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 <asm/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 void svc_rdma_recv_cid_init(struct svcxprt_rdma *rdma, 119 struct rpc_rdma_cid *cid) 120 { 121 cid->ci_queue_id = rdma->sc_rq_cq->res.id; 122 cid->ci_completion_id = atomic_inc_return(&rdma->sc_completion_ids); 123 } 124 125 static struct svc_rdma_recv_ctxt * 126 svc_rdma_recv_ctxt_alloc(struct svcxprt_rdma *rdma) 127 { 128 struct svc_rdma_recv_ctxt *ctxt; 129 dma_addr_t addr; 130 void *buffer; 131 132 ctxt = kmalloc(sizeof(*ctxt), GFP_KERNEL); 133 if (!ctxt) 134 goto fail0; 135 buffer = kmalloc(rdma->sc_max_req_size, GFP_KERNEL); 136 if (!buffer) 137 goto fail1; 138 addr = ib_dma_map_single(rdma->sc_pd->device, buffer, 139 rdma->sc_max_req_size, DMA_FROM_DEVICE); 140 if (ib_dma_mapping_error(rdma->sc_pd->device, addr)) 141 goto fail2; 142 143 svc_rdma_recv_cid_init(rdma, &ctxt->rc_cid); 144 pcl_init(&ctxt->rc_call_pcl); 145 pcl_init(&ctxt->rc_read_pcl); 146 pcl_init(&ctxt->rc_write_pcl); 147 pcl_init(&ctxt->rc_reply_pcl); 148 149 ctxt->rc_recv_wr.next = NULL; 150 ctxt->rc_recv_wr.wr_cqe = &ctxt->rc_cqe; 151 ctxt->rc_recv_wr.sg_list = &ctxt->rc_recv_sge; 152 ctxt->rc_recv_wr.num_sge = 1; 153 ctxt->rc_cqe.done = svc_rdma_wc_receive; 154 ctxt->rc_recv_sge.addr = addr; 155 ctxt->rc_recv_sge.length = rdma->sc_max_req_size; 156 ctxt->rc_recv_sge.lkey = rdma->sc_pd->local_dma_lkey; 157 ctxt->rc_recv_buf = buffer; 158 ctxt->rc_temp = false; 159 return ctxt; 160 161 fail2: 162 kfree(buffer); 163 fail1: 164 kfree(ctxt); 165 fail0: 166 return NULL; 167 } 168 169 static void svc_rdma_recv_ctxt_destroy(struct svcxprt_rdma *rdma, 170 struct svc_rdma_recv_ctxt *ctxt) 171 { 172 ib_dma_unmap_single(rdma->sc_pd->device, ctxt->rc_recv_sge.addr, 173 ctxt->rc_recv_sge.length, DMA_FROM_DEVICE); 174 kfree(ctxt->rc_recv_buf); 175 kfree(ctxt); 176 } 177 178 /** 179 * svc_rdma_recv_ctxts_destroy - Release all recv_ctxt's for an xprt 180 * @rdma: svcxprt_rdma being torn down 181 * 182 */ 183 void svc_rdma_recv_ctxts_destroy(struct svcxprt_rdma *rdma) 184 { 185 struct svc_rdma_recv_ctxt *ctxt; 186 struct llist_node *node; 187 188 while ((node = llist_del_first(&rdma->sc_recv_ctxts))) { 189 ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); 190 svc_rdma_recv_ctxt_destroy(rdma, ctxt); 191 } 192 } 193 194 /** 195 * svc_rdma_recv_ctxt_get - Allocate a recv_ctxt 196 * @rdma: controlling svcxprt_rdma 197 * 198 * Returns a recv_ctxt or (rarely) NULL if none are available. 199 */ 200 struct svc_rdma_recv_ctxt *svc_rdma_recv_ctxt_get(struct svcxprt_rdma *rdma) 201 { 202 struct svc_rdma_recv_ctxt *ctxt; 203 struct llist_node *node; 204 205 node = llist_del_first(&rdma->sc_recv_ctxts); 206 if (!node) 207 goto out_empty; 208 ctxt = llist_entry(node, struct svc_rdma_recv_ctxt, rc_node); 209 210 out: 211 ctxt->rc_page_count = 0; 212 return ctxt; 213 214 out_empty: 215 ctxt = svc_rdma_recv_ctxt_alloc(rdma); 216 if (!ctxt) 217 return NULL; 218 goto out; 219 } 220 221 /** 222 * svc_rdma_recv_ctxt_put - Return recv_ctxt to free list 223 * @rdma: controlling svcxprt_rdma 224 * @ctxt: object to return to the free list 225 * 226 */ 227 void svc_rdma_recv_ctxt_put(struct svcxprt_rdma *rdma, 228 struct svc_rdma_recv_ctxt *ctxt) 229 { 230 pcl_free(&ctxt->rc_call_pcl); 231 pcl_free(&ctxt->rc_read_pcl); 232 pcl_free(&ctxt->rc_write_pcl); 233 pcl_free(&ctxt->rc_reply_pcl); 234 235 if (!ctxt->rc_temp) 236 llist_add(&ctxt->rc_node, &rdma->sc_recv_ctxts); 237 else 238 svc_rdma_recv_ctxt_destroy(rdma, ctxt); 239 } 240 241 /** 242 * svc_rdma_release_rqst - Release transport-specific per-rqst resources 243 * @rqstp: svc_rqst being released 244 * 245 * Ensure that the recv_ctxt is released whether or not a Reply 246 * was sent. For example, the client could close the connection, 247 * or svc_process could drop an RPC, before the Reply is sent. 248 */ 249 void svc_rdma_release_rqst(struct svc_rqst *rqstp) 250 { 251 struct svc_rdma_recv_ctxt *ctxt = rqstp->rq_xprt_ctxt; 252 struct svc_xprt *xprt = rqstp->rq_xprt; 253 struct svcxprt_rdma *rdma = 254 container_of(xprt, struct svcxprt_rdma, sc_xprt); 255 256 rqstp->rq_xprt_ctxt = NULL; 257 if (ctxt) 258 svc_rdma_recv_ctxt_put(rdma, ctxt); 259 } 260 261 static bool svc_rdma_refresh_recvs(struct svcxprt_rdma *rdma, 262 unsigned int wanted, bool temp) 263 { 264 const struct ib_recv_wr *bad_wr = NULL; 265 struct svc_rdma_recv_ctxt *ctxt; 266 struct ib_recv_wr *recv_chain; 267 int ret; 268 269 if (test_bit(XPT_CLOSE, &rdma->sc_xprt.xpt_flags)) 270 return false; 271 272 recv_chain = NULL; 273 while (wanted--) { 274 ctxt = svc_rdma_recv_ctxt_get(rdma); 275 if (!ctxt) 276 break; 277 278 trace_svcrdma_post_recv(ctxt); 279 ctxt->rc_temp = temp; 280 ctxt->rc_recv_wr.next = recv_chain; 281 recv_chain = &ctxt->rc_recv_wr; 282 rdma->sc_pending_recvs++; 283 } 284 if (!recv_chain) 285 return false; 286 287 ret = ib_post_recv(rdma->sc_qp, recv_chain, &bad_wr); 288 if (ret) 289 goto err_free; 290 return true; 291 292 err_free: 293 trace_svcrdma_rq_post_err(rdma, ret); 294 while (bad_wr) { 295 ctxt = container_of(bad_wr, struct svc_rdma_recv_ctxt, 296 rc_recv_wr); 297 bad_wr = bad_wr->next; 298 svc_rdma_recv_ctxt_put(rdma, ctxt); 299 } 300 /* Since we're destroying the xprt, no need to reset 301 * sc_pending_recvs. */ 302 return false; 303 } 304 305 /** 306 * svc_rdma_post_recvs - Post initial set of Recv WRs 307 * @rdma: fresh svcxprt_rdma 308 * 309 * Returns true if successful, otherwise false. 310 */ 311 bool svc_rdma_post_recvs(struct svcxprt_rdma *rdma) 312 { 313 return svc_rdma_refresh_recvs(rdma, rdma->sc_max_requests, true); 314 } 315 316 /** 317 * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC 318 * @cq: Completion Queue context 319 * @wc: Work Completion object 320 * 321 */ 322 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) 323 { 324 struct svcxprt_rdma *rdma = cq->cq_context; 325 struct ib_cqe *cqe = wc->wr_cqe; 326 struct svc_rdma_recv_ctxt *ctxt; 327 328 rdma->sc_pending_recvs--; 329 330 /* WARNING: Only wc->wr_cqe and wc->status are reliable */ 331 ctxt = container_of(cqe, struct svc_rdma_recv_ctxt, rc_cqe); 332 333 if (wc->status != IB_WC_SUCCESS) 334 goto flushed; 335 trace_svcrdma_wc_recv(wc, &ctxt->rc_cid); 336 337 /* If receive posting fails, the connection is about to be 338 * lost anyway. The server will not be able to send a reply 339 * for this RPC, and the client will retransmit this RPC 340 * anyway when it reconnects. 341 * 342 * Therefore we drop the Receive, even if status was SUCCESS 343 * to reduce the likelihood of replayed requests once the 344 * client reconnects. 345 */ 346 if (rdma->sc_pending_recvs < rdma->sc_max_requests) 347 if (!svc_rdma_refresh_recvs(rdma, rdma->sc_recv_batch, false)) 348 goto dropped; 349 350 /* All wc fields are now known to be valid */ 351 ctxt->rc_byte_len = wc->byte_len; 352 353 spin_lock(&rdma->sc_rq_dto_lock); 354 list_add_tail(&ctxt->rc_list, &rdma->sc_rq_dto_q); 355 /* Note the unlock pairs with the smp_rmb in svc_xprt_ready: */ 356 set_bit(XPT_DATA, &rdma->sc_xprt.xpt_flags); 357 spin_unlock(&rdma->sc_rq_dto_lock); 358 if (!test_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags)) 359 svc_xprt_enqueue(&rdma->sc_xprt); 360 return; 361 362 flushed: 363 if (wc->status == IB_WC_WR_FLUSH_ERR) 364 trace_svcrdma_wc_recv_flush(wc, &ctxt->rc_cid); 365 else 366 trace_svcrdma_wc_recv_err(wc, &ctxt->rc_cid); 367 dropped: 368 svc_rdma_recv_ctxt_put(rdma, ctxt); 369 svc_xprt_deferred_close(&rdma->sc_xprt); 370 } 371 372 /** 373 * svc_rdma_flush_recv_queues - Drain pending Receive work 374 * @rdma: svcxprt_rdma being shut down 375 * 376 */ 377 void svc_rdma_flush_recv_queues(struct svcxprt_rdma *rdma) 378 { 379 struct svc_rdma_recv_ctxt *ctxt; 380 381 while ((ctxt = svc_rdma_next_recv_ctxt(&rdma->sc_rq_dto_q))) { 382 list_del(&ctxt->rc_list); 383 svc_rdma_recv_ctxt_put(rdma, ctxt); 384 } 385 } 386 387 static void svc_rdma_build_arg_xdr(struct svc_rqst *rqstp, 388 struct svc_rdma_recv_ctxt *ctxt) 389 { 390 struct xdr_buf *arg = &rqstp->rq_arg; 391 392 arg->head[0].iov_base = ctxt->rc_recv_buf; 393 arg->head[0].iov_len = ctxt->rc_byte_len; 394 arg->tail[0].iov_base = NULL; 395 arg->tail[0].iov_len = 0; 396 arg->page_len = 0; 397 arg->page_base = 0; 398 arg->buflen = ctxt->rc_byte_len; 399 arg->len = ctxt->rc_byte_len; 400 } 401 402 /** 403 * xdr_count_read_segments - Count number of Read segments in Read list 404 * @rctxt: Ingress receive context 405 * @p: Start of an un-decoded Read list 406 * 407 * Before allocating anything, ensure the ingress Read list is safe 408 * to use. 409 * 410 * The segment count is limited to how many segments can fit in the 411 * transport header without overflowing the buffer. That's about 40 412 * Read segments for a 1KB inline threshold. 413 * 414 * Return values: 415 * %true: Read list is valid. @rctxt's xdr_stream is updated to point 416 * to the first byte past the Read list. rc_read_pcl and 417 * rc_call_pcl cl_count fields are set to the number of 418 * Read segments in the list. 419 * %false: Read list is corrupt. @rctxt's xdr_stream is left in an 420 * unknown state. 421 */ 422 static bool xdr_count_read_segments(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) 423 { 424 rctxt->rc_call_pcl.cl_count = 0; 425 rctxt->rc_read_pcl.cl_count = 0; 426 while (xdr_item_is_present(p)) { 427 u32 position, handle, length; 428 u64 offset; 429 430 p = xdr_inline_decode(&rctxt->rc_stream, 431 rpcrdma_readseg_maxsz * sizeof(*p)); 432 if (!p) 433 return false; 434 435 xdr_decode_read_segment(p, &position, &handle, 436 &length, &offset); 437 if (position) { 438 if (position & 3) 439 return false; 440 ++rctxt->rc_read_pcl.cl_count; 441 } else { 442 ++rctxt->rc_call_pcl.cl_count; 443 } 444 445 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 446 if (!p) 447 return false; 448 } 449 return true; 450 } 451 452 /* Sanity check the Read list. 453 * 454 * Sanity checks: 455 * - Read list does not overflow Receive buffer. 456 * - Chunk size limited by largest NFS data payload. 457 * 458 * Return values: 459 * %true: Read list is valid. @rctxt's xdr_stream is updated 460 * to point to the first byte past the Read list. 461 * %false: Read list is corrupt. @rctxt's xdr_stream is left 462 * in an unknown state. 463 */ 464 static bool xdr_check_read_list(struct svc_rdma_recv_ctxt *rctxt) 465 { 466 __be32 *p; 467 468 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 469 if (!p) 470 return false; 471 if (!xdr_count_read_segments(rctxt, p)) 472 return false; 473 if (!pcl_alloc_call(rctxt, p)) 474 return false; 475 return pcl_alloc_read(rctxt, p); 476 } 477 478 static bool xdr_check_write_chunk(struct svc_rdma_recv_ctxt *rctxt) 479 { 480 u32 segcount; 481 __be32 *p; 482 483 if (xdr_stream_decode_u32(&rctxt->rc_stream, &segcount)) 484 return false; 485 486 /* A bogus segcount causes this buffer overflow check to fail. */ 487 p = xdr_inline_decode(&rctxt->rc_stream, 488 segcount * rpcrdma_segment_maxsz * sizeof(*p)); 489 return p != NULL; 490 } 491 492 /** 493 * xdr_count_write_chunks - Count number of Write chunks in Write list 494 * @rctxt: Received header and decoding state 495 * @p: start of an un-decoded Write list 496 * 497 * Before allocating anything, ensure the ingress Write list is 498 * safe to use. 499 * 500 * Return values: 501 * %true: Write list is valid. @rctxt's xdr_stream is updated 502 * to point to the first byte past the Write list, and 503 * the number of Write chunks is in rc_write_pcl.cl_count. 504 * %false: Write list is corrupt. @rctxt's xdr_stream is left 505 * in an indeterminate state. 506 */ 507 static bool xdr_count_write_chunks(struct svc_rdma_recv_ctxt *rctxt, __be32 *p) 508 { 509 rctxt->rc_write_pcl.cl_count = 0; 510 while (xdr_item_is_present(p)) { 511 if (!xdr_check_write_chunk(rctxt)) 512 return false; 513 ++rctxt->rc_write_pcl.cl_count; 514 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 515 if (!p) 516 return false; 517 } 518 return true; 519 } 520 521 /* Sanity check the Write list. 522 * 523 * Implementation limits: 524 * - This implementation currently supports only one Write chunk. 525 * 526 * Sanity checks: 527 * - Write list does not overflow Receive buffer. 528 * - Chunk size limited by largest NFS data payload. 529 * 530 * Return values: 531 * %true: Write list is valid. @rctxt's xdr_stream is updated 532 * to point to the first byte past the Write list. 533 * %false: Write list is corrupt. @rctxt's xdr_stream is left 534 * in an unknown state. 535 */ 536 static bool xdr_check_write_list(struct svc_rdma_recv_ctxt *rctxt) 537 { 538 __be32 *p; 539 540 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 541 if (!p) 542 return false; 543 if (!xdr_count_write_chunks(rctxt, p)) 544 return false; 545 if (!pcl_alloc_write(rctxt, &rctxt->rc_write_pcl, p)) 546 return false; 547 548 rctxt->rc_cur_result_payload = pcl_first_chunk(&rctxt->rc_write_pcl); 549 return true; 550 } 551 552 /* Sanity check the Reply chunk. 553 * 554 * Sanity checks: 555 * - Reply chunk does not overflow Receive buffer. 556 * - Chunk size limited by largest NFS data payload. 557 * 558 * Return values: 559 * %true: Reply chunk is valid. @rctxt's xdr_stream is updated 560 * to point to the first byte past the Reply chunk. 561 * %false: Reply chunk is corrupt. @rctxt's xdr_stream is left 562 * in an unknown state. 563 */ 564 static bool xdr_check_reply_chunk(struct svc_rdma_recv_ctxt *rctxt) 565 { 566 __be32 *p; 567 568 p = xdr_inline_decode(&rctxt->rc_stream, sizeof(*p)); 569 if (!p) 570 return false; 571 572 if (!xdr_item_is_present(p)) 573 return true; 574 if (!xdr_check_write_chunk(rctxt)) 575 return false; 576 577 rctxt->rc_reply_pcl.cl_count = 1; 578 return pcl_alloc_write(rctxt, &rctxt->rc_reply_pcl, p); 579 } 580 581 /* RPC-over-RDMA Version One private extension: Remote Invalidation. 582 * Responder's choice: requester signals it can handle Send With 583 * Invalidate, and responder chooses one R_key to invalidate. 584 * 585 * If there is exactly one distinct R_key in the received transport 586 * header, set rc_inv_rkey to that R_key. Otherwise, set it to zero. 587 */ 588 static void svc_rdma_get_inv_rkey(struct svcxprt_rdma *rdma, 589 struct svc_rdma_recv_ctxt *ctxt) 590 { 591 struct svc_rdma_segment *segment; 592 struct svc_rdma_chunk *chunk; 593 u32 inv_rkey; 594 595 ctxt->rc_inv_rkey = 0; 596 597 if (!rdma->sc_snd_w_inv) 598 return; 599 600 inv_rkey = 0; 601 pcl_for_each_chunk(chunk, &ctxt->rc_call_pcl) { 602 pcl_for_each_segment(segment, chunk) { 603 if (inv_rkey == 0) 604 inv_rkey = segment->rs_handle; 605 else if (inv_rkey != segment->rs_handle) 606 return; 607 } 608 } 609 pcl_for_each_chunk(chunk, &ctxt->rc_read_pcl) { 610 pcl_for_each_segment(segment, chunk) { 611 if (inv_rkey == 0) 612 inv_rkey = segment->rs_handle; 613 else if (inv_rkey != segment->rs_handle) 614 return; 615 } 616 } 617 pcl_for_each_chunk(chunk, &ctxt->rc_write_pcl) { 618 pcl_for_each_segment(segment, chunk) { 619 if (inv_rkey == 0) 620 inv_rkey = segment->rs_handle; 621 else if (inv_rkey != segment->rs_handle) 622 return; 623 } 624 } 625 pcl_for_each_chunk(chunk, &ctxt->rc_reply_pcl) { 626 pcl_for_each_segment(segment, chunk) { 627 if (inv_rkey == 0) 628 inv_rkey = segment->rs_handle; 629 else if (inv_rkey != segment->rs_handle) 630 return; 631 } 632 } 633 ctxt->rc_inv_rkey = inv_rkey; 634 } 635 636 /** 637 * svc_rdma_xdr_decode_req - Decode the transport header 638 * @rq_arg: xdr_buf containing ingress RPC/RDMA message 639 * @rctxt: state of decoding 640 * 641 * On entry, xdr->head[0].iov_base points to first byte of the 642 * RPC-over-RDMA transport header. 643 * 644 * On successful exit, head[0] points to first byte past the 645 * RPC-over-RDMA header. For RDMA_MSG, this is the RPC message. 646 * 647 * The length of the RPC-over-RDMA header is returned. 648 * 649 * Assumptions: 650 * - The transport header is entirely contained in the head iovec. 651 */ 652 static int svc_rdma_xdr_decode_req(struct xdr_buf *rq_arg, 653 struct svc_rdma_recv_ctxt *rctxt) 654 { 655 __be32 *p, *rdma_argp; 656 unsigned int hdr_len; 657 658 rdma_argp = rq_arg->head[0].iov_base; 659 xdr_init_decode(&rctxt->rc_stream, rq_arg, rdma_argp, NULL); 660 661 p = xdr_inline_decode(&rctxt->rc_stream, 662 rpcrdma_fixed_maxsz * sizeof(*p)); 663 if (unlikely(!p)) 664 goto out_short; 665 p++; 666 if (*p != rpcrdma_version) 667 goto out_version; 668 p += 2; 669 rctxt->rc_msgtype = *p; 670 switch (rctxt->rc_msgtype) { 671 case rdma_msg: 672 break; 673 case rdma_nomsg: 674 break; 675 case rdma_done: 676 goto out_drop; 677 case rdma_error: 678 goto out_drop; 679 default: 680 goto out_proc; 681 } 682 683 if (!xdr_check_read_list(rctxt)) 684 goto out_inval; 685 if (!xdr_check_write_list(rctxt)) 686 goto out_inval; 687 if (!xdr_check_reply_chunk(rctxt)) 688 goto out_inval; 689 690 rq_arg->head[0].iov_base = rctxt->rc_stream.p; 691 hdr_len = xdr_stream_pos(&rctxt->rc_stream); 692 rq_arg->head[0].iov_len -= hdr_len; 693 rq_arg->len -= hdr_len; 694 trace_svcrdma_decode_rqst(rctxt, rdma_argp, hdr_len); 695 return hdr_len; 696 697 out_short: 698 trace_svcrdma_decode_short_err(rctxt, rq_arg->len); 699 return -EINVAL; 700 701 out_version: 702 trace_svcrdma_decode_badvers_err(rctxt, rdma_argp); 703 return -EPROTONOSUPPORT; 704 705 out_drop: 706 trace_svcrdma_decode_drop_err(rctxt, rdma_argp); 707 return 0; 708 709 out_proc: 710 trace_svcrdma_decode_badproc_err(rctxt, rdma_argp); 711 return -EINVAL; 712 713 out_inval: 714 trace_svcrdma_decode_parse_err(rctxt, rdma_argp); 715 return -EINVAL; 716 } 717 718 static void svc_rdma_send_error(struct svcxprt_rdma *rdma, 719 struct svc_rdma_recv_ctxt *rctxt, 720 int status) 721 { 722 struct svc_rdma_send_ctxt *sctxt; 723 724 sctxt = svc_rdma_send_ctxt_get(rdma); 725 if (!sctxt) 726 return; 727 svc_rdma_send_error_msg(rdma, sctxt, rctxt, status); 728 } 729 730 /* By convention, backchannel calls arrive via rdma_msg type 731 * messages, and never populate the chunk lists. This makes 732 * the RPC/RDMA header small and fixed in size, so it is 733 * straightforward to check the RPC header's direction field. 734 */ 735 static bool svc_rdma_is_reverse_direction_reply(struct svc_xprt *xprt, 736 struct svc_rdma_recv_ctxt *rctxt) 737 { 738 __be32 *p = rctxt->rc_recv_buf; 739 740 if (!xprt->xpt_bc_xprt) 741 return false; 742 743 if (rctxt->rc_msgtype != rdma_msg) 744 return false; 745 746 if (!pcl_is_empty(&rctxt->rc_call_pcl)) 747 return false; 748 if (!pcl_is_empty(&rctxt->rc_read_pcl)) 749 return false; 750 if (!pcl_is_empty(&rctxt->rc_write_pcl)) 751 return false; 752 if (!pcl_is_empty(&rctxt->rc_reply_pcl)) 753 return false; 754 755 /* RPC call direction */ 756 if (*(p + 8) == cpu_to_be32(RPC_CALL)) 757 return false; 758 759 return true; 760 } 761 762 /** 763 * svc_rdma_recvfrom - Receive an RPC call 764 * @rqstp: request structure into which to receive an RPC Call 765 * 766 * Returns: 767 * The positive number of bytes in the RPC Call message, 768 * %0 if there were no Calls ready to return, 769 * %-EINVAL if the Read chunk data is too large, 770 * %-ENOMEM if rdma_rw context pool was exhausted, 771 * %-ENOTCONN if posting failed (connection is lost), 772 * %-EIO if rdma_rw initialization failed (DMA mapping, etc). 773 * 774 * Called in a loop when XPT_DATA is set. XPT_DATA is cleared only 775 * when there are no remaining ctxt's to process. 776 * 777 * The next ctxt is removed from the "receive" lists. 778 * 779 * - If the ctxt completes a Read, then finish assembling the Call 780 * message and return the number of bytes in the message. 781 * 782 * - If the ctxt completes a Receive, then construct the Call 783 * message from the contents of the Receive buffer. 784 * 785 * - If there are no Read chunks in this message, then finish 786 * assembling the Call message and return the number of bytes 787 * in the message. 788 * 789 * - If there are Read chunks in this message, post Read WRs to 790 * pull that payload and return 0. 791 */ 792 int svc_rdma_recvfrom(struct svc_rqst *rqstp) 793 { 794 struct svc_xprt *xprt = rqstp->rq_xprt; 795 struct svcxprt_rdma *rdma_xprt = 796 container_of(xprt, struct svcxprt_rdma, sc_xprt); 797 struct svc_rdma_recv_ctxt *ctxt; 798 int ret; 799 800 rqstp->rq_xprt_ctxt = NULL; 801 802 ctxt = NULL; 803 spin_lock(&rdma_xprt->sc_rq_dto_lock); 804 ctxt = svc_rdma_next_recv_ctxt(&rdma_xprt->sc_rq_dto_q); 805 if (ctxt) 806 list_del(&ctxt->rc_list); 807 else 808 /* No new incoming requests, terminate the loop */ 809 clear_bit(XPT_DATA, &xprt->xpt_flags); 810 spin_unlock(&rdma_xprt->sc_rq_dto_lock); 811 812 /* Unblock the transport for the next receive */ 813 svc_xprt_received(xprt); 814 if (!ctxt) 815 return 0; 816 817 percpu_counter_inc(&svcrdma_stat_recv); 818 ib_dma_sync_single_for_cpu(rdma_xprt->sc_pd->device, 819 ctxt->rc_recv_sge.addr, ctxt->rc_byte_len, 820 DMA_FROM_DEVICE); 821 svc_rdma_build_arg_xdr(rqstp, ctxt); 822 823 /* Prevent svc_xprt_release from releasing pages in rq_pages 824 * if we return 0 or an error. 825 */ 826 rqstp->rq_respages = rqstp->rq_pages; 827 rqstp->rq_next_page = rqstp->rq_respages; 828 829 ret = svc_rdma_xdr_decode_req(&rqstp->rq_arg, ctxt); 830 if (ret < 0) 831 goto out_err; 832 if (ret == 0) 833 goto out_drop; 834 rqstp->rq_xprt_hlen = ret; 835 836 if (svc_rdma_is_reverse_direction_reply(xprt, ctxt)) 837 goto out_backchannel; 838 839 svc_rdma_get_inv_rkey(rdma_xprt, ctxt); 840 841 if (!pcl_is_empty(&ctxt->rc_read_pcl) || 842 !pcl_is_empty(&ctxt->rc_call_pcl)) { 843 ret = svc_rdma_process_read_list(rdma_xprt, rqstp, ctxt); 844 if (ret < 0) 845 goto out_readfail; 846 } 847 848 rqstp->rq_xprt_ctxt = ctxt; 849 rqstp->rq_prot = IPPROTO_MAX; 850 svc_xprt_copy_addrs(rqstp, xprt); 851 return rqstp->rq_arg.len; 852 853 out_err: 854 svc_rdma_send_error(rdma_xprt, ctxt, ret); 855 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 856 return 0; 857 858 out_readfail: 859 if (ret == -EINVAL) 860 svc_rdma_send_error(rdma_xprt, ctxt, ret); 861 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 862 return ret; 863 864 out_backchannel: 865 svc_rdma_handle_bc_reply(rqstp, ctxt); 866 out_drop: 867 svc_rdma_recv_ctxt_put(rdma_xprt, ctxt); 868 return 0; 869 } 870