1 /*
2  * Copyright (c) 2014 Open Grid Computing, Inc. All rights reserved.
3  * Copyright (c) 2005-2007 Network Appliance, Inc. All rights reserved.
4  *
5  * This software is available to you under a choice of one of two
6  * licenses.  You may choose to be licensed under the terms of the GNU
7  * General Public License (GPL) Version 2, available from the file
8  * COPYING in the main directory of this source tree, or the BSD-type
9  * license below:
10  *
11  * Redistribution and use in source and binary forms, with or without
12  * modification, are permitted provided that the following conditions
13  * are met:
14  *
15  *      Redistributions of source code must retain the above copyright
16  *      notice, this list of conditions and the following disclaimer.
17  *
18  *      Redistributions in binary form must reproduce the above
19  *      copyright notice, this list of conditions and the following
20  *      disclaimer in the documentation and/or other materials provided
21  *      with the distribution.
22  *
23  *      Neither the name of the Network Appliance, Inc. nor the names of
24  *      its contributors may be used to endorse or promote products
25  *      derived from this software without specific prior written
26  *      permission.
27  *
28  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
29  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
30  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
31  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
32  * OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
33  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
34  * LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
35  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
36  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
37  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
38  * OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
39  *
40  * Author: Tom Tucker <tom@opengridcomputing.com>
41  */
42 
43 #include <linux/sunrpc/svc_xprt.h>
44 #include <linux/sunrpc/addr.h>
45 #include <linux/sunrpc/debug.h>
46 #include <linux/sunrpc/rpc_rdma.h>
47 #include <linux/interrupt.h>
48 #include <linux/sched.h>
49 #include <linux/slab.h>
50 #include <linux/spinlock.h>
51 #include <linux/workqueue.h>
52 #include <rdma/ib_verbs.h>
53 #include <rdma/rdma_cm.h>
54 #include <rdma/rw.h>
55 #include <linux/sunrpc/svc_rdma.h>
56 #include <linux/export.h>
57 #include "xprt_rdma.h"
58 
59 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
60 
61 static int svc_rdma_post_recv(struct svcxprt_rdma *xprt);
62 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *, int);
63 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
64 					struct net *net,
65 					struct sockaddr *sa, int salen,
66 					int flags);
67 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
68 static void svc_rdma_release_rqst(struct svc_rqst *);
69 static void svc_rdma_detach(struct svc_xprt *xprt);
70 static void svc_rdma_free(struct svc_xprt *xprt);
71 static int svc_rdma_has_wspace(struct svc_xprt *xprt);
72 static int svc_rdma_secure_port(struct svc_rqst *);
73 static void svc_rdma_kill_temp_xprt(struct svc_xprt *);
74 
75 static const struct svc_xprt_ops svc_rdma_ops = {
76 	.xpo_create = svc_rdma_create,
77 	.xpo_recvfrom = svc_rdma_recvfrom,
78 	.xpo_sendto = svc_rdma_sendto,
79 	.xpo_release_rqst = svc_rdma_release_rqst,
80 	.xpo_detach = svc_rdma_detach,
81 	.xpo_free = svc_rdma_free,
82 	.xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
83 	.xpo_has_wspace = svc_rdma_has_wspace,
84 	.xpo_accept = svc_rdma_accept,
85 	.xpo_secure_port = svc_rdma_secure_port,
86 	.xpo_kill_temp_xprt = svc_rdma_kill_temp_xprt,
87 };
88 
89 struct svc_xprt_class svc_rdma_class = {
90 	.xcl_name = "rdma",
91 	.xcl_owner = THIS_MODULE,
92 	.xcl_ops = &svc_rdma_ops,
93 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
94 	.xcl_ident = XPRT_TRANSPORT_RDMA,
95 };
96 
97 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
98 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *, struct net *,
99 					   struct sockaddr *, int, int);
100 static void svc_rdma_bc_detach(struct svc_xprt *);
101 static void svc_rdma_bc_free(struct svc_xprt *);
102 
103 static const struct svc_xprt_ops svc_rdma_bc_ops = {
104 	.xpo_create = svc_rdma_bc_create,
105 	.xpo_detach = svc_rdma_bc_detach,
106 	.xpo_free = svc_rdma_bc_free,
107 	.xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
108 	.xpo_secure_port = svc_rdma_secure_port,
109 };
110 
111 struct svc_xprt_class svc_rdma_bc_class = {
112 	.xcl_name = "rdma-bc",
113 	.xcl_owner = THIS_MODULE,
114 	.xcl_ops = &svc_rdma_bc_ops,
115 	.xcl_max_payload = (1024 - RPCRDMA_HDRLEN_MIN)
116 };
117 
118 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *serv,
119 					   struct net *net,
120 					   struct sockaddr *sa, int salen,
121 					   int flags)
122 {
123 	struct svcxprt_rdma *cma_xprt;
124 	struct svc_xprt *xprt;
125 
126 	cma_xprt = rdma_create_xprt(serv, 0);
127 	if (!cma_xprt)
128 		return ERR_PTR(-ENOMEM);
129 	xprt = &cma_xprt->sc_xprt;
130 
131 	svc_xprt_init(net, &svc_rdma_bc_class, xprt, serv);
132 	set_bit(XPT_CONG_CTRL, &xprt->xpt_flags);
133 	serv->sv_bc_xprt = xprt;
134 
135 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
136 	return xprt;
137 }
138 
139 static void svc_rdma_bc_detach(struct svc_xprt *xprt)
140 {
141 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
142 }
143 
144 static void svc_rdma_bc_free(struct svc_xprt *xprt)
145 {
146 	struct svcxprt_rdma *rdma =
147 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
148 
149 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
150 	if (xprt)
151 		kfree(rdma);
152 }
153 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
154 
155 static struct svc_rdma_op_ctxt *alloc_ctxt(struct svcxprt_rdma *xprt,
156 					   gfp_t flags)
157 {
158 	struct svc_rdma_op_ctxt *ctxt;
159 
160 	ctxt = kmalloc(sizeof(*ctxt), flags);
161 	if (ctxt) {
162 		ctxt->xprt = xprt;
163 		INIT_LIST_HEAD(&ctxt->list);
164 	}
165 	return ctxt;
166 }
167 
168 static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
169 {
170 	unsigned int i;
171 
172 	/* Each RPC/RDMA credit can consume one Receive and
173 	 * one Send WQE at the same time.
174 	 */
175 	i = xprt->sc_sq_depth + xprt->sc_rq_depth;
176 
177 	while (i--) {
178 		struct svc_rdma_op_ctxt *ctxt;
179 
180 		ctxt = alloc_ctxt(xprt, GFP_KERNEL);
181 		if (!ctxt) {
182 			dprintk("svcrdma: No memory for RDMA ctxt\n");
183 			return false;
184 		}
185 		list_add(&ctxt->list, &xprt->sc_ctxts);
186 	}
187 	return true;
188 }
189 
190 struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
191 {
192 	struct svc_rdma_op_ctxt *ctxt = NULL;
193 
194 	spin_lock(&xprt->sc_ctxt_lock);
195 	xprt->sc_ctxt_used++;
196 	if (list_empty(&xprt->sc_ctxts))
197 		goto out_empty;
198 
199 	ctxt = list_first_entry(&xprt->sc_ctxts,
200 				struct svc_rdma_op_ctxt, list);
201 	list_del(&ctxt->list);
202 	spin_unlock(&xprt->sc_ctxt_lock);
203 
204 out:
205 	ctxt->count = 0;
206 	ctxt->mapped_sges = 0;
207 	return ctxt;
208 
209 out_empty:
210 	/* Either pre-allocation missed the mark, or send
211 	 * queue accounting is broken.
212 	 */
213 	spin_unlock(&xprt->sc_ctxt_lock);
214 
215 	ctxt = alloc_ctxt(xprt, GFP_NOIO);
216 	if (ctxt)
217 		goto out;
218 
219 	spin_lock(&xprt->sc_ctxt_lock);
220 	xprt->sc_ctxt_used--;
221 	spin_unlock(&xprt->sc_ctxt_lock);
222 	WARN_ONCE(1, "svcrdma: empty RDMA ctxt list?\n");
223 	return NULL;
224 }
225 
226 void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
227 {
228 	struct svcxprt_rdma *xprt = ctxt->xprt;
229 	struct ib_device *device = xprt->sc_cm_id->device;
230 	unsigned int i;
231 
232 	for (i = 0; i < ctxt->mapped_sges; i++)
233 		ib_dma_unmap_page(device,
234 				  ctxt->sge[i].addr,
235 				  ctxt->sge[i].length,
236 				  ctxt->direction);
237 	ctxt->mapped_sges = 0;
238 }
239 
240 void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
241 {
242 	struct svcxprt_rdma *xprt = ctxt->xprt;
243 	int i;
244 
245 	if (free_pages)
246 		for (i = 0; i < ctxt->count; i++)
247 			put_page(ctxt->pages[i]);
248 
249 	spin_lock(&xprt->sc_ctxt_lock);
250 	xprt->sc_ctxt_used--;
251 	list_add(&ctxt->list, &xprt->sc_ctxts);
252 	spin_unlock(&xprt->sc_ctxt_lock);
253 }
254 
255 static void svc_rdma_destroy_ctxts(struct svcxprt_rdma *xprt)
256 {
257 	while (!list_empty(&xprt->sc_ctxts)) {
258 		struct svc_rdma_op_ctxt *ctxt;
259 
260 		ctxt = list_first_entry(&xprt->sc_ctxts,
261 					struct svc_rdma_op_ctxt, list);
262 		list_del(&ctxt->list);
263 		kfree(ctxt);
264 	}
265 }
266 
267 /* QP event handler */
268 static void qp_event_handler(struct ib_event *event, void *context)
269 {
270 	struct svc_xprt *xprt = context;
271 
272 	switch (event->event) {
273 	/* These are considered benign events */
274 	case IB_EVENT_PATH_MIG:
275 	case IB_EVENT_COMM_EST:
276 	case IB_EVENT_SQ_DRAINED:
277 	case IB_EVENT_QP_LAST_WQE_REACHED:
278 		dprintk("svcrdma: QP event %s (%d) received for QP=%p\n",
279 			ib_event_msg(event->event), event->event,
280 			event->element.qp);
281 		break;
282 	/* These are considered fatal events */
283 	case IB_EVENT_PATH_MIG_ERR:
284 	case IB_EVENT_QP_FATAL:
285 	case IB_EVENT_QP_REQ_ERR:
286 	case IB_EVENT_QP_ACCESS_ERR:
287 	case IB_EVENT_DEVICE_FATAL:
288 	default:
289 		dprintk("svcrdma: QP ERROR event %s (%d) received for QP=%p, "
290 			"closing transport\n",
291 			ib_event_msg(event->event), event->event,
292 			event->element.qp);
293 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
294 		svc_xprt_enqueue(xprt);
295 		break;
296 	}
297 }
298 
299 /**
300  * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
301  * @cq:        completion queue
302  * @wc:        completed WR
303  *
304  */
305 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
306 {
307 	struct svcxprt_rdma *xprt = cq->cq_context;
308 	struct ib_cqe *cqe = wc->wr_cqe;
309 	struct svc_rdma_op_ctxt *ctxt;
310 
311 	/* WARNING: Only wc->wr_cqe and wc->status are reliable */
312 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
313 	svc_rdma_unmap_dma(ctxt);
314 
315 	if (wc->status != IB_WC_SUCCESS)
316 		goto flushed;
317 
318 	/* All wc fields are now known to be valid */
319 	ctxt->byte_len = wc->byte_len;
320 	spin_lock(&xprt->sc_rq_dto_lock);
321 	list_add_tail(&ctxt->list, &xprt->sc_rq_dto_q);
322 	spin_unlock(&xprt->sc_rq_dto_lock);
323 
324 	svc_rdma_post_recv(xprt);
325 
326 	set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
327 	if (test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
328 		goto out;
329 	goto out_enqueue;
330 
331 flushed:
332 	if (wc->status != IB_WC_WR_FLUSH_ERR)
333 		pr_warn("svcrdma: receive: %s (%u/0x%x)\n",
334 			ib_wc_status_msg(wc->status),
335 			wc->status, wc->vendor_err);
336 	set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
337 	svc_rdma_put_context(ctxt, 1);
338 
339 out_enqueue:
340 	svc_xprt_enqueue(&xprt->sc_xprt);
341 out:
342 	svc_xprt_put(&xprt->sc_xprt);
343 }
344 
345 /**
346  * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC
347  * @cq:        completion queue
348  * @wc:        completed WR
349  *
350  */
351 void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
352 {
353 	struct svcxprt_rdma *xprt = cq->cq_context;
354 	struct ib_cqe *cqe = wc->wr_cqe;
355 	struct svc_rdma_op_ctxt *ctxt;
356 
357 	atomic_inc(&xprt->sc_sq_avail);
358 	wake_up(&xprt->sc_send_wait);
359 
360 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
361 	svc_rdma_unmap_dma(ctxt);
362 	svc_rdma_put_context(ctxt, 1);
363 
364 	if (unlikely(wc->status != IB_WC_SUCCESS)) {
365 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
366 		svc_xprt_enqueue(&xprt->sc_xprt);
367 		if (wc->status != IB_WC_WR_FLUSH_ERR)
368 			pr_err("svcrdma: Send: %s (%u/0x%x)\n",
369 			       ib_wc_status_msg(wc->status),
370 			       wc->status, wc->vendor_err);
371 	}
372 
373 	svc_xprt_put(&xprt->sc_xprt);
374 }
375 
376 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
377 					     int listener)
378 {
379 	struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
380 
381 	if (!cma_xprt)
382 		return NULL;
383 	svc_xprt_init(&init_net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
384 	INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
385 	INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
386 	INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
387 	INIT_LIST_HEAD(&cma_xprt->sc_ctxts);
388 	INIT_LIST_HEAD(&cma_xprt->sc_rw_ctxts);
389 	init_waitqueue_head(&cma_xprt->sc_send_wait);
390 
391 	spin_lock_init(&cma_xprt->sc_lock);
392 	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
393 	spin_lock_init(&cma_xprt->sc_ctxt_lock);
394 	spin_lock_init(&cma_xprt->sc_rw_ctxt_lock);
395 
396 	/*
397 	 * Note that this implies that the underlying transport support
398 	 * has some form of congestion control (see RFC 7530 section 3.1
399 	 * paragraph 2). For now, we assume that all supported RDMA
400 	 * transports are suitable here.
401 	 */
402 	set_bit(XPT_CONG_CTRL, &cma_xprt->sc_xprt.xpt_flags);
403 
404 	if (listener)
405 		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
406 
407 	return cma_xprt;
408 }
409 
410 static int
411 svc_rdma_post_recv(struct svcxprt_rdma *xprt)
412 {
413 	struct ib_recv_wr recv_wr, *bad_recv_wr;
414 	struct svc_rdma_op_ctxt *ctxt;
415 	struct page *page;
416 	dma_addr_t pa;
417 	int sge_no;
418 	int buflen;
419 	int ret;
420 
421 	ctxt = svc_rdma_get_context(xprt);
422 	buflen = 0;
423 	ctxt->direction = DMA_FROM_DEVICE;
424 	ctxt->cqe.done = svc_rdma_wc_receive;
425 	for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) {
426 		if (sge_no >= xprt->sc_max_sge) {
427 			pr_err("svcrdma: Too many sges (%d)\n", sge_no);
428 			goto err_put_ctxt;
429 		}
430 		page = alloc_page(GFP_KERNEL);
431 		if (!page)
432 			goto err_put_ctxt;
433 		ctxt->pages[sge_no] = page;
434 		pa = ib_dma_map_page(xprt->sc_cm_id->device,
435 				     page, 0, PAGE_SIZE,
436 				     DMA_FROM_DEVICE);
437 		if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
438 			goto err_put_ctxt;
439 		svc_rdma_count_mappings(xprt, ctxt);
440 		ctxt->sge[sge_no].addr = pa;
441 		ctxt->sge[sge_no].length = PAGE_SIZE;
442 		ctxt->sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
443 		ctxt->count = sge_no + 1;
444 		buflen += PAGE_SIZE;
445 	}
446 	recv_wr.next = NULL;
447 	recv_wr.sg_list = &ctxt->sge[0];
448 	recv_wr.num_sge = ctxt->count;
449 	recv_wr.wr_cqe = &ctxt->cqe;
450 
451 	svc_xprt_get(&xprt->sc_xprt);
452 	ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
453 	if (ret) {
454 		svc_rdma_unmap_dma(ctxt);
455 		svc_rdma_put_context(ctxt, 1);
456 		svc_xprt_put(&xprt->sc_xprt);
457 	}
458 	return ret;
459 
460  err_put_ctxt:
461 	svc_rdma_unmap_dma(ctxt);
462 	svc_rdma_put_context(ctxt, 1);
463 	return -ENOMEM;
464 }
465 
466 static void
467 svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
468 			       struct rdma_conn_param *param)
469 {
470 	const struct rpcrdma_connect_private *pmsg = param->private_data;
471 
472 	if (pmsg &&
473 	    pmsg->cp_magic == rpcrdma_cmp_magic &&
474 	    pmsg->cp_version == RPCRDMA_CMP_VERSION) {
475 		newxprt->sc_snd_w_inv = pmsg->cp_flags &
476 					RPCRDMA_CMP_F_SND_W_INV_OK;
477 
478 		dprintk("svcrdma: client send_size %u, recv_size %u "
479 			"remote inv %ssupported\n",
480 			rpcrdma_decode_buffer_size(pmsg->cp_send_size),
481 			rpcrdma_decode_buffer_size(pmsg->cp_recv_size),
482 			newxprt->sc_snd_w_inv ? "" : "un");
483 	}
484 }
485 
486 /*
487  * This function handles the CONNECT_REQUEST event on a listening
488  * endpoint. It is passed the cma_id for the _new_ connection. The context in
489  * this cma_id is inherited from the listening cma_id and is the svc_xprt
490  * structure for the listening endpoint.
491  *
492  * This function creates a new xprt for the new connection and enqueues it on
493  * the accept queue for the listent xprt. When the listen thread is kicked, it
494  * will call the recvfrom method on the listen xprt which will accept the new
495  * connection.
496  */
497 static void handle_connect_req(struct rdma_cm_id *new_cma_id,
498 			       struct rdma_conn_param *param)
499 {
500 	struct svcxprt_rdma *listen_xprt = new_cma_id->context;
501 	struct svcxprt_rdma *newxprt;
502 	struct sockaddr *sa;
503 
504 	/* Create a new transport */
505 	newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
506 	if (!newxprt) {
507 		dprintk("svcrdma: failed to create new transport\n");
508 		return;
509 	}
510 	newxprt->sc_cm_id = new_cma_id;
511 	new_cma_id->context = newxprt;
512 	dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
513 		newxprt, newxprt->sc_cm_id, listen_xprt);
514 	svc_rdma_parse_connect_private(newxprt, param);
515 
516 	/* Save client advertised inbound read limit for use later in accept. */
517 	newxprt->sc_ord = param->initiator_depth;
518 
519 	/* Set the local and remote addresses in the transport */
520 	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
521 	svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
522 	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
523 	svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
524 
525 	/*
526 	 * Enqueue the new transport on the accept queue of the listening
527 	 * transport
528 	 */
529 	spin_lock_bh(&listen_xprt->sc_lock);
530 	list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
531 	spin_unlock_bh(&listen_xprt->sc_lock);
532 
533 	set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
534 	svc_xprt_enqueue(&listen_xprt->sc_xprt);
535 }
536 
537 /*
538  * Handles events generated on the listening endpoint. These events will be
539  * either be incoming connect requests or adapter removal  events.
540  */
541 static int rdma_listen_handler(struct rdma_cm_id *cma_id,
542 			       struct rdma_cm_event *event)
543 {
544 	struct svcxprt_rdma *xprt = cma_id->context;
545 	int ret = 0;
546 
547 	switch (event->event) {
548 	case RDMA_CM_EVENT_CONNECT_REQUEST:
549 		dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
550 			"event = %s (%d)\n", cma_id, cma_id->context,
551 			rdma_event_msg(event->event), event->event);
552 		handle_connect_req(cma_id, &event->param.conn);
553 		break;
554 
555 	case RDMA_CM_EVENT_ESTABLISHED:
556 		/* Accept complete */
557 		dprintk("svcrdma: Connection completed on LISTEN xprt=%p, "
558 			"cm_id=%p\n", xprt, cma_id);
559 		break;
560 
561 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
562 		dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
563 			xprt, cma_id);
564 		if (xprt) {
565 			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
566 			svc_xprt_enqueue(&xprt->sc_xprt);
567 		}
568 		break;
569 
570 	default:
571 		dprintk("svcrdma: Unexpected event on listening endpoint %p, "
572 			"event = %s (%d)\n", cma_id,
573 			rdma_event_msg(event->event), event->event);
574 		break;
575 	}
576 
577 	return ret;
578 }
579 
580 static int rdma_cma_handler(struct rdma_cm_id *cma_id,
581 			    struct rdma_cm_event *event)
582 {
583 	struct svc_xprt *xprt = cma_id->context;
584 	struct svcxprt_rdma *rdma =
585 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
586 	switch (event->event) {
587 	case RDMA_CM_EVENT_ESTABLISHED:
588 		/* Accept complete */
589 		svc_xprt_get(xprt);
590 		dprintk("svcrdma: Connection completed on DTO xprt=%p, "
591 			"cm_id=%p\n", xprt, cma_id);
592 		clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
593 		svc_xprt_enqueue(xprt);
594 		break;
595 	case RDMA_CM_EVENT_DISCONNECTED:
596 		dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
597 			xprt, cma_id);
598 		if (xprt) {
599 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
600 			svc_xprt_enqueue(xprt);
601 			svc_xprt_put(xprt);
602 		}
603 		break;
604 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
605 		dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
606 			"event = %s (%d)\n", cma_id, xprt,
607 			rdma_event_msg(event->event), event->event);
608 		if (xprt) {
609 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
610 			svc_xprt_enqueue(xprt);
611 			svc_xprt_put(xprt);
612 		}
613 		break;
614 	default:
615 		dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
616 			"event = %s (%d)\n", cma_id,
617 			rdma_event_msg(event->event), event->event);
618 		break;
619 	}
620 	return 0;
621 }
622 
623 /*
624  * Create a listening RDMA service endpoint.
625  */
626 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
627 					struct net *net,
628 					struct sockaddr *sa, int salen,
629 					int flags)
630 {
631 	struct rdma_cm_id *listen_id;
632 	struct svcxprt_rdma *cma_xprt;
633 	int ret;
634 
635 	dprintk("svcrdma: Creating RDMA socket\n");
636 	if ((sa->sa_family != AF_INET) && (sa->sa_family != AF_INET6)) {
637 		dprintk("svcrdma: Address family %d is not supported.\n", sa->sa_family);
638 		return ERR_PTR(-EAFNOSUPPORT);
639 	}
640 	cma_xprt = rdma_create_xprt(serv, 1);
641 	if (!cma_xprt)
642 		return ERR_PTR(-ENOMEM);
643 
644 	listen_id = rdma_create_id(&init_net, rdma_listen_handler, cma_xprt,
645 				   RDMA_PS_TCP, IB_QPT_RC);
646 	if (IS_ERR(listen_id)) {
647 		ret = PTR_ERR(listen_id);
648 		dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
649 		goto err0;
650 	}
651 
652 	/* Allow both IPv4 and IPv6 sockets to bind a single port
653 	 * at the same time.
654 	 */
655 #if IS_ENABLED(CONFIG_IPV6)
656 	ret = rdma_set_afonly(listen_id, 1);
657 	if (ret) {
658 		dprintk("svcrdma: rdma_set_afonly failed = %d\n", ret);
659 		goto err1;
660 	}
661 #endif
662 	ret = rdma_bind_addr(listen_id, sa);
663 	if (ret) {
664 		dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
665 		goto err1;
666 	}
667 	cma_xprt->sc_cm_id = listen_id;
668 
669 	ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
670 	if (ret) {
671 		dprintk("svcrdma: rdma_listen failed = %d\n", ret);
672 		goto err1;
673 	}
674 
675 	/*
676 	 * We need to use the address from the cm_id in case the
677 	 * caller specified 0 for the port number.
678 	 */
679 	sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
680 	svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
681 
682 	return &cma_xprt->sc_xprt;
683 
684  err1:
685 	rdma_destroy_id(listen_id);
686  err0:
687 	kfree(cma_xprt);
688 	return ERR_PTR(ret);
689 }
690 
691 /*
692  * This is the xpo_recvfrom function for listening endpoints. Its
693  * purpose is to accept incoming connections. The CMA callback handler
694  * has already created a new transport and attached it to the new CMA
695  * ID.
696  *
697  * There is a queue of pending connections hung on the listening
698  * transport. This queue contains the new svc_xprt structure. This
699  * function takes svc_xprt structures off the accept_q and completes
700  * the connection.
701  */
702 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
703 {
704 	struct svcxprt_rdma *listen_rdma;
705 	struct svcxprt_rdma *newxprt = NULL;
706 	struct rdma_conn_param conn_param;
707 	struct rpcrdma_connect_private pmsg;
708 	struct ib_qp_init_attr qp_attr;
709 	struct ib_device *dev;
710 	struct sockaddr *sap;
711 	unsigned int i, ctxts;
712 	int ret = 0;
713 
714 	listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
715 	clear_bit(XPT_CONN, &xprt->xpt_flags);
716 	/* Get the next entry off the accept list */
717 	spin_lock_bh(&listen_rdma->sc_lock);
718 	if (!list_empty(&listen_rdma->sc_accept_q)) {
719 		newxprt = list_entry(listen_rdma->sc_accept_q.next,
720 				     struct svcxprt_rdma, sc_accept_q);
721 		list_del_init(&newxprt->sc_accept_q);
722 	}
723 	if (!list_empty(&listen_rdma->sc_accept_q))
724 		set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
725 	spin_unlock_bh(&listen_rdma->sc_lock);
726 	if (!newxprt)
727 		return NULL;
728 
729 	dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
730 		newxprt, newxprt->sc_cm_id);
731 
732 	dev = newxprt->sc_cm_id->device;
733 	newxprt->sc_port_num = newxprt->sc_cm_id->port_num;
734 
735 	/* Qualify the transport resource defaults with the
736 	 * capabilities of this particular device */
737 	newxprt->sc_max_sge = min((size_t)dev->attrs.max_sge,
738 				  (size_t)RPCSVC_MAXPAGES);
739 	newxprt->sc_max_req_size = svcrdma_max_req_size;
740 	newxprt->sc_max_requests = svcrdma_max_requests;
741 	newxprt->sc_max_bc_requests = svcrdma_max_bc_requests;
742 	newxprt->sc_rq_depth = newxprt->sc_max_requests +
743 			       newxprt->sc_max_bc_requests;
744 	if (newxprt->sc_rq_depth > dev->attrs.max_qp_wr) {
745 		pr_warn("svcrdma: reducing receive depth to %d\n",
746 			dev->attrs.max_qp_wr);
747 		newxprt->sc_rq_depth = dev->attrs.max_qp_wr;
748 		newxprt->sc_max_requests = newxprt->sc_rq_depth - 2;
749 		newxprt->sc_max_bc_requests = 2;
750 	}
751 	newxprt->sc_fc_credits = cpu_to_be32(newxprt->sc_max_requests);
752 	ctxts = rdma_rw_mr_factor(dev, newxprt->sc_port_num, RPCSVC_MAXPAGES);
753 	ctxts *= newxprt->sc_max_requests;
754 	newxprt->sc_sq_depth = newxprt->sc_rq_depth + ctxts;
755 	if (newxprt->sc_sq_depth > dev->attrs.max_qp_wr) {
756 		pr_warn("svcrdma: reducing send depth to %d\n",
757 			dev->attrs.max_qp_wr);
758 		newxprt->sc_sq_depth = dev->attrs.max_qp_wr;
759 	}
760 	atomic_set(&newxprt->sc_sq_avail, newxprt->sc_sq_depth);
761 
762 	if (!svc_rdma_prealloc_ctxts(newxprt))
763 		goto errout;
764 
765 	/*
766 	 * Limit ORD based on client limit, local device limit, and
767 	 * configured svcrdma limit.
768 	 */
769 	newxprt->sc_ord = min_t(size_t, dev->attrs.max_qp_rd_atom, newxprt->sc_ord);
770 	newxprt->sc_ord = min_t(size_t,	svcrdma_ord, newxprt->sc_ord);
771 
772 	newxprt->sc_pd = ib_alloc_pd(dev, 0);
773 	if (IS_ERR(newxprt->sc_pd)) {
774 		dprintk("svcrdma: error creating PD for connect request\n");
775 		goto errout;
776 	}
777 	newxprt->sc_sq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_sq_depth,
778 					0, IB_POLL_WORKQUEUE);
779 	if (IS_ERR(newxprt->sc_sq_cq)) {
780 		dprintk("svcrdma: error creating SQ CQ for connect request\n");
781 		goto errout;
782 	}
783 	newxprt->sc_rq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_rq_depth,
784 					0, IB_POLL_WORKQUEUE);
785 	if (IS_ERR(newxprt->sc_rq_cq)) {
786 		dprintk("svcrdma: error creating RQ CQ for connect request\n");
787 		goto errout;
788 	}
789 
790 	memset(&qp_attr, 0, sizeof qp_attr);
791 	qp_attr.event_handler = qp_event_handler;
792 	qp_attr.qp_context = &newxprt->sc_xprt;
793 	qp_attr.port_num = newxprt->sc_port_num;
794 	qp_attr.cap.max_rdma_ctxs = ctxts;
795 	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth - ctxts;
796 	qp_attr.cap.max_recv_wr = newxprt->sc_rq_depth;
797 	qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
798 	qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
799 	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
800 	qp_attr.qp_type = IB_QPT_RC;
801 	qp_attr.send_cq = newxprt->sc_sq_cq;
802 	qp_attr.recv_cq = newxprt->sc_rq_cq;
803 	dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n",
804 		newxprt->sc_cm_id, newxprt->sc_pd);
805 	dprintk("    cap.max_send_wr = %d, cap.max_recv_wr = %d\n",
806 		qp_attr.cap.max_send_wr, qp_attr.cap.max_recv_wr);
807 	dprintk("    cap.max_send_sge = %d, cap.max_recv_sge = %d\n",
808 		qp_attr.cap.max_send_sge, qp_attr.cap.max_recv_sge);
809 
810 	ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
811 	if (ret) {
812 		dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
813 		goto errout;
814 	}
815 	newxprt->sc_qp = newxprt->sc_cm_id->qp;
816 
817 	if (!(dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS))
818 		newxprt->sc_snd_w_inv = false;
819 	if (!rdma_protocol_iwarp(dev, newxprt->sc_port_num) &&
820 	    !rdma_ib_or_roce(dev, newxprt->sc_port_num))
821 		goto errout;
822 
823 	/* Post receive buffers */
824 	for (i = 0; i < newxprt->sc_max_requests; i++) {
825 		ret = svc_rdma_post_recv(newxprt);
826 		if (ret) {
827 			dprintk("svcrdma: failure posting receive buffers\n");
828 			goto errout;
829 		}
830 	}
831 
832 	/* Swap out the handler */
833 	newxprt->sc_cm_id->event_handler = rdma_cma_handler;
834 
835 	/* Construct RDMA-CM private message */
836 	pmsg.cp_magic = rpcrdma_cmp_magic;
837 	pmsg.cp_version = RPCRDMA_CMP_VERSION;
838 	pmsg.cp_flags = 0;
839 	pmsg.cp_send_size = pmsg.cp_recv_size =
840 		rpcrdma_encode_buffer_size(newxprt->sc_max_req_size);
841 
842 	/* Accept Connection */
843 	set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
844 	memset(&conn_param, 0, sizeof conn_param);
845 	conn_param.responder_resources = 0;
846 	conn_param.initiator_depth = newxprt->sc_ord;
847 	conn_param.private_data = &pmsg;
848 	conn_param.private_data_len = sizeof(pmsg);
849 	ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
850 	if (ret) {
851 		dprintk("svcrdma: failed to accept new connection, ret=%d\n",
852 		       ret);
853 		goto errout;
854 	}
855 
856 	dprintk("svcrdma: new connection %p accepted:\n", newxprt);
857 	sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
858 	dprintk("    local address   : %pIS:%u\n", sap, rpc_get_port(sap));
859 	sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
860 	dprintk("    remote address  : %pIS:%u\n", sap, rpc_get_port(sap));
861 	dprintk("    max_sge         : %d\n", newxprt->sc_max_sge);
862 	dprintk("    sq_depth        : %d\n", newxprt->sc_sq_depth);
863 	dprintk("    rdma_rw_ctxs    : %d\n", ctxts);
864 	dprintk("    max_requests    : %d\n", newxprt->sc_max_requests);
865 	dprintk("    ord             : %d\n", newxprt->sc_ord);
866 
867 	return &newxprt->sc_xprt;
868 
869  errout:
870 	dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
871 	/* Take a reference in case the DTO handler runs */
872 	svc_xprt_get(&newxprt->sc_xprt);
873 	if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
874 		ib_destroy_qp(newxprt->sc_qp);
875 	rdma_destroy_id(newxprt->sc_cm_id);
876 	/* This call to put will destroy the transport */
877 	svc_xprt_put(&newxprt->sc_xprt);
878 	return NULL;
879 }
880 
881 static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
882 {
883 }
884 
885 /*
886  * When connected, an svc_xprt has at least two references:
887  *
888  * - A reference held by the cm_id between the ESTABLISHED and
889  *   DISCONNECTED events. If the remote peer disconnected first, this
890  *   reference could be gone.
891  *
892  * - A reference held by the svc_recv code that called this function
893  *   as part of close processing.
894  *
895  * At a minimum one references should still be held.
896  */
897 static void svc_rdma_detach(struct svc_xprt *xprt)
898 {
899 	struct svcxprt_rdma *rdma =
900 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
901 	dprintk("svc: svc_rdma_detach(%p)\n", xprt);
902 
903 	/* Disconnect and flush posted WQE */
904 	rdma_disconnect(rdma->sc_cm_id);
905 }
906 
907 static void __svc_rdma_free(struct work_struct *work)
908 {
909 	struct svcxprt_rdma *rdma =
910 		container_of(work, struct svcxprt_rdma, sc_work);
911 	struct svc_xprt *xprt = &rdma->sc_xprt;
912 
913 	dprintk("svcrdma: %s(%p)\n", __func__, rdma);
914 
915 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
916 		ib_drain_qp(rdma->sc_qp);
917 
918 	/* We should only be called from kref_put */
919 	if (kref_read(&xprt->xpt_ref) != 0)
920 		pr_err("svcrdma: sc_xprt still in use? (%d)\n",
921 		       kref_read(&xprt->xpt_ref));
922 
923 	while (!list_empty(&rdma->sc_read_complete_q)) {
924 		struct svc_rdma_op_ctxt *ctxt;
925 		ctxt = list_first_entry(&rdma->sc_read_complete_q,
926 					struct svc_rdma_op_ctxt, list);
927 		list_del(&ctxt->list);
928 		svc_rdma_put_context(ctxt, 1);
929 	}
930 	while (!list_empty(&rdma->sc_rq_dto_q)) {
931 		struct svc_rdma_op_ctxt *ctxt;
932 		ctxt = list_first_entry(&rdma->sc_rq_dto_q,
933 					struct svc_rdma_op_ctxt, list);
934 		list_del(&ctxt->list);
935 		svc_rdma_put_context(ctxt, 1);
936 	}
937 
938 	/* Warn if we leaked a resource or under-referenced */
939 	if (rdma->sc_ctxt_used != 0)
940 		pr_err("svcrdma: ctxt still in use? (%d)\n",
941 		       rdma->sc_ctxt_used);
942 
943 	/* Final put of backchannel client transport */
944 	if (xprt->xpt_bc_xprt) {
945 		xprt_put(xprt->xpt_bc_xprt);
946 		xprt->xpt_bc_xprt = NULL;
947 	}
948 
949 	svc_rdma_destroy_rw_ctxts(rdma);
950 	svc_rdma_destroy_ctxts(rdma);
951 
952 	/* Destroy the QP if present (not a listener) */
953 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
954 		ib_destroy_qp(rdma->sc_qp);
955 
956 	if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
957 		ib_free_cq(rdma->sc_sq_cq);
958 
959 	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
960 		ib_free_cq(rdma->sc_rq_cq);
961 
962 	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
963 		ib_dealloc_pd(rdma->sc_pd);
964 
965 	/* Destroy the CM ID */
966 	rdma_destroy_id(rdma->sc_cm_id);
967 
968 	kfree(rdma);
969 }
970 
971 static void svc_rdma_free(struct svc_xprt *xprt)
972 {
973 	struct svcxprt_rdma *rdma =
974 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
975 	INIT_WORK(&rdma->sc_work, __svc_rdma_free);
976 	queue_work(svc_rdma_wq, &rdma->sc_work);
977 }
978 
979 static int svc_rdma_has_wspace(struct svc_xprt *xprt)
980 {
981 	struct svcxprt_rdma *rdma =
982 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
983 
984 	/*
985 	 * If there are already waiters on the SQ,
986 	 * return false.
987 	 */
988 	if (waitqueue_active(&rdma->sc_send_wait))
989 		return 0;
990 
991 	/* Otherwise return true. */
992 	return 1;
993 }
994 
995 static int svc_rdma_secure_port(struct svc_rqst *rqstp)
996 {
997 	return 1;
998 }
999 
1000 static void svc_rdma_kill_temp_xprt(struct svc_xprt *xprt)
1001 {
1002 }
1003 
1004 int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
1005 {
1006 	struct ib_send_wr *bad_wr, *n_wr;
1007 	int wr_count;
1008 	int i;
1009 	int ret;
1010 
1011 	if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1012 		return -ENOTCONN;
1013 
1014 	wr_count = 1;
1015 	for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
1016 		wr_count++;
1017 
1018 	/* If the SQ is full, wait until an SQ entry is available */
1019 	while (1) {
1020 		if ((atomic_sub_return(wr_count, &xprt->sc_sq_avail) < 0)) {
1021 			atomic_inc(&rdma_stat_sq_starve);
1022 
1023 			/* Wait until SQ WR available if SQ still full */
1024 			atomic_add(wr_count, &xprt->sc_sq_avail);
1025 			wait_event(xprt->sc_send_wait,
1026 				   atomic_read(&xprt->sc_sq_avail) > wr_count);
1027 			if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1028 				return -ENOTCONN;
1029 			continue;
1030 		}
1031 		/* Take a transport ref for each WR posted */
1032 		for (i = 0; i < wr_count; i++)
1033 			svc_xprt_get(&xprt->sc_xprt);
1034 
1035 		/* Bump used SQ WR count and post */
1036 		ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
1037 		if (ret) {
1038 			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
1039 			for (i = 0; i < wr_count; i ++)
1040 				svc_xprt_put(&xprt->sc_xprt);
1041 			dprintk("svcrdma: failed to post SQ WR rc=%d\n", ret);
1042 			dprintk("    sc_sq_avail=%d, sc_sq_depth=%d\n",
1043 				atomic_read(&xprt->sc_sq_avail),
1044 				xprt->sc_sq_depth);
1045 			wake_up(&xprt->sc_send_wait);
1046 		}
1047 		break;
1048 	}
1049 	return ret;
1050 }
1051