git.oblomov.eu Git - 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/seq_file.h>
  53
  54 #include "xprt_rdma.h"
  55
  56 #ifdef RPC_DEBUG
  57 # define RPCDBG_FACILITY        RPCDBG_TRANS
  58 #endif
  59
  60 MODULE_LICENSE("Dual BSD/GPL");
  61
  62 MODULE_DESCRIPTION("RPC/RDMA Transport for Linux kernel NFS");
  63 MODULE_AUTHOR("Network Appliance, Inc.");
  64
  65 /*
  66  * tunables
  67  */
  68
  69 static unsigned int xprt_rdma_slot_table_entries = RPCRDMA_DEF_SLOT_TABLE;
  70 static unsigned int xprt_rdma_max_inline_read = RPCRDMA_DEF_INLINE;
  71 static unsigned int xprt_rdma_max_inline_write = RPCRDMA_DEF_INLINE;
  72 static unsigned int xprt_rdma_inline_write_padding;
  73 #if !RPCRDMA_PERSISTENT_REGISTRATION
  74 static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_REGISTER; /* FMR? */
  75 #else
  76 static unsigned int xprt_rdma_memreg_strategy = RPCRDMA_ALLPHYSICAL;
  77 #endif
  78
  79 #ifdef RPC_DEBUG
  80
  81 static unsigned int min_slot_table_size = RPCRDMA_MIN_SLOT_TABLE;
  82 static unsigned int max_slot_table_size = RPCRDMA_MAX_SLOT_TABLE;
  83 static unsigned int zero;
  84 static unsigned int max_padding = PAGE_SIZE;
  85 static unsigned int min_memreg = RPCRDMA_BOUNCEBUFFERS;
  86 static unsigned int max_memreg = RPCRDMA_LAST - 1;
  87
  88 static struct ctl_table_header *sunrpc_table_header;
  89
  90 static ctl_table xr_tunables_table[] = {
  91         {
  92                 .ctl_name       = CTL_UNNUMBERED,
  93                 .procname       = "rdma_slot_table_entries",
  94                 .data           = &xprt_rdma_slot_table_entries,
  95                 .maxlen         = sizeof(unsigned int),
  96                 .mode           = 0644,
  97                 .proc_handler   = &proc_dointvec_minmax,
  98                 .strategy       = &sysctl_intvec,
  99                 .extra1         = &min_slot_table_size,
 100                 .extra2         = &max_slot_table_size
 101         },
 102         {
 103                 .ctl_name       = CTL_UNNUMBERED,
 104                 .procname       = "rdma_max_inline_read",
 105                 .data           = &xprt_rdma_max_inline_read,
 106                 .maxlen         = sizeof(unsigned int),
 107                 .mode           = 0644,
 108                 .proc_handler   = &proc_dointvec,
 109                 .strategy       = &sysctl_intvec,
 110         },
 111         {
 112                 .ctl_name       = CTL_UNNUMBERED,
 113                 .procname       = "rdma_max_inline_write",
 114                 .data           = &xprt_rdma_max_inline_write,
 115                 .maxlen         = sizeof(unsigned int),
 116                 .mode           = 0644,
 117                 .proc_handler   = &proc_dointvec,
 118                 .strategy       = &sysctl_intvec,
 119         },
 120         {
 121                 .ctl_name       = CTL_UNNUMBERED,
 122                 .procname       = "rdma_inline_write_padding",
 123                 .data           = &xprt_rdma_inline_write_padding,
 124                 .maxlen         = sizeof(unsigned int),
 125                 .mode           = 0644,
 126                 .proc_handler   = &proc_dointvec_minmax,
 127                 .strategy       = &sysctl_intvec,
 128                 .extra1         = &zero,
 129                 .extra2         = &max_padding,
 130         },
 131         {
 132                 .ctl_name       = CTL_UNNUMBERED,
 133                 .procname       = "rdma_memreg_strategy",
 134                 .data           = &xprt_rdma_memreg_strategy,
 135                 .maxlen         = sizeof(unsigned int),
 136                 .mode           = 0644,
 137                 .proc_handler   = &proc_dointvec_minmax,
 138                 .strategy       = &sysctl_intvec,
 139                 .extra1         = &min_memreg,
 140                 .extra2         = &max_memreg,
 141         },
 142         {
 143                 .ctl_name = 0,
 144         },
 145 };
 146
 147 static ctl_table sunrpc_table[] = {
 148         {
 149                 .ctl_name       = CTL_SUNRPC,
 150                 .procname       = "sunrpc",
 151                 .mode           = 0555,
 152                 .child          = xr_tunables_table
 153         },
 154         {
 155                 .ctl_name = 0,
 156         },
 157 };
 158
 159 #endif
 160
 161 static struct rpc_xprt_ops xprt_rdma_procs;     /* forward reference */
 162
 163 static void
 164 xprt_rdma_format_addresses(struct rpc_xprt *xprt)
 165 {
 166         struct sockaddr_in *addr = (struct sockaddr_in *)
 167                                         &rpcx_to_rdmad(xprt).addr;
 168         char *buf;
 169
 170         buf = kzalloc(20, GFP_KERNEL);
 171         if (buf)
 172                 snprintf(buf, 20, NIPQUAD_FMT, NIPQUAD(addr->sin_addr.s_addr));
 173         xprt->address_strings[RPC_DISPLAY_ADDR] = buf;
 174
 175         buf = kzalloc(8, GFP_KERNEL);
 176         if (buf)
 177                 snprintf(buf, 8, "%u", ntohs(addr->sin_port));
 178         xprt->address_strings[RPC_DISPLAY_PORT] = buf;
 179
 180         xprt->address_strings[RPC_DISPLAY_PROTO] = "rdma";
 181
 182         buf = kzalloc(48, GFP_KERNEL);
 183         if (buf)
 184                 snprintf(buf, 48, "addr="NIPQUAD_FMT" port=%u proto=%s",
 185                         NIPQUAD(addr->sin_addr.s_addr),
 186                         ntohs(addr->sin_port), "rdma");
 187         xprt->address_strings[RPC_DISPLAY_ALL] = buf;
 188
 189         buf = kzalloc(10, GFP_KERNEL);
 190         if (buf)
 191                 snprintf(buf, 10, "%02x%02x%02x%02x",
 192                         NIPQUAD(addr->sin_addr.s_addr));
 193         xprt->address_strings[RPC_DISPLAY_HEX_ADDR] = buf;
 194
 195         buf = kzalloc(8, GFP_KERNEL);
 196         if (buf)
 197                 snprintf(buf, 8, "%4hx", ntohs(addr->sin_port));
 198         xprt->address_strings[RPC_DISPLAY_HEX_PORT] = buf;
 199
 200         buf = kzalloc(30, GFP_KERNEL);
 201         if (buf)
 202                 snprintf(buf, 30, NIPQUAD_FMT".%u.%u",
 203                         NIPQUAD(addr->sin_addr.s_addr),
 204                         ntohs(addr->sin_port) >> 8,
 205                         ntohs(addr->sin_port) & 0xff);
 206         xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR] = buf;
 207
 208         /* netid */
 209         xprt->address_strings[RPC_DISPLAY_NETID] = "rdma";
 210 }
 211
 212 static void
 213 xprt_rdma_free_addresses(struct rpc_xprt *xprt)
 214 {
 215         kfree(xprt->address_strings[RPC_DISPLAY_ADDR]);
 216         kfree(xprt->address_strings[RPC_DISPLAY_PORT]);
 217         kfree(xprt->address_strings[RPC_DISPLAY_ALL]);
 218         kfree(xprt->address_strings[RPC_DISPLAY_HEX_ADDR]);
 219         kfree(xprt->address_strings[RPC_DISPLAY_HEX_PORT]);
 220         kfree(xprt->address_strings[RPC_DISPLAY_UNIVERSAL_ADDR]);
 221 }
 222
 223 static void
 224 xprt_rdma_connect_worker(struct work_struct *work)
 225 {
 226         struct rpcrdma_xprt *r_xprt =
 227                 container_of(work, struct rpcrdma_xprt, rdma_connect.work);
 228         struct rpc_xprt *xprt = &r_xprt->xprt;
 229         int rc = 0;
 230
 231         if (!xprt->shutdown) {
 232                 xprt_clear_connected(xprt);
 233
 234                 dprintk("RPC:       %s: %sconnect\n", __func__,
 235                                 r_xprt->rx_ep.rep_connected != 0 ? "re" : "");
 236                 rc = rpcrdma_ep_connect(&r_xprt->rx_ep, &r_xprt->rx_ia);
 237                 if (rc)
 238                         goto out;
 239         }
 240         goto out_clear;
 241
 242 out:
 243         xprt_wake_pending_tasks(xprt, rc);
 244
 245 out_clear:
 246         dprintk("RPC:       %s: exit\n", __func__);
 247         xprt_clear_connecting(xprt);
 248 }
 249
 250 /*
 251  * xprt_rdma_destroy
 252  *
 253  * Destroy the xprt.
 254  * Free all memory associated with the object, including its own.
 255  * NOTE: none of the *destroy methods free memory for their top-level
 256  * objects, even though they may have allocated it (they do free
 257  * private memory). It's up to the caller to handle it. In this
 258  * case (RDMA transport), all structure memory is inlined with the
 259  * struct rpcrdma_xprt.
 260  */
 261 static void
 262 xprt_rdma_destroy(struct rpc_xprt *xprt)
 263 {
 264         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 265         int rc;
 266
 267         dprintk("RPC:       %s: called\n", __func__);
 268
 269         cancel_delayed_work(&r_xprt->rdma_connect);
 270         flush_scheduled_work();
 271
 272         xprt_clear_connected(xprt);
 273
 274         rpcrdma_buffer_destroy(&r_xprt->rx_buf);
 275         rc = rpcrdma_ep_destroy(&r_xprt->rx_ep, &r_xprt->rx_ia);
 276         if (rc)
 277                 dprintk("RPC:       %s: rpcrdma_ep_destroy returned %i\n",
 278                         __func__, rc);
 279         rpcrdma_ia_close(&r_xprt->rx_ia);
 280
 281         xprt_rdma_free_addresses(xprt);
 282
 283         kfree(xprt->slot);
 284         xprt->slot = NULL;
 285         kfree(xprt);
 286
 287         dprintk("RPC:       %s: returning\n", __func__);
 288
 289         module_put(THIS_MODULE);
 290 }
 291
 292 /**
 293  * xprt_setup_rdma - Set up transport to use RDMA
 294  *
 295  * @args: rpc transport arguments
 296  */
 297 static struct rpc_xprt *
 298 xprt_setup_rdma(struct xprt_create *args)
 299 {
 300         struct rpcrdma_create_data_internal cdata;
 301         struct rpc_xprt *xprt;
 302         struct rpcrdma_xprt *new_xprt;
 303         struct rpcrdma_ep *new_ep;
 304         struct sockaddr_in *sin;
 305         int rc;
 306
 307         if (args->addrlen > sizeof(xprt->addr)) {
 308                 dprintk("RPC:       %s: address too large\n", __func__);
 309                 return ERR_PTR(-EBADF);
 310         }
 311
 312         xprt = kzalloc(sizeof(struct rpcrdma_xprt), GFP_KERNEL);
 313         if (xprt == NULL) {
 314                 dprintk("RPC:       %s: couldn't allocate rpcrdma_xprt\n",
 315                         __func__);
 316                 return ERR_PTR(-ENOMEM);
 317         }
 318
 319         xprt->max_reqs = xprt_rdma_slot_table_entries;
 320         xprt->slot = kcalloc(xprt->max_reqs,
 321                                 sizeof(struct rpc_rqst), GFP_KERNEL);
 322         if (xprt->slot == NULL) {
 323                 dprintk("RPC:       %s: couldn't allocate %d slots\n",
 324                         __func__, xprt->max_reqs);
 325                 kfree(xprt);
 326                 return ERR_PTR(-ENOMEM);
 327         }
 328
 329         /* 60 second timeout, no retries */
 330         xprt_set_timeout(&xprt->timeout, 0, 60UL * HZ);
 331         xprt->bind_timeout = (60U * HZ);
 332         xprt->connect_timeout = (60U * HZ);
 333         xprt->reestablish_timeout = (5U * HZ);
 334         xprt->idle_timeout = (5U * 60 * HZ);
 335
 336         xprt->resvport = 0;             /* privileged port not needed */
 337         xprt->tsh_size = 0;             /* RPC-RDMA handles framing */
 338         xprt->max_payload = RPCRDMA_MAX_DATA_SEGS * PAGE_SIZE;
 339         xprt->ops = &xprt_rdma_procs;
 340
 341         /*
 342          * Set up RDMA-specific connect data.
 343          */
 344
 345         /* Put server RDMA address in local cdata */
 346         memcpy(&cdata.addr, args->dstaddr, args->addrlen);
 347
 348         /* Ensure xprt->addr holds valid server TCP (not RDMA)
 349          * address, for any side protocols which peek at it */
 350         xprt->prot = IPPROTO_TCP;
 351         xprt->addrlen = args->addrlen;
 352         memcpy(&xprt->addr, &cdata.addr, xprt->addrlen);
 353
 354         sin = (struct sockaddr_in *)&cdata.addr;
 355         if (ntohs(sin->sin_port) != 0)
 356                 xprt_set_bound(xprt);
 357
 358         dprintk("RPC:       %s: %u.%u.%u.%u:%u\n", __func__,
 359                         NIPQUAD(sin->sin_addr.s_addr), ntohs(sin->sin_port));
 360
 361         /* Set max requests */
 362         cdata.max_requests = xprt->max_reqs;
 363
 364         /* Set some length limits */
 365         cdata.rsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA write max */
 366         cdata.wsize = RPCRDMA_MAX_SEGS * PAGE_SIZE; /* RDMA read max */
 367
 368         cdata.inline_wsize = xprt_rdma_max_inline_write;
 369         if (cdata.inline_wsize > cdata.wsize)
 370                 cdata.inline_wsize = cdata.wsize;
 371
 372         cdata.inline_rsize = xprt_rdma_max_inline_read;
 373         if (cdata.inline_rsize > cdata.rsize)
 374                 cdata.inline_rsize = cdata.rsize;
 375
 376         cdata.padding = xprt_rdma_inline_write_padding;
 377
 378         /*
 379          * Create new transport instance, which includes initialized
 380          *  o ia
 381          *  o endpoint
 382          *  o buffers
 383          */
 384
 385         new_xprt = rpcx_to_rdmax(xprt);
 386
 387         rc = rpcrdma_ia_open(new_xprt, (struct sockaddr *) &cdata.addr,
 388                                 xprt_rdma_memreg_strategy);
 389         if (rc)
 390                 goto out1;
 391
 392         /*
 393          * initialize and create ep
 394          */
 395         new_xprt->rx_data = cdata;
 396         new_ep = &new_xprt->rx_ep;
 397         new_ep->rep_remote_addr = cdata.addr;
 398
 399         rc = rpcrdma_ep_create(&new_xprt->rx_ep,
 400                                 &new_xprt->rx_ia, &new_xprt->rx_data);
 401         if (rc)
 402                 goto out2;
 403
 404         /*
 405          * Allocate pre-registered send and receive buffers for headers and
 406          * any inline data. Also specify any padding which will be provided
 407          * from a preregistered zero buffer.
 408          */
 409         rc = rpcrdma_buffer_create(&new_xprt->rx_buf, new_ep, &new_xprt->rx_ia,
 410                                 &new_xprt->rx_data);
 411         if (rc)
 412                 goto out3;
 413
 414         /*
 415          * Register a callback for connection events. This is necessary because
 416          * connection loss notification is async. We also catch connection loss
 417          * when reaping receives.
 418          */
 419         INIT_DELAYED_WORK(&new_xprt->rdma_connect, xprt_rdma_connect_worker);
 420         new_ep->rep_func = rpcrdma_conn_func;
 421         new_ep->rep_xprt = xprt;
 422
 423         xprt_rdma_format_addresses(xprt);
 424
 425         if (!try_module_get(THIS_MODULE))
 426                 goto out4;
 427
 428         return xprt;
 429
 430 out4:
 431         xprt_rdma_free_addresses(xprt);
 432         rc = -EINVAL;
 433 out3:
 434         (void) rpcrdma_ep_destroy(new_ep, &new_xprt->rx_ia);
 435 out2:
 436         rpcrdma_ia_close(&new_xprt->rx_ia);
 437 out1:
 438         kfree(xprt->slot);
 439         kfree(xprt);
 440         return ERR_PTR(rc);
 441 }
 442
 443 /*
 444  * Close a connection, during shutdown or timeout/reconnect
 445  */
 446 static void
 447 xprt_rdma_close(struct rpc_xprt *xprt)
 448 {
 449         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 450
 451         dprintk("RPC:       %s: closing\n", __func__);
 452         xprt_disconnect(xprt);
 453         (void) rpcrdma_ep_disconnect(&r_xprt->rx_ep, &r_xprt->rx_ia);
 454 }
 455
 456 static void
 457 xprt_rdma_set_port(struct rpc_xprt *xprt, u16 port)
 458 {
 459         struct sockaddr_in *sap;
 460
 461         sap = (struct sockaddr_in *)&xprt->addr;
 462         sap->sin_port = htons(port);
 463         sap = (struct sockaddr_in *)&rpcx_to_rdmad(xprt).addr;
 464         sap->sin_port = htons(port);
 465         dprintk("RPC:       %s: %u\n", __func__, port);
 466 }
 467
 468 static void
 469 xprt_rdma_connect(struct rpc_task *task)
 470 {
 471         struct rpc_xprt *xprt = (struct rpc_xprt *)task->tk_xprt;
 472         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 473
 474         if (!xprt_test_and_set_connecting(xprt)) {
 475                 if (r_xprt->rx_ep.rep_connected != 0) {
 476                         /* Reconnect */
 477                         schedule_delayed_work(&r_xprt->rdma_connect,
 478                                 xprt->reestablish_timeout);
 479                 } else {
 480                         schedule_delayed_work(&r_xprt->rdma_connect, 0);
 481                         if (!RPC_IS_ASYNC(task))
 482                                 flush_scheduled_work();
 483                 }
 484         }
 485 }
 486
 487 static int
 488 xprt_rdma_reserve_xprt(struct rpc_task *task)
 489 {
 490         struct rpc_xprt *xprt = task->tk_xprt;
 491         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 492         int credits = atomic_read(&r_xprt->rx_buf.rb_credits);
 493
 494         /* == RPC_CWNDSCALE @ init, but *after* setup */
 495         if (r_xprt->rx_buf.rb_cwndscale == 0UL) {
 496                 r_xprt->rx_buf.rb_cwndscale = xprt->cwnd;
 497                 dprintk("RPC:       %s: cwndscale %lu\n", __func__,
 498                         r_xprt->rx_buf.rb_cwndscale);
 499                 BUG_ON(r_xprt->rx_buf.rb_cwndscale <= 0);
 500         }
 501         xprt->cwnd = credits * r_xprt->rx_buf.rb_cwndscale;
 502         return xprt_reserve_xprt_cong(task);
 503 }
 504
 505 /*
 506  * The RDMA allocate/free functions need the task structure as a place
 507  * to hide the struct rpcrdma_req, which is necessary for the actual send/recv
 508  * sequence. For this reason, the recv buffers are attached to send
 509  * buffers for portions of the RPC. Note that the RPC layer allocates
 510  * both send and receive buffers in the same call. We may register
 511  * the receive buffer portion when using reply chunks.
 512  */
 513 static void *
 514 xprt_rdma_allocate(struct rpc_task *task, size_t size)
 515 {
 516         struct rpc_xprt *xprt = task->tk_xprt;
 517         struct rpcrdma_req *req, *nreq;
 518
 519         req = rpcrdma_buffer_get(&rpcx_to_rdmax(xprt)->rx_buf);
 520         BUG_ON(NULL == req);
 521
 522         if (size > req->rl_size) {
 523                 dprintk("RPC:       %s: size %zd too large for buffer[%zd]: "
 524                         "prog %d vers %d proc %d\n",
 525                         __func__, size, req->rl_size,
 526                         task->tk_client->cl_prog, task->tk_client->cl_vers,
 527                         task->tk_msg.rpc_proc->p_proc);
 528                 /*
 529                  * Outgoing length shortage. Our inline write max must have
 530                  * been configured to perform direct i/o.
 531                  *
 532                  * This is therefore a large metadata operation, and the
 533                  * allocate call was made on the maximum possible message,
 534                  * e.g. containing long filename(s) or symlink data. In
 535                  * fact, while these metadata operations *might* carry
 536                  * large outgoing payloads, they rarely *do*. However, we
 537                  * have to commit to the request here, so reallocate and
 538                  * register it now. The data path will never require this
 539                  * reallocation.
 540                  *
 541                  * If the allocation or registration fails, the RPC framework
 542                  * will (doggedly) retry.
 543                  */
 544                 if (rpcx_to_rdmax(xprt)->rx_ia.ri_memreg_strategy ==
 545                                 RPCRDMA_BOUNCEBUFFERS) {
 546                         /* forced to "pure inline" */
 547                         dprintk("RPC:       %s: too much data (%zd) for inline "
 548                                         "(r/w max %d/%d)\n", __func__, size,
 549                                         rpcx_to_rdmad(xprt).inline_rsize,
 550                                         rpcx_to_rdmad(xprt).inline_wsize);
 551                         size = req->rl_size;
 552                         rpc_exit(task, -EIO);           /* fail the operation */
 553                         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 554                         goto out;
 555                 }
 556                 if (task->tk_flags & RPC_TASK_SWAPPER)
 557                         nreq = kmalloc(sizeof *req + size, GFP_ATOMIC);
 558                 else
 559                         nreq = kmalloc(sizeof *req + size, GFP_NOFS);
 560                 if (nreq == NULL)
 561                         goto outfail;
 562
 563                 if (rpcrdma_register_internal(&rpcx_to_rdmax(xprt)->rx_ia,
 564                                 nreq->rl_base, size + sizeof(struct rpcrdma_req)
 565                                 - offsetof(struct rpcrdma_req, rl_base),
 566                                 &nreq->rl_handle, &nreq->rl_iov)) {
 567                         kfree(nreq);
 568                         goto outfail;
 569                 }
 570                 rpcx_to_rdmax(xprt)->rx_stats.hardway_register_count += size;
 571                 nreq->rl_size = size;
 572                 nreq->rl_niovs = 0;
 573                 nreq->rl_nchunks = 0;
 574                 nreq->rl_buffer = (struct rpcrdma_buffer *)req;
 575                 nreq->rl_reply = req->rl_reply;
 576                 memcpy(nreq->rl_segments,
 577                         req->rl_segments, sizeof nreq->rl_segments);
 578                 /* flag the swap with an unused field */
 579                 nreq->rl_iov.length = 0;
 580                 req->rl_reply = NULL;
 581                 req = nreq;
 582         }
 583         dprintk("RPC:       %s: size %zd, request 0x%p\n", __func__, size, req);
 584 out:
 585         return req->rl_xdr_buf;
 586
 587 outfail:
 588         rpcrdma_buffer_put(req);
 589         rpcx_to_rdmax(xprt)->rx_stats.failed_marshal_count++;
 590         return NULL;
 591 }
 592
 593 /*
 594  * This function returns all RDMA resources to the pool.
 595  */
 596 static void
 597 xprt_rdma_free(void *buffer)
 598 {
 599         struct rpcrdma_req *req;
 600         struct rpcrdma_xprt *r_xprt;
 601         struct rpcrdma_rep *rep;
 602         int i;
 603
 604         if (buffer == NULL)
 605                 return;
 606
 607         req = container_of(buffer, struct rpcrdma_req, rl_xdr_buf[0]);
 608         r_xprt = container_of(req->rl_buffer, struct rpcrdma_xprt, rx_buf);
 609         rep = req->rl_reply;
 610
 611         dprintk("RPC:       %s: called on 0x%p%s\n",
 612                 __func__, rep, (rep && rep->rr_func) ? " (with waiter)" : "");
 613
 614         /*
 615          * Finish the deregistration. When using mw bind, this was
 616          * begun in rpcrdma_reply_handler(). In all other modes, we
 617          * do it here, in thread context. The process is considered
 618          * complete when the rr_func vector becomes NULL - this
 619          * was put in place during rpcrdma_reply_handler() - the wait
 620          * call below will not block if the dereg is "done". If
 621          * interrupted, our framework will clean up.
 622          */
 623         for (i = 0; req->rl_nchunks;) {
 624                 --req->rl_nchunks;
 625                 i += rpcrdma_deregister_external(
 626                         &req->rl_segments[i], r_xprt, NULL);
 627         }
 628
 629         if (rep && wait_event_interruptible(rep->rr_unbind, !rep->rr_func)) {
 630                 rep->rr_func = NULL;    /* abandon the callback */
 631                 req->rl_reply = NULL;
 632         }
 633
 634         if (req->rl_iov.length == 0) {  /* see allocate above */
 635                 struct rpcrdma_req *oreq = (struct rpcrdma_req *)req->rl_buffer;
 636                 oreq->rl_reply = req->rl_reply;
 637                 (void) rpcrdma_deregister_internal(&r_xprt->rx_ia,
 638                                                    req->rl_handle,
 639                                                    &req->rl_iov);
 640                 kfree(req);
 641                 req = oreq;
 642         }
 643
 644         /* Put back request+reply buffers */
 645         rpcrdma_buffer_put(req);
 646 }
 647
 648 /*
 649  * send_request invokes the meat of RPC RDMA. It must do the following:
 650  *  1.  Marshal the RPC request into an RPC RDMA request, which means
 651  *      putting a header in front of data, and creating IOVs for RDMA
 652  *      from those in the request.
 653  *  2.  In marshaling, detect opportunities for RDMA, and use them.
 654  *  3.  Post a recv message to set up asynch completion, then send
 655  *      the request (rpcrdma_ep_post).
 656  *  4.  No partial sends are possible in the RPC-RDMA protocol (as in UDP).
 657  */
 658
 659 static int
 660 xprt_rdma_send_request(struct rpc_task *task)
 661 {
 662         struct rpc_rqst *rqst = task->tk_rqstp;
 663         struct rpc_xprt *xprt = task->tk_xprt;
 664         struct rpcrdma_req *req = rpcr_to_rdmar(rqst);
 665         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 666
 667         /* marshal the send itself */
 668         if (req->rl_niovs == 0 && rpcrdma_marshal_req(rqst) != 0) {
 669                 r_xprt->rx_stats.failed_marshal_count++;
 670                 dprintk("RPC:       %s: rpcrdma_marshal_req failed\n",
 671                         __func__);
 672                 return -EIO;
 673         }
 674
 675         if (req->rl_reply == NULL)              /* e.g. reconnection */
 676                 rpcrdma_recv_buffer_get(req);
 677
 678         if (req->rl_reply) {
 679                 req->rl_reply->rr_func = rpcrdma_reply_handler;
 680                 /* this need only be done once, but... */
 681                 req->rl_reply->rr_xprt = xprt;
 682         }
 683
 684         if (rpcrdma_ep_post(&r_xprt->rx_ia, &r_xprt->rx_ep, req)) {
 685                 xprt_disconnect(xprt);
 686                 return -ENOTCONN;       /* implies disconnect */
 687         }
 688
 689         rqst->rq_bytes_sent = 0;
 690         return 0;
 691 }
 692
 693 static void xprt_rdma_print_stats(struct rpc_xprt *xprt, struct seq_file *seq)
 694 {
 695         struct rpcrdma_xprt *r_xprt = rpcx_to_rdmax(xprt);
 696         long idle_time = 0;
 697
 698         if (xprt_connected(xprt))
 699                 idle_time = (long)(jiffies - xprt->last_used) / HZ;
 700
 701         seq_printf(seq,
 702           "\txprt:\trdma %u %lu %lu %lu %ld %lu %lu %lu %Lu %Lu "
 703           "%lu %lu %lu %Lu %Lu %Lu %Lu %lu %lu %lu\n",
 704
 705            0,   /* need a local port? */
 706            xprt->stat.bind_count,
 707            xprt->stat.connect_count,
 708            xprt->stat.connect_time,
 709            idle_time,
 710            xprt->stat.sends,
 711            xprt->stat.recvs,
 712            xprt->stat.bad_xids,
 713            xprt->stat.req_u,
 714            xprt->stat.bklog_u,
 715
 716            r_xprt->rx_stats.read_chunk_count,
 717            r_xprt->rx_stats.write_chunk_count,
 718            r_xprt->rx_stats.reply_chunk_count,
 719            r_xprt->rx_stats.total_rdma_request,
 720            r_xprt->rx_stats.total_rdma_reply,
 721            r_xprt->rx_stats.pullup_copy_count,
 722            r_xprt->rx_stats.fixup_copy_count,
 723            r_xprt->rx_stats.hardway_register_count,
 724            r_xprt->rx_stats.failed_marshal_count,
 725            r_xprt->rx_stats.bad_reply_count);
 726 }
 727
 728 /*
 729  * Plumbing for rpc transport switch and kernel module
 730  */
 731
 732 static struct rpc_xprt_ops xprt_rdma_procs = {
 733         .reserve_xprt           = xprt_rdma_reserve_xprt,
 734         .release_xprt           = xprt_release_xprt_cong, /* sunrpc/xprt.c */
 735         .release_request        = xprt_release_rqst_cong,       /* ditto */
 736         .set_retrans_timeout    = xprt_set_retrans_timeout_def, /* ditto */
 737         .rpcbind                = rpcb_getport_async,   /* sunrpc/rpcb_clnt.c */
 738         .set_port               = xprt_rdma_set_port,
 739         .connect                = xprt_rdma_connect,
 740         .buf_alloc              = xprt_rdma_allocate,
 741         .buf_free               = xprt_rdma_free,
 742         .send_request           = xprt_rdma_send_request,
 743         .close                  = xprt_rdma_close,
 744         .destroy                = xprt_rdma_destroy,
 745         .print_stats            = xprt_rdma_print_stats
 746 };
 747
 748 static struct xprt_class xprt_rdma = {
 749         .list                   = LIST_HEAD_INIT(xprt_rdma.list),
 750         .name                   = "rdma",
 751         .owner                  = THIS_MODULE,
 752         .ident                  = XPRT_TRANSPORT_RDMA,
 753         .setup                  = xprt_setup_rdma,
 754 };
 755
 756 static void __exit xprt_rdma_cleanup(void)
 757 {
 758         int rc;
 759
 760         dprintk("RPCRDMA Module Removed, deregister RPC RDMA transport\n");
 761 #ifdef RPC_DEBUG
 762         if (sunrpc_table_header) {
 763                 unregister_sysctl_table(sunrpc_table_header);
 764                 sunrpc_table_header = NULL;
 765         }
 766 #endif
 767         rc = xprt_unregister_transport(&xprt_rdma);
 768         if (rc)
 769                 dprintk("RPC:       %s: xprt_unregister returned %i\n",
 770                         __func__, rc);
 771 }
 772
 773 static int __init xprt_rdma_init(void)
 774 {
 775         int rc;
 776
 777         rc = xprt_register_transport(&xprt_rdma);
 778
 779         if (rc)
 780                 return rc;
 781
 782         dprintk(KERN_INFO "RPCRDMA Module Init, register RPC RDMA transport\n");
 783
 784         dprintk(KERN_INFO "Defaults:\n");
 785         dprintk(KERN_INFO "\tSlots %d\n"
 786                 "\tMaxInlineRead %d\n\tMaxInlineWrite %d\n",
 787                 xprt_rdma_slot_table_entries,
 788                 xprt_rdma_max_inline_read, xprt_rdma_max_inline_write);
 789         dprintk(KERN_INFO "\tPadding %d\n\tMemreg %d\n",
 790                 xprt_rdma_inline_write_padding, xprt_rdma_memreg_strategy);
 791
 792 #ifdef RPC_DEBUG
 793         if (!sunrpc_table_header)
 794                 sunrpc_table_header = register_sysctl_table(sunrpc_table);
 795 #endif
 796         return 0;
 797 }
 798
 799 module_init(xprt_rdma_init);
 800 module_exit(xprt_rdma_cleanup);