *xprt;
 275         struct rpcrdma_xprt *new_xprt;
 276         struct rpcrdma_ep *new_ep;
 277         struct sockaddr_in *sin;
 278         int rc;
 279
 280         if (args->addrlen > sizeof(xprt->addr)) {
 281                 dprintk("RPC:       %s: address too large\n", __func__);
 282                 return ERR_PTR(-EBADF);
 283         }
 284
 285         xprt = xprt_alloc(args->net, sizeof(struct rpcrdma_xprt),
 286                         xprt_rdma_slot_table_entries);
 287         if (xprt == NULL) {
 288                 dprintk("RPC:       %s: couldn't allocate rpcrdma_xprt\n",
 289                         __func__);
 290                 return ERR_PTR(-ENOMEM);
 291         }
 292
 293         /* 60 second timeout, no retries */
 294         xprt->timeout = &xprt_rdma_default_timeout;
 295         xprt->bind_timeout = (60U * HZ);
 296         xprt->reestablish_timeout = (5U * HZ);
 297         xprt->idle_timeout = (5U * 60 * HZ);
 298
 299         xprt->resvport = 0;             /* privileged port not needed */
 300         xprt->tsh_size = 0;             /* RPC-RDMA handles framing */
 301         xprt->max_payload = RPCRDMA_MAX_DATA_SEGS * PAGE_SIZE;
 302         xprt->ops = &xprt_rdma_procs;
 303
 304         /*
 305          * Set up RDMA-specific connect data.
 306          */
 307
 308         /* Put server RDMA address in local cdata */
 309         memcpy(&cdata.addr, args->dstaddr, args->addrlen);
 310
 311         /* Ensure xprt->addr holds valid server TCP (not RDMA)
 312          * address, for any side protocols which peek at it */
 313         xprt->prot = IPPROTO_TCP;
 314         xprt->addrlen = args->addrlen;
 315         memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
 316
 317         sin = (struct sockaddr_in *)&cdata.addr;
 318         if (ntohs(sin->sin_port) != 0)
 319                 xprt_set_bound(xprt);
 320
 321         dprintk("RPC:       %s: %pI4:%u\n",
 322                 __func__, &sin->sin_addr.s_addr, ntohs(sin->sin_port));
 323
 324         /* Set max requests */
 325         cdata.max_requests = xprt->max_reqs;
 326
 327         /* Set some length limits */
 328         cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
 329         cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
 330
 331         cdata.inline_wsize = xprt_rdma_max_inline_write;
 332         if (cdata.inline_wsize > cdata.wsize)
 333                 cdata.inline_wsize = cdata.wsize;
 334
 335         cdata.inline_rsize = xprt_rdma_max_inline_read;
 336         if (cdata.inline_rsize > cdata.rsize)
 337                 cdata.inline_rsize = cdata.rsize;
 338
 339         cdata.padding = xprt_rdma_inline_write_padding;
 340
 341         /*
 342          * Create new transport instance, which includes initialized
 343          *  o ia
 344          *  o endpoint
 345          *  o buffers
 346          */
 347
 348         new_xprt = rpcx_to_rdmax(xprt);
 349
 350         rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
 351                                 xprt_rdma_memreg_strategy);
 352         if (rc)
 353                 goto out1;
 354
 355         /*
 356          * initialize and create ep
 357          */
 358         new_xprt->rx_data = cdata;
 359         new_ep = &new_xprt->rx_ep;
 360         new_ep->rep_remote_addr = cdata.addr;
 361
 362         rc = rpcrdma_ep_create(&new_xprt->rx_ep,
 363                                 &new_xprt->rx_ia, &new_xprt->rx_data);
 364         if (rc)
 365                 goto out2;
 366
 367         /*
 368          * Allocate pre-registered send and receive buffers for headers and
 369          * any inline data. Also specify any padding which will be provided
 370          * from a preregistered zero buffer.
 371          */
 372         rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
 373                                 &new_xprt->rx_data);
 374         if (rc)
 375                 goto out3;
 376
 377         /*
 378          * Register a callback for connection events. This is necessary because
 379          * connection loss notification is async. We also catch connection loss
 380          * when reaping receives.
 381          */
 382         INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
 383         new_ep->rep_func = rpcrdma_conn_func;
 384         new_ep->rep_xprt = xprt;
 385
 386         xprt_rdma_format_addresses(xprt);
 387
 388         if (!try_module_get(THIS_MODULE))
 389                 goto out4;
 390
 391         return xprt;
 392
 393 out4:
 394         xprt_rdma_free_addresses(xprt);
 395         rc = -EINVAL;
 396 out3:
 397         (void) rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
 398 out2:
 399         rpcrdma_ia_close(&new_xprt->rx_ia);
 400 out1:
 401         xprt_free(xprt);
 402         return ERR_PTR(rc);
 403 }
 404
 405 /*
 406  * Close a connection, during shutdown or timeout/reconnect
 407  */
 408 static void
 409 xprt_rdma_close(struct rpc_xprt *xprt)
 410 {
 411         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 412
 413         dprintk("RPC:       %s: closing\n", __func__);
 414         if (r_xprt->rx_ep.rep_connected > 0)
 415                 xprt->reestablish_timeout = 0;
 416         xprt_disconnect_done(xprt);
 417         (void) rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
 418 }
 419
 420 static void
 421 xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
 422 {
 423         struct sockaddr_in *sap;
 424
 425         sap = (struct sockaddr_in *)&xprt->addr;
 426         sap->sin_port = htons(port);
 427         sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
 428         sap->sin_port = htons(port);
 429         dprintk("RPC:       %s: %u\n", __func__, port);
 430 }
 431
 432 static void
 433 xprt_rdma_connect(struct rpc_task *task)
 434 {
 435         struct rpc_xprt *xprt = (struct rpc_xprt *)task->tk_xprt;
 436         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 437
 438         if (r_xprt->rx_ep.rep_connected != 0) {
 439                 /* Reconnect */
 440                 schedule_delayed_work(&r_xprt->rdma_connect,
 441                         xprt->reestablish_timeout);
 442                 xprt->reestablish_timeout <<= 1;
 443                 if (xprt->reestablish_timeout > (30 * HZ))
 444                         xprt->reestablish_timeout = (30 * HZ);
 445                 else if (xprt->reestablish_timeout < (5 * HZ))
 446                         xprt->reestablish_timeout = (5 * HZ);
 447         } else {
 448                 schedule_delayed_work(&r_xprt->rdma_connect, 0);
 449                 if (!RPC_IS_ASYNC(task))
 450                         flush_delayed_work(&r_xprt->rdma_connect);
 451         }
 452 }
 453
 454 static int
 455 xprt_rdma_reserve_xprt(struct rpc_task *task)
 456 {
 457         struct rpc_xprt *xprt = task->tk_xprt;
 458         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 459         int credits = atomic_read(&r_xprt->rx_buf.rb_credits);
 460
 461         /* == RPC_CWNDSCALE @ init, but *after* setup */
 462         if (r_xprt->rx_buf.rb_cwndscale == 0UL) {
 463                 r_xprt->rx_buf.rb_cwndscale = xprt->cwnd;
 464                 dprintk("RPC:       %s: cwndscale %lu\n", __func__,
 465                         r_xprt->rx_buf.rb_cwndscale);
 466                 BUG_ON(r_xprt->rx_buf.rb_cwndscale <= 0);
 467         }
 468         xprt->cwnd = credits * r_xprt->rx_buf.rb_cwndscale;
 469         return xprt_reserve_xprt_cong(task);
 470 }
 471
 472 /*
 473  * The RDMA allocate/free functions need the task structure as a place
 474  * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
 475  * sequence. For this reason, the recv buffers are attached to send
 476  * buffers for portions of the RPC. Note that the RPC layer allocates
 477  * both send and receive buffers in the same call. We may register
 478  * the receive buffer portion when using reply chunks.
 479  */
 480 static void *
 481 xprt_rdma_allocate(struct rpc_task *task, size_t size)
 482 {
 483         struct rpc_xprt *xprt = task->tk_xprt;
 484         struct rpcrdma_req *req, *nreq;
 485
 486         req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
 487         BUG_ON(NULL == req);
 488
 489         if (size > req->rl_size) {
 490                 dprintk("RPC:       %s: size %zd too large for buffer[%zd]: "
 491                         "prog %d vers %d proc %d\n",
 492                         __func__, size, req->rl_size,
 493                         task->tk_client->cl_prog, task->tk_client->cl_vers,
 494                         task->tk_msg.rpc_proc->p_proc);
 495                 /*
 496                  * Outgoing length shortage. Our inline write max must have
 497                  * been configured to perform direct i/o.
 498                  *
 499                  * This is therefore a large metadata operation, and the
 500                  * allocate call was made on the maximum possible message,
 501                  * e.g. containing long filename(s) or symlink data. In
 502                  * fact, while these metadata operations *might* carry
 503                  * large outgoing payloads, they rarely *do*. However, we
 504                  * have to commit to the request here, so reallocate and
 505                  * register it now. The data path will never require this
 506                  * reallocation.
 507                  *
 508                  * If the allocation or registration fails, the RPC framework
 509                  * will (doggedly) retry.
 510                  */
 511                 if (rpcx_to_rdmax(xprt)->rx_ia.ri_memreg_strategy ==
 512                                 RPCRDMA_BOUNCEBUFFERS) {
 513                         /* forced to "pure inline" */
 514                         dprintk("RPC:       %s: too much data (%zd) for inline "
 515                                         "(r/w max %d/%d)\n", __func__, size,
 516                                         rpcx_to_rdmad(xprt).inline_rsize,
 517                                         rpcx_to_rdmad(xprt).inline_wsize);
 518                         size = req->rl_size;
 519                         rpc_exit(task, -EIO);           /* fail the operation */
 520                         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 521                         goto out;
 522                 }
 523                 if (task->tk_flags & RPC_TASK_SWAPPER)
 524                         nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
 525                 else
 526                         nreq = kmalloc(sizeof *req + size, GFP_NOFS);
 527                 if (nreq == NULL)
 528                         goto outfail;
 529
 530                 if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
 531                                 nreq->rl_base, size + sizeof(struct rpcrdma_req)
 532                                 - offsetof(struct rpcrdma_req, rl_base),
 533                                 &nreq->rl_handle, &nreq->rl_iov)) {
 534                         kfree(nreq);
 535                         goto outfail;
 536                 }
 537                 rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
 538                 nreq->rl_size = size;
 539                 nreq->rl_niovs = 0;
 540                 nreq->rl_nchunks = 0;
 541                 nreq->rl_buffer = (struct rpcrdma_buffer *)req;
 542                 nreq->rl_reply = req->rl_reply;
 543                 memcpy(nreq->rl_segments,
 544                         req->rl_segments, sizeof nreq->rl_segments);
 545                 /* flag the swap with an unused field */
 546                 nreq->rl_iov.length = 0;
 547                 req->rl_reply = NULL;
 548                 req = nreq;
 549         }
 550         dprintk("RPC:       %s: size %zd, request 0x%p\n", __func__, size, req);
 551 out:
 552         req->rl_connect_cookie = 0;     /* our reserved value */
 553         return req->rl_xdr_buf;
 554
 555 outfail:
 556         rpcrdma_buffer_put(req);
 557         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 558         return NULL;
 559 }
 560
 561 /*
 562  * This function returns all RDMA resources to the pool.
 563  */
 564 static void
 565 xprt_rdma_free(void *buffer)
 566 {
 567         struct rpcrdma_req *req;
 568         struct rpcrdma_xprt *r_xprt;
 569         struct rpcrdma_rep *rep;
 570         int i;
 571
 572         if (buffer == NULL)
 573                 return;
 574
 575         req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
 576         if (req->rl_iov.length == 0) {  /* see allocate above */
 577                 r_xprt = container_of(((struct rpcrdma_req *) req->rl_buffer)->rl_buffer,
 578                                       struct rpcrdma_xprt, rx_buf);
 579         } else
 580                 r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
 581         rep = req->rl_reply;
 582
 583         dprintk("RPC:       %s: called on 0x%p%s\n",
 584                 __func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
 585
 586         /*
 587          * Finish the deregistration. When using mw bind, this was
 588          * begun in rpcrdma_reply_handler(). In all other modes, we
 589          * do it here, in thread context. The process is considered
 590          * complete when the rr_func vector becomes NULL - this
 591          * was put in place during rpcrdma_reply_handler() - the wait
 592          * call below will not block if the dereg is "done". If
 593          * interrupted, our framework will clean up.
 594          */
 595         for (i = 0; req->rl_nchunks;) {
 596                 --req->rl_nchunks;
 597                 i += rpcrdma_deregister_external(
 598                         &req->rl_segments[i], r_xprt, NULL);
 599         }
 600
 601         if (rep && wait_event_interruptible(rep->rr_unbind, !rep->rr_func)) {
 602                 rep->rr_func = NULL;    /* abandon the callback */
 603                 req->rl_reply = NULL;
 604         }
 605
 606         if (req->rl_iov.length == 0) {  /* see allocate above */
 607                 struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
 608                 oreq->rl_reply = req->rl_reply;
 609                 (void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
 610                                                    req->rl_handle,
 611                                                    &req->rl_iov);
 612                 kfree(req);
 613                 req = oreq;
 614         }
 615
 616         /* Put back request+reply buffers */
 617         rpcrdma_buffer_put(req);
 618 }
 619
 620 /*
 621  * send_request invokes the meat of RPC RDMA. It must do the following:
 622  *  1.  Marshal the RPC request into an RPC RDMA request, which means
 623  *      putting a header in front of data, and creating IOVs for RDMA
 624  *      from those in the request.
 625  *  2.  In marshaling, detect opportunities for RDMA, and use them.
 626  *  3.  Post a recv message to set up asynch completion, then send
 627  *      the request (rpcrdma_ep_post).
 628  *  4.  No partial sends are possible in the RPC-RDMA protocol (as in UDP).
 629  */
 630
 631 static int
 632 xprt_rdma_send_request(struct rpc_task *task)
 633 {
 634         struct rpc_rqst *rqst = task->tk_rqstp;
 635         struct rpc_xprt *xprt = task->tk_xprt;
 636         struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 637         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 638
 639         /* marshal the send itself */
 640         if (req->rl_niovs == 0 && rpcrdma_marshal_req(rqst) != 0) {
 641                 r_xprt->rx_stats.failed_marshal_count++;
 642                 dprintk("RPC:       %s: rpcrdma_marshal_req failed\n",
 643                         __func__);
 644                 return -EIO;
 645         }
 646
 647         if (req->rl_reply == NULL)              /* e.g. reconnection */
 648                 rpcrdma_recv_buffer_get(req);
 649
 650         if (req->rl_reply) {
 651                 req->rl_reply->rr_func = rpcrdma_reply_handler;
 652                 /* this need only be done once, but... */
 653                 req->rl_reply->rr_xprt = xprt;
 654         }
 655
 656         /* Must suppress retransmit to maintain credits */
 657         if (req->rl_connect_cookie == xprt->connect_cookie)
 658                 goto drop_connection;
 659         req->rl_connect_cookie = xprt->connect_cookie;
 660
 661         if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req))
 662                 goto drop_connection;
 663
 664         rqst->rq_xmit_bytes_sent += rqst->rq_snd_buf.len;
 665         rqst->rq_bytes_sent = 0;
 666         return 0;
 667
 668 drop_connection:
 669         xprt_disconnect_done(xprt);
 670         return -ENOTCONN;       /* implies disconnect */
 671 }
 672
 673 static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
 674 {
 675         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 676         long idle_time = 0;
 677
 678         if (xprt_connected(xprt))
 679                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
 680
 681         seq_printf(seq,
 682           "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
 683           "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
 684
 685            0,   /* need a local port? */
 686            xprt->stat.bind_count,
 687            xprt->stat.connect_count,
 688            xprt->stat.connect_time,
 689            idle_time,
 690            xprt->stat.sends,
 691            xprt->stat.recvs,
 692            xprt->stat.bad_xids,
 693            xprt->stat.req_u,
 694            xprt->stat.bklog_u,
 695
 696            r_xprt->rx_stats.read_chunk_count,
 697            r_xprt->rx_stats.write_chunk_count,
 698            r_xprt->rx_stats.reply_chunk_count,
 699            r_xprt->rx_stats.total_rdma_request,
 700            r_xprt->rx_stats.total_rdma_reply,
 701            r_xprt->rx_stats.pullup_copy_count,
 702            r_xprt->rx_stats.fixup_copy_count,
 703            r_xprt->rx_stats.hardway_register_count,
 704            r_xprt->rx_stats.failed_marshal_count,
 705            r_xprt->rx_stats.bad_reply_count);
 706 }
 707
 708 /*
 709  * Plumbing for rpc transport switch and kernel module
 710  */
 711
 712 static struct rpc_xprt_ops xprt_rdma_procs = {
 713         .reserve_xprt           = xprt_rdma_reserve_xprt,
 714         .release_xprt           = xprt_release_xprt_cong, /* sunrpc/xprt.c */
 715         .release_request        = xprt_release_rqst_cong,       /* ditto */
 716         .set_retrans_timeout    = xprt_set_retrans_timeout_def, /* ditto */
 717         .rpcbind                = rpcb_getport_async,   /* sunrpc/rpcb_clnt.c */
 718         .set_port               = xprt_rdma_set_port,
 719         .connect                = xprt_rdma_connect,
 720         .buf_alloc              = xprt_rdma_allocate,
 721         .buf_free               = xprt_rdma_free,
 722         .send_request           = xprt_rdma_send_request,
 723         .close                  = xprt_rdma_close,
 724         .destroy                = xprt_rdma_destroy,
 725         .print_stats            = xprt_rdma_print_stats
 726 };
 727
 728 static struct xprt_class xprt_rdma = {
 729         .list                   = LIST_HEAD_INIT(xprt_rdma.list),
 730         .name                   = "rdma",
 731         .owner                  = THIS_MODULE,
 732         .ident                  = XPRT_TRANSPORT_RDMA,
 733         .setup                  = xprt_setup_rdma,
 734 };
 735
 736 static void __exit xprt_rdma_cleanup(void)
 737 {
 738         int rc;
 739
 740         dprintk(KERN_INFO "RPCRDMA Module Removed, deregister RPC RDMA transport\n");
 741 #ifdef RPC_DEBUG
 742         if (sunrpc_table_header) {
 743                 unregister_sysctl_table(sunrpc_table_header);
 744                 sunrpc_table_header = NULL;
 745         }
 746 #endif
 747         rc = xprt_unregister_transport(&xprt_rdma);
 748         if (rc)
 749                 dprintk("RPC:       %s: xprt_unregister returned %i\n",
 750                         __func__, rc);
 751 }
 752
 753 static int __init xprt_rdma_init(void)
 754 {
 755         int rc;
 756
 757         rc = xprt_register_transport(&xprt_rdma);
 758
 759         if (rc)
 760                 return rc;
 761
 762         dprintk(KERN_INFO "RPCRDMA Module Init, register RPC RDMA transport\n");
 763
 764         dprintk(KERN_INFO "Defaults:\n");
 765         dprintk(KERN_INFO "\tSlots %d\n"
 766                 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
 767                 xprt_rdma_slot_table_entries,
 768                 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
 769         dprintk(KERN_INFO "\tPadding %d\n\tMemreg %d\n",
 770                 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
 771
 772 #ifdef RPC_DEBUG
 773         if (!sunrpc_table_header)
 774                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
 775 #endif
 776         return 0;
 777 }
 778
 779 module_init(xprt_rdma_init);
 780 module_exit(xprt_rdma_cleanup);