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 <linux/sunrpc/svc_rdma.h>
55 #include <linux/export.h>
56 #include "xprt_rdma.h"
57 
58 #define RPCDBG_FACILITY	RPCDBG_SVCXPRT
59 
60 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *, int);
61 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
62 					struct net *net,
63 					struct sockaddr *sa, int salen,
64 					int flags);
65 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt);
66 static void svc_rdma_release_rqst(struct svc_rqst *);
67 static void svc_rdma_detach(struct svc_xprt *xprt);
68 static void svc_rdma_free(struct svc_xprt *xprt);
69 static int svc_rdma_has_wspace(struct svc_xprt *xprt);
70 static int svc_rdma_secure_port(struct svc_rqst *);
71 static void svc_rdma_kill_temp_xprt(struct svc_xprt *);
72 
73 static struct svc_xprt_ops svc_rdma_ops = {
74 	.xpo_create = svc_rdma_create,
75 	.xpo_recvfrom = svc_rdma_recvfrom,
76 	.xpo_sendto = svc_rdma_sendto,
77 	.xpo_release_rqst = svc_rdma_release_rqst,
78 	.xpo_detach = svc_rdma_detach,
79 	.xpo_free = svc_rdma_free,
80 	.xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
81 	.xpo_has_wspace = svc_rdma_has_wspace,
82 	.xpo_accept = svc_rdma_accept,
83 	.xpo_secure_port = svc_rdma_secure_port,
84 	.xpo_kill_temp_xprt = svc_rdma_kill_temp_xprt,
85 };
86 
87 struct svc_xprt_class svc_rdma_class = {
88 	.xcl_name = "rdma",
89 	.xcl_owner = THIS_MODULE,
90 	.xcl_ops = &svc_rdma_ops,
91 	.xcl_max_payload = RPCSVC_MAXPAYLOAD_RDMA,
92 	.xcl_ident = XPRT_TRANSPORT_RDMA,
93 };
94 
95 #if defined(CONFIG_SUNRPC_BACKCHANNEL)
96 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *, struct net *,
97 					   struct sockaddr *, int, int);
98 static void svc_rdma_bc_detach(struct svc_xprt *);
99 static void svc_rdma_bc_free(struct svc_xprt *);
100 
101 static struct svc_xprt_ops svc_rdma_bc_ops = {
102 	.xpo_create = svc_rdma_bc_create,
103 	.xpo_detach = svc_rdma_bc_detach,
104 	.xpo_free = svc_rdma_bc_free,
105 	.xpo_prep_reply_hdr = svc_rdma_prep_reply_hdr,
106 	.xpo_secure_port = svc_rdma_secure_port,
107 };
108 
109 struct svc_xprt_class svc_rdma_bc_class = {
110 	.xcl_name = "rdma-bc",
111 	.xcl_owner = THIS_MODULE,
112 	.xcl_ops = &svc_rdma_bc_ops,
113 	.xcl_max_payload = (1024 - RPCRDMA_HDRLEN_MIN)
114 };
115 
116 static struct svc_xprt *svc_rdma_bc_create(struct svc_serv *serv,
117 					   struct net *net,
118 					   struct sockaddr *sa, int salen,
119 					   int flags)
120 {
121 	struct svcxprt_rdma *cma_xprt;
122 	struct svc_xprt *xprt;
123 
124 	cma_xprt = rdma_create_xprt(serv, 0);
125 	if (!cma_xprt)
126 		return ERR_PTR(-ENOMEM);
127 	xprt = &cma_xprt->sc_xprt;
128 
129 	svc_xprt_init(net, &svc_rdma_bc_class, xprt, serv);
130 	serv->sv_bc_xprt = xprt;
131 
132 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
133 	return xprt;
134 }
135 
136 static void svc_rdma_bc_detach(struct svc_xprt *xprt)
137 {
138 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
139 }
140 
141 static void svc_rdma_bc_free(struct svc_xprt *xprt)
142 {
143 	struct svcxprt_rdma *rdma =
144 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
145 
146 	dprintk("svcrdma: %s(%p)\n", __func__, xprt);
147 	if (xprt)
148 		kfree(rdma);
149 }
150 #endif	/* CONFIG_SUNRPC_BACKCHANNEL */
151 
152 static struct svc_rdma_op_ctxt *alloc_ctxt(struct svcxprt_rdma *xprt,
153 					   gfp_t flags)
154 {
155 	struct svc_rdma_op_ctxt *ctxt;
156 
157 	ctxt = kmalloc(sizeof(*ctxt), flags);
158 	if (ctxt) {
159 		ctxt->xprt = xprt;
160 		INIT_LIST_HEAD(&ctxt->list);
161 	}
162 	return ctxt;
163 }
164 
165 static bool svc_rdma_prealloc_ctxts(struct svcxprt_rdma *xprt)
166 {
167 	unsigned int i;
168 
169 	/* Each RPC/RDMA credit can consume a number of send
170 	 * and receive WQEs. One ctxt is allocated for each.
171 	 */
172 	i = xprt->sc_sq_depth + xprt->sc_rq_depth;
173 
174 	while (i--) {
175 		struct svc_rdma_op_ctxt *ctxt;
176 
177 		ctxt = alloc_ctxt(xprt, GFP_KERNEL);
178 		if (!ctxt) {
179 			dprintk("svcrdma: No memory for RDMA ctxt\n");
180 			return false;
181 		}
182 		list_add(&ctxt->list, &xprt->sc_ctxts);
183 	}
184 	return true;
185 }
186 
187 struct svc_rdma_op_ctxt *svc_rdma_get_context(struct svcxprt_rdma *xprt)
188 {
189 	struct svc_rdma_op_ctxt *ctxt = NULL;
190 
191 	spin_lock(&xprt->sc_ctxt_lock);
192 	xprt->sc_ctxt_used++;
193 	if (list_empty(&xprt->sc_ctxts))
194 		goto out_empty;
195 
196 	ctxt = list_first_entry(&xprt->sc_ctxts,
197 				struct svc_rdma_op_ctxt, list);
198 	list_del(&ctxt->list);
199 	spin_unlock(&xprt->sc_ctxt_lock);
200 
201 out:
202 	ctxt->count = 0;
203 	ctxt->mapped_sges = 0;
204 	ctxt->frmr = NULL;
205 	return ctxt;
206 
207 out_empty:
208 	/* Either pre-allocation missed the mark, or send
209 	 * queue accounting is broken.
210 	 */
211 	spin_unlock(&xprt->sc_ctxt_lock);
212 
213 	ctxt = alloc_ctxt(xprt, GFP_NOIO);
214 	if (ctxt)
215 		goto out;
216 
217 	spin_lock(&xprt->sc_ctxt_lock);
218 	xprt->sc_ctxt_used--;
219 	spin_unlock(&xprt->sc_ctxt_lock);
220 	WARN_ONCE(1, "svcrdma: empty RDMA ctxt list?\n");
221 	return NULL;
222 }
223 
224 void svc_rdma_unmap_dma(struct svc_rdma_op_ctxt *ctxt)
225 {
226 	struct svcxprt_rdma *xprt = ctxt->xprt;
227 	struct ib_device *device = xprt->sc_cm_id->device;
228 	u32 lkey = xprt->sc_pd->local_dma_lkey;
229 	unsigned int i;
230 
231 	for (i = 0; i < ctxt->mapped_sges; i++) {
232 		/*
233 		 * Unmap the DMA addr in the SGE if the lkey matches
234 		 * the local_dma_lkey, otherwise, ignore it since it is
235 		 * an FRMR lkey and will be unmapped later when the
236 		 * last WR that uses it completes.
237 		 */
238 		if (ctxt->sge[i].lkey == lkey)
239 			ib_dma_unmap_page(device,
240 					    ctxt->sge[i].addr,
241 					    ctxt->sge[i].length,
242 					    ctxt->direction);
243 	}
244 	ctxt->mapped_sges = 0;
245 }
246 
247 void svc_rdma_put_context(struct svc_rdma_op_ctxt *ctxt, int free_pages)
248 {
249 	struct svcxprt_rdma *xprt = ctxt->xprt;
250 	int i;
251 
252 	if (free_pages)
253 		for (i = 0; i < ctxt->count; i++)
254 			put_page(ctxt->pages[i]);
255 
256 	spin_lock(&xprt->sc_ctxt_lock);
257 	xprt->sc_ctxt_used--;
258 	list_add(&ctxt->list, &xprt->sc_ctxts);
259 	spin_unlock(&xprt->sc_ctxt_lock);
260 }
261 
262 static void svc_rdma_destroy_ctxts(struct svcxprt_rdma *xprt)
263 {
264 	while (!list_empty(&xprt->sc_ctxts)) {
265 		struct svc_rdma_op_ctxt *ctxt;
266 
267 		ctxt = list_first_entry(&xprt->sc_ctxts,
268 					struct svc_rdma_op_ctxt, list);
269 		list_del(&ctxt->list);
270 		kfree(ctxt);
271 	}
272 }
273 
274 static struct svc_rdma_req_map *alloc_req_map(gfp_t flags)
275 {
276 	struct svc_rdma_req_map *map;
277 
278 	map = kmalloc(sizeof(*map), flags);
279 	if (map)
280 		INIT_LIST_HEAD(&map->free);
281 	return map;
282 }
283 
284 static bool svc_rdma_prealloc_maps(struct svcxprt_rdma *xprt)
285 {
286 	unsigned int i;
287 
288 	/* One for each receive buffer on this connection. */
289 	i = xprt->sc_max_requests;
290 
291 	while (i--) {
292 		struct svc_rdma_req_map *map;
293 
294 		map = alloc_req_map(GFP_KERNEL);
295 		if (!map) {
296 			dprintk("svcrdma: No memory for request map\n");
297 			return false;
298 		}
299 		list_add(&map->free, &xprt->sc_maps);
300 	}
301 	return true;
302 }
303 
304 struct svc_rdma_req_map *svc_rdma_get_req_map(struct svcxprt_rdma *xprt)
305 {
306 	struct svc_rdma_req_map *map = NULL;
307 
308 	spin_lock(&xprt->sc_map_lock);
309 	if (list_empty(&xprt->sc_maps))
310 		goto out_empty;
311 
312 	map = list_first_entry(&xprt->sc_maps,
313 			       struct svc_rdma_req_map, free);
314 	list_del_init(&map->free);
315 	spin_unlock(&xprt->sc_map_lock);
316 
317 out:
318 	map->count = 0;
319 	return map;
320 
321 out_empty:
322 	spin_unlock(&xprt->sc_map_lock);
323 
324 	/* Pre-allocation amount was incorrect */
325 	map = alloc_req_map(GFP_NOIO);
326 	if (map)
327 		goto out;
328 
329 	WARN_ONCE(1, "svcrdma: empty request map list?\n");
330 	return NULL;
331 }
332 
333 void svc_rdma_put_req_map(struct svcxprt_rdma *xprt,
334 			  struct svc_rdma_req_map *map)
335 {
336 	spin_lock(&xprt->sc_map_lock);
337 	list_add(&map->free, &xprt->sc_maps);
338 	spin_unlock(&xprt->sc_map_lock);
339 }
340 
341 static void svc_rdma_destroy_maps(struct svcxprt_rdma *xprt)
342 {
343 	while (!list_empty(&xprt->sc_maps)) {
344 		struct svc_rdma_req_map *map;
345 
346 		map = list_first_entry(&xprt->sc_maps,
347 				       struct svc_rdma_req_map, free);
348 		list_del(&map->free);
349 		kfree(map);
350 	}
351 }
352 
353 /* QP event handler */
354 static void qp_event_handler(struct ib_event *event, void *context)
355 {
356 	struct svc_xprt *xprt = context;
357 
358 	switch (event->event) {
359 	/* These are considered benign events */
360 	case IB_EVENT_PATH_MIG:
361 	case IB_EVENT_COMM_EST:
362 	case IB_EVENT_SQ_DRAINED:
363 	case IB_EVENT_QP_LAST_WQE_REACHED:
364 		dprintk("svcrdma: QP event %s (%d) received for QP=%p\n",
365 			ib_event_msg(event->event), event->event,
366 			event->element.qp);
367 		break;
368 	/* These are considered fatal events */
369 	case IB_EVENT_PATH_MIG_ERR:
370 	case IB_EVENT_QP_FATAL:
371 	case IB_EVENT_QP_REQ_ERR:
372 	case IB_EVENT_QP_ACCESS_ERR:
373 	case IB_EVENT_DEVICE_FATAL:
374 	default:
375 		dprintk("svcrdma: QP ERROR event %s (%d) received for QP=%p, "
376 			"closing transport\n",
377 			ib_event_msg(event->event), event->event,
378 			event->element.qp);
379 		set_bit(XPT_CLOSE, &xprt->xpt_flags);
380 		break;
381 	}
382 }
383 
384 /**
385  * svc_rdma_wc_receive - Invoked by RDMA provider for each polled Receive WC
386  * @cq:        completion queue
387  * @wc:        completed WR
388  *
389  */
390 static void svc_rdma_wc_receive(struct ib_cq *cq, struct ib_wc *wc)
391 {
392 	struct svcxprt_rdma *xprt = cq->cq_context;
393 	struct ib_cqe *cqe = wc->wr_cqe;
394 	struct svc_rdma_op_ctxt *ctxt;
395 
396 	/* WARNING: Only wc->wr_cqe and wc->status are reliable */
397 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
398 	svc_rdma_unmap_dma(ctxt);
399 
400 	if (wc->status != IB_WC_SUCCESS)
401 		goto flushed;
402 
403 	/* All wc fields are now known to be valid */
404 	ctxt->byte_len = wc->byte_len;
405 	spin_lock(&xprt->sc_rq_dto_lock);
406 	list_add_tail(&ctxt->list, &xprt->sc_rq_dto_q);
407 	spin_unlock(&xprt->sc_rq_dto_lock);
408 
409 	set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
410 	if (test_bit(RDMAXPRT_CONN_PENDING, &xprt->sc_flags))
411 		goto out;
412 	svc_xprt_enqueue(&xprt->sc_xprt);
413 	goto out;
414 
415 flushed:
416 	if (wc->status != IB_WC_WR_FLUSH_ERR)
417 		pr_warn("svcrdma: receive: %s (%u/0x%x)\n",
418 			ib_wc_status_msg(wc->status),
419 			wc->status, wc->vendor_err);
420 	set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
421 	svc_rdma_put_context(ctxt, 1);
422 
423 out:
424 	svc_xprt_put(&xprt->sc_xprt);
425 }
426 
427 static void svc_rdma_send_wc_common(struct svcxprt_rdma *xprt,
428 				    struct ib_wc *wc,
429 				    const char *opname)
430 {
431 	if (wc->status != IB_WC_SUCCESS)
432 		goto err;
433 
434 out:
435 	atomic_inc(&xprt->sc_sq_avail);
436 	wake_up(&xprt->sc_send_wait);
437 	return;
438 
439 err:
440 	set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
441 	if (wc->status != IB_WC_WR_FLUSH_ERR)
442 		pr_err("svcrdma: %s: %s (%u/0x%x)\n",
443 		       opname, ib_wc_status_msg(wc->status),
444 		       wc->status, wc->vendor_err);
445 	goto out;
446 }
447 
448 static void svc_rdma_send_wc_common_put(struct ib_cq *cq, struct ib_wc *wc,
449 					const char *opname)
450 {
451 	struct svcxprt_rdma *xprt = cq->cq_context;
452 
453 	svc_rdma_send_wc_common(xprt, wc, opname);
454 	svc_xprt_put(&xprt->sc_xprt);
455 }
456 
457 /**
458  * svc_rdma_wc_send - Invoked by RDMA provider for each polled Send WC
459  * @cq:        completion queue
460  * @wc:        completed WR
461  *
462  */
463 void svc_rdma_wc_send(struct ib_cq *cq, struct ib_wc *wc)
464 {
465 	struct ib_cqe *cqe = wc->wr_cqe;
466 	struct svc_rdma_op_ctxt *ctxt;
467 
468 	svc_rdma_send_wc_common_put(cq, wc, "send");
469 
470 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
471 	svc_rdma_unmap_dma(ctxt);
472 	svc_rdma_put_context(ctxt, 1);
473 }
474 
475 /**
476  * svc_rdma_wc_write - Invoked by RDMA provider for each polled Write WC
477  * @cq:        completion queue
478  * @wc:        completed WR
479  *
480  */
481 void svc_rdma_wc_write(struct ib_cq *cq, struct ib_wc *wc)
482 {
483 	struct ib_cqe *cqe = wc->wr_cqe;
484 	struct svc_rdma_op_ctxt *ctxt;
485 
486 	svc_rdma_send_wc_common_put(cq, wc, "write");
487 
488 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
489 	svc_rdma_unmap_dma(ctxt);
490 	svc_rdma_put_context(ctxt, 0);
491 }
492 
493 /**
494  * svc_rdma_wc_reg - Invoked by RDMA provider for each polled FASTREG WC
495  * @cq:        completion queue
496  * @wc:        completed WR
497  *
498  */
499 void svc_rdma_wc_reg(struct ib_cq *cq, struct ib_wc *wc)
500 {
501 	svc_rdma_send_wc_common_put(cq, wc, "fastreg");
502 }
503 
504 /**
505  * svc_rdma_wc_read - Invoked by RDMA provider for each polled Read WC
506  * @cq:        completion queue
507  * @wc:        completed WR
508  *
509  */
510 void svc_rdma_wc_read(struct ib_cq *cq, struct ib_wc *wc)
511 {
512 	struct svcxprt_rdma *xprt = cq->cq_context;
513 	struct ib_cqe *cqe = wc->wr_cqe;
514 	struct svc_rdma_op_ctxt *ctxt;
515 
516 	svc_rdma_send_wc_common(xprt, wc, "read");
517 
518 	ctxt = container_of(cqe, struct svc_rdma_op_ctxt, cqe);
519 	svc_rdma_unmap_dma(ctxt);
520 	svc_rdma_put_frmr(xprt, ctxt->frmr);
521 
522 	if (test_bit(RDMACTXT_F_LAST_CTXT, &ctxt->flags)) {
523 		struct svc_rdma_op_ctxt *read_hdr;
524 
525 		read_hdr = ctxt->read_hdr;
526 		spin_lock(&xprt->sc_rq_dto_lock);
527 		list_add_tail(&read_hdr->list,
528 			      &xprt->sc_read_complete_q);
529 		spin_unlock(&xprt->sc_rq_dto_lock);
530 
531 		set_bit(XPT_DATA, &xprt->sc_xprt.xpt_flags);
532 		svc_xprt_enqueue(&xprt->sc_xprt);
533 	}
534 
535 	svc_rdma_put_context(ctxt, 0);
536 	svc_xprt_put(&xprt->sc_xprt);
537 }
538 
539 /**
540  * svc_rdma_wc_inv - Invoked by RDMA provider for each polled LOCAL_INV WC
541  * @cq:        completion queue
542  * @wc:        completed WR
543  *
544  */
545 void svc_rdma_wc_inv(struct ib_cq *cq, struct ib_wc *wc)
546 {
547 	svc_rdma_send_wc_common_put(cq, wc, "localInv");
548 }
549 
550 static struct svcxprt_rdma *rdma_create_xprt(struct svc_serv *serv,
551 					     int listener)
552 {
553 	struct svcxprt_rdma *cma_xprt = kzalloc(sizeof *cma_xprt, GFP_KERNEL);
554 
555 	if (!cma_xprt)
556 		return NULL;
557 	svc_xprt_init(&init_net, &svc_rdma_class, &cma_xprt->sc_xprt, serv);
558 	INIT_LIST_HEAD(&cma_xprt->sc_accept_q);
559 	INIT_LIST_HEAD(&cma_xprt->sc_rq_dto_q);
560 	INIT_LIST_HEAD(&cma_xprt->sc_read_complete_q);
561 	INIT_LIST_HEAD(&cma_xprt->sc_frmr_q);
562 	INIT_LIST_HEAD(&cma_xprt->sc_ctxts);
563 	INIT_LIST_HEAD(&cma_xprt->sc_maps);
564 	init_waitqueue_head(&cma_xprt->sc_send_wait);
565 
566 	spin_lock_init(&cma_xprt->sc_lock);
567 	spin_lock_init(&cma_xprt->sc_rq_dto_lock);
568 	spin_lock_init(&cma_xprt->sc_frmr_q_lock);
569 	spin_lock_init(&cma_xprt->sc_ctxt_lock);
570 	spin_lock_init(&cma_xprt->sc_map_lock);
571 
572 	/*
573 	 * Note that this implies that the underlying transport support
574 	 * has some form of congestion control (see RFC 7530 section 3.1
575 	 * paragraph 2). For now, we assume that all supported RDMA
576 	 * transports are suitable here.
577 	 */
578 	set_bit(XPT_CONG_CTRL, &cma_xprt->sc_xprt.xpt_flags);
579 
580 	if (listener)
581 		set_bit(XPT_LISTENER, &cma_xprt->sc_xprt.xpt_flags);
582 
583 	return cma_xprt;
584 }
585 
586 int svc_rdma_post_recv(struct svcxprt_rdma *xprt, gfp_t flags)
587 {
588 	struct ib_recv_wr recv_wr, *bad_recv_wr;
589 	struct svc_rdma_op_ctxt *ctxt;
590 	struct page *page;
591 	dma_addr_t pa;
592 	int sge_no;
593 	int buflen;
594 	int ret;
595 
596 	ctxt = svc_rdma_get_context(xprt);
597 	buflen = 0;
598 	ctxt->direction = DMA_FROM_DEVICE;
599 	ctxt->cqe.done = svc_rdma_wc_receive;
600 	for (sge_no = 0; buflen < xprt->sc_max_req_size; sge_no++) {
601 		if (sge_no >= xprt->sc_max_sge) {
602 			pr_err("svcrdma: Too many sges (%d)\n", sge_no);
603 			goto err_put_ctxt;
604 		}
605 		page = alloc_page(flags);
606 		if (!page)
607 			goto err_put_ctxt;
608 		ctxt->pages[sge_no] = page;
609 		pa = ib_dma_map_page(xprt->sc_cm_id->device,
610 				     page, 0, PAGE_SIZE,
611 				     DMA_FROM_DEVICE);
612 		if (ib_dma_mapping_error(xprt->sc_cm_id->device, pa))
613 			goto err_put_ctxt;
614 		svc_rdma_count_mappings(xprt, ctxt);
615 		ctxt->sge[sge_no].addr = pa;
616 		ctxt->sge[sge_no].length = PAGE_SIZE;
617 		ctxt->sge[sge_no].lkey = xprt->sc_pd->local_dma_lkey;
618 		ctxt->count = sge_no + 1;
619 		buflen += PAGE_SIZE;
620 	}
621 	recv_wr.next = NULL;
622 	recv_wr.sg_list = &ctxt->sge[0];
623 	recv_wr.num_sge = ctxt->count;
624 	recv_wr.wr_cqe = &ctxt->cqe;
625 
626 	svc_xprt_get(&xprt->sc_xprt);
627 	ret = ib_post_recv(xprt->sc_qp, &recv_wr, &bad_recv_wr);
628 	if (ret) {
629 		svc_rdma_unmap_dma(ctxt);
630 		svc_rdma_put_context(ctxt, 1);
631 		svc_xprt_put(&xprt->sc_xprt);
632 	}
633 	return ret;
634 
635  err_put_ctxt:
636 	svc_rdma_unmap_dma(ctxt);
637 	svc_rdma_put_context(ctxt, 1);
638 	return -ENOMEM;
639 }
640 
641 int svc_rdma_repost_recv(struct svcxprt_rdma *xprt, gfp_t flags)
642 {
643 	int ret = 0;
644 
645 	ret = svc_rdma_post_recv(xprt, flags);
646 	if (ret) {
647 		pr_err("svcrdma: could not post a receive buffer, err=%d.\n",
648 		       ret);
649 		pr_err("svcrdma: closing transport %p.\n", xprt);
650 		set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
651 		ret = -ENOTCONN;
652 	}
653 	return ret;
654 }
655 
656 static void
657 svc_rdma_parse_connect_private(struct svcxprt_rdma *newxprt,
658 			       struct rdma_conn_param *param)
659 {
660 	const struct rpcrdma_connect_private *pmsg = param->private_data;
661 
662 	if (pmsg &&
663 	    pmsg->cp_magic == rpcrdma_cmp_magic &&
664 	    pmsg->cp_version == RPCRDMA_CMP_VERSION) {
665 		newxprt->sc_snd_w_inv = pmsg->cp_flags &
666 					RPCRDMA_CMP_F_SND_W_INV_OK;
667 
668 		dprintk("svcrdma: client send_size %u, recv_size %u "
669 			"remote inv %ssupported\n",
670 			rpcrdma_decode_buffer_size(pmsg->cp_send_size),
671 			rpcrdma_decode_buffer_size(pmsg->cp_recv_size),
672 			newxprt->sc_snd_w_inv ? "" : "un");
673 	}
674 }
675 
676 /*
677  * This function handles the CONNECT_REQUEST event on a listening
678  * endpoint. It is passed the cma_id for the _new_ connection. The context in
679  * this cma_id is inherited from the listening cma_id and is the svc_xprt
680  * structure for the listening endpoint.
681  *
682  * This function creates a new xprt for the new connection and enqueues it on
683  * the accept queue for the listent xprt. When the listen thread is kicked, it
684  * will call the recvfrom method on the listen xprt which will accept the new
685  * connection.
686  */
687 static void handle_connect_req(struct rdma_cm_id *new_cma_id,
688 			       struct rdma_conn_param *param)
689 {
690 	struct svcxprt_rdma *listen_xprt = new_cma_id->context;
691 	struct svcxprt_rdma *newxprt;
692 	struct sockaddr *sa;
693 
694 	/* Create a new transport */
695 	newxprt = rdma_create_xprt(listen_xprt->sc_xprt.xpt_server, 0);
696 	if (!newxprt) {
697 		dprintk("svcrdma: failed to create new transport\n");
698 		return;
699 	}
700 	newxprt->sc_cm_id = new_cma_id;
701 	new_cma_id->context = newxprt;
702 	dprintk("svcrdma: Creating newxprt=%p, cm_id=%p, listenxprt=%p\n",
703 		newxprt, newxprt->sc_cm_id, listen_xprt);
704 	svc_rdma_parse_connect_private(newxprt, param);
705 
706 	/* Save client advertised inbound read limit for use later in accept. */
707 	newxprt->sc_ord = param->initiator_depth;
708 
709 	/* Set the local and remote addresses in the transport */
710 	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
711 	svc_xprt_set_remote(&newxprt->sc_xprt, sa, svc_addr_len(sa));
712 	sa = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
713 	svc_xprt_set_local(&newxprt->sc_xprt, sa, svc_addr_len(sa));
714 
715 	/*
716 	 * Enqueue the new transport on the accept queue of the listening
717 	 * transport
718 	 */
719 	spin_lock_bh(&listen_xprt->sc_lock);
720 	list_add_tail(&newxprt->sc_accept_q, &listen_xprt->sc_accept_q);
721 	spin_unlock_bh(&listen_xprt->sc_lock);
722 
723 	set_bit(XPT_CONN, &listen_xprt->sc_xprt.xpt_flags);
724 	svc_xprt_enqueue(&listen_xprt->sc_xprt);
725 }
726 
727 /*
728  * Handles events generated on the listening endpoint. These events will be
729  * either be incoming connect requests or adapter removal  events.
730  */
731 static int rdma_listen_handler(struct rdma_cm_id *cma_id,
732 			       struct rdma_cm_event *event)
733 {
734 	struct svcxprt_rdma *xprt = cma_id->context;
735 	int ret = 0;
736 
737 	switch (event->event) {
738 	case RDMA_CM_EVENT_CONNECT_REQUEST:
739 		dprintk("svcrdma: Connect request on cma_id=%p, xprt = %p, "
740 			"event = %s (%d)\n", cma_id, cma_id->context,
741 			rdma_event_msg(event->event), event->event);
742 		handle_connect_req(cma_id, &event->param.conn);
743 		break;
744 
745 	case RDMA_CM_EVENT_ESTABLISHED:
746 		/* Accept complete */
747 		dprintk("svcrdma: Connection completed on LISTEN xprt=%p, "
748 			"cm_id=%p\n", xprt, cma_id);
749 		break;
750 
751 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
752 		dprintk("svcrdma: Device removal xprt=%p, cm_id=%p\n",
753 			xprt, cma_id);
754 		if (xprt)
755 			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
756 		break;
757 
758 	default:
759 		dprintk("svcrdma: Unexpected event on listening endpoint %p, "
760 			"event = %s (%d)\n", cma_id,
761 			rdma_event_msg(event->event), event->event);
762 		break;
763 	}
764 
765 	return ret;
766 }
767 
768 static int rdma_cma_handler(struct rdma_cm_id *cma_id,
769 			    struct rdma_cm_event *event)
770 {
771 	struct svc_xprt *xprt = cma_id->context;
772 	struct svcxprt_rdma *rdma =
773 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
774 	switch (event->event) {
775 	case RDMA_CM_EVENT_ESTABLISHED:
776 		/* Accept complete */
777 		svc_xprt_get(xprt);
778 		dprintk("svcrdma: Connection completed on DTO xprt=%p, "
779 			"cm_id=%p\n", xprt, cma_id);
780 		clear_bit(RDMAXPRT_CONN_PENDING, &rdma->sc_flags);
781 		svc_xprt_enqueue(xprt);
782 		break;
783 	case RDMA_CM_EVENT_DISCONNECTED:
784 		dprintk("svcrdma: Disconnect on DTO xprt=%p, cm_id=%p\n",
785 			xprt, cma_id);
786 		if (xprt) {
787 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
788 			svc_xprt_enqueue(xprt);
789 			svc_xprt_put(xprt);
790 		}
791 		break;
792 	case RDMA_CM_EVENT_DEVICE_REMOVAL:
793 		dprintk("svcrdma: Device removal cma_id=%p, xprt = %p, "
794 			"event = %s (%d)\n", cma_id, xprt,
795 			rdma_event_msg(event->event), event->event);
796 		if (xprt) {
797 			set_bit(XPT_CLOSE, &xprt->xpt_flags);
798 			svc_xprt_enqueue(xprt);
799 			svc_xprt_put(xprt);
800 		}
801 		break;
802 	default:
803 		dprintk("svcrdma: Unexpected event on DTO endpoint %p, "
804 			"event = %s (%d)\n", cma_id,
805 			rdma_event_msg(event->event), event->event);
806 		break;
807 	}
808 	return 0;
809 }
810 
811 /*
812  * Create a listening RDMA service endpoint.
813  */
814 static struct svc_xprt *svc_rdma_create(struct svc_serv *serv,
815 					struct net *net,
816 					struct sockaddr *sa, int salen,
817 					int flags)
818 {
819 	struct rdma_cm_id *listen_id;
820 	struct svcxprt_rdma *cma_xprt;
821 	int ret;
822 
823 	dprintk("svcrdma: Creating RDMA socket\n");
824 	if ((sa->sa_family != AF_INET) && (sa->sa_family != AF_INET6)) {
825 		dprintk("svcrdma: Address family %d is not supported.\n", sa->sa_family);
826 		return ERR_PTR(-EAFNOSUPPORT);
827 	}
828 	cma_xprt = rdma_create_xprt(serv, 1);
829 	if (!cma_xprt)
830 		return ERR_PTR(-ENOMEM);
831 
832 	listen_id = rdma_create_id(&init_net, rdma_listen_handler, cma_xprt,
833 				   RDMA_PS_TCP, IB_QPT_RC);
834 	if (IS_ERR(listen_id)) {
835 		ret = PTR_ERR(listen_id);
836 		dprintk("svcrdma: rdma_create_id failed = %d\n", ret);
837 		goto err0;
838 	}
839 
840 	/* Allow both IPv4 and IPv6 sockets to bind a single port
841 	 * at the same time.
842 	 */
843 #if IS_ENABLED(CONFIG_IPV6)
844 	ret = rdma_set_afonly(listen_id, 1);
845 	if (ret) {
846 		dprintk("svcrdma: rdma_set_afonly failed = %d\n", ret);
847 		goto err1;
848 	}
849 #endif
850 	ret = rdma_bind_addr(listen_id, sa);
851 	if (ret) {
852 		dprintk("svcrdma: rdma_bind_addr failed = %d\n", ret);
853 		goto err1;
854 	}
855 	cma_xprt->sc_cm_id = listen_id;
856 
857 	ret = rdma_listen(listen_id, RPCRDMA_LISTEN_BACKLOG);
858 	if (ret) {
859 		dprintk("svcrdma: rdma_listen failed = %d\n", ret);
860 		goto err1;
861 	}
862 
863 	/*
864 	 * We need to use the address from the cm_id in case the
865 	 * caller specified 0 for the port number.
866 	 */
867 	sa = (struct sockaddr *)&cma_xprt->sc_cm_id->route.addr.src_addr;
868 	svc_xprt_set_local(&cma_xprt->sc_xprt, sa, salen);
869 
870 	return &cma_xprt->sc_xprt;
871 
872  err1:
873 	rdma_destroy_id(listen_id);
874  err0:
875 	kfree(cma_xprt);
876 	return ERR_PTR(ret);
877 }
878 
879 static struct svc_rdma_fastreg_mr *rdma_alloc_frmr(struct svcxprt_rdma *xprt)
880 {
881 	struct ib_mr *mr;
882 	struct scatterlist *sg;
883 	struct svc_rdma_fastreg_mr *frmr;
884 	u32 num_sg;
885 
886 	frmr = kmalloc(sizeof(*frmr), GFP_KERNEL);
887 	if (!frmr)
888 		goto err;
889 
890 	num_sg = min_t(u32, RPCSVC_MAXPAGES, xprt->sc_frmr_pg_list_len);
891 	mr = ib_alloc_mr(xprt->sc_pd, IB_MR_TYPE_MEM_REG, num_sg);
892 	if (IS_ERR(mr))
893 		goto err_free_frmr;
894 
895 	sg = kcalloc(RPCSVC_MAXPAGES, sizeof(*sg), GFP_KERNEL);
896 	if (!sg)
897 		goto err_free_mr;
898 
899 	sg_init_table(sg, RPCSVC_MAXPAGES);
900 
901 	frmr->mr = mr;
902 	frmr->sg = sg;
903 	INIT_LIST_HEAD(&frmr->frmr_list);
904 	return frmr;
905 
906  err_free_mr:
907 	ib_dereg_mr(mr);
908  err_free_frmr:
909 	kfree(frmr);
910  err:
911 	return ERR_PTR(-ENOMEM);
912 }
913 
914 static void rdma_dealloc_frmr_q(struct svcxprt_rdma *xprt)
915 {
916 	struct svc_rdma_fastreg_mr *frmr;
917 
918 	while (!list_empty(&xprt->sc_frmr_q)) {
919 		frmr = list_entry(xprt->sc_frmr_q.next,
920 				  struct svc_rdma_fastreg_mr, frmr_list);
921 		list_del_init(&frmr->frmr_list);
922 		kfree(frmr->sg);
923 		ib_dereg_mr(frmr->mr);
924 		kfree(frmr);
925 	}
926 }
927 
928 struct svc_rdma_fastreg_mr *svc_rdma_get_frmr(struct svcxprt_rdma *rdma)
929 {
930 	struct svc_rdma_fastreg_mr *frmr = NULL;
931 
932 	spin_lock(&rdma->sc_frmr_q_lock);
933 	if (!list_empty(&rdma->sc_frmr_q)) {
934 		frmr = list_entry(rdma->sc_frmr_q.next,
935 				  struct svc_rdma_fastreg_mr, frmr_list);
936 		list_del_init(&frmr->frmr_list);
937 		frmr->sg_nents = 0;
938 	}
939 	spin_unlock(&rdma->sc_frmr_q_lock);
940 	if (frmr)
941 		return frmr;
942 
943 	return rdma_alloc_frmr(rdma);
944 }
945 
946 void svc_rdma_put_frmr(struct svcxprt_rdma *rdma,
947 		       struct svc_rdma_fastreg_mr *frmr)
948 {
949 	if (frmr) {
950 		ib_dma_unmap_sg(rdma->sc_cm_id->device,
951 				frmr->sg, frmr->sg_nents, frmr->direction);
952 		spin_lock(&rdma->sc_frmr_q_lock);
953 		WARN_ON_ONCE(!list_empty(&frmr->frmr_list));
954 		list_add(&frmr->frmr_list, &rdma->sc_frmr_q);
955 		spin_unlock(&rdma->sc_frmr_q_lock);
956 	}
957 }
958 
959 /*
960  * This is the xpo_recvfrom function for listening endpoints. Its
961  * purpose is to accept incoming connections. The CMA callback handler
962  * has already created a new transport and attached it to the new CMA
963  * ID.
964  *
965  * There is a queue of pending connections hung on the listening
966  * transport. This queue contains the new svc_xprt structure. This
967  * function takes svc_xprt structures off the accept_q and completes
968  * the connection.
969  */
970 static struct svc_xprt *svc_rdma_accept(struct svc_xprt *xprt)
971 {
972 	struct svcxprt_rdma *listen_rdma;
973 	struct svcxprt_rdma *newxprt = NULL;
974 	struct rdma_conn_param conn_param;
975 	struct rpcrdma_connect_private pmsg;
976 	struct ib_qp_init_attr qp_attr;
977 	struct ib_device *dev;
978 	struct sockaddr *sap;
979 	unsigned int i;
980 	int ret = 0;
981 
982 	listen_rdma = container_of(xprt, struct svcxprt_rdma, sc_xprt);
983 	clear_bit(XPT_CONN, &xprt->xpt_flags);
984 	/* Get the next entry off the accept list */
985 	spin_lock_bh(&listen_rdma->sc_lock);
986 	if (!list_empty(&listen_rdma->sc_accept_q)) {
987 		newxprt = list_entry(listen_rdma->sc_accept_q.next,
988 				     struct svcxprt_rdma, sc_accept_q);
989 		list_del_init(&newxprt->sc_accept_q);
990 	}
991 	if (!list_empty(&listen_rdma->sc_accept_q))
992 		set_bit(XPT_CONN, &listen_rdma->sc_xprt.xpt_flags);
993 	spin_unlock_bh(&listen_rdma->sc_lock);
994 	if (!newxprt)
995 		return NULL;
996 
997 	dprintk("svcrdma: newxprt from accept queue = %p, cm_id=%p\n",
998 		newxprt, newxprt->sc_cm_id);
999 
1000 	dev = newxprt->sc_cm_id->device;
1001 
1002 	/* Qualify the transport resource defaults with the
1003 	 * capabilities of this particular device */
1004 	newxprt->sc_max_sge = min((size_t)dev->attrs.max_sge,
1005 				  (size_t)RPCSVC_MAXPAGES);
1006 	newxprt->sc_max_sge_rd = min_t(size_t, dev->attrs.max_sge_rd,
1007 				       RPCSVC_MAXPAGES);
1008 	newxprt->sc_max_req_size = svcrdma_max_req_size;
1009 	newxprt->sc_max_requests = min_t(u32, dev->attrs.max_qp_wr,
1010 					 svcrdma_max_requests);
1011 	newxprt->sc_fc_credits = cpu_to_be32(newxprt->sc_max_requests);
1012 	newxprt->sc_max_bc_requests = min_t(u32, dev->attrs.max_qp_wr,
1013 					    svcrdma_max_bc_requests);
1014 	newxprt->sc_rq_depth = newxprt->sc_max_requests +
1015 			       newxprt->sc_max_bc_requests;
1016 	newxprt->sc_sq_depth = RPCRDMA_SQ_DEPTH_MULT * newxprt->sc_rq_depth;
1017 	atomic_set(&newxprt->sc_sq_avail, newxprt->sc_sq_depth);
1018 
1019 	if (!svc_rdma_prealloc_ctxts(newxprt))
1020 		goto errout;
1021 	if (!svc_rdma_prealloc_maps(newxprt))
1022 		goto errout;
1023 
1024 	/*
1025 	 * Limit ORD based on client limit, local device limit, and
1026 	 * configured svcrdma limit.
1027 	 */
1028 	newxprt->sc_ord = min_t(size_t, dev->attrs.max_qp_rd_atom, newxprt->sc_ord);
1029 	newxprt->sc_ord = min_t(size_t,	svcrdma_ord, newxprt->sc_ord);
1030 
1031 	newxprt->sc_pd = ib_alloc_pd(dev, 0);
1032 	if (IS_ERR(newxprt->sc_pd)) {
1033 		dprintk("svcrdma: error creating PD for connect request\n");
1034 		goto errout;
1035 	}
1036 	newxprt->sc_sq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_sq_depth,
1037 					0, IB_POLL_WORKQUEUE);
1038 	if (IS_ERR(newxprt->sc_sq_cq)) {
1039 		dprintk("svcrdma: error creating SQ CQ for connect request\n");
1040 		goto errout;
1041 	}
1042 	newxprt->sc_rq_cq = ib_alloc_cq(dev, newxprt, newxprt->sc_rq_depth,
1043 					0, IB_POLL_WORKQUEUE);
1044 	if (IS_ERR(newxprt->sc_rq_cq)) {
1045 		dprintk("svcrdma: error creating RQ CQ for connect request\n");
1046 		goto errout;
1047 	}
1048 
1049 	memset(&qp_attr, 0, sizeof qp_attr);
1050 	qp_attr.event_handler = qp_event_handler;
1051 	qp_attr.qp_context = &newxprt->sc_xprt;
1052 	qp_attr.cap.max_send_wr = newxprt->sc_sq_depth;
1053 	qp_attr.cap.max_recv_wr = newxprt->sc_rq_depth;
1054 	qp_attr.cap.max_send_sge = newxprt->sc_max_sge;
1055 	qp_attr.cap.max_recv_sge = newxprt->sc_max_sge;
1056 	qp_attr.sq_sig_type = IB_SIGNAL_REQ_WR;
1057 	qp_attr.qp_type = IB_QPT_RC;
1058 	qp_attr.send_cq = newxprt->sc_sq_cq;
1059 	qp_attr.recv_cq = newxprt->sc_rq_cq;
1060 	dprintk("svcrdma: newxprt->sc_cm_id=%p, newxprt->sc_pd=%p\n",
1061 		newxprt->sc_cm_id, newxprt->sc_pd);
1062 	dprintk("    cap.max_send_wr = %d, cap.max_recv_wr = %d\n",
1063 		qp_attr.cap.max_send_wr, qp_attr.cap.max_recv_wr);
1064 	dprintk("    cap.max_send_sge = %d, cap.max_recv_sge = %d\n",
1065 		qp_attr.cap.max_send_sge, qp_attr.cap.max_recv_sge);
1066 
1067 	ret = rdma_create_qp(newxprt->sc_cm_id, newxprt->sc_pd, &qp_attr);
1068 	if (ret) {
1069 		dprintk("svcrdma: failed to create QP, ret=%d\n", ret);
1070 		goto errout;
1071 	}
1072 	newxprt->sc_qp = newxprt->sc_cm_id->qp;
1073 
1074 	/*
1075 	 * Use the most secure set of MR resources based on the
1076 	 * transport type and available memory management features in
1077 	 * the device. Here's the table implemented below:
1078 	 *
1079 	 *		Fast	Global	DMA	Remote WR
1080 	 *		Reg	LKEY	MR	Access
1081 	 *		Sup'd	Sup'd	Needed	Needed
1082 	 *
1083 	 * IWARP	N	N	Y	Y
1084 	 *		N	Y	Y	Y
1085 	 *		Y	N	Y	N
1086 	 *		Y	Y	N	-
1087 	 *
1088 	 * IB		N	N	Y	N
1089 	 *		N	Y	N	-
1090 	 *		Y	N	Y	N
1091 	 *		Y	Y	N	-
1092 	 *
1093 	 * NB:	iWARP requires remote write access for the data sink
1094 	 *	of an RDMA_READ. IB does not.
1095 	 */
1096 	newxprt->sc_reader = rdma_read_chunk_lcl;
1097 	if (dev->attrs.device_cap_flags & IB_DEVICE_MEM_MGT_EXTENSIONS) {
1098 		newxprt->sc_frmr_pg_list_len =
1099 			dev->attrs.max_fast_reg_page_list_len;
1100 		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_FAST_REG;
1101 		newxprt->sc_reader = rdma_read_chunk_frmr;
1102 	} else
1103 		newxprt->sc_snd_w_inv = false;
1104 
1105 	/*
1106 	 * Determine if a DMA MR is required and if so, what privs are required
1107 	 */
1108 	if (!rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num) &&
1109 	    !rdma_ib_or_roce(dev, newxprt->sc_cm_id->port_num))
1110 		goto errout;
1111 
1112 	if (rdma_protocol_iwarp(dev, newxprt->sc_cm_id->port_num))
1113 		newxprt->sc_dev_caps |= SVCRDMA_DEVCAP_READ_W_INV;
1114 
1115 	/* Post receive buffers */
1116 	for (i = 0; i < newxprt->sc_max_requests; i++) {
1117 		ret = svc_rdma_post_recv(newxprt, GFP_KERNEL);
1118 		if (ret) {
1119 			dprintk("svcrdma: failure posting receive buffers\n");
1120 			goto errout;
1121 		}
1122 	}
1123 
1124 	/* Swap out the handler */
1125 	newxprt->sc_cm_id->event_handler = rdma_cma_handler;
1126 
1127 	/* Construct RDMA-CM private message */
1128 	pmsg.cp_magic = rpcrdma_cmp_magic;
1129 	pmsg.cp_version = RPCRDMA_CMP_VERSION;
1130 	pmsg.cp_flags = 0;
1131 	pmsg.cp_send_size = pmsg.cp_recv_size =
1132 		rpcrdma_encode_buffer_size(newxprt->sc_max_req_size);
1133 
1134 	/* Accept Connection */
1135 	set_bit(RDMAXPRT_CONN_PENDING, &newxprt->sc_flags);
1136 	memset(&conn_param, 0, sizeof conn_param);
1137 	conn_param.responder_resources = 0;
1138 	conn_param.initiator_depth = newxprt->sc_ord;
1139 	conn_param.private_data = &pmsg;
1140 	conn_param.private_data_len = sizeof(pmsg);
1141 	ret = rdma_accept(newxprt->sc_cm_id, &conn_param);
1142 	if (ret) {
1143 		dprintk("svcrdma: failed to accept new connection, ret=%d\n",
1144 		       ret);
1145 		goto errout;
1146 	}
1147 
1148 	dprintk("svcrdma: new connection %p accepted:\n", newxprt);
1149 	sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.src_addr;
1150 	dprintk("    local address   : %pIS:%u\n", sap, rpc_get_port(sap));
1151 	sap = (struct sockaddr *)&newxprt->sc_cm_id->route.addr.dst_addr;
1152 	dprintk("    remote address  : %pIS:%u\n", sap, rpc_get_port(sap));
1153 	dprintk("    max_sge         : %d\n", newxprt->sc_max_sge);
1154 	dprintk("    max_sge_rd      : %d\n", newxprt->sc_max_sge_rd);
1155 	dprintk("    sq_depth        : %d\n", newxprt->sc_sq_depth);
1156 	dprintk("    max_requests    : %d\n", newxprt->sc_max_requests);
1157 	dprintk("    ord             : %d\n", newxprt->sc_ord);
1158 
1159 	return &newxprt->sc_xprt;
1160 
1161  errout:
1162 	dprintk("svcrdma: failure accepting new connection rc=%d.\n", ret);
1163 	/* Take a reference in case the DTO handler runs */
1164 	svc_xprt_get(&newxprt->sc_xprt);
1165 	if (newxprt->sc_qp && !IS_ERR(newxprt->sc_qp))
1166 		ib_destroy_qp(newxprt->sc_qp);
1167 	rdma_destroy_id(newxprt->sc_cm_id);
1168 	/* This call to put will destroy the transport */
1169 	svc_xprt_put(&newxprt->sc_xprt);
1170 	return NULL;
1171 }
1172 
1173 static void svc_rdma_release_rqst(struct svc_rqst *rqstp)
1174 {
1175 }
1176 
1177 /*
1178  * When connected, an svc_xprt has at least two references:
1179  *
1180  * - A reference held by the cm_id between the ESTABLISHED and
1181  *   DISCONNECTED events. If the remote peer disconnected first, this
1182  *   reference could be gone.
1183  *
1184  * - A reference held by the svc_recv code that called this function
1185  *   as part of close processing.
1186  *
1187  * At a minimum one references should still be held.
1188  */
1189 static void svc_rdma_detach(struct svc_xprt *xprt)
1190 {
1191 	struct svcxprt_rdma *rdma =
1192 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
1193 	dprintk("svc: svc_rdma_detach(%p)\n", xprt);
1194 
1195 	/* Disconnect and flush posted WQE */
1196 	rdma_disconnect(rdma->sc_cm_id);
1197 }
1198 
1199 static void __svc_rdma_free(struct work_struct *work)
1200 {
1201 	struct svcxprt_rdma *rdma =
1202 		container_of(work, struct svcxprt_rdma, sc_work);
1203 	struct svc_xprt *xprt = &rdma->sc_xprt;
1204 
1205 	dprintk("svcrdma: %s(%p)\n", __func__, rdma);
1206 
1207 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
1208 		ib_drain_qp(rdma->sc_qp);
1209 
1210 	/* We should only be called from kref_put */
1211 	if (kref_read(&xprt->xpt_ref) != 0)
1212 		pr_err("svcrdma: sc_xprt still in use? (%d)\n",
1213 		       kref_read(&xprt->xpt_ref));
1214 
1215 	/*
1216 	 * Destroy queued, but not processed read completions. Note
1217 	 * that this cleanup has to be done before destroying the
1218 	 * cm_id because the device ptr is needed to unmap the dma in
1219 	 * svc_rdma_put_context.
1220 	 */
1221 	while (!list_empty(&rdma->sc_read_complete_q)) {
1222 		struct svc_rdma_op_ctxt *ctxt;
1223 		ctxt = list_first_entry(&rdma->sc_read_complete_q,
1224 					struct svc_rdma_op_ctxt, list);
1225 		list_del(&ctxt->list);
1226 		svc_rdma_put_context(ctxt, 1);
1227 	}
1228 
1229 	/* Destroy queued, but not processed recv completions */
1230 	while (!list_empty(&rdma->sc_rq_dto_q)) {
1231 		struct svc_rdma_op_ctxt *ctxt;
1232 		ctxt = list_first_entry(&rdma->sc_rq_dto_q,
1233 					struct svc_rdma_op_ctxt, list);
1234 		list_del(&ctxt->list);
1235 		svc_rdma_put_context(ctxt, 1);
1236 	}
1237 
1238 	/* Warn if we leaked a resource or under-referenced */
1239 	if (rdma->sc_ctxt_used != 0)
1240 		pr_err("svcrdma: ctxt still in use? (%d)\n",
1241 		       rdma->sc_ctxt_used);
1242 
1243 	/* Final put of backchannel client transport */
1244 	if (xprt->xpt_bc_xprt) {
1245 		xprt_put(xprt->xpt_bc_xprt);
1246 		xprt->xpt_bc_xprt = NULL;
1247 	}
1248 
1249 	rdma_dealloc_frmr_q(rdma);
1250 	svc_rdma_destroy_ctxts(rdma);
1251 	svc_rdma_destroy_maps(rdma);
1252 
1253 	/* Destroy the QP if present (not a listener) */
1254 	if (rdma->sc_qp && !IS_ERR(rdma->sc_qp))
1255 		ib_destroy_qp(rdma->sc_qp);
1256 
1257 	if (rdma->sc_sq_cq && !IS_ERR(rdma->sc_sq_cq))
1258 		ib_free_cq(rdma->sc_sq_cq);
1259 
1260 	if (rdma->sc_rq_cq && !IS_ERR(rdma->sc_rq_cq))
1261 		ib_free_cq(rdma->sc_rq_cq);
1262 
1263 	if (rdma->sc_pd && !IS_ERR(rdma->sc_pd))
1264 		ib_dealloc_pd(rdma->sc_pd);
1265 
1266 	/* Destroy the CM ID */
1267 	rdma_destroy_id(rdma->sc_cm_id);
1268 
1269 	kfree(rdma);
1270 }
1271 
1272 static void svc_rdma_free(struct svc_xprt *xprt)
1273 {
1274 	struct svcxprt_rdma *rdma =
1275 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
1276 	INIT_WORK(&rdma->sc_work, __svc_rdma_free);
1277 	queue_work(svc_rdma_wq, &rdma->sc_work);
1278 }
1279 
1280 static int svc_rdma_has_wspace(struct svc_xprt *xprt)
1281 {
1282 	struct svcxprt_rdma *rdma =
1283 		container_of(xprt, struct svcxprt_rdma, sc_xprt);
1284 
1285 	/*
1286 	 * If there are already waiters on the SQ,
1287 	 * return false.
1288 	 */
1289 	if (waitqueue_active(&rdma->sc_send_wait))
1290 		return 0;
1291 
1292 	/* Otherwise return true. */
1293 	return 1;
1294 }
1295 
1296 static int svc_rdma_secure_port(struct svc_rqst *rqstp)
1297 {
1298 	return 1;
1299 }
1300 
1301 static void svc_rdma_kill_temp_xprt(struct svc_xprt *xprt)
1302 {
1303 }
1304 
1305 int svc_rdma_send(struct svcxprt_rdma *xprt, struct ib_send_wr *wr)
1306 {
1307 	struct ib_send_wr *bad_wr, *n_wr;
1308 	int wr_count;
1309 	int i;
1310 	int ret;
1311 
1312 	if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1313 		return -ENOTCONN;
1314 
1315 	wr_count = 1;
1316 	for (n_wr = wr->next; n_wr; n_wr = n_wr->next)
1317 		wr_count++;
1318 
1319 	/* If the SQ is full, wait until an SQ entry is available */
1320 	while (1) {
1321 		if ((atomic_sub_return(wr_count, &xprt->sc_sq_avail) < 0)) {
1322 			atomic_inc(&rdma_stat_sq_starve);
1323 
1324 			/* Wait until SQ WR available if SQ still full */
1325 			atomic_add(wr_count, &xprt->sc_sq_avail);
1326 			wait_event(xprt->sc_send_wait,
1327 				   atomic_read(&xprt->sc_sq_avail) > wr_count);
1328 			if (test_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags))
1329 				return -ENOTCONN;
1330 			continue;
1331 		}
1332 		/* Take a transport ref for each WR posted */
1333 		for (i = 0; i < wr_count; i++)
1334 			svc_xprt_get(&xprt->sc_xprt);
1335 
1336 		/* Bump used SQ WR count and post */
1337 		ret = ib_post_send(xprt->sc_qp, wr, &bad_wr);
1338 		if (ret) {
1339 			set_bit(XPT_CLOSE, &xprt->sc_xprt.xpt_flags);
1340 			for (i = 0; i < wr_count; i ++)
1341 				svc_xprt_put(&xprt->sc_xprt);
1342 			dprintk("svcrdma: failed to post SQ WR rc=%d\n", ret);
1343 			dprintk("    sc_sq_avail=%d, sc_sq_depth=%d\n",
1344 				atomic_read(&xprt->sc_sq_avail),
1345 				xprt->sc_sq_depth);
1346 			wake_up(&xprt->sc_send_wait);
1347 		}
1348 		break;
1349 	}
1350 	return ret;
1351 }
1352