kernel/net/sunrpc/xprtrdma/svc_rdma_recvfrom.c

   1 /*
   2  * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
   3  * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved.
   4  *
   5  * This software is available to you under a choice of one of two
   6  * licenses.  You may choose to be licensed under the terms of the GNU
   7  * General Public License (GPL) Version 2, available from the file
   8  * COPYING in the main directory of this source tree, or the BSD-type
   9  * license below:
  10  *
  11  * Redistribution and use in source and binary forms, with or without
  12  * modification, are permitted provided that the following conditions
  13  * are met:
  14  *
  15  *      Redistributions of source code must retain the above copyright
  16  *      notice, this list of conditions and the following disclaimer.
  17  *
  18  *      Redistributions in binary form must reproduce the above
  19  *      copyright notice, this list of conditions and the following
  20  *      disclaimer in the documentation and/or other materials provided
  21  *      with the distribution.
  22  *
  23  *      Neither the name of the Network Appliance, Inc. nor the names of
  24  *      its contributors may be used to endorse or promote products
  25  *      derived from this software without specific prior written
  26  *      permission.
  27  *
  28  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  29  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  30  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
  31  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
  32  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
  33  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
  34  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
  35  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
  36  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
  37  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
  38  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  39  *
  40  * Author: Tom Tucker <tom@opengridcomputing.com>
  41  */
  42
  43 #include <linux/sunrpc/debug.h>
  44 #include <linux/sunrpc/rpc_rdma.h>
  45 #include <linux/spinlock.h>
  46 #include <asm/unaligned.h>
  47 #include <rdma/ib_verbs.h>
  48 #include <rdma/rdma_cm.h>
  49 #include <linux/sunrpc/svc_rdma.h>
  50
  51 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
  52
  53 /*
  54  * Replace the pages in the rq_argpages array with the pages from the SGE in
  55  * the RDMA_RECV completion. The SGL should contain full pages up until the
  56  * last one.
  57  */
  58 static void rdma_build_arg_xdr(struct svc_rqst *rqstp,
  59                                struct svc_rdma_op_ctxt *ctxt,
  60                                u32 byte_count)
  61 {
  62         struct rpcrdma_msg *rmsgp;
  63         struct page *page;
  64         u32 bc;
  65         int sge_no;
  66
  67         /* Swap the page in the SGE with the page in argpages */
  68         page = ctxt->pages[0];
  69         put_page(rqstp->rq_pages[0]);
  70         rqstp->rq_pages[0] = page;
  71
  72         /* Set up the XDR head */
  73         rqstp->rq_arg.head[0].iov_base = page_address(page);
  74         rqstp->rq_arg.head[0].iov_len =
  75                 min_t(size_t, byte_count, ctxt->sge[0].length);
  76         rqstp->rq_arg.len = byte_count;
  77         rqstp->rq_arg.buflen = byte_count;
  78
  79         /* Compute bytes past head in the SGL */
  80         bc = byte_count - rqstp->rq_arg.head[0].iov_len;
  81
  82         /* If data remains, store it in the pagelist */
  83         rqstp->rq_arg.page_len = bc;
  84         rqstp->rq_arg.page_base = 0;
  85
  86         /* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
  87         rmsgp = (struct rpcrdma_msg *)rqstp->rq_arg.head[0].iov_base;
  88         if (be32_to_cpu(rmsgp->rm_type) == RDMA_NOMSG)
  89                 rqstp->rq_arg.pages = &rqstp->rq_pages[0];
  90         else
  91                 rqstp->rq_arg.pages = &rqstp->rq_pages[1];
  92
  93         sge_no = 1;
  94         while (bc && sge_no < ctxt->count) {
  95                 page = ctxt->pages[sge_no];
  96                 put_page(rqstp->rq_pages[sge_no]);
  97                 rqstp->rq_pages[sge_no] = page;
  98                 bc -= min_t(u32, bc, ctxt->sge[sge_no].length);
  99                 rqstp->rq_arg.buflen += ctxt->sge[sge_no].length;
 100                 sge_no++;
 101         }
 102         rqstp->rq_respages = &rqstp->rq_pages[sge_no];
 103         rqstp->rq_next_page = rqstp->rq_respages + 1;
 104
 105         /* If not all pages were used from the SGL, free the remaining ones */
 106         bc = sge_no;
 107         while (sge_no < ctxt->count) {
 108                 page = ctxt->pages[sge_no++];
 109                 put_page(page);
 110         }
 111         ctxt->count = bc;
 112
 113         /* Set up tail */
 114         rqstp->rq_arg.tail[0].iov_base = NULL;
 115         rqstp->rq_arg.tail[0].iov_len = 0;
 116 }
 117
 118 static int rdma_read_max_sge(struct svcxprt_rdma *xprt, int sge_count)
 119 {
 120         if (rdma_node_get_transport(xprt->sc_cm_id->device->node_type) ==
 121              RDMA_TRANSPORT_IWARP)
 122                 return 1;
 123         else
 124                 return min_t(int, sge_count, xprt->sc_max_sge);
 125 }
 126
 127 /* Issue an RDMA_READ using the local lkey to map the data sink */
 128 int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt,
 129                         struct svc_rqst *rqstp,
 130                         struct svc_rdma_op_ctxt *head,
 131                         int *page_no,
 132                         u32 *page_offset,
 133                         u32 rs_handle,
 134                         u32 rs_length,
 135                         u64 rs_offset,
 136                         bool last)
 137 {
 138         struct ib_send_wr read_wr;
 139         int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
 140         struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
 141         int ret, read, pno;
 142         u32 pg_off = *page_offset;
 143         u32 pg_no = *page_no;
 144
 145         ctxt->direction = DMA_FROM_DEVICE;
 146         ctxt->read_hdr = head;
 147         pages_needed =
 148                 min_t(int, pages_needed, rdma_read_max_sge(xprt, pages_needed));
 149         read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
 150
 151         for (pno = 0; pno < pages_needed; pno++) {
 152                 int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
 153
 154                 head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
 155                 head->arg.page_len += len;
 156                 head->arg.len += len;
 157                 if (!pg_off)
 158                         head->count++;
 159                 rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
 160                 rqstp->rq_next_page = rqstp->rq_respages + 1;
 161                 ctxt->sge[pno].addr =
 162                         ib_dma_map_page(xprt->sc_cm_id->device,
 163                                         head->arg.pages[pg_no], pg_off,
 164                                         PAGE_SIZE - pg_off,
 165                                         DMA_FROM_DEVICE);
 166                 ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
 167                                            ctxt->sge[pno].addr);
 168                 if (ret)
 169                         goto err;
 170                 atomic_inc(&xprt->sc_dma_used);
 171
 172                 /* The lkey here is either a local dma lkey or a dma_mr lkey */
 173                 ctxt->sge[pno].lkey = xprt->sc_dma_lkey;
 174                 ctxt->sge[pno].length = len;
 175                 ctxt->count++;
 176
 177                 /* adjust offset and wrap to next page if needed */
 178                 pg_off += len;
 179                 if (pg_off == PAGE_SIZE) {
 180                         pg_off = 0;
 181                         pg_no++;
 182                 }
 183                 rs_length -= len;
 184         }
 185
 186         if (last && rs_length == 0)
 187                 set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
 188         else
 189                 clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
 190
 191         memset(&read_wr, 0, sizeof(read_wr));
 192         read_wr.wr_id = (unsigned long)ctxt;
 193         read_wr.opcode = IB_WR_RDMA_READ;
 194         ctxt->wr_op = read_wr.opcode;
 195         read_wr.send_flags = IB_SEND_SIGNALED;
 196         read_wr.wr.rdma.rkey = rs_handle;
 197         read_wr.wr.rdma.remote_addr = rs_offset;
 198         read_wr.sg_list = ctxt->sge;
 199         read_wr.num_sge = pages_needed;
 200
 201         ret = svc_rdma_send(xprt, &read_wr);
 202         if (ret) {
 203                 pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
 204                 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
 205                 goto err;
 206         }
 207
 208         /* return current location in page array */
 209         *page_no = pg_no;
 210         *page_offset = pg_off;
 211         ret = read;
 212         atomic_inc(&rdma_stat_read);
 213         return ret;
 214  err:
 215         svc_rdma_unmap_dma(ctxt);
 216         svc_rdma_put_context(ctxt, 0);
 217         return ret;
 218 }
 219
 220 /* Issue an RDMA_READ using an FRMR to map the data sink */
 221 int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt,
 222                          struct svc_rqst *rqstp,
 223                          struct svc_rdma_op_ctxt *head,
 224                          int *page_no,
 225                          u32 *page_offset,
 226                          u32 rs_handle,
 227                          u32 rs_length,
 228                          u64 rs_offset,
 229                          bool last)
 230 {
 231         struct ib_send_wr read_wr;
 232         struct ib_send_wr inv_wr;
 233         struct ib_send_wr fastreg_wr;
 234         u8 key;
 235         int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT;
 236         struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt);
 237         struct svc_rdma_fastreg_mr *frmr = svc_rdma_get_frmr(xprt);
 238         int ret, read, pno;
 239         u32 pg_off = *page_offset;
 240         u32 pg_no = *page_no;
 241
 242         if (IS_ERR(frmr))
 243                 return -ENOMEM;
 244
 245         ctxt->direction = DMA_FROM_DEVICE;
 246         ctxt->frmr = frmr;
 247         pages_needed = min_t(int, pages_needed, xprt->sc_frmr_pg_list_len);
 248         read = min_t(int, pages_needed << PAGE_SHIFT, rs_length);
 249
 250         frmr->kva = page_address(rqstp->rq_arg.pages[pg_no]);
 251         frmr->direction = DMA_FROM_DEVICE;
 252         frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE);
 253         frmr->map_len = pages_needed << PAGE_SHIFT;
 254         frmr->page_list_len = pages_needed;
 255
 256         for (pno = 0; pno < pages_needed; pno++) {
 257                 int len = min_t(int, rs_length, PAGE_SIZE - pg_off);
 258
 259                 head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no];
 260                 head->arg.page_len += len;
 261                 head->arg.len += len;
 262                 if (!pg_off)
 263                         head->count++;
 264                 rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1];
 265                 rqstp->rq_next_page = rqstp->rq_respages + 1;
 266                 frmr->page_list->page_list[pno] =
 267                         ib_dma_map_page(xprt->sc_cm_id->device,
 268                                         head->arg.pages[pg_no], 0,
 269                                         PAGE_SIZE, DMA_FROM_DEVICE);
 270                 ret = ib_dma_mapping_error(xprt->sc_cm_id->device,
 271                                            frmr->page_list->page_list[pno]);
 272                 if (ret)
 273                         goto err;
 274                 atomic_inc(&xprt->sc_dma_used);
 275
 276                 /* adjust offset and wrap to next page if needed */
 277                 pg_off += len;
 278                 if (pg_off == PAGE_SIZE) {
 279                         pg_off = 0;
 280                         pg_no++;
 281                 }
 282                 rs_length -= len;
 283         }
 284
 285         if (last && rs_length == 0)
 286                 set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
 287         else
 288                 clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags);
 289
 290         /* Bump the key */
 291         key = (u8)(frmr->mr->lkey & 0x000000FF);
 292         ib_update_fast_reg_key(frmr->mr, ++key);
 293
 294         ctxt->sge[0].addr = (unsigned long)frmr->kva + *page_offset;
 295         ctxt->sge[0].lkey = frmr->mr->lkey;
 296         ctxt->sge[0].length = read;
 297         ctxt->count = 1;
 298         ctxt->read_hdr = head;
 299
 300         /* Prepare FASTREG WR */
 301         memset(&fastreg_wr, 0, sizeof(fastreg_wr));
 302         fastreg_wr.opcode = IB_WR_FAST_REG_MR;
 303         fastreg_wr.send_flags = IB_SEND_SIGNALED;
 304         fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva;
 305         fastreg_wr.wr.fast_reg.page_list = frmr->page_list;
 306         fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len;
 307         fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT;
 308         fastreg_wr.wr.fast_reg.length = frmr->map_len;
 309         fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags;
 310         fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey;
 311         fastreg_wr.next = &read_wr;
 312
 313         /* Prepare RDMA_READ */
 314         memset(&read_wr, 0, sizeof(read_wr));
 315         read_wr.send_flags = IB_SEND_SIGNALED;
 316         read_wr.wr.rdma.rkey = rs_handle;
 317         read_wr.wr.rdma.remote_addr = rs_offset;
 318         read_wr.sg_list = ctxt->sge;
 319         read_wr.num_sge = 1;
 320         if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) {
 321                 read_wr.opcode = IB_WR_RDMA_READ_WITH_INV;
 322                 read_wr.wr_id = (unsigned long)ctxt;
 323                 read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey;
 324         } else {
 325                 read_wr.opcode = IB_WR_RDMA_READ;
 326                 read_wr.next = &inv_wr;
 327                 /* Prepare invalidate */
 328                 memset(&inv_wr, 0, sizeof(inv_wr));
 329                 inv_wr.wr_id = (unsigned long)ctxt;
 330                 inv_wr.opcode = IB_WR_LOCAL_INV;
 331                 inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE;
 332                 inv_wr.ex.invalidate_rkey = frmr->mr->lkey;
 333         }
 334         ctxt->wr_op = read_wr.opcode;
 335
 336         /* Post the chain */
 337         ret = svc_rdma_send(xprt, &fastreg_wr);
 338         if (ret) {
 339                 pr_err("svcrdma: Error %d posting RDMA_READ\n", ret);
 340                 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
 341                 goto err;
 342         }
 343
 344         /* return current location in page array */
 345         *page_no = pg_no;
 346         *page_offset = pg_off;
 347         ret = read;
 348         atomic_inc(&rdma_stat_read);
 349         return ret;
 350  err:
 351         svc_rdma_unmap_dma(ctxt);
 352         svc_rdma_put_context(ctxt, 0);
 353         svc_rdma_put_frmr(xprt, frmr);
 354         return ret;
 355 }
 356
 357 static unsigned int
 358 rdma_rcl_chunk_count(struct rpcrdma_read_chunk *ch)
 359 {
 360         unsigned int count;
 361
 362         for (count = 0; ch->rc_discrim != xdr_zero; ch++)
 363                 count++;
 364         return count;
 365 }
 366
 367 /* If there was additional inline content, append it to the end of arg.pages.
 368  * Tail copy has to be done after the reader function has determined how many
 369  * pages are needed for RDMA READ.
 370  */
 371 static int
 372 rdma_copy_tail(struct svc_rqst *rqstp, struct svc_rdma_op_ctxt *head,
 373                u32 position, u32 byte_count, u32 page_offset, int page_no)
 374 {
 375         char *srcp, *destp;
 376         int ret;
 377
 378         ret = 0;
 379         srcp = head->arg.head[0].iov_base + position;
 380         byte_count = head->arg.head[0].iov_len - position;
 381         if (byte_count > PAGE_SIZE) {
 382                 dprintk("svcrdma: large tail unsupported\n");
 383                 return 0;
 384         }
 385
 386         /* Fit as much of the tail on the current page as possible */
 387         if (page_offset != PAGE_SIZE) {
 388                 destp = page_address(rqstp->rq_arg.pages[page_no]);
 389                 destp += page_offset;
 390                 while (byte_count--) {
 391                         *destp++ = *srcp++;
 392                         page_offset++;
 393                         if (page_offset == PAGE_SIZE && byte_count)
 394                                 goto more;
 395                 }
 396                 goto done;
 397         }
 398
 399 more:
 400         /* Fit the rest on the next page */
 401         page_no++;
 402         destp = page_address(rqstp->rq_arg.pages[page_no]);
 403         while (byte_count--)
 404                 *destp++ = *srcp++;
 405
 406         rqstp->rq_respages = &rqstp->rq_arg.pages[page_no+1];
 407         rqstp->rq_next_page = rqstp->rq_respages + 1;
 408
 409 done:
 410         byte_count = head->arg.head[0].iov_len - position;
 411         head->arg.page_len += byte_count;
 412         head->arg.len += byte_count;
 413         head->arg.buflen += byte_count;
 414         return 1;
 415 }
 416
 417 static int rdma_read_chunks(struct svcxprt_rdma *xprt,
 418                             struct rpcrdma_msg *rmsgp,
 419                             struct svc_rqst *rqstp,
 420                             struct svc_rdma_op_ctxt *head)
 421 {
 422         int page_no, ret;
 423         struct rpcrdma_read_chunk *ch;
 424         u32 handle, page_offset, byte_count;
 425         u32 position;
 426         u64 rs_offset;
 427         bool last;
 428
 429         /* If no read list is present, return 0 */
 430         ch = svc_rdma_get_read_chunk(rmsgp);
 431         if (!ch)
 432                 return 0;
 433
 434         if (rdma_rcl_chunk_count(ch) > RPCSVC_MAXPAGES)
 435                 return -EINVAL;
 436
 437         /* The request is completed when the RDMA_READs complete. The
 438          * head context keeps all the pages that comprise the
 439          * request.
 440          */
 441         head->arg.head[0] = rqstp->rq_arg.head[0];
 442         head->arg.tail[0] = rqstp->rq_arg.tail[0];
 443         head->hdr_count = head->count;
 444         head->arg.page_base = 0;
 445         head->arg.page_len = 0;
 446         head->arg.len = rqstp->rq_arg.len;
 447         head->arg.buflen = rqstp->rq_arg.buflen;
 448
 449         ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0];
 450         position = be32_to_cpu(ch->rc_position);
 451
 452         /* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */
 453         if (position == 0) {
 454                 head->arg.pages = &head->pages[0];
 455                 page_offset = head->byte_len;
 456         } else {
 457                 head->arg.pages = &head->pages[head->count];
 458                 page_offset = 0;
 459         }
 460
 461         ret = 0;
 462         page_no = 0;
 463         for (; ch->rc_discrim != xdr_zero; ch++) {
 464                 if (be32_to_cpu(ch->rc_position) != position)
 465                         goto err;
 466
 467                 handle = be32_to_cpu(ch->rc_target.rs_handle),
 468                 byte_count = be32_to_cpu(ch->rc_target.rs_length);
 469                 xdr_decode_hyper((__be32 *)&ch->rc_target.rs_offset,
 470                                  &rs_offset);
 471
 472                 while (byte_count > 0) {
 473                         last = (ch + 1)->rc_discrim == xdr_zero;
 474                         ret = xprt->sc_reader(xprt, rqstp, head,
 475                                               &page_no, &page_offset,
 476                                               handle, byte_count,
 477                                               rs_offset, last);
 478                         if (ret < 0)
 479                                 goto err;
 480                         byte_count -= ret;
 481                         rs_offset += ret;
 482                         head->arg.buflen += ret;
 483                 }
 484         }
 485
 486         /* Read list may need XDR round-up (see RFC 5666, s. 3.7) */
 487         if (page_offset & 3) {
 488                 u32 pad = 4 - (page_offset & 3);
 489
 490                 head->arg.page_len += pad;
 491                 head->arg.len += pad;
 492                 head->arg.buflen += pad;
 493                 page_offset += pad;
 494         }
 495
 496         ret = 1;
 497         if (position && position < head->arg.head[0].iov_len)
 498                 ret = rdma_copy_tail(rqstp, head, position,
 499                                      byte_count, page_offset, page_no);
 500         head->arg.head[0].iov_len = position;
 501         head->position = position;
 502
 503  err:
 504         /* Detach arg pages. svc_recv will replenish them */
 505         for (page_no = 0;
 506              &rqstp->rq_pages[page_no] < rqstp->rq_respages; page_no++)
 507                 rqstp->rq_pages[page_no] = NULL;
 508
 509         return ret;
 510 }
 511
 512 static int rdma_read_complete(struct svc_rqst *rqstp,
 513                               struct svc_rdma_op_ctxt *head)
 514 {
 515         int page_no;
 516         int ret;
 517
 518         /* Copy RPC pages */
 519         for (page_no = 0; page_no < head->count; page_no++) {
 520                 put_page(rqstp->rq_pages[page_no]);
 521                 rqstp->rq_pages[page_no] = head->pages[page_no];
 522         }
 523
 524         /* Adjustments made for RDMA_NOMSG type requests */
 525         if (head->position == 0) {
 526                 if (head->arg.len <= head->sge[0].length) {
 527                         head->arg.head[0].iov_len = head->arg.len -
 528                                                         head->byte_len;
 529                         head->arg.page_len = 0;
 530                 } else {
 531                         head->arg.head[0].iov_len = head->sge[0].length -
 532                                                                 head->byte_len;
 533                         head->arg.page_len = head->arg.len -
 534                                                 head->sge[0].length;
 535                 }
 536         }
 537
 538         /* Point rq_arg.pages past header */
 539         rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count];
 540         rqstp->rq_arg.page_len = head->arg.page_len;
 541         rqstp->rq_arg.page_base = head->arg.page_base;
 542
 543         /* rq_respages starts after the last arg page */
 544         rqstp->rq_respages = &rqstp->rq_arg.pages[page_no];
 545         rqstp->rq_next_page = rqstp->rq_respages + 1;
 546
 547         /* Rebuild rq_arg head and tail. */
 548         rqstp->rq_arg.head[0] = head->arg.head[0];
 549         rqstp->rq_arg.tail[0] = head->arg.tail[0];
 550         rqstp->rq_arg.len = head->arg.len;
 551         rqstp->rq_arg.buflen = head->arg.buflen;
 552
 553         /* Free the context */
 554         svc_rdma_put_context(head, 0);
 555
 556         /* XXX: What should this be? */
 557         rqstp->rq_prot = IPPROTO_MAX;
 558         svc_xprt_copy_addrs(rqstp, rqstp->rq_xprt);
 559
 560         ret = rqstp->rq_arg.head[0].iov_len
 561                 + rqstp->rq_arg.page_len
 562                 + rqstp->rq_arg.tail[0].iov_len;
 563         dprintk("svcrdma: deferred read ret=%d, rq_arg.len=%u, "
 564                 "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zu\n",
 565                 ret, rqstp->rq_arg.len, rqstp->rq_arg.head[0].iov_base,
 566                 rqstp->rq_arg.head[0].iov_len);
 567
 568         return ret;
 569 }
 570
 571 /*
 572  * Set up the rqstp thread context to point to the RQ buffer. If
 573  * necessary, pull additional data from the client with an RDMA_READ
 574  * request.
 575  */
 576 int svc_rdma_recvfrom(struct svc_rqst *rqstp)
 577 {
 578         struct svc_xprt *xprt = rqstp->rq_xprt;
 579         struct svcxprt_rdma *rdma_xprt =
 580                 container_of(xprt, struct svcxprt_rdma, sc_xprt);
 581         struct svc_rdma_op_ctxt *ctxt = NULL;
 582         struct rpcrdma_msg *rmsgp;
 583         int ret = 0;
 584         int len;
 585
 586         dprintk("svcrdma: rqstp=%p\n", rqstp);
 587
 588         spin_lock_bh(&rdma_xprt->sc_rq_dto_lock);
 589         if (!list_empty(&rdma_xprt->sc_read_complete_q)) {
 590                 ctxt = list_entry(rdma_xprt->sc_read_complete_q.next,
 591                                   struct svc_rdma_op_ctxt,
 592                                   dto_q);
 593                 list_del_init(&ctxt->dto_q);
 594                 spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
 595                 return rdma_read_complete(rqstp, ctxt);
 596         } else if (!list_empty(&rdma_xprt->sc_rq_dto_q)) {
 597                 ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next,
 598                                   struct svc_rdma_op_ctxt,
 599                                   dto_q);
 600                 list_del_init(&ctxt->dto_q);
 601         } else {
 602                 atomic_inc(&rdma_stat_rq_starve);
 603                 clear_bit(XPT_DATA, &xprt->xpt_flags);
 604                 ctxt = NULL;
 605         }
 606         spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock);
 607         if (!ctxt) {
 608                 /* This is the EAGAIN path. The svc_recv routine will
 609                  * return -EAGAIN, the nfsd thread will go to call into
 610                  * svc_recv again and we shouldn't be on the active
 611                  * transport list
 612                  */
 613                 if (test_bit(XPT_CLOSE, &xprt->xpt_flags))
 614                         goto close_out;
 615
 616                 goto out;
 617         }
 618         dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n",
 619                 ctxt, rdma_xprt, rqstp, ctxt->wc_status);
 620         atomic_inc(&rdma_stat_recv);
 621
 622         /* Build up the XDR from the receive buffers. */
 623         rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len);
 624
 625         /* Decode the RDMA header. */
 626         len = svc_rdma_xdr_decode_req(&rmsgp, rqstp);
 627         rqstp->rq_xprt_hlen = len;
 628
 629         /* If the request is invalid, reply with an error */
 630         if (len < 0) {
 631                 if (len == -ENOSYS)
 632                         svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS);
 633                 goto close_out;
 634         }
 635
 636         /* Read read-list data. */
 637         ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt);
 638         if (ret > 0) {
 639                 /* read-list posted, defer until data received from client. */
 640                 goto defer;
 641         } else if (ret < 0) {
 642                 /* Post of read-list failed, free context. */
 643                 svc_rdma_put_context(ctxt, 1);
 644                 return 0;
 645         }
 646
 647         ret = rqstp->rq_arg.head[0].iov_len
 648                 + rqstp->rq_arg.page_len
 649                 + rqstp->rq_arg.tail[0].iov_len;
 650         svc_rdma_put_context(ctxt, 0);
 651  out:
 652         dprintk("svcrdma: ret=%d, rq_arg.len=%u, "
 653                 "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zd\n",
 654                 ret, rqstp->rq_arg.len,
 655                 rqstp->rq_arg.head[0].iov_base,
 656                 rqstp->rq_arg.head[0].iov_len);
 657         rqstp->rq_prot = IPPROTO_MAX;
 658         svc_xprt_copy_addrs(rqstp, xprt);
 659         return ret;
 660
 661  close_out:
 662         if (ctxt)
 663                 svc_rdma_put_context(ctxt, 1);
 664         dprintk("svcrdma: transport %p is closing\n", xprt);
 665         /*
 666          * Set the close bit and enqueue it. svc_recv will see the
 667          * close bit and call svc_xprt_delete
 668          */
 669         set_bit(XPT_CLOSE, &xprt->xpt_flags);
 670 defer:
 671         return 0;
 672 }