| 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 (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 | /* Issue an RDMA_READ using the local lkey to map the data sink */ |
| 119 | int rdma_read_chunk_lcl(struct svcxprt_rdma *xprt, |
| 120 | struct svc_rqst *rqstp, |
| 121 | struct svc_rdma_op_ctxt *head, |
| 122 | int *page_no, |
| 123 | u32 *page_offset, |
| 124 | u32 rs_handle, |
| 125 | u32 rs_length, |
| 126 | u64 rs_offset, |
| 127 | bool last) |
| 128 | { |
| 129 | struct ib_send_wr read_wr; |
| 130 | int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT; |
| 131 | struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt); |
| 132 | int ret, read, pno; |
| 133 | u32 pg_off = *page_offset; |
| 134 | u32 pg_no = *page_no; |
| 135 | |
| 136 | ctxt->direction = DMA_FROM_DEVICE; |
| 137 | ctxt->read_hdr = head; |
| 138 | pages_needed = min_t(int, pages_needed, xprt->sc_max_sge_rd); |
| 139 | read = min_t(int, (pages_needed << PAGE_SHIFT) - *page_offset, |
| 140 | rs_length); |
| 141 | |
| 142 | for (pno = 0; pno < pages_needed; pno++) { |
| 143 | int len = min_t(int, rs_length, PAGE_SIZE - pg_off); |
| 144 | |
| 145 | head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no]; |
| 146 | head->arg.page_len += len; |
| 147 | head->arg.len += len; |
| 148 | if (!pg_off) |
| 149 | head->count++; |
| 150 | rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1]; |
| 151 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
| 152 | ctxt->sge[pno].addr = |
| 153 | ib_dma_map_page(xprt->sc_cm_id->device, |
| 154 | head->arg.pages[pg_no], pg_off, |
| 155 | PAGE_SIZE - pg_off, |
| 156 | DMA_FROM_DEVICE); |
| 157 | ret = ib_dma_mapping_error(xprt->sc_cm_id->device, |
| 158 | ctxt->sge[pno].addr); |
| 159 | if (ret) |
| 160 | goto err; |
| 161 | atomic_inc(&xprt->sc_dma_used); |
| 162 | |
| 163 | /* The lkey here is either a local dma lkey or a dma_mr lkey */ |
| 164 | ctxt->sge[pno].lkey = xprt->sc_dma_lkey; |
| 165 | ctxt->sge[pno].length = len; |
| 166 | ctxt->count++; |
| 167 | |
| 168 | /* adjust offset and wrap to next page if needed */ |
| 169 | pg_off += len; |
| 170 | if (pg_off == PAGE_SIZE) { |
| 171 | pg_off = 0; |
| 172 | pg_no++; |
| 173 | } |
| 174 | rs_length -= len; |
| 175 | } |
| 176 | |
| 177 | if (last && rs_length == 0) |
| 178 | set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); |
| 179 | else |
| 180 | clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); |
| 181 | |
| 182 | memset(&read_wr, 0, sizeof(read_wr)); |
| 183 | read_wr.wr_id = (unsigned long)ctxt; |
| 184 | read_wr.opcode = IB_WR_RDMA_READ; |
| 185 | ctxt->wr_op = read_wr.opcode; |
| 186 | read_wr.send_flags = IB_SEND_SIGNALED; |
| 187 | read_wr.wr.rdma.rkey = rs_handle; |
| 188 | read_wr.wr.rdma.remote_addr = rs_offset; |
| 189 | read_wr.sg_list = ctxt->sge; |
| 190 | read_wr.num_sge = pages_needed; |
| 191 | |
| 192 | ret = svc_rdma_send(xprt, &read_wr); |
| 193 | if (ret) { |
| 194 | pr_err("svcrdma: Error %d posting RDMA_READ\n", ret); |
| 195 | set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); |
| 196 | goto err; |
| 197 | } |
| 198 | |
| 199 | /* return current location in page array */ |
| 200 | *page_no = pg_no; |
| 201 | *page_offset = pg_off; |
| 202 | ret = read; |
| 203 | atomic_inc(&rdma_stat_read); |
| 204 | return ret; |
| 205 | err: |
| 206 | svc_rdma_unmap_dma(ctxt); |
| 207 | svc_rdma_put_context(ctxt, 0); |
| 208 | return ret; |
| 209 | } |
| 210 | |
| 211 | /* Issue an RDMA_READ using an FRMR to map the data sink */ |
| 212 | int rdma_read_chunk_frmr(struct svcxprt_rdma *xprt, |
| 213 | struct svc_rqst *rqstp, |
| 214 | struct svc_rdma_op_ctxt *head, |
| 215 | int *page_no, |
| 216 | u32 *page_offset, |
| 217 | u32 rs_handle, |
| 218 | u32 rs_length, |
| 219 | u64 rs_offset, |
| 220 | bool last) |
| 221 | { |
| 222 | struct ib_send_wr read_wr; |
| 223 | struct ib_send_wr inv_wr; |
| 224 | struct ib_send_wr fastreg_wr; |
| 225 | u8 key; |
| 226 | int pages_needed = PAGE_ALIGN(*page_offset + rs_length) >> PAGE_SHIFT; |
| 227 | struct svc_rdma_op_ctxt *ctxt = svc_rdma_get_context(xprt); |
| 228 | struct svc_rdma_fastreg_mr *frmr = svc_rdma_get_frmr(xprt); |
| 229 | int ret, read, pno; |
| 230 | u32 pg_off = *page_offset; |
| 231 | u32 pg_no = *page_no; |
| 232 | |
| 233 | if (IS_ERR(frmr)) |
| 234 | return -ENOMEM; |
| 235 | |
| 236 | ctxt->direction = DMA_FROM_DEVICE; |
| 237 | ctxt->frmr = frmr; |
| 238 | pages_needed = min_t(int, pages_needed, xprt->sc_frmr_pg_list_len); |
| 239 | read = min_t(int, (pages_needed << PAGE_SHIFT) - *page_offset, |
| 240 | rs_length); |
| 241 | |
| 242 | frmr->kva = page_address(rqstp->rq_arg.pages[pg_no]); |
| 243 | frmr->direction = DMA_FROM_DEVICE; |
| 244 | frmr->access_flags = (IB_ACCESS_LOCAL_WRITE|IB_ACCESS_REMOTE_WRITE); |
| 245 | frmr->map_len = pages_needed << PAGE_SHIFT; |
| 246 | frmr->page_list_len = pages_needed; |
| 247 | |
| 248 | for (pno = 0; pno < pages_needed; pno++) { |
| 249 | int len = min_t(int, rs_length, PAGE_SIZE - pg_off); |
| 250 | |
| 251 | head->arg.pages[pg_no] = rqstp->rq_arg.pages[pg_no]; |
| 252 | head->arg.page_len += len; |
| 253 | head->arg.len += len; |
| 254 | if (!pg_off) |
| 255 | head->count++; |
| 256 | rqstp->rq_respages = &rqstp->rq_arg.pages[pg_no+1]; |
| 257 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
| 258 | frmr->page_list->page_list[pno] = |
| 259 | ib_dma_map_page(xprt->sc_cm_id->device, |
| 260 | head->arg.pages[pg_no], 0, |
| 261 | PAGE_SIZE, DMA_FROM_DEVICE); |
| 262 | ret = ib_dma_mapping_error(xprt->sc_cm_id->device, |
| 263 | frmr->page_list->page_list[pno]); |
| 264 | if (ret) |
| 265 | goto err; |
| 266 | atomic_inc(&xprt->sc_dma_used); |
| 267 | |
| 268 | /* adjust offset and wrap to next page if needed */ |
| 269 | pg_off += len; |
| 270 | if (pg_off == PAGE_SIZE) { |
| 271 | pg_off = 0; |
| 272 | pg_no++; |
| 273 | } |
| 274 | rs_length -= len; |
| 275 | } |
| 276 | |
| 277 | if (last && rs_length == 0) |
| 278 | set_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); |
| 279 | else |
| 280 | clear_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags); |
| 281 | |
| 282 | /* Bump the key */ |
| 283 | key = (u8)(frmr->mr->lkey & 0x000000FF); |
| 284 | ib_update_fast_reg_key(frmr->mr, ++key); |
| 285 | |
| 286 | ctxt->sge[0].addr = (unsigned long)frmr->kva + *page_offset; |
| 287 | ctxt->sge[0].lkey = frmr->mr->lkey; |
| 288 | ctxt->sge[0].length = read; |
| 289 | ctxt->count = 1; |
| 290 | ctxt->read_hdr = head; |
| 291 | |
| 292 | /* Prepare FASTREG WR */ |
| 293 | memset(&fastreg_wr, 0, sizeof(fastreg_wr)); |
| 294 | fastreg_wr.opcode = IB_WR_FAST_REG_MR; |
| 295 | fastreg_wr.send_flags = IB_SEND_SIGNALED; |
| 296 | fastreg_wr.wr.fast_reg.iova_start = (unsigned long)frmr->kva; |
| 297 | fastreg_wr.wr.fast_reg.page_list = frmr->page_list; |
| 298 | fastreg_wr.wr.fast_reg.page_list_len = frmr->page_list_len; |
| 299 | fastreg_wr.wr.fast_reg.page_shift = PAGE_SHIFT; |
| 300 | fastreg_wr.wr.fast_reg.length = frmr->map_len; |
| 301 | fastreg_wr.wr.fast_reg.access_flags = frmr->access_flags; |
| 302 | fastreg_wr.wr.fast_reg.rkey = frmr->mr->lkey; |
| 303 | fastreg_wr.next = &read_wr; |
| 304 | |
| 305 | /* Prepare RDMA_READ */ |
| 306 | memset(&read_wr, 0, sizeof(read_wr)); |
| 307 | read_wr.send_flags = IB_SEND_SIGNALED; |
| 308 | read_wr.wr.rdma.rkey = rs_handle; |
| 309 | read_wr.wr.rdma.remote_addr = rs_offset; |
| 310 | read_wr.sg_list = ctxt->sge; |
| 311 | read_wr.num_sge = 1; |
| 312 | if (xprt->sc_dev_caps & SVCRDMA_DEVCAP_READ_W_INV) { |
| 313 | read_wr.opcode = IB_WR_RDMA_READ_WITH_INV; |
| 314 | read_wr.wr_id = (unsigned long)ctxt; |
| 315 | read_wr.ex.invalidate_rkey = ctxt->frmr->mr->lkey; |
| 316 | } else { |
| 317 | read_wr.opcode = IB_WR_RDMA_READ; |
| 318 | read_wr.next = &inv_wr; |
| 319 | /* Prepare invalidate */ |
| 320 | memset(&inv_wr, 0, sizeof(inv_wr)); |
| 321 | inv_wr.wr_id = (unsigned long)ctxt; |
| 322 | inv_wr.opcode = IB_WR_LOCAL_INV; |
| 323 | inv_wr.send_flags = IB_SEND_SIGNALED | IB_SEND_FENCE; |
| 324 | inv_wr.ex.invalidate_rkey = frmr->mr->lkey; |
| 325 | } |
| 326 | ctxt->wr_op = read_wr.opcode; |
| 327 | |
| 328 | /* Post the chain */ |
| 329 | ret = svc_rdma_send(xprt, &fastreg_wr); |
| 330 | if (ret) { |
| 331 | pr_err("svcrdma: Error %d posting RDMA_READ\n", ret); |
| 332 | set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags); |
| 333 | goto err; |
| 334 | } |
| 335 | |
| 336 | /* return current location in page array */ |
| 337 | *page_no = pg_no; |
| 338 | *page_offset = pg_off; |
| 339 | ret = read; |
| 340 | atomic_inc(&rdma_stat_read); |
| 341 | return ret; |
| 342 | err: |
| 343 | svc_rdma_unmap_dma(ctxt); |
| 344 | svc_rdma_put_context(ctxt, 0); |
| 345 | svc_rdma_put_frmr(xprt, frmr); |
| 346 | return ret; |
| 347 | } |
| 348 | |
| 349 | static unsigned int |
| 350 | rdma_rcl_chunk_count(struct rpcrdma_read_chunk *ch) |
| 351 | { |
| 352 | unsigned int count; |
| 353 | |
| 354 | for (count = 0; ch->rc_discrim != xdr_zero; ch++) |
| 355 | count++; |
| 356 | return count; |
| 357 | } |
| 358 | |
| 359 | /* If there was additional inline content, append it to the end of arg.pages. |
| 360 | * Tail copy has to be done after the reader function has determined how many |
| 361 | * pages are needed for RDMA READ. |
| 362 | */ |
| 363 | static int |
| 364 | rdma_copy_tail(struct svc_rqst *rqstp, struct svc_rdma_op_ctxt *head, |
| 365 | u32 position, u32 byte_count, u32 page_offset, int page_no) |
| 366 | { |
| 367 | char *srcp, *destp; |
| 368 | int ret; |
| 369 | |
| 370 | ret = 0; |
| 371 | srcp = head->arg.head[0].iov_base + position; |
| 372 | byte_count = head->arg.head[0].iov_len - position; |
| 373 | if (byte_count > PAGE_SIZE) { |
| 374 | dprintk("svcrdma: large tail unsupported\n"); |
| 375 | return 0; |
| 376 | } |
| 377 | |
| 378 | /* Fit as much of the tail on the current page as possible */ |
| 379 | if (page_offset != PAGE_SIZE) { |
| 380 | destp = page_address(rqstp->rq_arg.pages[page_no]); |
| 381 | destp += page_offset; |
| 382 | while (byte_count--) { |
| 383 | *destp++ = *srcp++; |
| 384 | page_offset++; |
| 385 | if (page_offset == PAGE_SIZE && byte_count) |
| 386 | goto more; |
| 387 | } |
| 388 | goto done; |
| 389 | } |
| 390 | |
| 391 | more: |
| 392 | /* Fit the rest on the next page */ |
| 393 | page_no++; |
| 394 | destp = page_address(rqstp->rq_arg.pages[page_no]); |
| 395 | while (byte_count--) |
| 396 | *destp++ = *srcp++; |
| 397 | |
| 398 | rqstp->rq_respages = &rqstp->rq_arg.pages[page_no+1]; |
| 399 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
| 400 | |
| 401 | done: |
| 402 | byte_count = head->arg.head[0].iov_len - position; |
| 403 | head->arg.page_len += byte_count; |
| 404 | head->arg.len += byte_count; |
| 405 | head->arg.buflen += byte_count; |
| 406 | return 1; |
| 407 | } |
| 408 | |
| 409 | static int rdma_read_chunks(struct svcxprt_rdma *xprt, |
| 410 | struct rpcrdma_msg *rmsgp, |
| 411 | struct svc_rqst *rqstp, |
| 412 | struct svc_rdma_op_ctxt *head) |
| 413 | { |
| 414 | int page_no, ret; |
| 415 | struct rpcrdma_read_chunk *ch; |
| 416 | u32 handle, page_offset, byte_count; |
| 417 | u32 position; |
| 418 | u64 rs_offset; |
| 419 | bool last; |
| 420 | |
| 421 | /* If no read list is present, return 0 */ |
| 422 | ch = svc_rdma_get_read_chunk(rmsgp); |
| 423 | if (!ch) |
| 424 | return 0; |
| 425 | |
| 426 | if (rdma_rcl_chunk_count(ch) > RPCSVC_MAXPAGES) |
| 427 | return -EINVAL; |
| 428 | |
| 429 | /* The request is completed when the RDMA_READs complete. The |
| 430 | * head context keeps all the pages that comprise the |
| 431 | * request. |
| 432 | */ |
| 433 | head->arg.head[0] = rqstp->rq_arg.head[0]; |
| 434 | head->arg.tail[0] = rqstp->rq_arg.tail[0]; |
| 435 | head->hdr_count = head->count; |
| 436 | head->arg.page_base = 0; |
| 437 | head->arg.page_len = 0; |
| 438 | head->arg.len = rqstp->rq_arg.len; |
| 439 | head->arg.buflen = rqstp->rq_arg.buflen; |
| 440 | |
| 441 | ch = (struct rpcrdma_read_chunk *)&rmsgp->rm_body.rm_chunks[0]; |
| 442 | position = be32_to_cpu(ch->rc_position); |
| 443 | |
| 444 | /* RDMA_NOMSG: RDMA READ data should land just after RDMA RECV data */ |
| 445 | if (position == 0) { |
| 446 | head->arg.pages = &head->pages[0]; |
| 447 | page_offset = head->byte_len; |
| 448 | } else { |
| 449 | head->arg.pages = &head->pages[head->count]; |
| 450 | page_offset = 0; |
| 451 | } |
| 452 | |
| 453 | ret = 0; |
| 454 | page_no = 0; |
| 455 | for (; ch->rc_discrim != xdr_zero; ch++) { |
| 456 | if (be32_to_cpu(ch->rc_position) != position) |
| 457 | goto err; |
| 458 | |
| 459 | handle = be32_to_cpu(ch->rc_target.rs_handle), |
| 460 | byte_count = be32_to_cpu(ch->rc_target.rs_length); |
| 461 | xdr_decode_hyper((__be32 *)&ch->rc_target.rs_offset, |
| 462 | &rs_offset); |
| 463 | |
| 464 | while (byte_count > 0) { |
| 465 | last = (ch + 1)->rc_discrim == xdr_zero; |
| 466 | ret = xprt->sc_reader(xprt, rqstp, head, |
| 467 | &page_no, &page_offset, |
| 468 | handle, byte_count, |
| 469 | rs_offset, last); |
| 470 | if (ret < 0) |
| 471 | goto err; |
| 472 | byte_count -= ret; |
| 473 | rs_offset += ret; |
| 474 | head->arg.buflen += ret; |
| 475 | } |
| 476 | } |
| 477 | |
| 478 | /* Read list may need XDR round-up (see RFC 5666, s. 3.7) */ |
| 479 | if (page_offset & 3) { |
| 480 | u32 pad = 4 - (page_offset & 3); |
| 481 | |
| 482 | head->arg.page_len += pad; |
| 483 | head->arg.len += pad; |
| 484 | head->arg.buflen += pad; |
| 485 | page_offset += pad; |
| 486 | } |
| 487 | |
| 488 | ret = 1; |
| 489 | if (position && position < head->arg.head[0].iov_len) |
| 490 | ret = rdma_copy_tail(rqstp, head, position, |
| 491 | byte_count, page_offset, page_no); |
| 492 | head->arg.head[0].iov_len = position; |
| 493 | head->position = position; |
| 494 | |
| 495 | err: |
| 496 | /* Detach arg pages. svc_recv will replenish them */ |
| 497 | for (page_no = 0; |
| 498 | &rqstp->rq_pages[page_no] < rqstp->rq_respages; page_no++) |
| 499 | rqstp->rq_pages[page_no] = NULL; |
| 500 | |
| 501 | return ret; |
| 502 | } |
| 503 | |
| 504 | static int rdma_read_complete(struct svc_rqst *rqstp, |
| 505 | struct svc_rdma_op_ctxt *head) |
| 506 | { |
| 507 | int page_no; |
| 508 | int ret; |
| 509 | |
| 510 | /* Copy RPC pages */ |
| 511 | for (page_no = 0; page_no < head->count; page_no++) { |
| 512 | put_page(rqstp->rq_pages[page_no]); |
| 513 | rqstp->rq_pages[page_no] = head->pages[page_no]; |
| 514 | } |
| 515 | |
| 516 | /* Adjustments made for RDMA_NOMSG type requests */ |
| 517 | if (head->position == 0) { |
| 518 | if (head->arg.len <= head->sge[0].length) { |
| 519 | head->arg.head[0].iov_len = head->arg.len - |
| 520 | head->byte_len; |
| 521 | head->arg.page_len = 0; |
| 522 | } else { |
| 523 | head->arg.head[0].iov_len = head->sge[0].length - |
| 524 | head->byte_len; |
| 525 | head->arg.page_len = head->arg.len - |
| 526 | head->sge[0].length; |
| 527 | } |
| 528 | } |
| 529 | |
| 530 | /* Point rq_arg.pages past header */ |
| 531 | rqstp->rq_arg.pages = &rqstp->rq_pages[head->hdr_count]; |
| 532 | rqstp->rq_arg.page_len = head->arg.page_len; |
| 533 | rqstp->rq_arg.page_base = head->arg.page_base; |
| 534 | |
| 535 | /* rq_respages starts after the last arg page */ |
| 536 | rqstp->rq_respages = &rqstp->rq_pages[page_no]; |
| 537 | rqstp->rq_next_page = rqstp->rq_respages + 1; |
| 538 | |
| 539 | /* Rebuild rq_arg head and tail. */ |
| 540 | rqstp->rq_arg.head[0] = head->arg.head[0]; |
| 541 | rqstp->rq_arg.tail[0] = head->arg.tail[0]; |
| 542 | rqstp->rq_arg.len = head->arg.len; |
| 543 | rqstp->rq_arg.buflen = head->arg.buflen; |
| 544 | |
| 545 | /* Free the context */ |
| 546 | svc_rdma_put_context(head, 0); |
| 547 | |
| 548 | /* XXX: What should this be? */ |
| 549 | rqstp->rq_prot = IPPROTO_MAX; |
| 550 | svc_xprt_copy_addrs(rqstp, rqstp->rq_xprt); |
| 551 | |
| 552 | ret = rqstp->rq_arg.head[0].iov_len |
| 553 | + rqstp->rq_arg.page_len |
| 554 | + rqstp->rq_arg.tail[0].iov_len; |
| 555 | dprintk("svcrdma: deferred read ret=%d, rq_arg.len=%u, " |
| 556 | "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zu\n", |
| 557 | ret, rqstp->rq_arg.len, rqstp->rq_arg.head[0].iov_base, |
| 558 | rqstp->rq_arg.head[0].iov_len); |
| 559 | |
| 560 | return ret; |
| 561 | } |
| 562 | |
| 563 | /* |
| 564 | * Set up the rqstp thread context to point to the RQ buffer. If |
| 565 | * necessary, pull additional data from the client with an RDMA_READ |
| 566 | * request. |
| 567 | */ |
| 568 | int svc_rdma_recvfrom(struct svc_rqst *rqstp) |
| 569 | { |
| 570 | struct svc_xprt *xprt = rqstp->rq_xprt; |
| 571 | struct svcxprt_rdma *rdma_xprt = |
| 572 | container_of(xprt, struct svcxprt_rdma, sc_xprt); |
| 573 | struct svc_rdma_op_ctxt *ctxt = NULL; |
| 574 | struct rpcrdma_msg *rmsgp; |
| 575 | int ret = 0; |
| 576 | int len; |
| 577 | |
| 578 | dprintk("svcrdma: rqstp=%p\n", rqstp); |
| 579 | |
| 580 | spin_lock_bh(&rdma_xprt->sc_rq_dto_lock); |
| 581 | if (!list_empty(&rdma_xprt->sc_read_complete_q)) { |
| 582 | ctxt = list_entry(rdma_xprt->sc_read_complete_q.next, |
| 583 | struct svc_rdma_op_ctxt, |
| 584 | dto_q); |
| 585 | list_del_init(&ctxt->dto_q); |
| 586 | spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock); |
| 587 | return rdma_read_complete(rqstp, ctxt); |
| 588 | } else if (!list_empty(&rdma_xprt->sc_rq_dto_q)) { |
| 589 | ctxt = list_entry(rdma_xprt->sc_rq_dto_q.next, |
| 590 | struct svc_rdma_op_ctxt, |
| 591 | dto_q); |
| 592 | list_del_init(&ctxt->dto_q); |
| 593 | } else { |
| 594 | atomic_inc(&rdma_stat_rq_starve); |
| 595 | clear_bit(XPT_DATA, &xprt->xpt_flags); |
| 596 | ctxt = NULL; |
| 597 | } |
| 598 | spin_unlock_bh(&rdma_xprt->sc_rq_dto_lock); |
| 599 | if (!ctxt) { |
| 600 | /* This is the EAGAIN path. The svc_recv routine will |
| 601 | * return -EAGAIN, the nfsd thread will go to call into |
| 602 | * svc_recv again and we shouldn't be on the active |
| 603 | * transport list |
| 604 | */ |
| 605 | if (test_bit(XPT_CLOSE, &xprt->xpt_flags)) |
| 606 | goto close_out; |
| 607 | |
| 608 | goto out; |
| 609 | } |
| 610 | dprintk("svcrdma: processing ctxt=%p on xprt=%p, rqstp=%p, status=%d\n", |
| 611 | ctxt, rdma_xprt, rqstp, ctxt->wc_status); |
| 612 | atomic_inc(&rdma_stat_recv); |
| 613 | |
| 614 | /* Build up the XDR from the receive buffers. */ |
| 615 | rdma_build_arg_xdr(rqstp, ctxt, ctxt->byte_len); |
| 616 | |
| 617 | /* Decode the RDMA header. */ |
| 618 | len = svc_rdma_xdr_decode_req(&rmsgp, rqstp); |
| 619 | rqstp->rq_xprt_hlen = len; |
| 620 | |
| 621 | /* If the request is invalid, reply with an error */ |
| 622 | if (len < 0) { |
| 623 | if (len == -ENOSYS) |
| 624 | svc_rdma_send_error(rdma_xprt, rmsgp, ERR_VERS); |
| 625 | goto close_out; |
| 626 | } |
| 627 | |
| 628 | /* Read read-list data. */ |
| 629 | ret = rdma_read_chunks(rdma_xprt, rmsgp, rqstp, ctxt); |
| 630 | if (ret > 0) { |
| 631 | /* read-list posted, defer until data received from client. */ |
| 632 | goto defer; |
| 633 | } else if (ret < 0) { |
| 634 | /* Post of read-list failed, free context. */ |
| 635 | svc_rdma_put_context(ctxt, 1); |
| 636 | return 0; |
| 637 | } |
| 638 | |
| 639 | ret = rqstp->rq_arg.head[0].iov_len |
| 640 | + rqstp->rq_arg.page_len |
| 641 | + rqstp->rq_arg.tail[0].iov_len; |
| 642 | svc_rdma_put_context(ctxt, 0); |
| 643 | out: |
| 644 | dprintk("svcrdma: ret=%d, rq_arg.len=%u, " |
| 645 | "rq_arg.head[0].iov_base=%p, rq_arg.head[0].iov_len=%zd\n", |
| 646 | ret, rqstp->rq_arg.len, |
| 647 | rqstp->rq_arg.head[0].iov_base, |
| 648 | rqstp->rq_arg.head[0].iov_len); |
| 649 | rqstp->rq_prot = IPPROTO_MAX; |
| 650 | svc_xprt_copy_addrs(rqstp, xprt); |
| 651 | return ret; |
| 652 | |
| 653 | close_out: |
| 654 | if (ctxt) |
| 655 | svc_rdma_put_context(ctxt, 1); |
| 656 | dprintk("svcrdma: transport %p is closing\n", xprt); |
| 657 | /* |
| 658 | * Set the close bit and enqueue it. svc_recv will see the |
| 659 | * close bit and call svc_xprt_delete |
| 660 | */ |
| 661 | set_bit(XPT_CLOSE, &xprt->xpt_flags); |
| 662 | defer: |
| 663 | return 0; |
| 664 | } |