2 * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
3 * Copyright (c) 2005-2007 Network Appliance, Inc. All rights reserved.
5 * This software is available to you under a choice of one of two
6 * licenses. You may choose to be licensed under the terms of the GNU
7 * General Public License (GPL) Version 2, available from the file
8 * COPYING in the main directory of this source tree, or the BSD-type
11 * Redistribution and use in source and binary forms, with or without
12 * modification, are permitted provided that the following conditions
15 * Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
18 * Redistributions in binary form must reproduce the above
19 * copyright notice, this list of conditions and the following
20 * disclaimer in the documentation and/or other materials provided
21 * with the distribution.
23 * Neither the name of the Network Appliance, Inc. nor the names of
24 * its contributors may be used to endorse or promote products
25 * derived from this software without specific prior written
28 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29 * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30 * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31 * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32 * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34 * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38 * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
40 * Author: Tom Tucker <tom@opengridcomputing.com>
43 #include <linux/sunrpc/svc_xprt.h>
44 #include <linux/sunrpc/debug.h>
45 #include <linux/sunrpc/rpc_rdma.h>
46 #include <linux/interrupt.h>
47 #include <linux/sched.h>
48 #include <linux/slab.h>
49 #include <linux/spinlock.h>
50 #include <linux/workqueue.h>
51 #include <rdma/ib_verbs.h>
52 #include <rdma/rdma_cm.h>
53 #include <linux/sunrpc/svc_rdma.h>
54 #include <linux/export.h>
55 #include "xprt_rdma.h"
57 #define RPCDBG_FACILITY RPCDBG_SVCXPRT
59 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *, int);
60 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
62 struct sockaddr *sa, int salen,
64 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
65 static void svc_rdma_release_rqst(struct svc_rqst *);
66 static void svc_rdma_detach(struct svc_xprt *xprt);
67 static void svc_rdma_free(struct svc_xprt *xprt);
68 static int svc_rdma_has_wspace(struct svc_xprt *xprt);
69 static int svc_rdma_secure_port(struct svc_rqst *);
71 static struct svc_xprt_ops svc_rdma_ops = {
72 .xpo_create = svc_rdma_create,
73 .xpo_recvfrom = svc_rdma_recvfrom,
74 .xpo_sendto = svc_rdma_sendto,
75 .xpo_release_rqst = svc_rdma_release_rqst,
76 .xpo_detach = svc_rdma_detach,
77 .xpo_free = svc_rdma_free,
78 .xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
79 .xpo_has_wspace = svc_rdma_has_wspace,
80 .xpo_accept = svc_rdma_accept,
81 .xpo_secure_port = svc_rdma_secure_port,
84 struct svc_xprt_class svc_rdma_class = {
86 .xcl_owner = THIS_MODULE,
87 .xcl_ops = &svc_rdma_ops,
88 .xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
89 .xcl_ident = XPRT_TRANSPORT_RDMA,
92 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
93 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *, struct net *,
94 struct sockaddr *, int, int);
95 static void svc_rdma_bc_detach(struct svc_xprt *);
96 static void svc_rdma_bc_free(struct svc_xprt *);
98 static struct svc_xprt_ops svc_rdma_bc_ops = {
99 .xpo_create = svc_rdma_bc_create,
100 .xpo_detach = svc_rdma_bc_detach,
101 .xpo_free = svc_rdma_bc_free,
102 .xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
103 .xpo_secure_port = svc_rdma_secure_port,
106 struct svc_xprt_class svc_rdma_bc_class = {
107 .xcl_name = "rdma-bc",
108 .xcl_owner = THIS_MODULE,
109 .xcl_ops = &svc_rdma_bc_ops,
110 .xcl_max_payload = (1024 - RPCRDMA_HDRLEN_MIN)
113 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *serv,
115 struct sockaddr *sa, int salen,
118 struct svcxprt_rdma *cma_xprt;
119 struct svc_xprt *xprt;
121 cma_xprt = rdma_create_xprt(serv, 0);
123 return ERR_PTR(-ENOMEM);
124 xprt = &cma_xprt->sc_xprt;
126 svc_xprt_init(net, &svc_rdma_bc_class, xprt, serv);
127 serv->sv_bc_xprt = xprt;
129 dprintk("svcrdma: %s(%p)\n", __func__, xprt);
133 static void svc_rdma_bc_detach(struct svc_xprt *xprt)
135 dprintk("svcrdma: %s(%p)\n", __func__, xprt);
138 static void svc_rdma_bc_free(struct svc_xprt *xprt)
140 struct svcxprt_rdma *rdma =
141 container_of(xprt, struct svcxprt_rdma, sc_xprt);
143 dprintk("svcrdma: %s(%p)\n", __func__, xprt);
147 #endif /* CONFIG_SUNRPC_BACKCHANNEL */
149 static struct svc_rdma_op_ctxt *alloc_ctxt(struct svcxprt_rdma *xprt,
152 struct svc_rdma_op_ctxt *ctxt;
154 ctxt = kmalloc(sizeof(*ctxt), flags);
157 INIT_LIST_HEAD(&ctxt->free);
158 INIT_LIST_HEAD(&ctxt->dto_q);
163 static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
167 /* Each RPC/RDMA credit can consume a number of send
168 * and receive WQEs. One ctxt is allocated for each.
170 i = xprt->sc_sq_depth + xprt->sc_rq_depth;
173 struct svc_rdma_op_ctxt *ctxt;
175 ctxt = alloc_ctxt(xprt, GFP_KERNEL);
177 dprintk("svcrdma: No memory for RDMA ctxt\n");
180 list_add(&ctxt->free, &xprt->sc_ctxts);
185 struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
187 struct svc_rdma_op_ctxt *ctxt = NULL;
189 spin_lock_bh(&xprt->sc_ctxt_lock);
190 xprt->sc_ctxt_used++;
191 if (list_empty(&xprt->sc_ctxts))
194 ctxt = list_first_entry(&xprt->sc_ctxts,
195 struct svc_rdma_op_ctxt, free);
196 list_del_init(&ctxt->free);
197 spin_unlock_bh(&xprt->sc_ctxt_lock);
201 ctxt->mapped_sges = 0;
206 /* Either pre-allocation missed the mark, or send
207 * queue accounting is broken.
209 spin_unlock_bh(&xprt->sc_ctxt_lock);
211 ctxt = alloc_ctxt(xprt, GFP_NOIO);
215 spin_lock_bh(&xprt->sc_ctxt_lock);
216 xprt->sc_ctxt_used--;
217 spin_unlock_bh(&xprt->sc_ctxt_lock);
218 WARN_ONCE(1, "svcrdma: empty RDMA ctxt list?\n");
222 void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
224 struct svcxprt_rdma *xprt = ctxt->xprt;
225 struct ib_device *device = xprt->sc_cm_id->device;
226 u32 lkey = xprt->sc_pd->local_dma_lkey;
227 unsigned int i, count;
229 for (count = 0, i = 0; i < ctxt->mapped_sges; i++) {
231 * Unmap the DMA addr in the SGE if the lkey matches
232 * the local_dma_lkey, otherwise, ignore it since it is
233 * an FRMR lkey and will be unmapped later when the
234 * last WR that uses it completes.
236 if (ctxt->sge[i].lkey == lkey) {
238 ib_dma_unmap_page(device,
244 ctxt->mapped_sges = 0;
245 atomic_sub(count, &xprt->sc_dma_used);
248 void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
250 struct svcxprt_rdma *xprt = ctxt->xprt;
254 for (i = 0; i < ctxt->count; i++)
255 put_page(ctxt->pages[i]);
257 spin_lock_bh(&xprt->sc_ctxt_lock);
258 xprt->sc_ctxt_used--;
259 list_add(&ctxt->free, &xprt->sc_ctxts);
260 spin_unlock_bh(&xprt->sc_ctxt_lock);
263 static void svc_rdma_destroy_ctxts(struct svcxprt_rdma *xprt)
265 while (!list_empty(&xprt->sc_ctxts)) {
266 struct svc_rdma_op_ctxt *ctxt;
268 ctxt = list_first_entry(&xprt->sc_ctxts,
269 struct svc_rdma_op_ctxt, free);
270 list_del(&ctxt->free);
275 static struct svc_rdma_req_map *alloc_req_map(gfp_t flags)
277 struct svc_rdma_req_map *map;
279 map = kmalloc(sizeof(*map), flags);
281 INIT_LIST_HEAD(&map->free);
285 static bool svc_rdma_prealloc_maps(struct svcxprt_rdma *xprt)
289 /* One for each receive buffer on this connection. */
290 i = xprt->sc_max_requests;
293 struct svc_rdma_req_map *map;
295 map = alloc_req_map(GFP_KERNEL);
297 dprintk("svcrdma: No memory for request map\n");
300 list_add(&map->free, &xprt->sc_maps);
305 struct svc_rdma_req_map *svc_rdma_get_req_map(struct svcxprt_rdma *xprt)
307 struct svc_rdma_req_map *map = NULL;
309 spin_lock(&xprt->sc_map_lock);
310 if (list_empty(&xprt->sc_maps))
313 map = list_first_entry(&xprt->sc_maps,
314 struct svc_rdma_req_map, free);
315 list_del_init(&map->free);
316 spin_unlock(&xprt->sc_map_lock);
323 spin_unlock(&xprt->sc_map_lock);
325 /* Pre-allocation amount was incorrect */
326 map = alloc_req_map(GFP_NOIO);
330 WARN_ONCE(1, "svcrdma: empty request map list?\n");
334 void svc_rdma_put_req_map(struct svcxprt_rdma *xprt,
335 struct svc_rdma_req_map *map)
337 spin_lock(&xprt->sc_map_lock);
338 list_add(&map->free, &xprt->sc_maps);
339 spin_unlock(&xprt->sc_map_lock);
342 static void svc_rdma_destroy_maps(struct svcxprt_rdma *xprt)
344 while (!list_empty(&xprt->sc_maps)) {
345 struct svc_rdma_req_map *map;
347 map = list_first_entry(&xprt->sc_maps,
348 struct svc_rdma_req_map, free);
349 list_del(&map->free);
354 /* QP event handler */
355 static void qp_event_handler(struct ib_event *event, void *context)
357 struct svc_xprt *xprt = context;
359 switch (event->event) {
360 /* These are considered benign events */
361 case IB_EVENT_PATH_MIG:
362 case IB_EVENT_COMM_EST:
363 case IB_EVENT_SQ_DRAINED:
364 case IB_EVENT_QP_LAST_WQE_REACHED:
365 dprintk("svcrdma: QP event %s (%d) received for QP=%p\n",
366 ib_event_msg(event->event), event->event,
369 /* These are considered fatal events */
370 case IB_EVENT_PATH_MIG_ERR:
371 case IB_EVENT_QP_FATAL:
372 case IB_EVENT_QP_REQ_ERR:
373 case IB_EVENT_QP_ACCESS_ERR:
374 case IB_EVENT_DEVICE_FATAL:
376 dprintk("svcrdma: QP ERROR event %s (%d) received for QP=%p, "
377 "closing transport\n",
378 ib_event_msg(event->event), event->event,
380 set_bit(XPT_CLOSE, &xprt->xpt_flags);
386 * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
387 * @cq: completion queue
391 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
393 struct svcxprt_rdma *xprt = cq->cq_context;
394 struct ib_cqe *cqe = wc->wr_cqe;
395 struct svc_rdma_op_ctxt *ctxt;
397 /* WARNING: Only wc->wr_cqe and wc->status are reliable */
398 ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
399 ctxt->wc_status = wc->status;
400 svc_rdma_unmap_dma(ctxt);
402 if (wc->status != IB_WC_SUCCESS)
405 /* All wc fields are now known to be valid */
406 ctxt->byte_len = wc->byte_len;
407 spin_lock(&xprt->sc_rq_dto_lock);
408 list_add_tail(&ctxt->dto_q, &xprt->sc_rq_dto_q);
409 spin_unlock(&xprt->sc_rq_dto_lock);
411 set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
412 if (test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
414 svc_xprt_enqueue(&xprt->sc_xprt);
418 if (wc->status != IB_WC_WR_FLUSH_ERR)
419 pr_warn("svcrdma: receive: %s (%u/0x%x)\n",
420 ib_wc_status_msg(wc->status),
421 wc->status, wc->vendor_err);
422 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
423 svc_rdma_put_context(ctxt, 1);
426 svc_xprt_put(&xprt->sc_xprt);
429 static void svc_rdma_send_wc_common(struct svcxprt_rdma *xprt,
433 if (wc->status != IB_WC_SUCCESS)
437 atomic_dec(&xprt->sc_sq_count);
438 wake_up(&xprt->sc_send_wait);
442 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
443 if (wc->status != IB_WC_WR_FLUSH_ERR)
444 pr_err("svcrdma: %s: %s (%u/0x%x)\n",
445 opname, ib_wc_status_msg(wc->status),
446 wc->status, wc->vendor_err);
450 static void svc_rdma_send_wc_common_put(struct ib_cq *cq, struct ib_wc *wc,
453 struct svcxprt_rdma *xprt = cq->cq_context;
455 svc_rdma_send_wc_common(xprt, wc, opname);
456 svc_xprt_put(&xprt->sc_xprt);
460 * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC
461 * @cq: completion queue
465 void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
467 struct ib_cqe *cqe = wc->wr_cqe;
468 struct svc_rdma_op_ctxt *ctxt;
470 svc_rdma_send_wc_common_put(cq, wc, "send");
472 ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
473 svc_rdma_unmap_dma(ctxt);
474 svc_rdma_put_context(ctxt, 1);
478 * svc_rdma_wc_write - Invoked by RDMA provider for each polled Write WC
479 * @cq: completion queue
483 void svc_rdma_wc_write(struct ib_cq *cq, struct ib_wc *wc)
485 struct ib_cqe *cqe = wc->wr_cqe;
486 struct svc_rdma_op_ctxt *ctxt;
488 svc_rdma_send_wc_common_put(cq, wc, "write");
490 ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
491 svc_rdma_unmap_dma(ctxt);
492 svc_rdma_put_context(ctxt, 0);
496 * svc_rdma_wc_reg - Invoked by RDMA provider for each polled FASTREG WC
497 * @cq: completion queue
501 void svc_rdma_wc_reg(struct ib_cq *cq, struct ib_wc *wc)
503 svc_rdma_send_wc_common_put(cq, wc, "fastreg");
507 * svc_rdma_wc_read - Invoked by RDMA provider for each polled Read WC
508 * @cq: completion queue
512 void svc_rdma_wc_read(struct ib_cq *cq, struct ib_wc *wc)
514 struct svcxprt_rdma *xprt = cq->cq_context;
515 struct ib_cqe *cqe = wc->wr_cqe;
516 struct svc_rdma_op_ctxt *ctxt;
518 svc_rdma_send_wc_common(xprt, wc, "read");
520 ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
521 svc_rdma_unmap_dma(ctxt);
522 svc_rdma_put_frmr(xprt, ctxt->frmr);
524 if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
525 struct svc_rdma_op_ctxt *read_hdr;
527 read_hdr = ctxt->read_hdr;
528 spin_lock(&xprt->sc_rq_dto_lock);
529 list_add_tail(&read_hdr->dto_q,
530 &xprt->sc_read_complete_q);
531 spin_unlock(&xprt->sc_rq_dto_lock);
533 set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
534 svc_xprt_enqueue(&xprt->sc_xprt);
537 svc_rdma_put_context(ctxt, 0);
538 svc_xprt_put(&xprt->sc_xprt);
542 * svc_rdma_wc_inv - Invoked by RDMA provider for each polled LOCAL_INV WC
543 * @cq: completion queue
547 void svc_rdma_wc_inv(struct ib_cq *cq, struct ib_wc *wc)
549 svc_rdma_send_wc_common_put(cq, wc, "localInv");
552 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
555 struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
559 svc_xprt_init(&init_net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
560 INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
561 INIT_LIST_HEAD(&cma_xprt->sc_dto_q);
562 INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
563 INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
564 INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
565 INIT_LIST_HEAD(&cma_xprt->sc_ctxts);
566 INIT_LIST_HEAD(&cma_xprt->sc_maps);
567 init_waitqueue_head(&cma_xprt->sc_send_wait);
569 spin_lock_init(&cma_xprt->sc_lock);
570 spin_lock_init(&cma_xprt->sc_rq_dto_lock);
571 spin_lock_init(&cma_xprt->sc_frmr_q_lock);
572 spin_lock_init(&cma_xprt->sc_ctxt_lock);
573 spin_lock_init(&cma_xprt->sc_map_lock);
576 set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
581 int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags)
583 struct ib_recv_wr recv_wr, *bad_recv_wr;
584 struct svc_rdma_op_ctxt *ctxt;
591 ctxt = svc_rdma_get_context(xprt);
593 ctxt->direction = DMA_FROM_DEVICE;
594 ctxt->cqe.done = svc_rdma_wc_receive;
595 for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) {
596 if (sge_no >= xprt->sc_max_sge) {
597 pr_err("svcrdma: Too many sges (%d)\n", sge_no);
600 page = alloc_page(flags);
603 ctxt->pages[sge_no] = page;
604 pa = ib_dma_map_page(xprt->sc_cm_id->device,
607 if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
609 svc_rdma_count_mappings(xprt, ctxt);
610 ctxt->sge[sge_no].addr = pa;
611 ctxt->sge[sge_no].length = PAGE_SIZE;
612 ctxt->sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
613 ctxt->count = sge_no + 1;
617 recv_wr.sg_list = &ctxt->sge[0];
618 recv_wr.num_sge = ctxt->count;
619 recv_wr.wr_cqe = &ctxt->cqe;
621 svc_xprt_get(&xprt->sc_xprt);
622 ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
624 svc_rdma_unmap_dma(ctxt);
625 svc_rdma_put_context(ctxt, 1);
626 svc_xprt_put(&xprt->sc_xprt);
631 svc_rdma_unmap_dma(ctxt);
632 svc_rdma_put_context(ctxt, 1);
636 int svc_rdma_repost_recv(struct svcxprt_rdma *xprt, gfp_t flags)
640 ret = svc_rdma_post_recv(xprt, flags);
642 pr_err("svcrdma: could not post a receive buffer, err=%d.\n",
644 pr_err("svcrdma: closing transport %p.\n", xprt);
645 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
652 svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
653 struct rdma_conn_param *param)
655 const struct rpcrdma_connect_private *pmsg = param->private_data;
658 pmsg->cp_magic == rpcrdma_cmp_magic &&
659 pmsg->cp_version == RPCRDMA_CMP_VERSION) {
660 newxprt->sc_snd_w_inv = pmsg->cp_flags &
661 RPCRDMA_CMP_F_SND_W_INV_OK;
663 dprintk("svcrdma: client send_size %u, recv_size %u "
664 "remote inv %ssupported\n",
665 rpcrdma_decode_buffer_size(pmsg->cp_send_size),
666 rpcrdma_decode_buffer_size(pmsg->cp_recv_size),
667 newxprt->sc_snd_w_inv ? "" : "un");
672 * This function handles the CONNECT_REQUEST event on a listening
673 * endpoint. It is passed the cma_id for the _new_ connection. The context in
674 * this cma_id is inherited from the listening cma_id and is the svc_xprt
675 * structure for the listening endpoint.
677 * This function creates a new xprt for the new connection and enqueues it on
678 * the accept queue for the listent xprt. When the listen thread is kicked, it
679 * will call the recvfrom method on the listen xprt which will accept the new
682 static void handle_connect_req(struct rdma_cm_id *new_cma_id,
683 struct rdma_conn_param *param)
685 struct svcxprt_rdma *listen_xprt = new_cma_id->context;
686 struct svcxprt_rdma *newxprt;
689 /* Create a new transport */
690 newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
692 dprintk("svcrdma: failed to create new transport\n");
695 newxprt->sc_cm_id = new_cma_id;
696 new_cma_id->context = newxprt;
697 dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
698 newxprt, newxprt->sc_cm_id, listen_xprt);
699 svc_rdma_parse_connect_private(newxprt, param);
701 /* Save client advertised inbound read limit for use later in accept. */
702 newxprt->sc_ord = param->initiator_depth;
704 /* Set the local and remote addresses in the transport */
705 sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
706 svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
707 sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
708 svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
711 * Enqueue the new transport on the accept queue of the listening
714 spin_lock_bh(&listen_xprt->sc_lock);
715 list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
716 spin_unlock_bh(&listen_xprt->sc_lock);
718 set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
719 svc_xprt_enqueue(&listen_xprt->sc_xprt);
723 * Handles events generated on the listening endpoint. These events will be
724 * either be incoming connect requests or adapter removal events.
726 static int rdma_listen_handler(struct rdma_cm_id *cma_id,
727 struct rdma_cm_event *event)
729 struct svcxprt_rdma *xprt = cma_id->context;
732 switch (event->event) {
733 case RDMA_CM_EVENT_CONNECT_REQUEST:
734 dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
735 "event = %s (%d)\n", cma_id, cma_id->context,
736 rdma_event_msg(event->event), event->event);
737 handle_connect_req(cma_id, &event->param.conn);
740 case RDMA_CM_EVENT_ESTABLISHED:
741 /* Accept complete */
742 dprintk("svcrdma: Connection completed on LISTEN xprt=%p, "
743 "cm_id=%p\n", xprt, cma_id);
746 case RDMA_CM_EVENT_DEVICE_REMOVAL:
747 dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
750 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
754 dprintk("svcrdma: Unexpected event on listening endpoint %p, "
755 "event = %s (%d)\n", cma_id,
756 rdma_event_msg(event->event), event->event);
763 static int rdma_cma_handler(struct rdma_cm_id *cma_id,
764 struct rdma_cm_event *event)
766 struct svc_xprt *xprt = cma_id->context;
767 struct svcxprt_rdma *rdma =
768 container_of(xprt, struct svcxprt_rdma, sc_xprt);
769 switch (event->event) {
770 case RDMA_CM_EVENT_ESTABLISHED:
771 /* Accept complete */
773 dprintk("svcrdma: Connection completed on DTO xprt=%p, "
774 "cm_id=%p\n", xprt, cma_id);
775 clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
776 svc_xprt_enqueue(xprt);
778 case RDMA_CM_EVENT_DISCONNECTED:
779 dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
782 set_bit(XPT_CLOSE, &xprt->xpt_flags);
783 svc_xprt_enqueue(xprt);
787 case RDMA_CM_EVENT_DEVICE_REMOVAL:
788 dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
789 "event = %s (%d)\n", cma_id, xprt,
790 rdma_event_msg(event->event), event->event);
792 set_bit(XPT_CLOSE, &xprt->xpt_flags);
793 svc_xprt_enqueue(xprt);
798 dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
799 "event = %s (%d)\n", cma_id,
800 rdma_event_msg(event->event), event->event);
807 * Create a listening RDMA service endpoint.
809 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
811 struct sockaddr *sa, int salen,
814 struct rdma_cm_id *listen_id;
815 struct svcxprt_rdma *cma_xprt;
818 dprintk("svcrdma: Creating RDMA socket\n");
819 if ((sa->sa_family != AF_INET) && (sa->sa_family != AF_INET6)) {
820 dprintk("svcrdma: Address family %d is not supported.\n", sa->sa_family);
821 return ERR_PTR(-EAFNOSUPPORT);
823 cma_xprt = rdma_create_xprt(serv, 1);
825 return ERR_PTR(-ENOMEM);
827 listen_id = rdma_create_id(&init_net, rdma_listen_handler, cma_xprt,
828 RDMA_PS_TCP, IB_QPT_RC);
829 if (IS_ERR(listen_id)) {
830 ret = PTR_ERR(listen_id);
831 dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
835 /* Allow both IPv4 and IPv6 sockets to bind a single port
838 #if IS_ENABLED(CONFIG_IPV6)
839 ret = rdma_set_afonly(listen_id, 1);
841 dprintk("svcrdma: rdma_set_afonly failed = %d\n", ret);
845 ret = rdma_bind_addr(listen_id, sa);
847 dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
850 cma_xprt->sc_cm_id = listen_id;
852 ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
854 dprintk("svcrdma: rdma_listen failed = %d\n", ret);
859 * We need to use the address from the cm_id in case the
860 * caller specified 0 for the port number.
862 sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
863 svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
865 return &cma_xprt->sc_xprt;
868 rdma_destroy_id(listen_id);
874 static struct svc_rdma_fastreg_mr *rdma_alloc_frmr(struct svcxprt_rdma *xprt)
877 struct scatterlist *sg;
878 struct svc_rdma_fastreg_mr *frmr;
881 frmr = kmalloc(sizeof(*frmr), GFP_KERNEL);
885 num_sg = min_t(u32, RPCSVC_MAXPAGES, xprt->sc_frmr_pg_list_len);
886 mr = ib_alloc_mr(xprt->sc_pd, IB_MR_TYPE_MEM_REG, num_sg);
890 sg = kcalloc(RPCSVC_MAXPAGES, sizeof(*sg), GFP_KERNEL);
894 sg_init_table(sg, RPCSVC_MAXPAGES);
898 INIT_LIST_HEAD(&frmr->frmr_list);
906 return ERR_PTR(-ENOMEM);
909 static void rdma_dealloc_frmr_q(struct svcxprt_rdma *xprt)
911 struct svc_rdma_fastreg_mr *frmr;
913 while (!list_empty(&xprt->sc_frmr_q)) {
914 frmr = list_entry(xprt->sc_frmr_q.next,
915 struct svc_rdma_fastreg_mr, frmr_list);
916 list_del_init(&frmr->frmr_list);
918 ib_dereg_mr(frmr->mr);
923 struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *rdma)
925 struct svc_rdma_fastreg_mr *frmr = NULL;
927 spin_lock_bh(&rdma->sc_frmr_q_lock);
928 if (!list_empty(&rdma->sc_frmr_q)) {
929 frmr = list_entry(rdma->sc_frmr_q.next,
930 struct svc_rdma_fastreg_mr, frmr_list);
931 list_del_init(&frmr->frmr_list);
934 spin_unlock_bh(&rdma->sc_frmr_q_lock);
938 return rdma_alloc_frmr(rdma);
941 void svc_rdma_put_frmr(struct svcxprt_rdma *rdma,
942 struct svc_rdma_fastreg_mr *frmr)
945 ib_dma_unmap_sg(rdma->sc_cm_id->device,
946 frmr->sg, frmr->sg_nents, frmr->direction);
947 atomic_dec(&rdma->sc_dma_used);
948 spin_lock_bh(&rdma->sc_frmr_q_lock);
949 WARN_ON_ONCE(!list_empty(&frmr->frmr_list));
950 list_add(&frmr->frmr_list, &rdma->sc_frmr_q);
951 spin_unlock_bh(&rdma->sc_frmr_q_lock);
956 * This is the xpo_recvfrom function for listening endpoints. Its
957 * purpose is to accept incoming connections. The CMA callback handler
958 * has already created a new transport and attached it to the new CMA
961 * There is a queue of pending connections hung on the listening
962 * transport. This queue contains the new svc_xprt structure. This
963 * function takes svc_xprt structures off the accept_q and completes
966 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
968 struct svcxprt_rdma *listen_rdma;
969 struct svcxprt_rdma *newxprt = NULL;
970 struct rdma_conn_param conn_param;
971 struct rpcrdma_connect_private pmsg;
972 struct ib_qp_init_attr qp_attr;
973 struct ib_device *dev;
977 listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
978 clear_bit(XPT_CONN, &xprt->xpt_flags);
979 /* Get the next entry off the accept list */
980 spin_lock_bh(&listen_rdma->sc_lock);
981 if (!list_empty(&listen_rdma->sc_accept_q)) {
982 newxprt = list_entry(listen_rdma->sc_accept_q.next,
983 struct svcxprt_rdma, sc_accept_q);
984 list_del_init(&newxprt->sc_accept_q);
986 if (!list_empty(&listen_rdma->sc_accept_q))
987 set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
988 spin_unlock_bh(&listen_rdma->sc_lock);
992 dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
993 newxprt, newxprt->sc_cm_id);
995 dev = newxprt->sc_cm_id->device;
997 /* Qualify the transport resource defaults with the
998 * capabilities of this particular device */
999 newxprt->sc_max_sge = min((size_t)dev->attrs.max_sge,
1000 (size_t)RPCSVC_MAXPAGES);
1001 newxprt->sc_max_sge_rd = min_t(size_t, dev->attrs.max_sge_rd,
1003 newxprt->sc_max_req_size = svcrdma_max_req_size;
1004 newxprt->sc_max_requests = min_t(u32, dev->attrs.max_qp_wr,
1005 svcrdma_max_requests);
1006 newxprt->sc_max_bc_requests = min_t(u32, dev->attrs.max_qp_wr,
1007 svcrdma_max_bc_requests);
1008 newxprt->sc_rq_depth = newxprt->sc_max_requests +
1009 newxprt->sc_max_bc_requests;
1010 newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_rq_depth;
1012 if (!svc_rdma_prealloc_ctxts(newxprt))
1014 if (!svc_rdma_prealloc_maps(newxprt))
1018 * Limit ORD based on client limit, local device limit, and
1019 * configured svcrdma limit.
1021 newxprt->sc_ord = min_t(size_t, dev->attrs.max_qp_rd_atom, newxprt->sc_ord);
1022 newxprt->sc_ord = min_t(size_t, svcrdma_ord, newxprt->sc_ord);
1024 newxprt->sc_pd = ib_alloc_pd(dev, 0);
1025 if (IS_ERR(newxprt->sc_pd)) {
1026 dprintk("svcrdma: error creating PD for connect request\n");
1029 newxprt->sc_sq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_sq_depth,
1030 0, IB_POLL_SOFTIRQ);
1031 if (IS_ERR(newxprt->sc_sq_cq)) {
1032 dprintk("svcrdma: error creating SQ CQ for connect request\n");
1035 newxprt->sc_rq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_rq_depth,
1036 0, IB_POLL_SOFTIRQ);
1037 if (IS_ERR(newxprt->sc_rq_cq)) {
1038 dprintk("svcrdma: error creating RQ CQ for connect request\n");
1042 memset(&qp_attr, 0, sizeof qp_attr);
1043 qp_attr.event_handler = qp_event_handler;
1044 qp_attr.qp_context = &newxprt->sc_xprt;
1045 qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
1046 qp_attr.cap.max_recv_wr = newxprt->sc_rq_depth;
1047 qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
1048 qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
1049 qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
1050 qp_attr.qp_type = IB_QPT_RC;
1051 qp_attr.send_cq = newxprt->sc_sq_cq;
1052 qp_attr.recv_cq = newxprt->sc_rq_cq;
1053 dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n"
1054 " cm_id->device=%p, sc_pd->device=%p\n"
1055 " cap.max_send_wr = %d\n"
1056 " cap.max_recv_wr = %d\n"
1057 " cap.max_send_sge = %d\n"
1058 " cap.max_recv_sge = %d\n",
1059 newxprt->sc_cm_id, newxprt->sc_pd,
1060 dev, newxprt->sc_pd->device,
1061 qp_attr.cap.max_send_wr,
1062 qp_attr.cap.max_recv_wr,
1063 qp_attr.cap.max_send_sge,
1064 qp_attr.cap.max_recv_sge);
1066 ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
1068 dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
1071 newxprt->sc_qp = newxprt->sc_cm_id->qp;
1074 * Use the most secure set of MR resources based on the
1075 * transport type and available memory management features in
1076 * the device. Here's the table implemented below:
1078 * Fast Global DMA Remote WR
1079 * Reg LKEY MR Access
1080 * Sup'd Sup'd Needed Needed
1092 * NB: iWARP requires remote write access for the data sink
1093 * of an RDMA_READ. IB does not.
1095 newxprt->sc_reader = rdma_read_chunk_lcl;
1096 if (dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
1097 newxprt->sc_frmr_pg_list_len =
1098 dev->attrs.max_fast_reg_page_list_len;
1099 newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
1100 newxprt->sc_reader = rdma_read_chunk_frmr;
1102 newxprt->sc_snd_w_inv = false;
1105 * Determine if a DMA MR is required and if so, what privs are required
1107 if (!rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num) &&
1108 !rdma_ib_or_roce(dev, newxprt->sc_cm_id->port_num))
1111 if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num))
1112 newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
1114 /* Post receive buffers */
1115 for (i = 0; i < newxprt->sc_max_requests; i++) {
1116 ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
1118 dprintk("svcrdma: failure posting receive buffers\n");
1123 /* Swap out the handler */
1124 newxprt->sc_cm_id->event_handler = rdma_cma_handler;
1126 /* Construct RDMA-CM private message */
1127 pmsg.cp_magic = rpcrdma_cmp_magic;
1128 pmsg.cp_version = RPCRDMA_CMP_VERSION;
1130 pmsg.cp_send_size = pmsg.cp_recv_size =
1131 rpcrdma_encode_buffer_size(newxprt->sc_max_req_size);
1133 /* Accept Connection */
1134 set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
1135 memset(&conn_param, 0, sizeof conn_param);
1136 conn_param.responder_resources = 0;
1137 conn_param.initiator_depth = newxprt->sc_ord;
1138 conn_param.private_data = &pmsg;
1139 conn_param.private_data_len = sizeof(pmsg);
1140 ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
1142 dprintk("svcrdma: failed to accept new connection, ret=%d\n",
1147 dprintk("svcrdma: new connection %p accepted with the following "
1149 " local_ip : %pI4\n"
1150 " local_port : %d\n"
1151 " remote_ip : %pI4\n"
1152 " remote_port : %d\n"
1154 " max_sge_rd : %d\n"
1156 " max_requests : %d\n"
1159 &((struct sockaddr_in *)&newxprt->sc_cm_id->
1160 route.addr.src_addr)->sin_addr.s_addr,
1161 ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
1162 route.addr.src_addr)->sin_port),
1163 &((struct sockaddr_in *)&newxprt->sc_cm_id->
1164 route.addr.dst_addr)->sin_addr.s_addr,
1165 ntohs(((struct sockaddr_in *)&newxprt->sc_cm_id->
1166 route.addr.dst_addr)->sin_port),
1167 newxprt->sc_max_sge,
1168 newxprt->sc_max_sge_rd,
1169 newxprt->sc_sq_depth,
1170 newxprt->sc_max_requests,
1173 return &newxprt->sc_xprt;
1176 dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
1177 /* Take a reference in case the DTO handler runs */
1178 svc_xprt_get(&newxprt->sc_xprt);
1179 if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
1180 ib_destroy_qp(newxprt->sc_qp);
1181 rdma_destroy_id(newxprt->sc_cm_id);
1182 /* This call to put will destroy the transport */
1183 svc_xprt_put(&newxprt->sc_xprt);
1187 static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
1192 * When connected, an svc_xprt has at least two references:
1194 * - A reference held by the cm_id between the ESTABLISHED and
1195 * DISCONNECTED events. If the remote peer disconnected first, this
1196 * reference could be gone.
1198 * - A reference held by the svc_recv code that called this function
1199 * as part of close processing.
1201 * At a minimum one references should still be held.
1203 static void svc_rdma_detach(struct svc_xprt *xprt)
1205 struct svcxprt_rdma *rdma =
1206 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1207 dprintk("svc: svc_rdma_detach(%p)\n", xprt);
1209 /* Disconnect and flush posted WQE */
1210 rdma_disconnect(rdma->sc_cm_id);
1213 static void __svc_rdma_free(struct work_struct *work)
1215 struct svcxprt_rdma *rdma =
1216 container_of(work, struct svcxprt_rdma, sc_work);
1217 struct svc_xprt *xprt = &rdma->sc_xprt;
1219 dprintk("svcrdma: %s(%p)\n", __func__, rdma);
1221 if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
1222 ib_drain_qp(rdma->sc_qp);
1224 /* We should only be called from kref_put */
1225 if (atomic_read(&xprt->xpt_ref.refcount) != 0)
1226 pr_err("svcrdma: sc_xprt still in use? (%d)\n",
1227 atomic_read(&xprt->xpt_ref.refcount));
1230 * Destroy queued, but not processed read completions. Note
1231 * that this cleanup has to be done before destroying the
1232 * cm_id because the device ptr is needed to unmap the dma in
1233 * svc_rdma_put_context.
1235 while (!list_empty(&rdma->sc_read_complete_q)) {
1236 struct svc_rdma_op_ctxt *ctxt;
1237 ctxt = list_entry(rdma->sc_read_complete_q.next,
1238 struct svc_rdma_op_ctxt,
1240 list_del_init(&ctxt->dto_q);
1241 svc_rdma_put_context(ctxt, 1);
1244 /* Destroy queued, but not processed recv completions */
1245 while (!list_empty(&rdma->sc_rq_dto_q)) {
1246 struct svc_rdma_op_ctxt *ctxt;
1247 ctxt = list_entry(rdma->sc_rq_dto_q.next,
1248 struct svc_rdma_op_ctxt,
1250 list_del_init(&ctxt->dto_q);
1251 svc_rdma_put_context(ctxt, 1);
1254 /* Warn if we leaked a resource or under-referenced */
1255 if (rdma->sc_ctxt_used != 0)
1256 pr_err("svcrdma: ctxt still in use? (%d)\n",
1257 rdma->sc_ctxt_used);
1258 if (atomic_read(&rdma->sc_dma_used) != 0)
1259 pr_err("svcrdma: dma still in use? (%d)\n",
1260 atomic_read(&rdma->sc_dma_used));
1262 /* Final put of backchannel client transport */
1263 if (xprt->xpt_bc_xprt) {
1264 xprt_put(xprt->xpt_bc_xprt);
1265 xprt->xpt_bc_xprt = NULL;
1268 rdma_dealloc_frmr_q(rdma);
1269 svc_rdma_destroy_ctxts(rdma);
1270 svc_rdma_destroy_maps(rdma);
1272 /* Destroy the QP if present (not a listener) */
1273 if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
1274 ib_destroy_qp(rdma->sc_qp);
1276 if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
1277 ib_free_cq(rdma->sc_sq_cq);
1279 if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
1280 ib_free_cq(rdma->sc_rq_cq);
1282 if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
1283 ib_dealloc_pd(rdma->sc_pd);
1285 /* Destroy the CM ID */
1286 rdma_destroy_id(rdma->sc_cm_id);
1291 static void svc_rdma_free(struct svc_xprt *xprt)
1293 struct svcxprt_rdma *rdma =
1294 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1295 INIT_WORK(&rdma->sc_work, __svc_rdma_free);
1296 queue_work(svc_rdma_wq, &rdma->sc_work);
1299 static int svc_rdma_has_wspace(struct svc_xprt *xprt)
1301 struct svcxprt_rdma *rdma =
1302 container_of(xprt, struct svcxprt_rdma, sc_xprt);
1305 * If there are already waiters on the SQ,
1308 if (waitqueue_active(&rdma->sc_send_wait))
1311 /* Otherwise return true. */
1315 static int svc_rdma_secure_port(struct svc_rqst *rqstp)
1320 int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
1322 struct ib_send_wr *bad_wr, *n_wr;
1327 if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1331 for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
1334 /* If the SQ is full, wait until an SQ entry is available */
1336 spin_lock_bh(&xprt->sc_lock);
1337 if (xprt->sc_sq_depth < atomic_read(&xprt->sc_sq_count) + wr_count) {
1338 spin_unlock_bh(&xprt->sc_lock);
1339 atomic_inc(&rdma_stat_sq_starve);
1341 /* Wait until SQ WR available if SQ still full */
1342 wait_event(xprt->sc_send_wait,
1343 atomic_read(&xprt->sc_sq_count) <
1345 if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1349 /* Take a transport ref for each WR posted */
1350 for (i = 0; i < wr_count; i++)
1351 svc_xprt_get(&xprt->sc_xprt);
1353 /* Bump used SQ WR count and post */
1354 atomic_add(wr_count, &xprt->sc_sq_count);
1355 ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
1357 set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
1358 atomic_sub(wr_count, &xprt->sc_sq_count);
1359 for (i = 0; i < wr_count; i ++)
1360 svc_xprt_put(&xprt->sc_xprt);
1361 dprintk("svcrdma: failed to post SQ WR rc=%d, "
1362 "sc_sq_count=%d, sc_sq_depth=%d\n",
1363 ret, atomic_read(&xprt->sc_sq_count),
1366 spin_unlock_bh(&xprt->sc_lock);
1368 wake_up(&xprt->sc_send_wait);