1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause 2 /* 3 * Copyright (c) 2016-2018 Oracle. All rights reserved. 4 * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved. 5 * Copyright (c) 2005-2006 Network Appliance, Inc. All rights reserved. 6 * 7 * This software is available to you under a choice of one of two 8 * licenses. You may choose to be licensed under the terms of the GNU 9 * General Public License (GPL) Version 2, available from the file 10 * COPYING in the main directory of this source tree, or the BSD-type 11 * license below: 12 * 13 * Redistribution and use in source and binary forms, with or without 14 * modification, are permitted provided that the following conditions 15 * are met: 16 * 17 * Redistributions of source code must retain the above copyright 18 * notice, this list of conditions and the following disclaimer. 19 * 20 * Redistributions in binary form must reproduce the above 21 * copyright notice, this list of conditions and the following 22 * disclaimer in the documentation and/or other materials provided 23 * with the distribution. 24 * 25 * Neither the name of the Network Appliance, Inc. nor the names of 26 * its contributors may be used to endorse or promote products 27 * derived from this software without specific prior written 28 * permission. 29 * 30 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 31 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 32 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 33 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 34 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 35 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 36 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 37 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 38 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 39 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 40 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 41 * 42 * Author: Tom Tucker <tom@opengridcomputing.com> 43 */ 44 45 /* Operation 46 * 47 * The main entry point is svc_rdma_recvfrom. This is called from 48 * svc_recv when the transport indicates there is incoming data to 49 * be read. "Data Ready" is signaled when an RDMA Receive completes, 50 * or when a set of RDMA Reads complete. 51 * 52 * An svc_rqst is passed in. This structure contains an array of 53 * free pages (rq_pages) that will contain the incoming RPC message. 54 * 55 * Short messages are moved directly into svc_rqst::rq_arg, and 56 * the RPC Call is ready to be processed by the Upper Layer. 57 * svc_rdma_recvfrom returns the length of the RPC Call message, 58 * completing the reception of the RPC Call. 59 * 60 * However, when an incoming message has Read chunks, 61 * svc_rdma_recvfrom must post RDMA Reads to pull the RPC Call's 62 * data payload from the client. svc_rdma_recvfrom sets up the 63 * RDMA Reads using pages in svc_rqst::rq_pages, which are 64 * transferred to an svc_rdma_recv_ctxt for the duration of the 65 * I/O. svc_rdma_recvfrom then returns zero, since the RPC message 66 * is still not yet ready. 67 * 68 * When the Read chunk payloads have become available on the 69 * server, "Data Ready" is raised again, and svc_recv calls 70 * svc_rdma_recvfrom again. This second call may use a different 71 * svc_rqst than the first one, thus any information that needs 72 * to be preserved across these two calls is kept in an 73 * svc_rdma_recv_ctxt. 74 * 75 * The second call to svc_rdma_recvfrom performs final assembly 76 * of the RPC Call message, using the RDMA Read sink pages kept in 77 * the svc_rdma_recv_ctxt. The xdr_buf is copied from the 78 * svc_rdma_recv_ctxt to the second svc_rqst. The second call returns 79 * the length of the completed RPC Call message. 80 * 81 * Page Management 82 * 83 * Pages under I/O must be transferred from the first svc_rqst to an 84 * svc_rdma_recv_ctxt before the first svc_rdma_recvfrom call returns. 85 * 86 * The first svc_rqst supplies pages for RDMA Reads. These are moved 87 * from rqstp::rq_pages into ctxt::pages. The consumed elements of 88 * the rq_pages array are set to NULL and refilled with the first 89 * svc_rdma_recvfrom call returns. 90 * 91 * During the second svc_rdma_recvfrom call, RDMA Read sink pages 92 * are transferred from the svc_rdma_recv_ctxt to the second svc_rqst. 93 */ 94 95 #include <linux/slab.h> 96 #include <linux/spinlock.h> 97 #include <linux/unaligned.h> 98 #include <rdma/ib_verbs.h> 99 #include <rdma/rdma_cm.h> 100 101 #include <linux/sunrpc/xdr.h> 102 #include <linux/sunrpc/debug.h> 103 #include <linux/sunrpc/rpc_rdma.h> 104 #include <linux/sunrpc/svc_rdma.h> 105 106 #include "xprt_rdma.h" 107 #include <trace/events/rpcrdma.h> 108 109 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc); 110 111 static inline struct svc_rdma_recv_ctxt * 112 svc_rdma_next_recv_ctxt(struct list_head *list) 113 { 114 return list_first_entry_or_null(list, struct svc_rdma_recv_ctxt, 115 rc_list); 116 } 117 118 static struct svc_rdma_recv_ctxt * 119 svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma) 120 { 121 int node = ibdev_to_node(rdma->sc_cm_id->device); 122 struct svc_rdma_recv_ctxt *ctxt; 123 dma_addr_t addr; 124 void *buffer; 125 126 ctxt = kzalloc_node(sizeof(*ctxt), GFP_KERNEL, node); 127 if (!ctxt) 128 goto fail0; 129 buffer = kmalloc_node(rdma->sc_max_req_size, GFP_KERNEL, node); 130 if (!buffer) 131 goto fail1; 132 addr = ib_dma_map_single(rdma->sc_pd->device, buffer, 133 rdma->sc_max_req_size, DMA_FROM_DEVICE); 134 if (ib_dma_mapping_error(rdma->sc_pd->device, addr)) 135 goto fail2; 136 137 svc_rdma_recv_cid_init(rdma, &ctxt->rc_cid); 138 pcl_init(&ctxt->rc_call_pcl); 139 pcl_init(&ctxt->rc_read_pcl); 140 pcl_init(&ctxt->rc_write_pcl); 141 pcl_init(&ctxt->rc_reply_pcl); 142 143 ctxt->rc_recv_wr.next = NULL; 144 ctxt->rc_recv_wr.wr_cqe = &ctxt->rc_cqe; 145 ctxt->rc_recv_wr.sg_list = &ctxt->rc_recv_sge; 146 ctxt->rc_recv_wr.num_sge = 1; 147 ctxt->rc_cqe.done = svc_rdma_wc_receive; 148 ctxt->rc_recv_sge.addr = addr; 149 ctxt->rc_recv_sge.length = rdma->sc_max_req_size; 150 ctxt->rc_recv_sge.lkey = rdma->sc_pd->local_dma_lkey; 151 ctxt->rc_recv_buf = buffer; 152 svc_rdma_cc_init(rdma, &ctxt->rc_cc); 153 return ctxt; 154 155 fail2: 156 kfree(buffer); 157 fail1: 158 kfree(ctxt); 159 fail0: 160 return NULL; 161 } 162 163 static void svc_rdma_recv_ctxt_destroy(struct svcxprt_rdma *rdma, 164 struct svc_rdma_recv_ctxt *ctxt) 165 { 166 ib_dma_unmap_single(rdma->sc_pd->device, ctxt->rc_recv_sge.addr, 167 ctxt->rc_recv_sge.length, DMA_FROM_DEVICE); 168 kfree(ctxt->rc_recv_buf); 169 kfree(ctxt); 170 } 171 172 /** 173 * svc_rdma_recv_ctxts_destroy - Release all recv_ctxt's for an xprt 174 * @rdma: svcxprt_rdma being torn down 175 * 176 */ 177 void svc_rdma_recv_ctxts_destroy(struct svcxprt_rdma *rdma) 178 { 179 struct svc_rdma_recv_ctxt *ctxt; 180 struct llist_node *node; 181 182 while ((node = llist_del_first(&rdma->sc_recv_ctxts))) { 183 ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); 184 svc_rdma_recv_ctxt_destroy(rdma, ctxt); 185 } 186 } 187 188 /** 189 * svc_rdma_recv_ctxt_get - Allocate a recv_ctxt 190 * @rdma: controlling svcxprt_rdma 191 * 192 * Returns a recv_ctxt or (rarely) NULL if none are available. 193 */ 194 struct svc_rdma_recv_ctxt *svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma) 195 { 196 struct svc_rdma_recv_ctxt *ctxt; 197 struct llist_node *node; 198 199 node = llist_del_first(&rdma->sc_recv_ctxts); 200 if (!node) 201 return NULL; 202 203 ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); 204 ctxt->rc_page_count = 0; 205 return ctxt; 206 } 207 208 /** 209 * svc_rdma_recv_ctxt_put - Return recv_ctxt to free list 210 * @rdma: controlling svcxprt_rdma 211 * @ctxt: object to return to the free list 212 * 213 */ 214 void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma, 215 struct svc_rdma_recv_ctxt *ctxt) 216 { 217 svc_rdma_cc_release(rdma, &ctxt->rc_cc, DMA_FROM_DEVICE); 218 219 /* @rc_page_count is normally zero here, but error flows 220 * can leave pages in @rc_pages. 221 */ 222 release_pages(ctxt->rc_pages, ctxt->rc_page_count); 223 224 pcl_free(&ctxt->rc_call_pcl); 225 pcl_free(&ctxt->rc_read_pcl); 226 pcl_free(&ctxt->rc_write_pcl); 227 pcl_free(&ctxt->rc_reply_pcl); 228 229 llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts); 230 } 231 232 /** 233 * svc_rdma_release_ctxt - Release transport-specific per-rqst resources 234 * @xprt: the transport which owned the context 235 * @vctxt: the context from rqstp->rq_xprt_ctxt or dr->xprt_ctxt 236 * 237 * Ensure that the recv_ctxt is released whether or not a Reply 238 * was sent. For example, the client could close the connection, 239 * or svc_process could drop an RPC, before the Reply is sent. 240 */ 241 void svc_rdma_release_ctxt(struct svc_xprt *xprt, void *vctxt) 242 { 243 struct svc_rdma_recv_ctxt *ctxt = vctxt; 244 struct svcxprt_rdma *rdma = 245 container_of(xprt, struct svcxprt_rdma, sc_xprt); 246 247 if (ctxt) 248 svc_rdma_recv_ctxt_put(rdma, ctxt); 249 } 250 251 static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma, 252 unsigned int wanted) 253 { 254 const struct ib_recv_wr *bad_wr = NULL; 255 struct svc_rdma_recv_ctxt *ctxt; 256 struct ib_recv_wr *recv_chain; 257 int ret; 258 259 if (test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags)) 260 return false; 261 262 recv_chain = NULL; 263 while (wanted--) { 264 ctxt = svc_rdma_recv_ctxt_get(rdma); 265 if (!ctxt) 266 break; 267 268 trace_svcrdma_post_recv(&ctxt->rc_cid); 269 ctxt->rc_recv_wr.next = recv_chain; 270 recv_chain = &ctxt->rc_recv_wr; 271 rdma->sc_pending_recvs++; 272 } 273 if (!recv_chain) 274 return true; 275 276 ret = ib_post_recv(rdma->sc_qp, recv_chain, &bad_wr); 277 if (ret) 278 goto err_free; 279 return true; 280 281 err_free: 282 trace_svcrdma_rq_post_err(rdma, ret); 283 while (bad_wr) { 284 ctxt = container_of(bad_wr, struct svc_rdma_recv_ctxt, 285 rc_recv_wr); 286 bad_wr = bad_wr->next; 287 svc_rdma_recv_ctxt_put(rdma, ctxt); 288 } 289 /* Since we're destroying the xprt, no need to reset 290 * sc_pending_recvs. */ 291 return false; 292 } 293 294 /** 295 * svc_rdma_post_recvs - Post initial set of Recv WRs 296 * @rdma: fresh svcxprt_rdma 297 * 298 * Return values: 299 * %true: Receive Queue initialization successful 300 * %false: memory allocation or DMA error 301 */ 302 bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma) 303 { 304 unsigned int total; 305 306 /* For each credit, allocate enough recv_ctxts for one 307 * posted Receive and one RPC in process. 308 */ 309 total = (rdma->sc_max_requests * 2) + rdma->sc_recv_batch; 310 while (total--) { 311 struct svc_rdma_recv_ctxt *ctxt; 312 313 ctxt = svc_rdma_recv_ctxt_alloc(rdma); 314 if (!ctxt) 315 return false; 316 llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts); 317 } 318 319 return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests); 320 } 321 322 /** 323 * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC 324 * @cq: Completion Queue context 325 * @wc: Work Completion object 326 * 327 */ 328 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) 329 { 330 struct svcxprt_rdma *rdma = cq->cq_context; 331 struct ib_cqe *cqe = wc->wr_cqe; 332 struct svc_rdma_recv_ctxt *ctxt; 333 334 rdma->sc_pending_recvs--; 335 336 /* WARNING: Only wc->wr_cqe and wc->status are reliable */ 337 ctxt = container_of(cqe, struct svc_rdma_recv_ctxt, rc_cqe); 338 339 if (wc->status != IB_WC_SUCCESS) 340 goto flushed; 341 trace_svcrdma_wc_recv(wc, &ctxt->rc_cid); 342 343 /* If receive posting fails, the connection is about to be 344 * lost anyway. The server will not be able to send a reply 345 * for this RPC, and the client will retransmit this RPC 346 * anyway when it reconnects. 347 * 348 * Therefore we drop the Receive, even if status was SUCCESS 349 * to reduce the likelihood of replayed requests once the 350 * client reconnects. 351 */ 352 if (rdma->sc_pending_recvs < rdma->sc_max_requests) 353 if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch)) 354 goto dropped; 355 356 /* All wc fields are now known to be valid */ 357 ctxt->rc_byte_len = wc->byte_len; 358 359 spin_lock(&rdma->sc_rq_dto_lock); 360 list_add_tail(&ctxt->rc_list, &rdma->sc_rq_dto_q); 361 /* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */ 362 set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags); 363 spin_unlock(&rdma->sc_rq_dto_lock); 364 if (!test_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags)) 365 svc_xprt_enqueue(&rdma->sc_xprt); 366 return; 367 368 flushed: 369 if (wc->status == IB_WC_WR_FLUSH_ERR) 370 trace_svcrdma_wc_recv_flush(wc, &ctxt->rc_cid); 371 else 372 trace_svcrdma_wc_recv_err(wc, &ctxt->rc_cid); 373 dropped: 374 svc_rdma_recv_ctxt_put(rdma, ctxt); 375 svc_xprt_deferred_close(&rdma->sc_xprt); 376 } 377 378 /** 379 * svc_rdma_flush_recv_queues - Drain pending Receive work 380 * @rdma: svcxprt_rdma being shut down 381 * 382 */ 383 void svc_rdma_flush_recv_queues(struct svcxprt_rdma *rdma) 384 { 385 struct svc_rdma_recv_ctxt *ctxt; 386 387 while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_read_complete_q))) { 388 list_del(&ctxt->rc_list); 389 svc_rdma_recv_ctxt_put(rdma, ctxt); 390 } 391 while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_rq_dto_q))) { 392 list_del(&ctxt->rc_list); 393 svc_rdma_recv_ctxt_put(rdma, ctxt); 394 } 395 } 396 397 static void svc_rdma_build_arg_xdr(struct svc_rqst *rqstp, 398 struct svc_rdma_recv_ctxt *ctxt) 399 { 400 struct xdr_buf *arg = &rqstp->rq_arg; 401 402 arg->head[0].iov_base = ctxt->rc_recv_buf; 403 arg->head[0].iov_len = ctxt->rc_byte_len; 404 arg->tail[0].iov_base = NULL; 405 arg->tail[0].iov_len = 0; 406 arg->page_len = 0; 407 arg->page_base = 0; 408 arg->buflen = ctxt->rc_byte_len; 409 arg->len = ctxt->rc_byte_len; 410 } 411 412 /** 413 * xdr_count_read_segments - Count number of Read segments in Read list 414 * @rctxt: Ingress receive context 415 * @p: Start of an un-decoded Read list 416 * 417 * Before allocating anything, ensure the ingress Read list is safe 418 * to use. 419 * 420 * The segment count is limited to how many segments can fit in the 421 * transport header without overflowing the buffer. That's about 40 422 * Read segments for a 1KB inline threshold. 423 * 424 * Return values: 425 * %true: Read list is valid. @rctxt's xdr_stream is updated to point 426 * to the first byte past the Read list. rc_read_pcl and 427 * rc_call_pcl cl_count fields are set to the number of 428 * Read segments in the list. 429 * %false: Read list is corrupt. @rctxt's xdr_stream is left in an 430 * unknown state. 431 */ 432 static bool xdr_count_read_segments(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) 433 { 434 rctxt->rc_call_pcl.cl_count = 0; 435 rctxt->rc_read_pcl.cl_count = 0; 436 while (xdr_item_is_present(p)) { 437 u32 position, handle, length; 438 u64 offset; 439 440 p = xdr_inline_decode(&rctxt->rc_stream, 441 rpcrdma_readseg_maxsz * sizeof(*p)); 442 if (!p) 443 return false; 444 445 xdr_decode_read_segment(p, &position, &handle, 446 &length, &offset); 447 if (position) { 448 if (position & 3) 449 return false; 450 ++rctxt->rc_read_pcl.cl_count; 451 } else { 452 ++rctxt->rc_call_pcl.cl_count; 453 } 454 455 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 456 if (!p) 457 return false; 458 } 459 return true; 460 } 461 462 /* Sanity check the Read list. 463 * 464 * Sanity checks: 465 * - Read list does not overflow Receive buffer. 466 * - Chunk size limited by largest NFS data payload. 467 * 468 * Return values: 469 * %true: Read list is valid. @rctxt's xdr_stream is updated 470 * to point to the first byte past the Read list. 471 * %false: Read list is corrupt. @rctxt's xdr_stream is left 472 * in an unknown state. 473 */ 474 static bool xdr_check_read_list(struct svc_rdma_recv_ctxt *rctxt) 475 { 476 __be32 *p; 477 478 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 479 if (!p) 480 return false; 481 if (!xdr_count_read_segments(rctxt, p)) 482 return false; 483 if (!pcl_alloc_call(rctxt, p)) 484 return false; 485 return pcl_alloc_read(rctxt, p); 486 } 487 488 static bool xdr_check_write_chunk(struct svc_rdma_recv_ctxt *rctxt) 489 { 490 u32 segcount; 491 __be32 *p; 492 493 if (xdr_stream_decode_u32(&rctxt->rc_stream, &segcount)) 494 return false; 495 496 /* A bogus segcount causes this buffer overflow check to fail. */ 497 p = xdr_inline_decode(&rctxt->rc_stream, 498 segcount * rpcrdma_segment_maxsz * sizeof(*p)); 499 return p != NULL; 500 } 501 502 /** 503 * xdr_count_write_chunks - Count number of Write chunks in Write list 504 * @rctxt: Received header and decoding state 505 * @p: start of an un-decoded Write list 506 * 507 * Before allocating anything, ensure the ingress Write list is 508 * safe to use. 509 * 510 * Return values: 511 * %true: Write list is valid. @rctxt's xdr_stream is updated 512 * to point to the first byte past the Write list, and 513 * the number of Write chunks is in rc_write_pcl.cl_count. 514 * %false: Write list is corrupt. @rctxt's xdr_stream is left 515 * in an indeterminate state. 516 */ 517 static bool xdr_count_write_chunks(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) 518 { 519 rctxt->rc_write_pcl.cl_count = 0; 520 while (xdr_item_is_present(p)) { 521 if (!xdr_check_write_chunk(rctxt)) 522 return false; 523 ++rctxt->rc_write_pcl.cl_count; 524 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 525 if (!p) 526 return false; 527 } 528 return true; 529 } 530 531 /* Sanity check the Write list. 532 * 533 * Implementation limits: 534 * - This implementation currently supports only one Write chunk. 535 * 536 * Sanity checks: 537 * - Write list does not overflow Receive buffer. 538 * - Chunk size limited by largest NFS data payload. 539 * 540 * Return values: 541 * %true: Write list is valid. @rctxt's xdr_stream is updated 542 * to point to the first byte past the Write list. 543 * %false: Write list is corrupt. @rctxt's xdr_stream is left 544 * in an unknown state. 545 */ 546 static bool xdr_check_write_list(struct svc_rdma_recv_ctxt *rctxt) 547 { 548 __be32 *p; 549 550 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 551 if (!p) 552 return false; 553 if (!xdr_count_write_chunks(rctxt, p)) 554 return false; 555 if (!pcl_alloc_write(rctxt, &rctxt->rc_write_pcl, p)) 556 return false; 557 558 rctxt->rc_cur_result_payload = pcl_first_chunk(&rctxt->rc_write_pcl); 559 return true; 560 } 561 562 /* Sanity check the Reply chunk. 563 * 564 * Sanity checks: 565 * - Reply chunk does not overflow Receive buffer. 566 * - Chunk size limited by largest NFS data payload. 567 * 568 * Return values: 569 * %true: Reply chunk is valid. @rctxt's xdr_stream is updated 570 * to point to the first byte past the Reply chunk. 571 * %false: Reply chunk is corrupt. @rctxt's xdr_stream is left 572 * in an unknown state. 573 */ 574 static bool xdr_check_reply_chunk(struct svc_rdma_recv_ctxt *rctxt) 575 { 576 __be32 *p; 577 578 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 579 if (!p) 580 return false; 581 582 if (!xdr_item_is_present(p)) 583 return true; 584 if (!xdr_check_write_chunk(rctxt)) 585 return false; 586 587 rctxt->rc_reply_pcl.cl_count = 1; 588 return pcl_alloc_write(rctxt, &rctxt->rc_reply_pcl, p); 589 } 590 591 /* RPC-over-RDMA Version One private extension: Remote Invalidation. 592 * Responder's choice: requester signals it can handle Send With 593 * Invalidate, and responder chooses one R_key to invalidate. 594 * 595 * If there is exactly one distinct R_key in the received transport 596 * header, set rc_inv_rkey to that R_key. Otherwise, set it to zero. 597 */ 598 static void svc_rdma_get_inv_rkey(struct svcxprt_rdma *rdma, 599 struct svc_rdma_recv_ctxt *ctxt) 600 { 601 struct svc_rdma_segment *segment; 602 struct svc_rdma_chunk *chunk; 603 u32 inv_rkey; 604 605 ctxt->rc_inv_rkey = 0; 606 607 if (!rdma->sc_snd_w_inv) 608 return; 609 610 inv_rkey = 0; 611 pcl_for_each_chunk(chunk, &ctxt->rc_call_pcl) { 612 pcl_for_each_segment(segment, chunk) { 613 if (inv_rkey == 0) 614 inv_rkey = segment->rs_handle; 615 else if (inv_rkey != segment->rs_handle) 616 return; 617 } 618 } 619 pcl_for_each_chunk(chunk, &ctxt->rc_read_pcl) { 620 pcl_for_each_segment(segment, chunk) { 621 if (inv_rkey == 0) 622 inv_rkey = segment->rs_handle; 623 else if (inv_rkey != segment->rs_handle) 624 return; 625 } 626 } 627 pcl_for_each_chunk(chunk, &ctxt->rc_write_pcl) { 628 pcl_for_each_segment(segment, chunk) { 629 if (inv_rkey == 0) 630 inv_rkey = segment->rs_handle; 631 else if (inv_rkey != segment->rs_handle) 632 return; 633 } 634 } 635 pcl_for_each_chunk(chunk, &ctxt->rc_reply_pcl) { 636 pcl_for_each_segment(segment, chunk) { 637 if (inv_rkey == 0) 638 inv_rkey = segment->rs_handle; 639 else if (inv_rkey != segment->rs_handle) 640 return; 641 } 642 } 643 ctxt->rc_inv_rkey = inv_rkey; 644 } 645 646 /** 647 * svc_rdma_xdr_decode_req - Decode the transport header 648 * @rq_arg: xdr_buf containing ingress RPC/RDMA message 649 * @rctxt: state of decoding 650 * 651 * On entry, xdr->head[0].iov_base points to first byte of the 652 * RPC-over-RDMA transport header. 653 * 654 * On successful exit, head[0] points to first byte past the 655 * RPC-over-RDMA header. For RDMA_MSG, this is the RPC message. 656 * 657 * The length of the RPC-over-RDMA header is returned. 658 * 659 * Assumptions: 660 * - The transport header is entirely contained in the head iovec. 661 */ 662 static int svc_rdma_xdr_decode_req(struct xdr_buf *rq_arg, 663 struct svc_rdma_recv_ctxt *rctxt) 664 { 665 __be32 *p, *rdma_argp; 666 unsigned int hdr_len; 667 668 rdma_argp = rq_arg->head[0].iov_base; 669 xdr_init_decode(&rctxt->rc_stream, rq_arg, rdma_argp, NULL); 670 671 p = xdr_inline_decode(&rctxt->rc_stream, 672 rpcrdma_fixed_maxsz * sizeof(*p)); 673 if (unlikely(!p)) 674 goto out_short; 675 p++; 676 if (*p != rpcrdma_version) 677 goto out_version; 678 p += 2; 679 rctxt->rc_msgtype = *p; 680 switch (rctxt->rc_msgtype) { 681 case rdma_msg: 682 break; 683 case rdma_nomsg: 684 break; 685 case rdma_done: 686 goto out_drop; 687 case rdma_error: 688 goto out_drop; 689 default: 690 goto out_proc; 691 } 692 693 if (!xdr_check_read_list(rctxt)) 694 goto out_inval; 695 if (!xdr_check_write_list(rctxt)) 696 goto out_inval; 697 if (!xdr_check_reply_chunk(rctxt)) 698 goto out_inval; 699 700 rq_arg->head[0].iov_base = rctxt->rc_stream.p; 701 hdr_len = xdr_stream_pos(&rctxt->rc_stream); 702 rq_arg->head[0].iov_len -= hdr_len; 703 rq_arg->len -= hdr_len; 704 trace_svcrdma_decode_rqst(rctxt, rdma_argp, hdr_len); 705 return hdr_len; 706 707 out_short: 708 trace_svcrdma_decode_short_err(rctxt, rq_arg->len); 709 return -EINVAL; 710 711 out_version: 712 trace_svcrdma_decode_badvers_err(rctxt, rdma_argp); 713 return -EPROTONOSUPPORT; 714 715 out_drop: 716 trace_svcrdma_decode_drop_err(rctxt, rdma_argp); 717 return 0; 718 719 out_proc: 720 trace_svcrdma_decode_badproc_err(rctxt, rdma_argp); 721 return -EINVAL; 722 723 out_inval: 724 trace_svcrdma_decode_parse_err(rctxt, rdma_argp); 725 return -EINVAL; 726 } 727 728 static void svc_rdma_send_error(struct svcxprt_rdma *rdma, 729 struct svc_rdma_recv_ctxt *rctxt, 730 int status) 731 { 732 struct svc_rdma_send_ctxt *sctxt; 733 734 sctxt = svc_rdma_send_ctxt_get(rdma); 735 if (!sctxt) 736 return; 737 svc_rdma_send_error_msg(rdma, sctxt, rctxt, status); 738 } 739 740 /* By convention, backchannel calls arrive via rdma_msg type 741 * messages, and never populate the chunk lists. This makes 742 * the RPC/RDMA header small and fixed in size, so it is 743 * straightforward to check the RPC header's direction field. 744 */ 745 static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt, 746 struct svc_rdma_recv_ctxt *rctxt) 747 { 748 __be32 *p = rctxt->rc_recv_buf; 749 750 if (!xprt->xpt_bc_xprt) 751 return false; 752 753 if (rctxt->rc_msgtype != rdma_msg) 754 return false; 755 756 if (!pcl_is_empty(&rctxt->rc_call_pcl)) 757 return false; 758 if (!pcl_is_empty(&rctxt->rc_read_pcl)) 759 return false; 760 if (!pcl_is_empty(&rctxt->rc_write_pcl)) 761 return false; 762 if (!pcl_is_empty(&rctxt->rc_reply_pcl)) 763 return false; 764 765 /* RPC call direction */ 766 if (*(p + 8) == cpu_to_be32(RPC_CALL)) 767 return false; 768 769 return true; 770 } 771 772 /* Finish constructing the RPC Call message in rqstp::rq_arg. 773 * 774 * The incoming RPC/RDMA message is an RDMA_MSG type message 775 * with a single Read chunk (only the upper layer data payload 776 * was conveyed via RDMA Read). 777 */ 778 static void svc_rdma_read_complete_one(struct svc_rqst *rqstp, 779 struct svc_rdma_recv_ctxt *ctxt) 780 { 781 struct svc_rdma_chunk *chunk = pcl_first_chunk(&ctxt->rc_read_pcl); 782 struct xdr_buf *buf = &rqstp->rq_arg; 783 unsigned int length; 784 785 /* Split the Receive buffer between the head and tail 786 * buffers at Read chunk's position. XDR roundup of the 787 * chunk is not included in either the pagelist or in 788 * the tail. 789 */ 790 buf->tail[0].iov_base = buf->head[0].iov_base + chunk->ch_position; 791 buf->tail[0].iov_len = buf->head[0].iov_len - chunk->ch_position; 792 buf->head[0].iov_len = chunk->ch_position; 793 794 /* Read chunk may need XDR roundup (see RFC 8166, s. 3.4.5.2). 795 * 796 * If the client already rounded up the chunk length, the 797 * length does not change. Otherwise, the length of the page 798 * list is increased to include XDR round-up. 799 * 800 * Currently these chunks always start at page offset 0, 801 * thus the rounded-up length never crosses a page boundary. 802 */ 803 buf->pages = &rqstp->rq_pages[0]; 804 length = xdr_align_size(chunk->ch_length); 805 buf->page_len = length; 806 buf->len += length; 807 buf->buflen += length; 808 } 809 810 /* Finish constructing the RPC Call message in rqstp::rq_arg. 811 * 812 * The incoming RPC/RDMA message is an RDMA_MSG type message 813 * with payload in multiple Read chunks and no PZRC. 814 */ 815 static void svc_rdma_read_complete_multiple(struct svc_rqst *rqstp, 816 struct svc_rdma_recv_ctxt *ctxt) 817 { 818 struct xdr_buf *buf = &rqstp->rq_arg; 819 820 buf->len += ctxt->rc_readbytes; 821 buf->buflen += ctxt->rc_readbytes; 822 823 buf->head[0].iov_base = page_address(rqstp->rq_pages[0]); 824 buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes); 825 buf->pages = &rqstp->rq_pages[1]; 826 buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len; 827 } 828 829 /* Finish constructing the RPC Call message in rqstp::rq_arg. 830 * 831 * The incoming RPC/RDMA message is an RDMA_NOMSG type message 832 * (the RPC message body was conveyed via RDMA Read). 833 */ 834 static void svc_rdma_read_complete_pzrc(struct svc_rqst *rqstp, 835 struct svc_rdma_recv_ctxt *ctxt) 836 { 837 struct xdr_buf *buf = &rqstp->rq_arg; 838 839 buf->len += ctxt->rc_readbytes; 840 buf->buflen += ctxt->rc_readbytes; 841 842 buf->head[0].iov_base = page_address(rqstp->rq_pages[0]); 843 buf->head[0].iov_len = min_t(size_t, PAGE_SIZE, ctxt->rc_readbytes); 844 buf->pages = &rqstp->rq_pages[1]; 845 buf->page_len = ctxt->rc_readbytes - buf->head[0].iov_len; 846 } 847 848 static noinline void svc_rdma_read_complete(struct svc_rqst *rqstp, 849 struct svc_rdma_recv_ctxt *ctxt) 850 { 851 unsigned int i; 852 853 /* Transfer the Read chunk pages into @rqstp.rq_pages, replacing 854 * the rq_pages that were already allocated for this rqstp. 855 */ 856 release_pages(rqstp->rq_respages, ctxt->rc_page_count); 857 for (i = 0; i < ctxt->rc_page_count; i++) 858 rqstp->rq_pages[i] = ctxt->rc_pages[i]; 859 860 /* Update @rqstp's result send buffer to start after the 861 * last page in the RDMA Read payload. 862 */ 863 rqstp->rq_respages = &rqstp->rq_pages[ctxt->rc_page_count]; 864 rqstp->rq_next_page = rqstp->rq_respages + 1; 865 866 /* Prevent svc_rdma_recv_ctxt_put() from releasing the 867 * pages in ctxt::rc_pages a second time. 868 */ 869 ctxt->rc_page_count = 0; 870 871 /* Finish constructing the RPC Call message. The exact 872 * procedure for that depends on what kind of RPC/RDMA 873 * chunks were provided by the client. 874 */ 875 rqstp->rq_arg = ctxt->rc_saved_arg; 876 if (pcl_is_empty(&ctxt->rc_call_pcl)) { 877 if (ctxt->rc_read_pcl.cl_count == 1) 878 svc_rdma_read_complete_one(rqstp, ctxt); 879 else 880 svc_rdma_read_complete_multiple(rqstp, ctxt); 881 } else { 882 svc_rdma_read_complete_pzrc(rqstp, ctxt); 883 } 884 885 trace_svcrdma_read_finished(&ctxt->rc_cid); 886 } 887 888 /** 889 * svc_rdma_recvfrom - Receive an RPC call 890 * @rqstp: request structure into which to receive an RPC Call 891 * 892 * Returns: 893 * The positive number of bytes in the RPC Call message, 894 * %0 if there were no Calls ready to return, 895 * %-EINVAL if the Read chunk data is too large, 896 * %-ENOMEM if rdma_rw context pool was exhausted, 897 * %-ENOTCONN if posting failed (connection is lost), 898 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 899 * 900 * Called in a loop when XPT_DATA is set. XPT_DATA is cleared only 901 * when there are no remaining ctxt's to process. 902 * 903 * The next ctxt is removed from the "receive" lists. 904 * 905 * - If the ctxt completes a Receive, then construct the Call 906 * message from the contents of the Receive buffer. 907 * 908 * - If there are no Read chunks in this message, then finish 909 * assembling the Call message and return the number of bytes 910 * in the message. 911 * 912 * - If there are Read chunks in this message, post Read WRs to 913 * pull that payload. When the Read WRs complete, build the 914 * full message and return the number of bytes in it. 915 */ 916 int svc_rdma_recvfrom(struct svc_rqst *rqstp) 917 { 918 struct svc_xprt *xprt = rqstp->rq_xprt; 919 struct svcxprt_rdma *rdma_xprt = 920 container_of(xprt, struct svcxprt_rdma, sc_xprt); 921 struct svc_rdma_recv_ctxt *ctxt; 922 int ret; 923 924 /* Prevent svc_xprt_release() from releasing pages in rq_pages 925 * when returning 0 or an error. 926 */ 927 rqstp->rq_respages = rqstp->rq_pages; 928 rqstp->rq_next_page = rqstp->rq_respages; 929 930 rqstp->rq_xprt_ctxt = NULL; 931 932 spin_lock(&rdma_xprt->sc_rq_dto_lock); 933 ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_read_complete_q); 934 if (ctxt) { 935 list_del(&ctxt->rc_list); 936 spin_unlock(&rdma_xprt->sc_rq_dto_lock); 937 svc_xprt_received(xprt); 938 svc_rdma_read_complete(rqstp, ctxt); 939 goto complete; 940 } 941 ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_rq_dto_q); 942 if (ctxt) 943 list_del(&ctxt->rc_list); 944 else 945 /* No new incoming requests, terminate the loop */ 946 clear_bit(XPT_DATA, &xprt->xpt_flags); 947 spin_unlock(&rdma_xprt->sc_rq_dto_lock); 948 949 /* Unblock the transport for the next receive */ 950 svc_xprt_received(xprt); 951 if (!ctxt) 952 return 0; 953 954 percpu_counter_inc(&svcrdma_stat_recv); 955 ib_dma_sync_single_for_cpu(rdma_xprt->sc_pd->device, 956 ctxt->rc_recv_sge.addr, ctxt->rc_byte_len, 957 DMA_FROM_DEVICE); 958 svc_rdma_build_arg_xdr(rqstp, ctxt); 959 960 ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg, ctxt); 961 if (ret < 0) 962 goto out_err; 963 if (ret == 0) 964 goto out_drop; 965 966 if (svc_rdma_is_reverse_direction_reply(xprt, ctxt)) 967 goto out_backchannel; 968 969 svc_rdma_get_inv_rkey(rdma_xprt, ctxt); 970 971 if (!pcl_is_empty(&ctxt->rc_read_pcl) || 972 !pcl_is_empty(&ctxt->rc_call_pcl)) 973 goto out_readlist; 974 975 complete: 976 rqstp->rq_xprt_ctxt = ctxt; 977 rqstp->rq_prot = IPPROTO_MAX; 978 svc_xprt_copy_addrs(rqstp, xprt); 979 set_bit(RQ_SECURE, &rqstp->rq_flags); 980 return rqstp->rq_arg.len; 981 982 out_err: 983 svc_rdma_send_error(rdma_xprt, ctxt, ret); 984 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 985 return 0; 986 987 out_readlist: 988 /* This @rqstp is about to be recycled. Save the work 989 * already done constructing the Call message in rq_arg 990 * so it can be restored when the RDMA Reads have 991 * completed. 992 */ 993 ctxt->rc_saved_arg = rqstp->rq_arg; 994 995 ret = svc_rdma_process_read_list(rdma_xprt, rqstp, ctxt); 996 if (ret < 0) { 997 if (ret == -EINVAL) 998 svc_rdma_send_error(rdma_xprt, ctxt, ret); 999 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 1000 svc_xprt_deferred_close(xprt); 1001 return ret; 1002 } 1003 return 0; 1004 1005 out_backchannel: 1006 svc_rdma_handle_bc_reply(rqstp, ctxt); 1007 out_drop: 1008 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 1009 return 0; 1010 } 1011