1 // SPDX-License-Identifier: GPL-2.0 OR BSD-3-Clause 2 /* 3 * Copyright (c) 2014-2017 Oracle. All rights reserved. 4 * Copyright (c) 2003-2007 Network Appliance, Inc. All rights reserved. 5 * 6 * This software is available to you under a choice of one of two 7 * licenses. You may choose to be licensed under the terms of the GNU 8 * General Public License (GPL) Version 2, available from the file 9 * COPYING in the main directory of this source tree, or the BSD-type 10 * license below: 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 16 * Redistributions of source code must retain the above copyright 17 * notice, this list of conditions and the following disclaimer. 18 * 19 * Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials provided 22 * with the distribution. 23 * 24 * Neither the name of the Network Appliance, Inc. nor the names of 25 * its contributors may be used to endorse or promote products 26 * derived from this software without specific prior written 27 * permission. 28 * 29 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 30 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT 31 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR 32 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT 33 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 34 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT 35 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, 36 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY 37 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT 38 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE 39 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 40 */ 41 42 /* 43 * verbs.c 44 * 45 * Encapsulates the major functions managing: 46 * o adapters 47 * o endpoints 48 * o connections 49 * o buffer memory 50 */ 51 52 #include <linux/interrupt.h> 53 #include <linux/slab.h> 54 #include <linux/sunrpc/addr.h> 55 #include <linux/sunrpc/svc_rdma.h> 56 #include <linux/log2.h> 57 58 #include <asm-generic/barrier.h> 59 #include <asm/bitops.h> 60 61 #include <rdma/ib_cm.h> 62 63 #include "xprt_rdma.h" 64 #include <trace/events/rpcrdma.h> 65 66 /* 67 * Globals/Macros 68 */ 69 70 #if IS_ENABLED(CONFIG_SUNRPC_DEBUG) 71 # define RPCDBG_FACILITY RPCDBG_TRANS 72 #endif 73 74 /* 75 * internal functions 76 */ 77 static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt); 78 static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt); 79 static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt, 80 struct rpcrdma_sendctx *sc); 81 static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt); 82 static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt); 83 static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep); 84 static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt); 85 static void rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt); 86 static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt); 87 static void rpcrdma_ep_get(struct rpcrdma_ep *ep); 88 static int rpcrdma_ep_put(struct rpcrdma_ep *ep); 89 static struct rpcrdma_regbuf * 90 rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction, 91 gfp_t flags); 92 static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb); 93 static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb); 94 95 /* Wait for outstanding transport work to finish. ib_drain_qp 96 * handles the drains in the wrong order for us, so open code 97 * them here. 98 */ 99 static void rpcrdma_xprt_drain(struct rpcrdma_xprt *r_xprt) 100 { 101 struct rpcrdma_ep *ep = r_xprt->rx_ep; 102 struct rdma_cm_id *id = ep->re_id; 103 104 /* Flush Receives, then wait for deferred Reply work 105 * to complete. 106 */ 107 ib_drain_rq(id->qp); 108 109 /* Deferred Reply processing might have scheduled 110 * local invalidations. 111 */ 112 ib_drain_sq(id->qp); 113 114 rpcrdma_ep_put(ep); 115 } 116 117 /** 118 * rpcrdma_qp_event_handler - Handle one QP event (error notification) 119 * @event: details of the event 120 * @context: ep that owns QP where event occurred 121 * 122 * Called from the RDMA provider (device driver) possibly in an interrupt 123 * context. The QP is always destroyed before the ID, so the ID will be 124 * reliably available when this handler is invoked. 125 */ 126 static void rpcrdma_qp_event_handler(struct ib_event *event, void *context) 127 { 128 struct rpcrdma_ep *ep = context; 129 130 trace_xprtrdma_qp_event(ep, event); 131 } 132 133 /* Ensure xprt_force_disconnect() is invoked exactly once when a 134 * connection is closed or lost. (The important thing is it needs 135 * to be invoked "at least" once). 136 */ 137 static void rpcrdma_force_disconnect(struct rpcrdma_ep *ep) 138 { 139 if (atomic_add_unless(&ep->re_force_disconnect, 1, 1)) 140 xprt_force_disconnect(ep->re_xprt); 141 } 142 143 /** 144 * rpcrdma_flush_disconnect - Disconnect on flushed completion 145 * @r_xprt: transport to disconnect 146 * @wc: work completion entry 147 * 148 * Must be called in process context. 149 */ 150 void rpcrdma_flush_disconnect(struct rpcrdma_xprt *r_xprt, struct ib_wc *wc) 151 { 152 if (wc->status != IB_WC_SUCCESS) 153 rpcrdma_force_disconnect(r_xprt->rx_ep); 154 } 155 156 /** 157 * rpcrdma_wc_send - Invoked by RDMA provider for each polled Send WC 158 * @cq: completion queue 159 * @wc: WCE for a completed Send WR 160 * 161 */ 162 static void rpcrdma_wc_send(struct ib_cq *cq, struct ib_wc *wc) 163 { 164 struct ib_cqe *cqe = wc->wr_cqe; 165 struct rpcrdma_sendctx *sc = 166 container_of(cqe, struct rpcrdma_sendctx, sc_cqe); 167 struct rpcrdma_xprt *r_xprt = cq->cq_context; 168 169 /* WARNING: Only wr_cqe and status are reliable at this point */ 170 trace_xprtrdma_wc_send(sc, wc); 171 rpcrdma_sendctx_put_locked(r_xprt, sc); 172 rpcrdma_flush_disconnect(r_xprt, wc); 173 } 174 175 /** 176 * rpcrdma_wc_receive - Invoked by RDMA provider for each polled Receive WC 177 * @cq: completion queue 178 * @wc: WCE for a completed Receive WR 179 * 180 */ 181 static void rpcrdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc) 182 { 183 struct ib_cqe *cqe = wc->wr_cqe; 184 struct rpcrdma_rep *rep = container_of(cqe, struct rpcrdma_rep, 185 rr_cqe); 186 struct rpcrdma_xprt *r_xprt = cq->cq_context; 187 188 /* WARNING: Only wr_cqe and status are reliable at this point */ 189 trace_xprtrdma_wc_receive(wc); 190 --r_xprt->rx_ep->re_receive_count; 191 if (wc->status != IB_WC_SUCCESS) 192 goto out_flushed; 193 194 /* status == SUCCESS means all fields in wc are trustworthy */ 195 rpcrdma_set_xdrlen(&rep->rr_hdrbuf, wc->byte_len); 196 rep->rr_wc_flags = wc->wc_flags; 197 rep->rr_inv_rkey = wc->ex.invalidate_rkey; 198 199 ib_dma_sync_single_for_cpu(rdmab_device(rep->rr_rdmabuf), 200 rdmab_addr(rep->rr_rdmabuf), 201 wc->byte_len, DMA_FROM_DEVICE); 202 203 rpcrdma_reply_handler(rep); 204 return; 205 206 out_flushed: 207 rpcrdma_flush_disconnect(r_xprt, wc); 208 rpcrdma_rep_destroy(rep); 209 } 210 211 static void rpcrdma_update_cm_private(struct rpcrdma_ep *ep, 212 struct rdma_conn_param *param) 213 { 214 const struct rpcrdma_connect_private *pmsg = param->private_data; 215 unsigned int rsize, wsize; 216 217 /* Default settings for RPC-over-RDMA Version One */ 218 ep->re_implicit_roundup = xprt_rdma_pad_optimize; 219 rsize = RPCRDMA_V1_DEF_INLINE_SIZE; 220 wsize = RPCRDMA_V1_DEF_INLINE_SIZE; 221 222 if (pmsg && 223 pmsg->cp_magic == rpcrdma_cmp_magic && 224 pmsg->cp_version == RPCRDMA_CMP_VERSION) { 225 ep->re_implicit_roundup = true; 226 rsize = rpcrdma_decode_buffer_size(pmsg->cp_send_size); 227 wsize = rpcrdma_decode_buffer_size(pmsg->cp_recv_size); 228 } 229 230 if (rsize < ep->re_inline_recv) 231 ep->re_inline_recv = rsize; 232 if (wsize < ep->re_inline_send) 233 ep->re_inline_send = wsize; 234 235 rpcrdma_set_max_header_sizes(ep); 236 } 237 238 /** 239 * rpcrdma_cm_event_handler - Handle RDMA CM events 240 * @id: rdma_cm_id on which an event has occurred 241 * @event: details of the event 242 * 243 * Called with @id's mutex held. Returns 1 if caller should 244 * destroy @id, otherwise 0. 245 */ 246 static int 247 rpcrdma_cm_event_handler(struct rdma_cm_id *id, struct rdma_cm_event *event) 248 { 249 struct sockaddr *sap = (struct sockaddr *)&id->route.addr.dst_addr; 250 struct rpcrdma_ep *ep = id->context; 251 252 might_sleep(); 253 254 switch (event->event) { 255 case RDMA_CM_EVENT_ADDR_RESOLVED: 256 case RDMA_CM_EVENT_ROUTE_RESOLVED: 257 ep->re_async_rc = 0; 258 complete(&ep->re_done); 259 return 0; 260 case RDMA_CM_EVENT_ADDR_ERROR: 261 ep->re_async_rc = -EPROTO; 262 complete(&ep->re_done); 263 return 0; 264 case RDMA_CM_EVENT_ROUTE_ERROR: 265 ep->re_async_rc = -ENETUNREACH; 266 complete(&ep->re_done); 267 return 0; 268 case RDMA_CM_EVENT_DEVICE_REMOVAL: 269 pr_info("rpcrdma: removing device %s for %pISpc\n", 270 ep->re_id->device->name, sap); 271 fallthrough; 272 case RDMA_CM_EVENT_ADDR_CHANGE: 273 ep->re_connect_status = -ENODEV; 274 goto disconnected; 275 case RDMA_CM_EVENT_ESTABLISHED: 276 rpcrdma_ep_get(ep); 277 ep->re_connect_status = 1; 278 rpcrdma_update_cm_private(ep, &event->param.conn); 279 trace_xprtrdma_inline_thresh(ep); 280 wake_up_all(&ep->re_connect_wait); 281 break; 282 case RDMA_CM_EVENT_CONNECT_ERROR: 283 ep->re_connect_status = -ENOTCONN; 284 goto wake_connect_worker; 285 case RDMA_CM_EVENT_UNREACHABLE: 286 ep->re_connect_status = -ENETUNREACH; 287 goto wake_connect_worker; 288 case RDMA_CM_EVENT_REJECTED: 289 dprintk("rpcrdma: connection to %pISpc rejected: %s\n", 290 sap, rdma_reject_msg(id, event->status)); 291 ep->re_connect_status = -ECONNREFUSED; 292 if (event->status == IB_CM_REJ_STALE_CONN) 293 ep->re_connect_status = -ENOTCONN; 294 wake_connect_worker: 295 wake_up_all(&ep->re_connect_wait); 296 return 0; 297 case RDMA_CM_EVENT_DISCONNECTED: 298 ep->re_connect_status = -ECONNABORTED; 299 disconnected: 300 rpcrdma_force_disconnect(ep); 301 return rpcrdma_ep_put(ep); 302 default: 303 break; 304 } 305 306 dprintk("RPC: %s: %pISpc on %s/frwr: %s\n", __func__, sap, 307 ep->re_id->device->name, rdma_event_msg(event->event)); 308 return 0; 309 } 310 311 static struct rdma_cm_id *rpcrdma_create_id(struct rpcrdma_xprt *r_xprt, 312 struct rpcrdma_ep *ep) 313 { 314 unsigned long wtimeout = msecs_to_jiffies(RDMA_RESOLVE_TIMEOUT) + 1; 315 struct rpc_xprt *xprt = &r_xprt->rx_xprt; 316 struct rdma_cm_id *id; 317 int rc; 318 319 init_completion(&ep->re_done); 320 321 id = rdma_create_id(xprt->xprt_net, rpcrdma_cm_event_handler, ep, 322 RDMA_PS_TCP, IB_QPT_RC); 323 if (IS_ERR(id)) 324 return id; 325 326 ep->re_async_rc = -ETIMEDOUT; 327 rc = rdma_resolve_addr(id, NULL, (struct sockaddr *)&xprt->addr, 328 RDMA_RESOLVE_TIMEOUT); 329 if (rc) 330 goto out; 331 rc = wait_for_completion_interruptible_timeout(&ep->re_done, wtimeout); 332 if (rc < 0) 333 goto out; 334 335 rc = ep->re_async_rc; 336 if (rc) 337 goto out; 338 339 ep->re_async_rc = -ETIMEDOUT; 340 rc = rdma_resolve_route(id, RDMA_RESOLVE_TIMEOUT); 341 if (rc) 342 goto out; 343 rc = wait_for_completion_interruptible_timeout(&ep->re_done, wtimeout); 344 if (rc < 0) 345 goto out; 346 rc = ep->re_async_rc; 347 if (rc) 348 goto out; 349 350 return id; 351 352 out: 353 rdma_destroy_id(id); 354 return ERR_PTR(rc); 355 } 356 357 static void rpcrdma_ep_destroy(struct kref *kref) 358 { 359 struct rpcrdma_ep *ep = container_of(kref, struct rpcrdma_ep, re_kref); 360 361 if (ep->re_id->qp) { 362 rdma_destroy_qp(ep->re_id); 363 ep->re_id->qp = NULL; 364 } 365 366 if (ep->re_attr.recv_cq) 367 ib_free_cq(ep->re_attr.recv_cq); 368 ep->re_attr.recv_cq = NULL; 369 if (ep->re_attr.send_cq) 370 ib_free_cq(ep->re_attr.send_cq); 371 ep->re_attr.send_cq = NULL; 372 373 if (ep->re_pd) 374 ib_dealloc_pd(ep->re_pd); 375 ep->re_pd = NULL; 376 377 kfree(ep); 378 module_put(THIS_MODULE); 379 } 380 381 static noinline void rpcrdma_ep_get(struct rpcrdma_ep *ep) 382 { 383 kref_get(&ep->re_kref); 384 } 385 386 /* Returns: 387 * %0 if @ep still has a positive kref count, or 388 * %1 if @ep was destroyed successfully. 389 */ 390 static noinline int rpcrdma_ep_put(struct rpcrdma_ep *ep) 391 { 392 return kref_put(&ep->re_kref, rpcrdma_ep_destroy); 393 } 394 395 static int rpcrdma_ep_create(struct rpcrdma_xprt *r_xprt) 396 { 397 struct rpcrdma_connect_private *pmsg; 398 struct ib_device *device; 399 struct rdma_cm_id *id; 400 struct rpcrdma_ep *ep; 401 int rc; 402 403 ep = kzalloc(sizeof(*ep), GFP_NOFS); 404 if (!ep) 405 return -ENOTCONN; 406 ep->re_xprt = &r_xprt->rx_xprt; 407 kref_init(&ep->re_kref); 408 409 id = rpcrdma_create_id(r_xprt, ep); 410 if (IS_ERR(id)) { 411 kfree(ep); 412 return PTR_ERR(id); 413 } 414 __module_get(THIS_MODULE); 415 device = id->device; 416 ep->re_id = id; 417 418 ep->re_max_requests = r_xprt->rx_xprt.max_reqs; 419 ep->re_inline_send = xprt_rdma_max_inline_write; 420 ep->re_inline_recv = xprt_rdma_max_inline_read; 421 rc = frwr_query_device(ep, device); 422 if (rc) 423 goto out_destroy; 424 425 r_xprt->rx_buf.rb_max_requests = cpu_to_be32(ep->re_max_requests); 426 427 ep->re_attr.event_handler = rpcrdma_qp_event_handler; 428 ep->re_attr.qp_context = ep; 429 ep->re_attr.srq = NULL; 430 ep->re_attr.cap.max_inline_data = 0; 431 ep->re_attr.sq_sig_type = IB_SIGNAL_REQ_WR; 432 ep->re_attr.qp_type = IB_QPT_RC; 433 ep->re_attr.port_num = ~0; 434 435 dprintk("RPC: %s: requested max: dtos: send %d recv %d; " 436 "iovs: send %d recv %d\n", 437 __func__, 438 ep->re_attr.cap.max_send_wr, 439 ep->re_attr.cap.max_recv_wr, 440 ep->re_attr.cap.max_send_sge, 441 ep->re_attr.cap.max_recv_sge); 442 443 ep->re_send_batch = ep->re_max_requests >> 3; 444 ep->re_send_count = ep->re_send_batch; 445 init_waitqueue_head(&ep->re_connect_wait); 446 447 ep->re_attr.send_cq = ib_alloc_cq_any(device, r_xprt, 448 ep->re_attr.cap.max_send_wr, 449 IB_POLL_WORKQUEUE); 450 if (IS_ERR(ep->re_attr.send_cq)) { 451 rc = PTR_ERR(ep->re_attr.send_cq); 452 goto out_destroy; 453 } 454 455 ep->re_attr.recv_cq = ib_alloc_cq_any(device, r_xprt, 456 ep->re_attr.cap.max_recv_wr, 457 IB_POLL_WORKQUEUE); 458 if (IS_ERR(ep->re_attr.recv_cq)) { 459 rc = PTR_ERR(ep->re_attr.recv_cq); 460 goto out_destroy; 461 } 462 ep->re_receive_count = 0; 463 464 /* Initialize cma parameters */ 465 memset(&ep->re_remote_cma, 0, sizeof(ep->re_remote_cma)); 466 467 /* Prepare RDMA-CM private message */ 468 pmsg = &ep->re_cm_private; 469 pmsg->cp_magic = rpcrdma_cmp_magic; 470 pmsg->cp_version = RPCRDMA_CMP_VERSION; 471 pmsg->cp_flags |= RPCRDMA_CMP_F_SND_W_INV_OK; 472 pmsg->cp_send_size = rpcrdma_encode_buffer_size(ep->re_inline_send); 473 pmsg->cp_recv_size = rpcrdma_encode_buffer_size(ep->re_inline_recv); 474 ep->re_remote_cma.private_data = pmsg; 475 ep->re_remote_cma.private_data_len = sizeof(*pmsg); 476 477 /* Client offers RDMA Read but does not initiate */ 478 ep->re_remote_cma.initiator_depth = 0; 479 ep->re_remote_cma.responder_resources = 480 min_t(int, U8_MAX, device->attrs.max_qp_rd_atom); 481 482 /* Limit transport retries so client can detect server 483 * GID changes quickly. RPC layer handles re-establishing 484 * transport connection and retransmission. 485 */ 486 ep->re_remote_cma.retry_count = 6; 487 488 /* RPC-over-RDMA handles its own flow control. In addition, 489 * make all RNR NAKs visible so we know that RPC-over-RDMA 490 * flow control is working correctly (no NAKs should be seen). 491 */ 492 ep->re_remote_cma.flow_control = 0; 493 ep->re_remote_cma.rnr_retry_count = 0; 494 495 ep->re_pd = ib_alloc_pd(device, 0); 496 if (IS_ERR(ep->re_pd)) { 497 rc = PTR_ERR(ep->re_pd); 498 goto out_destroy; 499 } 500 501 rc = rdma_create_qp(id, ep->re_pd, &ep->re_attr); 502 if (rc) 503 goto out_destroy; 504 505 r_xprt->rx_ep = ep; 506 return 0; 507 508 out_destroy: 509 rpcrdma_ep_put(ep); 510 rdma_destroy_id(id); 511 return rc; 512 } 513 514 /** 515 * rpcrdma_xprt_connect - Connect an unconnected transport 516 * @r_xprt: controlling transport instance 517 * 518 * Returns 0 on success or a negative errno. 519 */ 520 int rpcrdma_xprt_connect(struct rpcrdma_xprt *r_xprt) 521 { 522 struct rpc_xprt *xprt = &r_xprt->rx_xprt; 523 struct rpcrdma_ep *ep; 524 int rc; 525 526 rc = rpcrdma_ep_create(r_xprt); 527 if (rc) 528 return rc; 529 ep = r_xprt->rx_ep; 530 531 xprt_clear_connected(xprt); 532 rpcrdma_reset_cwnd(r_xprt); 533 534 /* Bump the ep's reference count while there are 535 * outstanding Receives. 536 */ 537 rpcrdma_ep_get(ep); 538 rpcrdma_post_recvs(r_xprt, true); 539 540 rc = rdma_connect(ep->re_id, &ep->re_remote_cma); 541 if (rc) 542 goto out; 543 544 if (xprt->reestablish_timeout < RPCRDMA_INIT_REEST_TO) 545 xprt->reestablish_timeout = RPCRDMA_INIT_REEST_TO; 546 wait_event_interruptible(ep->re_connect_wait, 547 ep->re_connect_status != 0); 548 if (ep->re_connect_status <= 0) { 549 rc = ep->re_connect_status; 550 goto out; 551 } 552 553 rc = rpcrdma_sendctxs_create(r_xprt); 554 if (rc) { 555 rc = -ENOTCONN; 556 goto out; 557 } 558 559 rc = rpcrdma_reqs_setup(r_xprt); 560 if (rc) { 561 rc = -ENOTCONN; 562 goto out; 563 } 564 rpcrdma_mrs_create(r_xprt); 565 566 out: 567 trace_xprtrdma_connect(r_xprt, rc); 568 return rc; 569 } 570 571 /** 572 * rpcrdma_xprt_disconnect - Disconnect underlying transport 573 * @r_xprt: controlling transport instance 574 * 575 * Caller serializes. Either the transport send lock is held, 576 * or we're being called to destroy the transport. 577 * 578 * On return, @r_xprt is completely divested of all hardware 579 * resources and prepared for the next ->connect operation. 580 */ 581 void rpcrdma_xprt_disconnect(struct rpcrdma_xprt *r_xprt) 582 { 583 struct rpcrdma_ep *ep = r_xprt->rx_ep; 584 struct rdma_cm_id *id; 585 int rc; 586 587 if (!ep) 588 return; 589 590 id = ep->re_id; 591 rc = rdma_disconnect(id); 592 trace_xprtrdma_disconnect(r_xprt, rc); 593 594 rpcrdma_xprt_drain(r_xprt); 595 rpcrdma_reps_unmap(r_xprt); 596 rpcrdma_reqs_reset(r_xprt); 597 rpcrdma_mrs_destroy(r_xprt); 598 rpcrdma_sendctxs_destroy(r_xprt); 599 600 if (rpcrdma_ep_put(ep)) 601 rdma_destroy_id(id); 602 603 r_xprt->rx_ep = NULL; 604 } 605 606 /* Fixed-size circular FIFO queue. This implementation is wait-free and 607 * lock-free. 608 * 609 * Consumer is the code path that posts Sends. This path dequeues a 610 * sendctx for use by a Send operation. Multiple consumer threads 611 * are serialized by the RPC transport lock, which allows only one 612 * ->send_request call at a time. 613 * 614 * Producer is the code path that handles Send completions. This path 615 * enqueues a sendctx that has been completed. Multiple producer 616 * threads are serialized by the ib_poll_cq() function. 617 */ 618 619 /* rpcrdma_sendctxs_destroy() assumes caller has already quiesced 620 * queue activity, and rpcrdma_xprt_drain has flushed all remaining 621 * Send requests. 622 */ 623 static void rpcrdma_sendctxs_destroy(struct rpcrdma_xprt *r_xprt) 624 { 625 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 626 unsigned long i; 627 628 if (!buf->rb_sc_ctxs) 629 return; 630 for (i = 0; i <= buf->rb_sc_last; i++) 631 kfree(buf->rb_sc_ctxs[i]); 632 kfree(buf->rb_sc_ctxs); 633 buf->rb_sc_ctxs = NULL; 634 } 635 636 static struct rpcrdma_sendctx *rpcrdma_sendctx_create(struct rpcrdma_ep *ep) 637 { 638 struct rpcrdma_sendctx *sc; 639 640 sc = kzalloc(struct_size(sc, sc_sges, ep->re_attr.cap.max_send_sge), 641 GFP_KERNEL); 642 if (!sc) 643 return NULL; 644 645 sc->sc_cqe.done = rpcrdma_wc_send; 646 return sc; 647 } 648 649 static int rpcrdma_sendctxs_create(struct rpcrdma_xprt *r_xprt) 650 { 651 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 652 struct rpcrdma_sendctx *sc; 653 unsigned long i; 654 655 /* Maximum number of concurrent outstanding Send WRs. Capping 656 * the circular queue size stops Send Queue overflow by causing 657 * the ->send_request call to fail temporarily before too many 658 * Sends are posted. 659 */ 660 i = r_xprt->rx_ep->re_max_requests + RPCRDMA_MAX_BC_REQUESTS; 661 buf->rb_sc_ctxs = kcalloc(i, sizeof(sc), GFP_KERNEL); 662 if (!buf->rb_sc_ctxs) 663 return -ENOMEM; 664 665 buf->rb_sc_last = i - 1; 666 for (i = 0; i <= buf->rb_sc_last; i++) { 667 sc = rpcrdma_sendctx_create(r_xprt->rx_ep); 668 if (!sc) 669 return -ENOMEM; 670 671 buf->rb_sc_ctxs[i] = sc; 672 } 673 674 buf->rb_sc_head = 0; 675 buf->rb_sc_tail = 0; 676 return 0; 677 } 678 679 /* The sendctx queue is not guaranteed to have a size that is a 680 * power of two, thus the helpers in circ_buf.h cannot be used. 681 * The other option is to use modulus (%), which can be expensive. 682 */ 683 static unsigned long rpcrdma_sendctx_next(struct rpcrdma_buffer *buf, 684 unsigned long item) 685 { 686 return likely(item < buf->rb_sc_last) ? item + 1 : 0; 687 } 688 689 /** 690 * rpcrdma_sendctx_get_locked - Acquire a send context 691 * @r_xprt: controlling transport instance 692 * 693 * Returns pointer to a free send completion context; or NULL if 694 * the queue is empty. 695 * 696 * Usage: Called to acquire an SGE array before preparing a Send WR. 697 * 698 * The caller serializes calls to this function (per transport), and 699 * provides an effective memory barrier that flushes the new value 700 * of rb_sc_head. 701 */ 702 struct rpcrdma_sendctx *rpcrdma_sendctx_get_locked(struct rpcrdma_xprt *r_xprt) 703 { 704 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 705 struct rpcrdma_sendctx *sc; 706 unsigned long next_head; 707 708 next_head = rpcrdma_sendctx_next(buf, buf->rb_sc_head); 709 710 if (next_head == READ_ONCE(buf->rb_sc_tail)) 711 goto out_emptyq; 712 713 /* ORDER: item must be accessed _before_ head is updated */ 714 sc = buf->rb_sc_ctxs[next_head]; 715 716 /* Releasing the lock in the caller acts as a memory 717 * barrier that flushes rb_sc_head. 718 */ 719 buf->rb_sc_head = next_head; 720 721 return sc; 722 723 out_emptyq: 724 /* The queue is "empty" if there have not been enough Send 725 * completions recently. This is a sign the Send Queue is 726 * backing up. Cause the caller to pause and try again. 727 */ 728 xprt_wait_for_buffer_space(&r_xprt->rx_xprt); 729 r_xprt->rx_stats.empty_sendctx_q++; 730 return NULL; 731 } 732 733 /** 734 * rpcrdma_sendctx_put_locked - Release a send context 735 * @r_xprt: controlling transport instance 736 * @sc: send context to release 737 * 738 * Usage: Called from Send completion to return a sendctxt 739 * to the queue. 740 * 741 * The caller serializes calls to this function (per transport). 742 */ 743 static void rpcrdma_sendctx_put_locked(struct rpcrdma_xprt *r_xprt, 744 struct rpcrdma_sendctx *sc) 745 { 746 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 747 unsigned long next_tail; 748 749 /* Unmap SGEs of previously completed but unsignaled 750 * Sends by walking up the queue until @sc is found. 751 */ 752 next_tail = buf->rb_sc_tail; 753 do { 754 next_tail = rpcrdma_sendctx_next(buf, next_tail); 755 756 /* ORDER: item must be accessed _before_ tail is updated */ 757 rpcrdma_sendctx_unmap(buf->rb_sc_ctxs[next_tail]); 758 759 } while (buf->rb_sc_ctxs[next_tail] != sc); 760 761 /* Paired with READ_ONCE */ 762 smp_store_release(&buf->rb_sc_tail, next_tail); 763 764 xprt_write_space(&r_xprt->rx_xprt); 765 } 766 767 static void 768 rpcrdma_mrs_create(struct rpcrdma_xprt *r_xprt) 769 { 770 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 771 struct rpcrdma_ep *ep = r_xprt->rx_ep; 772 unsigned int count; 773 774 for (count = 0; count < ep->re_max_rdma_segs; count++) { 775 struct rpcrdma_mr *mr; 776 int rc; 777 778 mr = kzalloc(sizeof(*mr), GFP_NOFS); 779 if (!mr) 780 break; 781 782 rc = frwr_mr_init(r_xprt, mr); 783 if (rc) { 784 kfree(mr); 785 break; 786 } 787 788 spin_lock(&buf->rb_lock); 789 rpcrdma_mr_push(mr, &buf->rb_mrs); 790 list_add(&mr->mr_all, &buf->rb_all_mrs); 791 spin_unlock(&buf->rb_lock); 792 } 793 794 r_xprt->rx_stats.mrs_allocated += count; 795 trace_xprtrdma_createmrs(r_xprt, count); 796 } 797 798 static void 799 rpcrdma_mr_refresh_worker(struct work_struct *work) 800 { 801 struct rpcrdma_buffer *buf = container_of(work, struct rpcrdma_buffer, 802 rb_refresh_worker); 803 struct rpcrdma_xprt *r_xprt = container_of(buf, struct rpcrdma_xprt, 804 rx_buf); 805 806 rpcrdma_mrs_create(r_xprt); 807 xprt_write_space(&r_xprt->rx_xprt); 808 } 809 810 /** 811 * rpcrdma_mrs_refresh - Wake the MR refresh worker 812 * @r_xprt: controlling transport instance 813 * 814 */ 815 void rpcrdma_mrs_refresh(struct rpcrdma_xprt *r_xprt) 816 { 817 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 818 struct rpcrdma_ep *ep = r_xprt->rx_ep; 819 820 /* If there is no underlying connection, it's no use 821 * to wake the refresh worker. 822 */ 823 if (ep->re_connect_status == 1) { 824 /* The work is scheduled on a WQ_MEM_RECLAIM 825 * workqueue in order to prevent MR allocation 826 * from recursing into NFS during direct reclaim. 827 */ 828 queue_work(xprtiod_workqueue, &buf->rb_refresh_worker); 829 } 830 } 831 832 /** 833 * rpcrdma_req_create - Allocate an rpcrdma_req object 834 * @r_xprt: controlling r_xprt 835 * @size: initial size, in bytes, of send and receive buffers 836 * @flags: GFP flags passed to memory allocators 837 * 838 * Returns an allocated and fully initialized rpcrdma_req or NULL. 839 */ 840 struct rpcrdma_req *rpcrdma_req_create(struct rpcrdma_xprt *r_xprt, size_t size, 841 gfp_t flags) 842 { 843 struct rpcrdma_buffer *buffer = &r_xprt->rx_buf; 844 struct rpcrdma_req *req; 845 846 req = kzalloc(sizeof(*req), flags); 847 if (req == NULL) 848 goto out1; 849 850 req->rl_sendbuf = rpcrdma_regbuf_alloc(size, DMA_TO_DEVICE, flags); 851 if (!req->rl_sendbuf) 852 goto out2; 853 854 req->rl_recvbuf = rpcrdma_regbuf_alloc(size, DMA_NONE, flags); 855 if (!req->rl_recvbuf) 856 goto out3; 857 858 INIT_LIST_HEAD(&req->rl_free_mrs); 859 INIT_LIST_HEAD(&req->rl_registered); 860 spin_lock(&buffer->rb_lock); 861 list_add(&req->rl_all, &buffer->rb_allreqs); 862 spin_unlock(&buffer->rb_lock); 863 return req; 864 865 out3: 866 kfree(req->rl_sendbuf); 867 out2: 868 kfree(req); 869 out1: 870 return NULL; 871 } 872 873 /** 874 * rpcrdma_req_setup - Per-connection instance setup of an rpcrdma_req object 875 * @r_xprt: controlling transport instance 876 * @req: rpcrdma_req object to set up 877 * 878 * Returns zero on success, and a negative errno on failure. 879 */ 880 int rpcrdma_req_setup(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) 881 { 882 struct rpcrdma_regbuf *rb; 883 size_t maxhdrsize; 884 885 /* Compute maximum header buffer size in bytes */ 886 maxhdrsize = rpcrdma_fixed_maxsz + 3 + 887 r_xprt->rx_ep->re_max_rdma_segs * rpcrdma_readchunk_maxsz; 888 maxhdrsize *= sizeof(__be32); 889 rb = rpcrdma_regbuf_alloc(__roundup_pow_of_two(maxhdrsize), 890 DMA_TO_DEVICE, GFP_KERNEL); 891 if (!rb) 892 goto out; 893 894 if (!__rpcrdma_regbuf_dma_map(r_xprt, rb)) 895 goto out_free; 896 897 req->rl_rdmabuf = rb; 898 xdr_buf_init(&req->rl_hdrbuf, rdmab_data(rb), rdmab_length(rb)); 899 return 0; 900 901 out_free: 902 rpcrdma_regbuf_free(rb); 903 out: 904 return -ENOMEM; 905 } 906 907 /* ASSUMPTION: the rb_allreqs list is stable for the duration, 908 * and thus can be walked without holding rb_lock. Eg. the 909 * caller is holding the transport send lock to exclude 910 * device removal or disconnection. 911 */ 912 static int rpcrdma_reqs_setup(struct rpcrdma_xprt *r_xprt) 913 { 914 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 915 struct rpcrdma_req *req; 916 int rc; 917 918 list_for_each_entry(req, &buf->rb_allreqs, rl_all) { 919 rc = rpcrdma_req_setup(r_xprt, req); 920 if (rc) 921 return rc; 922 } 923 return 0; 924 } 925 926 static void rpcrdma_req_reset(struct rpcrdma_req *req) 927 { 928 /* Credits are valid for only one connection */ 929 req->rl_slot.rq_cong = 0; 930 931 rpcrdma_regbuf_free(req->rl_rdmabuf); 932 req->rl_rdmabuf = NULL; 933 934 rpcrdma_regbuf_dma_unmap(req->rl_sendbuf); 935 rpcrdma_regbuf_dma_unmap(req->rl_recvbuf); 936 937 frwr_reset(req); 938 } 939 940 /* ASSUMPTION: the rb_allreqs list is stable for the duration, 941 * and thus can be walked without holding rb_lock. Eg. the 942 * caller is holding the transport send lock to exclude 943 * device removal or disconnection. 944 */ 945 static void rpcrdma_reqs_reset(struct rpcrdma_xprt *r_xprt) 946 { 947 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 948 struct rpcrdma_req *req; 949 950 list_for_each_entry(req, &buf->rb_allreqs, rl_all) 951 rpcrdma_req_reset(req); 952 } 953 954 /* No locking needed here. This function is called only by the 955 * Receive completion handler. 956 */ 957 static noinline 958 struct rpcrdma_rep *rpcrdma_rep_create(struct rpcrdma_xprt *r_xprt, 959 bool temp) 960 { 961 struct rpcrdma_rep *rep; 962 963 rep = kzalloc(sizeof(*rep), GFP_KERNEL); 964 if (rep == NULL) 965 goto out; 966 967 rep->rr_rdmabuf = rpcrdma_regbuf_alloc(r_xprt->rx_ep->re_inline_recv, 968 DMA_FROM_DEVICE, GFP_KERNEL); 969 if (!rep->rr_rdmabuf) 970 goto out_free; 971 972 if (!rpcrdma_regbuf_dma_map(r_xprt, rep->rr_rdmabuf)) 973 goto out_free_regbuf; 974 975 xdr_buf_init(&rep->rr_hdrbuf, rdmab_data(rep->rr_rdmabuf), 976 rdmab_length(rep->rr_rdmabuf)); 977 rep->rr_cqe.done = rpcrdma_wc_receive; 978 rep->rr_rxprt = r_xprt; 979 rep->rr_recv_wr.next = NULL; 980 rep->rr_recv_wr.wr_cqe = &rep->rr_cqe; 981 rep->rr_recv_wr.sg_list = &rep->rr_rdmabuf->rg_iov; 982 rep->rr_recv_wr.num_sge = 1; 983 rep->rr_temp = temp; 984 list_add(&rep->rr_all, &r_xprt->rx_buf.rb_all_reps); 985 return rep; 986 987 out_free_regbuf: 988 rpcrdma_regbuf_free(rep->rr_rdmabuf); 989 out_free: 990 kfree(rep); 991 out: 992 return NULL; 993 } 994 995 /* No locking needed here. This function is invoked only by the 996 * Receive completion handler, or during transport shutdown. 997 */ 998 static void rpcrdma_rep_destroy(struct rpcrdma_rep *rep) 999 { 1000 list_del(&rep->rr_all); 1001 rpcrdma_regbuf_free(rep->rr_rdmabuf); 1002 kfree(rep); 1003 } 1004 1005 static struct rpcrdma_rep *rpcrdma_rep_get_locked(struct rpcrdma_buffer *buf) 1006 { 1007 struct llist_node *node; 1008 1009 /* Calls to llist_del_first are required to be serialized */ 1010 node = llist_del_first(&buf->rb_free_reps); 1011 if (!node) 1012 return NULL; 1013 return llist_entry(node, struct rpcrdma_rep, rr_node); 1014 } 1015 1016 static void rpcrdma_rep_put(struct rpcrdma_buffer *buf, 1017 struct rpcrdma_rep *rep) 1018 { 1019 llist_add(&rep->rr_node, &buf->rb_free_reps); 1020 } 1021 1022 static void rpcrdma_reps_unmap(struct rpcrdma_xprt *r_xprt) 1023 { 1024 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 1025 struct rpcrdma_rep *rep; 1026 1027 list_for_each_entry(rep, &buf->rb_all_reps, rr_all) { 1028 rpcrdma_regbuf_dma_unmap(rep->rr_rdmabuf); 1029 rep->rr_temp = true; 1030 } 1031 } 1032 1033 static void rpcrdma_reps_destroy(struct rpcrdma_buffer *buf) 1034 { 1035 struct rpcrdma_rep *rep; 1036 1037 while ((rep = rpcrdma_rep_get_locked(buf)) != NULL) 1038 rpcrdma_rep_destroy(rep); 1039 } 1040 1041 /** 1042 * rpcrdma_buffer_create - Create initial set of req/rep objects 1043 * @r_xprt: transport instance to (re)initialize 1044 * 1045 * Returns zero on success, otherwise a negative errno. 1046 */ 1047 int rpcrdma_buffer_create(struct rpcrdma_xprt *r_xprt) 1048 { 1049 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 1050 int i, rc; 1051 1052 buf->rb_bc_srv_max_requests = 0; 1053 spin_lock_init(&buf->rb_lock); 1054 INIT_LIST_HEAD(&buf->rb_mrs); 1055 INIT_LIST_HEAD(&buf->rb_all_mrs); 1056 INIT_WORK(&buf->rb_refresh_worker, rpcrdma_mr_refresh_worker); 1057 1058 INIT_LIST_HEAD(&buf->rb_send_bufs); 1059 INIT_LIST_HEAD(&buf->rb_allreqs); 1060 INIT_LIST_HEAD(&buf->rb_all_reps); 1061 1062 rc = -ENOMEM; 1063 for (i = 0; i < r_xprt->rx_xprt.max_reqs; i++) { 1064 struct rpcrdma_req *req; 1065 1066 req = rpcrdma_req_create(r_xprt, RPCRDMA_V1_DEF_INLINE_SIZE * 2, 1067 GFP_KERNEL); 1068 if (!req) 1069 goto out; 1070 list_add(&req->rl_list, &buf->rb_send_bufs); 1071 } 1072 1073 init_llist_head(&buf->rb_free_reps); 1074 1075 return 0; 1076 out: 1077 rpcrdma_buffer_destroy(buf); 1078 return rc; 1079 } 1080 1081 /** 1082 * rpcrdma_req_destroy - Destroy an rpcrdma_req object 1083 * @req: unused object to be destroyed 1084 * 1085 * Relies on caller holding the transport send lock to protect 1086 * removing req->rl_all from buf->rb_all_reqs safely. 1087 */ 1088 void rpcrdma_req_destroy(struct rpcrdma_req *req) 1089 { 1090 struct rpcrdma_mr *mr; 1091 1092 list_del(&req->rl_all); 1093 1094 while ((mr = rpcrdma_mr_pop(&req->rl_free_mrs))) { 1095 struct rpcrdma_buffer *buf = &mr->mr_xprt->rx_buf; 1096 1097 spin_lock(&buf->rb_lock); 1098 list_del(&mr->mr_all); 1099 spin_unlock(&buf->rb_lock); 1100 1101 frwr_release_mr(mr); 1102 } 1103 1104 rpcrdma_regbuf_free(req->rl_recvbuf); 1105 rpcrdma_regbuf_free(req->rl_sendbuf); 1106 rpcrdma_regbuf_free(req->rl_rdmabuf); 1107 kfree(req); 1108 } 1109 1110 /** 1111 * rpcrdma_mrs_destroy - Release all of a transport's MRs 1112 * @r_xprt: controlling transport instance 1113 * 1114 * Relies on caller holding the transport send lock to protect 1115 * removing mr->mr_list from req->rl_free_mrs safely. 1116 */ 1117 static void rpcrdma_mrs_destroy(struct rpcrdma_xprt *r_xprt) 1118 { 1119 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 1120 struct rpcrdma_mr *mr; 1121 1122 cancel_work_sync(&buf->rb_refresh_worker); 1123 1124 spin_lock(&buf->rb_lock); 1125 while ((mr = list_first_entry_or_null(&buf->rb_all_mrs, 1126 struct rpcrdma_mr, 1127 mr_all)) != NULL) { 1128 list_del(&mr->mr_list); 1129 list_del(&mr->mr_all); 1130 spin_unlock(&buf->rb_lock); 1131 1132 frwr_release_mr(mr); 1133 1134 spin_lock(&buf->rb_lock); 1135 } 1136 spin_unlock(&buf->rb_lock); 1137 } 1138 1139 /** 1140 * rpcrdma_buffer_destroy - Release all hw resources 1141 * @buf: root control block for resources 1142 * 1143 * ORDERING: relies on a prior rpcrdma_xprt_drain : 1144 * - No more Send or Receive completions can occur 1145 * - All MRs, reps, and reqs are returned to their free lists 1146 */ 1147 void 1148 rpcrdma_buffer_destroy(struct rpcrdma_buffer *buf) 1149 { 1150 rpcrdma_reps_destroy(buf); 1151 1152 while (!list_empty(&buf->rb_send_bufs)) { 1153 struct rpcrdma_req *req; 1154 1155 req = list_first_entry(&buf->rb_send_bufs, 1156 struct rpcrdma_req, rl_list); 1157 list_del(&req->rl_list); 1158 rpcrdma_req_destroy(req); 1159 } 1160 } 1161 1162 /** 1163 * rpcrdma_mr_get - Allocate an rpcrdma_mr object 1164 * @r_xprt: controlling transport 1165 * 1166 * Returns an initialized rpcrdma_mr or NULL if no free 1167 * rpcrdma_mr objects are available. 1168 */ 1169 struct rpcrdma_mr * 1170 rpcrdma_mr_get(struct rpcrdma_xprt *r_xprt) 1171 { 1172 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 1173 struct rpcrdma_mr *mr; 1174 1175 spin_lock(&buf->rb_lock); 1176 mr = rpcrdma_mr_pop(&buf->rb_mrs); 1177 spin_unlock(&buf->rb_lock); 1178 return mr; 1179 } 1180 1181 /** 1182 * rpcrdma_mr_put - DMA unmap an MR and release it 1183 * @mr: MR to release 1184 * 1185 */ 1186 void rpcrdma_mr_put(struct rpcrdma_mr *mr) 1187 { 1188 struct rpcrdma_xprt *r_xprt = mr->mr_xprt; 1189 1190 if (mr->mr_dir != DMA_NONE) { 1191 trace_xprtrdma_mr_unmap(mr); 1192 ib_dma_unmap_sg(r_xprt->rx_ep->re_id->device, 1193 mr->mr_sg, mr->mr_nents, mr->mr_dir); 1194 mr->mr_dir = DMA_NONE; 1195 } 1196 1197 rpcrdma_mr_push(mr, &mr->mr_req->rl_free_mrs); 1198 } 1199 1200 /** 1201 * rpcrdma_buffer_get - Get a request buffer 1202 * @buffers: Buffer pool from which to obtain a buffer 1203 * 1204 * Returns a fresh rpcrdma_req, or NULL if none are available. 1205 */ 1206 struct rpcrdma_req * 1207 rpcrdma_buffer_get(struct rpcrdma_buffer *buffers) 1208 { 1209 struct rpcrdma_req *req; 1210 1211 spin_lock(&buffers->rb_lock); 1212 req = list_first_entry_or_null(&buffers->rb_send_bufs, 1213 struct rpcrdma_req, rl_list); 1214 if (req) 1215 list_del_init(&req->rl_list); 1216 spin_unlock(&buffers->rb_lock); 1217 return req; 1218 } 1219 1220 /** 1221 * rpcrdma_buffer_put - Put request/reply buffers back into pool 1222 * @buffers: buffer pool 1223 * @req: object to return 1224 * 1225 */ 1226 void rpcrdma_buffer_put(struct rpcrdma_buffer *buffers, struct rpcrdma_req *req) 1227 { 1228 if (req->rl_reply) 1229 rpcrdma_rep_put(buffers, req->rl_reply); 1230 req->rl_reply = NULL; 1231 1232 spin_lock(&buffers->rb_lock); 1233 list_add(&req->rl_list, &buffers->rb_send_bufs); 1234 spin_unlock(&buffers->rb_lock); 1235 } 1236 1237 /** 1238 * rpcrdma_recv_buffer_put - Release rpcrdma_rep back to free list 1239 * @rep: rep to release 1240 * 1241 * Used after error conditions. 1242 */ 1243 void rpcrdma_recv_buffer_put(struct rpcrdma_rep *rep) 1244 { 1245 rpcrdma_rep_put(&rep->rr_rxprt->rx_buf, rep); 1246 } 1247 1248 /* Returns a pointer to a rpcrdma_regbuf object, or NULL. 1249 * 1250 * xprtrdma uses a regbuf for posting an outgoing RDMA SEND, or for 1251 * receiving the payload of RDMA RECV operations. During Long Calls 1252 * or Replies they may be registered externally via frwr_map. 1253 */ 1254 static struct rpcrdma_regbuf * 1255 rpcrdma_regbuf_alloc(size_t size, enum dma_data_direction direction, 1256 gfp_t flags) 1257 { 1258 struct rpcrdma_regbuf *rb; 1259 1260 rb = kmalloc(sizeof(*rb), flags); 1261 if (!rb) 1262 return NULL; 1263 rb->rg_data = kmalloc(size, flags); 1264 if (!rb->rg_data) { 1265 kfree(rb); 1266 return NULL; 1267 } 1268 1269 rb->rg_device = NULL; 1270 rb->rg_direction = direction; 1271 rb->rg_iov.length = size; 1272 return rb; 1273 } 1274 1275 /** 1276 * rpcrdma_regbuf_realloc - re-allocate a SEND/RECV buffer 1277 * @rb: regbuf to reallocate 1278 * @size: size of buffer to be allocated, in bytes 1279 * @flags: GFP flags 1280 * 1281 * Returns true if reallocation was successful. If false is 1282 * returned, @rb is left untouched. 1283 */ 1284 bool rpcrdma_regbuf_realloc(struct rpcrdma_regbuf *rb, size_t size, gfp_t flags) 1285 { 1286 void *buf; 1287 1288 buf = kmalloc(size, flags); 1289 if (!buf) 1290 return false; 1291 1292 rpcrdma_regbuf_dma_unmap(rb); 1293 kfree(rb->rg_data); 1294 1295 rb->rg_data = buf; 1296 rb->rg_iov.length = size; 1297 return true; 1298 } 1299 1300 /** 1301 * __rpcrdma_regbuf_dma_map - DMA-map a regbuf 1302 * @r_xprt: controlling transport instance 1303 * @rb: regbuf to be mapped 1304 * 1305 * Returns true if the buffer is now DMA mapped to @r_xprt's device 1306 */ 1307 bool __rpcrdma_regbuf_dma_map(struct rpcrdma_xprt *r_xprt, 1308 struct rpcrdma_regbuf *rb) 1309 { 1310 struct ib_device *device = r_xprt->rx_ep->re_id->device; 1311 1312 if (rb->rg_direction == DMA_NONE) 1313 return false; 1314 1315 rb->rg_iov.addr = ib_dma_map_single(device, rdmab_data(rb), 1316 rdmab_length(rb), rb->rg_direction); 1317 if (ib_dma_mapping_error(device, rdmab_addr(rb))) { 1318 trace_xprtrdma_dma_maperr(rdmab_addr(rb)); 1319 return false; 1320 } 1321 1322 rb->rg_device = device; 1323 rb->rg_iov.lkey = r_xprt->rx_ep->re_pd->local_dma_lkey; 1324 return true; 1325 } 1326 1327 static void rpcrdma_regbuf_dma_unmap(struct rpcrdma_regbuf *rb) 1328 { 1329 if (!rb) 1330 return; 1331 1332 if (!rpcrdma_regbuf_is_mapped(rb)) 1333 return; 1334 1335 ib_dma_unmap_single(rb->rg_device, rdmab_addr(rb), rdmab_length(rb), 1336 rb->rg_direction); 1337 rb->rg_device = NULL; 1338 } 1339 1340 static void rpcrdma_regbuf_free(struct rpcrdma_regbuf *rb) 1341 { 1342 rpcrdma_regbuf_dma_unmap(rb); 1343 if (rb) 1344 kfree(rb->rg_data); 1345 kfree(rb); 1346 } 1347 1348 /** 1349 * rpcrdma_post_sends - Post WRs to a transport's Send Queue 1350 * @r_xprt: controlling transport instance 1351 * @req: rpcrdma_req containing the Send WR to post 1352 * 1353 * Returns 0 if the post was successful, otherwise -ENOTCONN 1354 * is returned. 1355 */ 1356 int rpcrdma_post_sends(struct rpcrdma_xprt *r_xprt, struct rpcrdma_req *req) 1357 { 1358 struct ib_send_wr *send_wr = &req->rl_wr; 1359 struct rpcrdma_ep *ep = r_xprt->rx_ep; 1360 int rc; 1361 1362 if (!ep->re_send_count || kref_read(&req->rl_kref) > 1) { 1363 send_wr->send_flags |= IB_SEND_SIGNALED; 1364 ep->re_send_count = ep->re_send_batch; 1365 } else { 1366 send_wr->send_flags &= ~IB_SEND_SIGNALED; 1367 --ep->re_send_count; 1368 } 1369 1370 trace_xprtrdma_post_send(req); 1371 rc = frwr_send(r_xprt, req); 1372 if (rc) 1373 return -ENOTCONN; 1374 return 0; 1375 } 1376 1377 /** 1378 * rpcrdma_post_recvs - Refill the Receive Queue 1379 * @r_xprt: controlling transport instance 1380 * @temp: mark Receive buffers to be deleted after use 1381 * 1382 */ 1383 void rpcrdma_post_recvs(struct rpcrdma_xprt *r_xprt, bool temp) 1384 { 1385 struct rpcrdma_buffer *buf = &r_xprt->rx_buf; 1386 struct rpcrdma_ep *ep = r_xprt->rx_ep; 1387 struct ib_recv_wr *wr, *bad_wr; 1388 struct rpcrdma_rep *rep; 1389 int needed, count, rc; 1390 1391 rc = 0; 1392 count = 0; 1393 1394 needed = buf->rb_credits + (buf->rb_bc_srv_max_requests << 1); 1395 if (likely(ep->re_receive_count > needed)) 1396 goto out; 1397 needed -= ep->re_receive_count; 1398 if (!temp) 1399 needed += RPCRDMA_MAX_RECV_BATCH; 1400 1401 /* fast path: all needed reps can be found on the free list */ 1402 wr = NULL; 1403 while (needed) { 1404 rep = rpcrdma_rep_get_locked(buf); 1405 if (rep && rep->rr_temp) { 1406 rpcrdma_rep_destroy(rep); 1407 continue; 1408 } 1409 if (!rep) 1410 rep = rpcrdma_rep_create(r_xprt, temp); 1411 if (!rep) 1412 break; 1413 1414 trace_xprtrdma_post_recv(rep); 1415 rep->rr_recv_wr.next = wr; 1416 wr = &rep->rr_recv_wr; 1417 --needed; 1418 ++count; 1419 } 1420 if (!wr) 1421 goto out; 1422 1423 rc = ib_post_recv(ep->re_id->qp, wr, 1424 (const struct ib_recv_wr **)&bad_wr); 1425 out: 1426 trace_xprtrdma_post_recvs(r_xprt, count, rc); 1427 if (rc) { 1428 for (wr = bad_wr; wr;) { 1429 struct rpcrdma_rep *rep; 1430 1431 rep = container_of(wr, struct rpcrdma_rep, rr_recv_wr); 1432 wr = wr->next; 1433 rpcrdma_recv_buffer_put(rep); 1434 --count; 1435 } 1436 } 1437 ep->re_receive_count += count; 1438 return; 1439 } 1440