xref: /openbmc/linux/drivers/infiniband/sw/siw/siw_cm.c (revision a20eefae)
1 // SPDX-License-Identifier: GPL-2.0 or BSD-3-Clause
2 
3 /* Authors: Bernard Metzler <bmt@zurich.ibm.com> */
4 /*          Fredy Neeser */
5 /*          Greg Joyce <greg@opengridcomputing.com> */
6 /* Copyright (c) 2008-2019, IBM Corporation */
7 /* Copyright (c) 2017, Open Grid Computing, Inc. */
8 
9 #include <linux/errno.h>
10 #include <linux/types.h>
11 #include <linux/net.h>
12 #include <linux/inetdevice.h>
13 #include <net/addrconf.h>
14 #include <linux/workqueue.h>
15 #include <net/sock.h>
16 #include <net/tcp.h>
17 #include <linux/inet.h>
18 #include <linux/tcp.h>
19 
20 #include <rdma/iw_cm.h>
21 #include <rdma/ib_verbs.h>
22 #include <rdma/ib_user_verbs.h>
23 
24 #include "siw.h"
25 #include "siw_cm.h"
26 
27 /*
28  * Set to any combination of
29  * MPA_V2_RDMA_NO_RTR, MPA_V2_RDMA_READ_RTR, MPA_V2_RDMA_WRITE_RTR
30  */
31 static __be16 rtr_type = MPA_V2_RDMA_READ_RTR | MPA_V2_RDMA_WRITE_RTR;
32 static const bool relaxed_ird_negotiation = 1;
33 
34 static void siw_cm_llp_state_change(struct sock *s);
35 static void siw_cm_llp_data_ready(struct sock *s);
36 static void siw_cm_llp_write_space(struct sock *s);
37 static void siw_cm_llp_error_report(struct sock *s);
38 static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
39 			 int status);
40 
41 static void siw_sk_assign_cm_upcalls(struct sock *sk)
42 {
43 	write_lock_bh(&sk->sk_callback_lock);
44 	sk->sk_state_change = siw_cm_llp_state_change;
45 	sk->sk_data_ready = siw_cm_llp_data_ready;
46 	sk->sk_write_space = siw_cm_llp_write_space;
47 	sk->sk_error_report = siw_cm_llp_error_report;
48 	write_unlock_bh(&sk->sk_callback_lock);
49 }
50 
51 static void siw_sk_save_upcalls(struct sock *sk)
52 {
53 	struct siw_cep *cep = sk_to_cep(sk);
54 
55 	write_lock_bh(&sk->sk_callback_lock);
56 	cep->sk_state_change = sk->sk_state_change;
57 	cep->sk_data_ready = sk->sk_data_ready;
58 	cep->sk_write_space = sk->sk_write_space;
59 	cep->sk_error_report = sk->sk_error_report;
60 	write_unlock_bh(&sk->sk_callback_lock);
61 }
62 
63 static void siw_sk_restore_upcalls(struct sock *sk, struct siw_cep *cep)
64 {
65 	sk->sk_state_change = cep->sk_state_change;
66 	sk->sk_data_ready = cep->sk_data_ready;
67 	sk->sk_write_space = cep->sk_write_space;
68 	sk->sk_error_report = cep->sk_error_report;
69 	sk->sk_user_data = NULL;
70 }
71 
72 static void siw_qp_socket_assoc(struct siw_cep *cep, struct siw_qp *qp)
73 {
74 	struct socket *s = cep->sock;
75 	struct sock *sk = s->sk;
76 
77 	write_lock_bh(&sk->sk_callback_lock);
78 
79 	qp->attrs.sk = s;
80 	sk->sk_data_ready = siw_qp_llp_data_ready;
81 	sk->sk_write_space = siw_qp_llp_write_space;
82 
83 	write_unlock_bh(&sk->sk_callback_lock);
84 }
85 
86 static void siw_socket_disassoc(struct socket *s)
87 {
88 	struct sock *sk = s->sk;
89 	struct siw_cep *cep;
90 
91 	if (sk) {
92 		write_lock_bh(&sk->sk_callback_lock);
93 		cep = sk_to_cep(sk);
94 		if (cep) {
95 			siw_sk_restore_upcalls(sk, cep);
96 			siw_cep_put(cep);
97 		} else {
98 			pr_warn("siw: cannot restore sk callbacks: no ep\n");
99 		}
100 		write_unlock_bh(&sk->sk_callback_lock);
101 	} else {
102 		pr_warn("siw: cannot restore sk callbacks: no sk\n");
103 	}
104 }
105 
106 static void siw_rtr_data_ready(struct sock *sk)
107 {
108 	struct siw_cep *cep;
109 	struct siw_qp *qp = NULL;
110 	read_descriptor_t rd_desc;
111 
112 	read_lock(&sk->sk_callback_lock);
113 
114 	cep = sk_to_cep(sk);
115 	if (!cep) {
116 		WARN(1, "No connection endpoint\n");
117 		goto out;
118 	}
119 	qp = sk_to_qp(sk);
120 
121 	memset(&rd_desc, 0, sizeof(rd_desc));
122 	rd_desc.arg.data = qp;
123 	rd_desc.count = 1;
124 
125 	tcp_read_sock(sk, &rd_desc, siw_tcp_rx_data);
126 	/*
127 	 * Check if first frame was successfully processed.
128 	 * Signal connection full establishment if yes.
129 	 * Failed data processing would have already scheduled
130 	 * connection drop.
131 	 */
132 	if (!qp->rx_stream.rx_suspend)
133 		siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
134 out:
135 	read_unlock(&sk->sk_callback_lock);
136 	if (qp)
137 		siw_qp_socket_assoc(cep, qp);
138 }
139 
140 static void siw_sk_assign_rtr_upcalls(struct siw_cep *cep)
141 {
142 	struct sock *sk = cep->sock->sk;
143 
144 	write_lock_bh(&sk->sk_callback_lock);
145 	sk->sk_data_ready = siw_rtr_data_ready;
146 	sk->sk_write_space = siw_qp_llp_write_space;
147 	write_unlock_bh(&sk->sk_callback_lock);
148 }
149 
150 static void siw_cep_socket_assoc(struct siw_cep *cep, struct socket *s)
151 {
152 	cep->sock = s;
153 	siw_cep_get(cep);
154 	s->sk->sk_user_data = cep;
155 
156 	siw_sk_save_upcalls(s->sk);
157 	siw_sk_assign_cm_upcalls(s->sk);
158 }
159 
160 static struct siw_cep *siw_cep_alloc(struct siw_device *sdev)
161 {
162 	struct siw_cep *cep = kzalloc(sizeof(*cep), GFP_KERNEL);
163 	unsigned long flags;
164 
165 	if (!cep)
166 		return NULL;
167 
168 	INIT_LIST_HEAD(&cep->listenq);
169 	INIT_LIST_HEAD(&cep->devq);
170 	INIT_LIST_HEAD(&cep->work_freelist);
171 
172 	kref_init(&cep->ref);
173 	cep->state = SIW_EPSTATE_IDLE;
174 	init_waitqueue_head(&cep->waitq);
175 	spin_lock_init(&cep->lock);
176 	cep->sdev = sdev;
177 	cep->enhanced_rdma_conn_est = false;
178 
179 	spin_lock_irqsave(&sdev->lock, flags);
180 	list_add_tail(&cep->devq, &sdev->cep_list);
181 	spin_unlock_irqrestore(&sdev->lock, flags);
182 
183 	siw_dbg_cep(cep, "new endpoint\n");
184 	return cep;
185 }
186 
187 static void siw_cm_free_work(struct siw_cep *cep)
188 {
189 	struct list_head *w, *tmp;
190 	struct siw_cm_work *work;
191 
192 	list_for_each_safe(w, tmp, &cep->work_freelist) {
193 		work = list_entry(w, struct siw_cm_work, list);
194 		list_del(&work->list);
195 		kfree(work);
196 	}
197 }
198 
199 static void siw_cancel_mpatimer(struct siw_cep *cep)
200 {
201 	spin_lock_bh(&cep->lock);
202 	if (cep->mpa_timer) {
203 		if (cancel_delayed_work(&cep->mpa_timer->work)) {
204 			siw_cep_put(cep);
205 			kfree(cep->mpa_timer); /* not needed again */
206 		}
207 		cep->mpa_timer = NULL;
208 	}
209 	spin_unlock_bh(&cep->lock);
210 }
211 
212 static void siw_put_work(struct siw_cm_work *work)
213 {
214 	INIT_LIST_HEAD(&work->list);
215 	spin_lock_bh(&work->cep->lock);
216 	list_add(&work->list, &work->cep->work_freelist);
217 	spin_unlock_bh(&work->cep->lock);
218 }
219 
220 static void siw_cep_set_inuse(struct siw_cep *cep)
221 {
222 	unsigned long flags;
223 	int rv;
224 retry:
225 	spin_lock_irqsave(&cep->lock, flags);
226 
227 	if (cep->in_use) {
228 		spin_unlock_irqrestore(&cep->lock, flags);
229 		rv = wait_event_interruptible(cep->waitq, !cep->in_use);
230 		if (signal_pending(current))
231 			flush_signals(current);
232 		goto retry;
233 	} else {
234 		cep->in_use = 1;
235 		spin_unlock_irqrestore(&cep->lock, flags);
236 	}
237 }
238 
239 static void siw_cep_set_free(struct siw_cep *cep)
240 {
241 	unsigned long flags;
242 
243 	spin_lock_irqsave(&cep->lock, flags);
244 	cep->in_use = 0;
245 	spin_unlock_irqrestore(&cep->lock, flags);
246 
247 	wake_up(&cep->waitq);
248 }
249 
250 static void __siw_cep_dealloc(struct kref *ref)
251 {
252 	struct siw_cep *cep = container_of(ref, struct siw_cep, ref);
253 	struct siw_device *sdev = cep->sdev;
254 	unsigned long flags;
255 
256 	WARN_ON(cep->listen_cep);
257 
258 	/* kfree(NULL) is safe */
259 	kfree(cep->mpa.pdata);
260 	spin_lock_bh(&cep->lock);
261 	if (!list_empty(&cep->work_freelist))
262 		siw_cm_free_work(cep);
263 	spin_unlock_bh(&cep->lock);
264 
265 	spin_lock_irqsave(&sdev->lock, flags);
266 	list_del(&cep->devq);
267 	spin_unlock_irqrestore(&sdev->lock, flags);
268 
269 	siw_dbg_cep(cep, "free endpoint\n");
270 	kfree(cep);
271 }
272 
273 static struct siw_cm_work *siw_get_work(struct siw_cep *cep)
274 {
275 	struct siw_cm_work *work = NULL;
276 
277 	spin_lock_bh(&cep->lock);
278 	if (!list_empty(&cep->work_freelist)) {
279 		work = list_entry(cep->work_freelist.next, struct siw_cm_work,
280 				  list);
281 		list_del_init(&work->list);
282 	}
283 	spin_unlock_bh(&cep->lock);
284 	return work;
285 }
286 
287 static int siw_cm_alloc_work(struct siw_cep *cep, int num)
288 {
289 	struct siw_cm_work *work;
290 
291 	while (num--) {
292 		work = kmalloc(sizeof(*work), GFP_KERNEL);
293 		if (!work) {
294 			if (!(list_empty(&cep->work_freelist)))
295 				siw_cm_free_work(cep);
296 			return -ENOMEM;
297 		}
298 		work->cep = cep;
299 		INIT_LIST_HEAD(&work->list);
300 		list_add(&work->list, &cep->work_freelist);
301 	}
302 	return 0;
303 }
304 
305 /*
306  * siw_cm_upcall()
307  *
308  * Upcall to IWCM to inform about async connection events
309  */
310 static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
311 			 int status)
312 {
313 	struct iw_cm_event event;
314 	struct iw_cm_id *id;
315 
316 	memset(&event, 0, sizeof(event));
317 	event.status = status;
318 	event.event = reason;
319 
320 	if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
321 		event.provider_data = cep;
322 		id = cep->listen_cep->cm_id;
323 	} else {
324 		id = cep->cm_id;
325 	}
326 	/* Signal IRD and ORD */
327 	if (reason == IW_CM_EVENT_ESTABLISHED ||
328 	    reason == IW_CM_EVENT_CONNECT_REPLY) {
329 		/* Signal negotiated IRD/ORD values we will use */
330 		event.ird = cep->ird;
331 		event.ord = cep->ord;
332 	} else if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
333 		event.ird = cep->ord;
334 		event.ord = cep->ird;
335 	}
336 	/* Signal private data and address information */
337 	if (reason == IW_CM_EVENT_CONNECT_REQUEST ||
338 	    reason == IW_CM_EVENT_CONNECT_REPLY) {
339 		u16 pd_len = be16_to_cpu(cep->mpa.hdr.params.pd_len);
340 
341 		if (pd_len) {
342 			/*
343 			 * hand over MPA private data
344 			 */
345 			event.private_data_len = pd_len;
346 			event.private_data = cep->mpa.pdata;
347 
348 			/* Hide MPA V2 IRD/ORD control */
349 			if (cep->enhanced_rdma_conn_est) {
350 				event.private_data_len -=
351 					sizeof(struct mpa_v2_data);
352 				event.private_data +=
353 					sizeof(struct mpa_v2_data);
354 			}
355 		}
356 		getname_local(cep->sock, &event.local_addr);
357 		getname_peer(cep->sock, &event.remote_addr);
358 	}
359 	siw_dbg_cep(cep, "[QP %u]: id 0x%p, reason=%d, status=%d\n",
360 		    cep->qp ? qp_id(cep->qp) : -1, id, reason, status);
361 
362 	return id->event_handler(id, &event);
363 }
364 
365 /*
366  * siw_qp_cm_drop()
367  *
368  * Drops established LLP connection if present and not already
369  * scheduled for dropping. Called from user context, SQ workqueue
370  * or receive IRQ. Caller signals if socket can be immediately
371  * closed (basically, if not in IRQ).
372  */
373 void siw_qp_cm_drop(struct siw_qp *qp, int schedule)
374 {
375 	struct siw_cep *cep = qp->cep;
376 
377 	qp->rx_stream.rx_suspend = 1;
378 	qp->tx_ctx.tx_suspend = 1;
379 
380 	if (!qp->cep)
381 		return;
382 
383 	if (schedule) {
384 		siw_cm_queue_work(cep, SIW_CM_WORK_CLOSE_LLP);
385 	} else {
386 		siw_cep_set_inuse(cep);
387 
388 		if (cep->state == SIW_EPSTATE_CLOSED) {
389 			siw_dbg_cep(cep, "already closed\n");
390 			goto out;
391 		}
392 		siw_dbg_cep(cep, "immediate close, state %d\n", cep->state);
393 
394 		if (qp->term_info.valid)
395 			siw_send_terminate(qp);
396 
397 		if (cep->cm_id) {
398 			switch (cep->state) {
399 			case SIW_EPSTATE_AWAIT_MPAREP:
400 				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
401 					      -EINVAL);
402 				break;
403 
404 			case SIW_EPSTATE_RDMA_MODE:
405 				siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
406 				break;
407 
408 			case SIW_EPSTATE_IDLE:
409 			case SIW_EPSTATE_LISTENING:
410 			case SIW_EPSTATE_CONNECTING:
411 			case SIW_EPSTATE_AWAIT_MPAREQ:
412 			case SIW_EPSTATE_RECVD_MPAREQ:
413 			case SIW_EPSTATE_CLOSED:
414 			default:
415 				break;
416 			}
417 			cep->cm_id->rem_ref(cep->cm_id);
418 			cep->cm_id = NULL;
419 			siw_cep_put(cep);
420 		}
421 		cep->state = SIW_EPSTATE_CLOSED;
422 
423 		if (cep->sock) {
424 			siw_socket_disassoc(cep->sock);
425 			/*
426 			 * Immediately close socket
427 			 */
428 			sock_release(cep->sock);
429 			cep->sock = NULL;
430 		}
431 		if (cep->qp) {
432 			cep->qp = NULL;
433 			siw_qp_put(qp);
434 		}
435 out:
436 		siw_cep_set_free(cep);
437 	}
438 }
439 
440 void siw_cep_put(struct siw_cep *cep)
441 {
442 	WARN_ON(kref_read(&cep->ref) < 1);
443 	kref_put(&cep->ref, __siw_cep_dealloc);
444 }
445 
446 void siw_cep_get(struct siw_cep *cep)
447 {
448 	kref_get(&cep->ref);
449 }
450 
451 /*
452  * Expects params->pd_len in host byte order
453  */
454 static int siw_send_mpareqrep(struct siw_cep *cep, const void *pdata, u8 pd_len)
455 {
456 	struct socket *s = cep->sock;
457 	struct mpa_rr *rr = &cep->mpa.hdr;
458 	struct kvec iov[3];
459 	struct msghdr msg;
460 	int rv;
461 	int iovec_num = 0;
462 	int mpa_len;
463 
464 	memset(&msg, 0, sizeof(msg));
465 
466 	iov[iovec_num].iov_base = rr;
467 	iov[iovec_num].iov_len = sizeof(*rr);
468 	mpa_len = sizeof(*rr);
469 
470 	if (cep->enhanced_rdma_conn_est) {
471 		iovec_num++;
472 		iov[iovec_num].iov_base = &cep->mpa.v2_ctrl;
473 		iov[iovec_num].iov_len = sizeof(cep->mpa.v2_ctrl);
474 		mpa_len += sizeof(cep->mpa.v2_ctrl);
475 	}
476 	if (pd_len) {
477 		iovec_num++;
478 		iov[iovec_num].iov_base = (char *)pdata;
479 		iov[iovec_num].iov_len = pd_len;
480 		mpa_len += pd_len;
481 	}
482 	if (cep->enhanced_rdma_conn_est)
483 		pd_len += sizeof(cep->mpa.v2_ctrl);
484 
485 	rr->params.pd_len = cpu_to_be16(pd_len);
486 
487 	rv = kernel_sendmsg(s, &msg, iov, iovec_num + 1, mpa_len);
488 
489 	return rv < 0 ? rv : 0;
490 }
491 
492 /*
493  * Receive MPA Request/Reply header.
494  *
495  * Returns 0 if complete MPA Request/Reply header including
496  * eventual private data was received. Returns -EAGAIN if
497  * header was partially received or negative error code otherwise.
498  *
499  * Context: May be called in process context only
500  */
501 static int siw_recv_mpa_rr(struct siw_cep *cep)
502 {
503 	struct mpa_rr *hdr = &cep->mpa.hdr;
504 	struct socket *s = cep->sock;
505 	u16 pd_len;
506 	int rcvd, to_rcv;
507 
508 	if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr)) {
509 		rcvd = ksock_recv(s, (char *)hdr + cep->mpa.bytes_rcvd,
510 				  sizeof(struct mpa_rr) - cep->mpa.bytes_rcvd,
511 				  0);
512 		if (rcvd <= 0)
513 			return -ECONNABORTED;
514 
515 		cep->mpa.bytes_rcvd += rcvd;
516 
517 		if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr))
518 			return -EAGAIN;
519 
520 		if (be16_to_cpu(hdr->params.pd_len) > MPA_MAX_PRIVDATA)
521 			return -EPROTO;
522 	}
523 	pd_len = be16_to_cpu(hdr->params.pd_len);
524 
525 	/*
526 	 * At least the MPA Request/Reply header (frame not including
527 	 * private data) has been received.
528 	 * Receive (or continue receiving) any private data.
529 	 */
530 	to_rcv = pd_len - (cep->mpa.bytes_rcvd - sizeof(struct mpa_rr));
531 
532 	if (!to_rcv) {
533 		/*
534 		 * We must have hdr->params.pd_len == 0 and thus received a
535 		 * complete MPA Request/Reply frame.
536 		 * Check against peer protocol violation.
537 		 */
538 		u32 word;
539 
540 		rcvd = ksock_recv(s, (char *)&word, sizeof(word), MSG_DONTWAIT);
541 		if (rcvd == -EAGAIN)
542 			return 0;
543 
544 		if (rcvd == 0) {
545 			siw_dbg_cep(cep, "peer EOF\n");
546 			return -EPIPE;
547 		}
548 		if (rcvd < 0) {
549 			siw_dbg_cep(cep, "error: %d\n", rcvd);
550 			return rcvd;
551 		}
552 		siw_dbg_cep(cep, "peer sent extra data: %d\n", rcvd);
553 
554 		return -EPROTO;
555 	}
556 
557 	/*
558 	 * At this point, we must have hdr->params.pd_len != 0.
559 	 * A private data buffer gets allocated if hdr->params.pd_len != 0.
560 	 */
561 	if (!cep->mpa.pdata) {
562 		cep->mpa.pdata = kmalloc(pd_len + 4, GFP_KERNEL);
563 		if (!cep->mpa.pdata)
564 			return -ENOMEM;
565 	}
566 	rcvd = ksock_recv(
567 		s, cep->mpa.pdata + cep->mpa.bytes_rcvd - sizeof(struct mpa_rr),
568 		to_rcv + 4, MSG_DONTWAIT);
569 
570 	if (rcvd < 0)
571 		return rcvd;
572 
573 	if (rcvd > to_rcv)
574 		return -EPROTO;
575 
576 	cep->mpa.bytes_rcvd += rcvd;
577 
578 	if (to_rcv == rcvd) {
579 		siw_dbg_cep(cep, "%d bytes private data received\n", pd_len);
580 		return 0;
581 	}
582 	return -EAGAIN;
583 }
584 
585 /*
586  * siw_proc_mpareq()
587  *
588  * Read MPA Request from socket and signal new connection to IWCM
589  * if success. Caller must hold lock on corresponding listening CEP.
590  */
591 static int siw_proc_mpareq(struct siw_cep *cep)
592 {
593 	struct mpa_rr *req;
594 	int version, rv;
595 	u16 pd_len;
596 
597 	rv = siw_recv_mpa_rr(cep);
598 	if (rv)
599 		return rv;
600 
601 	req = &cep->mpa.hdr;
602 
603 	version = __mpa_rr_revision(req->params.bits);
604 	pd_len = be16_to_cpu(req->params.pd_len);
605 
606 	if (version > MPA_REVISION_2)
607 		/* allow for 0, 1, and 2 only */
608 		return -EPROTO;
609 
610 	if (memcmp(req->key, MPA_KEY_REQ, 16))
611 		return -EPROTO;
612 
613 	/* Prepare for sending MPA reply */
614 	memcpy(req->key, MPA_KEY_REP, 16);
615 
616 	if (version == MPA_REVISION_2 &&
617 	    (req->params.bits & MPA_RR_FLAG_ENHANCED)) {
618 		/*
619 		 * MPA version 2 must signal IRD/ORD values and P2P mode
620 		 * in private data if header flag MPA_RR_FLAG_ENHANCED
621 		 * is set.
622 		 */
623 		if (pd_len < sizeof(struct mpa_v2_data))
624 			goto reject_conn;
625 
626 		cep->enhanced_rdma_conn_est = true;
627 	}
628 
629 	/* MPA Markers: currently not supported. Marker TX to be added. */
630 	if (req->params.bits & MPA_RR_FLAG_MARKERS)
631 		goto reject_conn;
632 
633 	if (req->params.bits & MPA_RR_FLAG_CRC) {
634 		/*
635 		 * RFC 5044, page 27: CRC MUST be used if peer requests it.
636 		 * siw specific: 'mpa_crc_strict' parameter to reject
637 		 * connection with CRC if local CRC off enforced by
638 		 * 'mpa_crc_strict' module parameter.
639 		 */
640 		if (!mpa_crc_required && mpa_crc_strict)
641 			goto reject_conn;
642 
643 		/* Enable CRC if requested by module parameter */
644 		if (mpa_crc_required)
645 			req->params.bits |= MPA_RR_FLAG_CRC;
646 	}
647 	if (cep->enhanced_rdma_conn_est) {
648 		struct mpa_v2_data *v2 = (struct mpa_v2_data *)cep->mpa.pdata;
649 
650 		/*
651 		 * Peer requested ORD becomes requested local IRD,
652 		 * peer requested IRD becomes requested local ORD.
653 		 * IRD and ORD get limited by global maximum values.
654 		 */
655 		cep->ord = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
656 		cep->ord = min(cep->ord, SIW_MAX_ORD_QP);
657 		cep->ird = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
658 		cep->ird = min(cep->ird, SIW_MAX_IRD_QP);
659 
660 		/* May get overwritten by locally negotiated values */
661 		cep->mpa.v2_ctrl.ird = htons(cep->ird);
662 		cep->mpa.v2_ctrl.ord = htons(cep->ord);
663 
664 		/*
665 		 * Support for peer sent zero length Write or Read to
666 		 * let local side enter RTS. Writes are preferred.
667 		 * Sends would require pre-posting a Receive and are
668 		 * not supported.
669 		 * Propose zero length Write if none of Read and Write
670 		 * is indicated.
671 		 */
672 		if (v2->ird & MPA_V2_PEER_TO_PEER) {
673 			cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
674 
675 			if (v2->ord & MPA_V2_RDMA_WRITE_RTR)
676 				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
677 			else if (v2->ord & MPA_V2_RDMA_READ_RTR)
678 				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_READ_RTR;
679 			else
680 				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
681 		}
682 	}
683 
684 	cep->state = SIW_EPSTATE_RECVD_MPAREQ;
685 
686 	/* Keep reference until IWCM accepts/rejects */
687 	siw_cep_get(cep);
688 	rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REQUEST, 0);
689 	if (rv)
690 		siw_cep_put(cep);
691 
692 	return rv;
693 
694 reject_conn:
695 	siw_dbg_cep(cep, "reject: crc %d:%d:%d, m %d:%d\n",
696 		    req->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
697 		    mpa_crc_required, mpa_crc_strict,
698 		    req->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
699 
700 	req->params.bits &= ~MPA_RR_FLAG_MARKERS;
701 	req->params.bits |= MPA_RR_FLAG_REJECT;
702 
703 	if (!mpa_crc_required && mpa_crc_strict)
704 		req->params.bits &= ~MPA_RR_FLAG_CRC;
705 
706 	if (pd_len)
707 		kfree(cep->mpa.pdata);
708 
709 	cep->mpa.pdata = NULL;
710 
711 	siw_send_mpareqrep(cep, NULL, 0);
712 
713 	return -EOPNOTSUPP;
714 }
715 
716 static int siw_proc_mpareply(struct siw_cep *cep)
717 {
718 	struct siw_qp_attrs qp_attrs;
719 	enum siw_qp_attr_mask qp_attr_mask;
720 	struct siw_qp *qp = cep->qp;
721 	struct mpa_rr *rep;
722 	int rv;
723 	u16 rep_ord;
724 	u16 rep_ird;
725 	bool ird_insufficient = false;
726 	enum mpa_v2_ctrl mpa_p2p_mode = MPA_V2_RDMA_NO_RTR;
727 
728 	rv = siw_recv_mpa_rr(cep);
729 	if (rv != -EAGAIN)
730 		siw_cancel_mpatimer(cep);
731 	if (rv)
732 		goto out_err;
733 
734 	rep = &cep->mpa.hdr;
735 
736 	if (__mpa_rr_revision(rep->params.bits) > MPA_REVISION_2) {
737 		/* allow for 0, 1,  and 2 only */
738 		rv = -EPROTO;
739 		goto out_err;
740 	}
741 	if (memcmp(rep->key, MPA_KEY_REP, 16)) {
742 		siw_init_terminate(qp, TERM_ERROR_LAYER_LLP, LLP_ETYPE_MPA,
743 				   LLP_ECODE_INVALID_REQ_RESP, 0);
744 		siw_send_terminate(qp);
745 		rv = -EPROTO;
746 		goto out_err;
747 	}
748 	if (rep->params.bits & MPA_RR_FLAG_REJECT) {
749 		siw_dbg_cep(cep, "got mpa reject\n");
750 		siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNRESET);
751 
752 		return -ECONNRESET;
753 	}
754 	if (try_gso && rep->params.bits & MPA_RR_FLAG_GSO_EXP) {
755 		siw_dbg_cep(cep, "peer allows GSO on TX\n");
756 		qp->tx_ctx.gso_seg_limit = 0;
757 	}
758 	if ((rep->params.bits & MPA_RR_FLAG_MARKERS) ||
759 	    (mpa_crc_required && !(rep->params.bits & MPA_RR_FLAG_CRC)) ||
760 	    (mpa_crc_strict && !mpa_crc_required &&
761 	     (rep->params.bits & MPA_RR_FLAG_CRC))) {
762 		siw_dbg_cep(cep, "reply unsupp: crc %d:%d:%d, m %d:%d\n",
763 			    rep->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
764 			    mpa_crc_required, mpa_crc_strict,
765 			    rep->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
766 
767 		siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
768 
769 		return -EINVAL;
770 	}
771 	if (cep->enhanced_rdma_conn_est) {
772 		struct mpa_v2_data *v2;
773 
774 		if (__mpa_rr_revision(rep->params.bits) < MPA_REVISION_2 ||
775 		    !(rep->params.bits & MPA_RR_FLAG_ENHANCED)) {
776 			/*
777 			 * Protocol failure: The responder MUST reply with
778 			 * MPA version 2 and MUST set MPA_RR_FLAG_ENHANCED.
779 			 */
780 			siw_dbg_cep(cep, "mpa reply error: vers %d, enhcd %d\n",
781 				    __mpa_rr_revision(rep->params.bits),
782 				    rep->params.bits & MPA_RR_FLAG_ENHANCED ?
783 					    1 :
784 					    0);
785 
786 			siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
787 				      -ECONNRESET);
788 			return -EINVAL;
789 		}
790 		v2 = (struct mpa_v2_data *)cep->mpa.pdata;
791 		rep_ird = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
792 		rep_ord = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
793 
794 		if (cep->ird < rep_ord &&
795 		    (relaxed_ird_negotiation == false ||
796 		     rep_ord > cep->sdev->attrs.max_ird)) {
797 			siw_dbg_cep(cep, "ird %d, rep_ord %d, max_ord %d\n",
798 				    cep->ird, rep_ord,
799 				    cep->sdev->attrs.max_ord);
800 			ird_insufficient = true;
801 		}
802 		if (cep->ord > rep_ird && relaxed_ird_negotiation == false) {
803 			siw_dbg_cep(cep, "ord %d, rep_ird %d\n", cep->ord,
804 				    rep_ird);
805 			ird_insufficient = true;
806 		}
807 		/*
808 		 * Always report negotiated peer values to user,
809 		 * even if IRD/ORD negotiation failed
810 		 */
811 		cep->ird = rep_ord;
812 		cep->ord = rep_ird;
813 
814 		if (ird_insufficient) {
815 			/*
816 			 * If the initiator IRD is insuffient for the
817 			 * responder ORD, send a TERM.
818 			 */
819 			siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
820 					   LLP_ETYPE_MPA,
821 					   LLP_ECODE_INSUFFICIENT_IRD, 0);
822 			siw_send_terminate(qp);
823 			rv = -ENOMEM;
824 			goto out_err;
825 		}
826 		if (cep->mpa.v2_ctrl_req.ird & MPA_V2_PEER_TO_PEER)
827 			mpa_p2p_mode =
828 				cep->mpa.v2_ctrl_req.ord &
829 				(MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR);
830 
831 		/*
832 		 * Check if we requested P2P mode, and if peer agrees
833 		 */
834 		if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
835 			if ((mpa_p2p_mode & v2->ord) == 0) {
836 				/*
837 				 * We requested RTR mode(s), but the peer
838 				 * did not pick any mode we support.
839 				 */
840 				siw_dbg_cep(cep,
841 					    "rtr mode:  req %2x, got %2x\n",
842 					    mpa_p2p_mode,
843 					    v2->ord & (MPA_V2_RDMA_WRITE_RTR |
844 						       MPA_V2_RDMA_READ_RTR));
845 
846 				siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
847 						   LLP_ETYPE_MPA,
848 						   LLP_ECODE_NO_MATCHING_RTR,
849 						   0);
850 				siw_send_terminate(qp);
851 				rv = -EPROTO;
852 				goto out_err;
853 			}
854 			mpa_p2p_mode = v2->ord & (MPA_V2_RDMA_WRITE_RTR |
855 						  MPA_V2_RDMA_READ_RTR);
856 		}
857 	}
858 	memset(&qp_attrs, 0, sizeof(qp_attrs));
859 
860 	if (rep->params.bits & MPA_RR_FLAG_CRC)
861 		qp_attrs.flags = SIW_MPA_CRC;
862 
863 	qp_attrs.irq_size = cep->ird;
864 	qp_attrs.orq_size = cep->ord;
865 	qp_attrs.sk = cep->sock;
866 	qp_attrs.state = SIW_QP_STATE_RTS;
867 
868 	qp_attr_mask = SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
869 		       SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD | SIW_QP_ATTR_MPA;
870 
871 	/* Move socket RX/TX under QP control */
872 	down_write(&qp->state_lock);
873 	if (qp->attrs.state > SIW_QP_STATE_RTR) {
874 		rv = -EINVAL;
875 		up_write(&qp->state_lock);
876 		goto out_err;
877 	}
878 	rv = siw_qp_modify(qp, &qp_attrs, qp_attr_mask);
879 
880 	siw_qp_socket_assoc(cep, qp);
881 
882 	up_write(&qp->state_lock);
883 
884 	/* Send extra RDMA frame to trigger peer RTS if negotiated */
885 	if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
886 		rv = siw_qp_mpa_rts(qp, mpa_p2p_mode);
887 		if (rv)
888 			goto out_err;
889 	}
890 	if (!rv) {
891 		rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, 0);
892 		if (!rv)
893 			cep->state = SIW_EPSTATE_RDMA_MODE;
894 
895 		return 0;
896 	}
897 
898 out_err:
899 	siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -EINVAL);
900 
901 	return rv;
902 }
903 
904 /*
905  * siw_accept_newconn - accept an incoming pending connection
906  *
907  */
908 static void siw_accept_newconn(struct siw_cep *cep)
909 {
910 	struct socket *s = cep->sock;
911 	struct socket *new_s = NULL;
912 	struct siw_cep *new_cep = NULL;
913 	int rv = 0; /* debug only. should disappear */
914 
915 	if (cep->state != SIW_EPSTATE_LISTENING)
916 		goto error;
917 
918 	new_cep = siw_cep_alloc(cep->sdev);
919 	if (!new_cep)
920 		goto error;
921 
922 	/*
923 	 * 4: Allocate a sufficient number of work elements
924 	 * to allow concurrent handling of local + peer close
925 	 * events, MPA header processing + MPA timeout.
926 	 */
927 	if (siw_cm_alloc_work(new_cep, 4) != 0)
928 		goto error;
929 
930 	/*
931 	 * Copy saved socket callbacks from listening CEP
932 	 * and assign new socket with new CEP
933 	 */
934 	new_cep->sk_state_change = cep->sk_state_change;
935 	new_cep->sk_data_ready = cep->sk_data_ready;
936 	new_cep->sk_write_space = cep->sk_write_space;
937 	new_cep->sk_error_report = cep->sk_error_report;
938 
939 	rv = kernel_accept(s, &new_s, O_NONBLOCK);
940 	if (rv != 0) {
941 		/*
942 		 * Connection already aborted by peer..?
943 		 */
944 		siw_dbg_cep(cep, "kernel_accept() error: %d\n", rv);
945 		goto error;
946 	}
947 	new_cep->sock = new_s;
948 	siw_cep_get(new_cep);
949 	new_s->sk->sk_user_data = new_cep;
950 
951 	siw_dbg_cep(cep, "listen socket 0x%p, new 0x%p\n", s, new_s);
952 
953 	if (siw_tcp_nagle == false) {
954 		int val = 1;
955 
956 		rv = kernel_setsockopt(new_s, SOL_TCP, TCP_NODELAY,
957 				       (char *)&val, sizeof(val));
958 		if (rv) {
959 			siw_dbg_cep(cep, "setsockopt NODELAY error: %d\n", rv);
960 			goto error;
961 		}
962 	}
963 	new_cep->state = SIW_EPSTATE_AWAIT_MPAREQ;
964 
965 	rv = siw_cm_queue_work(new_cep, SIW_CM_WORK_MPATIMEOUT);
966 	if (rv)
967 		goto error;
968 	/*
969 	 * See siw_proc_mpareq() etc. for the use of new_cep->listen_cep.
970 	 */
971 	new_cep->listen_cep = cep;
972 	siw_cep_get(cep);
973 
974 	if (atomic_read(&new_s->sk->sk_rmem_alloc)) {
975 		/*
976 		 * MPA REQ already queued
977 		 */
978 		siw_dbg_cep(cep, "immediate mpa request\n");
979 
980 		siw_cep_set_inuse(new_cep);
981 		rv = siw_proc_mpareq(new_cep);
982 		siw_cep_set_free(new_cep);
983 
984 		if (rv != -EAGAIN) {
985 			siw_cep_put(cep);
986 			new_cep->listen_cep = NULL;
987 			if (rv)
988 				goto error;
989 		}
990 	}
991 	return;
992 
993 error:
994 	if (new_cep)
995 		siw_cep_put(new_cep);
996 
997 	if (new_s) {
998 		siw_socket_disassoc(new_s);
999 		sock_release(new_s);
1000 		new_cep->sock = NULL;
1001 	}
1002 	siw_dbg_cep(cep, "error %d\n", rv);
1003 }
1004 
1005 static void siw_cm_work_handler(struct work_struct *w)
1006 {
1007 	struct siw_cm_work *work;
1008 	struct siw_cep *cep;
1009 	int release_cep = 0, rv = 0;
1010 
1011 	work = container_of(w, struct siw_cm_work, work.work);
1012 	cep = work->cep;
1013 
1014 	siw_dbg_cep(cep, "[QP %u]: work type: %d, state %d\n",
1015 		    cep->qp ? qp_id(cep->qp) : -1, work->type, cep->state);
1016 
1017 	siw_cep_set_inuse(cep);
1018 
1019 	switch (work->type) {
1020 	case SIW_CM_WORK_ACCEPT:
1021 		siw_accept_newconn(cep);
1022 		break;
1023 
1024 	case SIW_CM_WORK_READ_MPAHDR:
1025 		if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1026 			if (cep->listen_cep) {
1027 				siw_cep_set_inuse(cep->listen_cep);
1028 
1029 				if (cep->listen_cep->state ==
1030 				    SIW_EPSTATE_LISTENING)
1031 					rv = siw_proc_mpareq(cep);
1032 				else
1033 					rv = -EFAULT;
1034 
1035 				siw_cep_set_free(cep->listen_cep);
1036 
1037 				if (rv != -EAGAIN) {
1038 					siw_cep_put(cep->listen_cep);
1039 					cep->listen_cep = NULL;
1040 					if (rv)
1041 						siw_cep_put(cep);
1042 				}
1043 			}
1044 		} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1045 			rv = siw_proc_mpareply(cep);
1046 		} else {
1047 			/*
1048 			 * CEP already moved out of MPA handshake.
1049 			 * any connection management already done.
1050 			 * silently ignore the mpa packet.
1051 			 */
1052 			if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1053 				cep->sock->sk->sk_data_ready(cep->sock->sk);
1054 				siw_dbg_cep(cep, "already in RDMA mode");
1055 			} else {
1056 				siw_dbg_cep(cep, "out of state: %d\n",
1057 					    cep->state);
1058 			}
1059 		}
1060 		if (rv && rv != EAGAIN)
1061 			release_cep = 1;
1062 		break;
1063 
1064 	case SIW_CM_WORK_CLOSE_LLP:
1065 		/*
1066 		 * QP scheduled LLP close
1067 		 */
1068 		if (cep->qp && cep->qp->term_info.valid)
1069 			siw_send_terminate(cep->qp);
1070 
1071 		if (cep->cm_id)
1072 			siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1073 
1074 		release_cep = 1;
1075 		break;
1076 
1077 	case SIW_CM_WORK_PEER_CLOSE:
1078 		if (cep->cm_id) {
1079 			if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1080 				/*
1081 				 * MPA reply not received, but connection drop
1082 				 */
1083 				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1084 					      -ECONNRESET);
1085 			} else if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1086 				/*
1087 				 * NOTE: IW_CM_EVENT_DISCONNECT is given just
1088 				 *       to transition IWCM into CLOSING.
1089 				 */
1090 				siw_cm_upcall(cep, IW_CM_EVENT_DISCONNECT, 0);
1091 				siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1092 			}
1093 			/*
1094 			 * for other states there is no connection
1095 			 * known to the IWCM.
1096 			 */
1097 		} else {
1098 			if (cep->state == SIW_EPSTATE_RECVD_MPAREQ) {
1099 				/*
1100 				 * Wait for the ulp/CM to call accept/reject
1101 				 */
1102 				siw_dbg_cep(cep,
1103 					    "mpa req recvd, wait for ULP\n");
1104 			} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1105 				/*
1106 				 * Socket close before MPA request received.
1107 				 */
1108 				siw_dbg_cep(cep, "no mpareq: drop listener\n");
1109 				siw_cep_put(cep->listen_cep);
1110 				cep->listen_cep = NULL;
1111 			}
1112 		}
1113 		release_cep = 1;
1114 		break;
1115 
1116 	case SIW_CM_WORK_MPATIMEOUT:
1117 		cep->mpa_timer = NULL;
1118 
1119 		if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1120 			/*
1121 			 * MPA request timed out:
1122 			 * Hide any partially received private data and signal
1123 			 * timeout
1124 			 */
1125 			cep->mpa.hdr.params.pd_len = 0;
1126 
1127 			if (cep->cm_id)
1128 				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1129 					      -ETIMEDOUT);
1130 			release_cep = 1;
1131 
1132 		} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1133 			/*
1134 			 * No MPA request received after peer TCP stream setup.
1135 			 */
1136 			if (cep->listen_cep) {
1137 				siw_cep_put(cep->listen_cep);
1138 				cep->listen_cep = NULL;
1139 			}
1140 			release_cep = 1;
1141 		}
1142 		break;
1143 
1144 	default:
1145 		WARN(1, "Undefined CM work type: %d\n", work->type);
1146 	}
1147 	if (release_cep) {
1148 		siw_dbg_cep(cep,
1149 			    "release: timer=%s, QP[%u], id 0x%p\n",
1150 			    cep->mpa_timer ? "y" : "n",
1151 			    cep->qp ? qp_id(cep->qp) : -1, cep->cm_id);
1152 
1153 		siw_cancel_mpatimer(cep);
1154 
1155 		cep->state = SIW_EPSTATE_CLOSED;
1156 
1157 		if (cep->qp) {
1158 			struct siw_qp *qp = cep->qp;
1159 			/*
1160 			 * Serialize a potential race with application
1161 			 * closing the QP and calling siw_qp_cm_drop()
1162 			 */
1163 			siw_qp_get(qp);
1164 			siw_cep_set_free(cep);
1165 
1166 			siw_qp_llp_close(qp);
1167 			siw_qp_put(qp);
1168 
1169 			siw_cep_set_inuse(cep);
1170 			cep->qp = NULL;
1171 			siw_qp_put(qp);
1172 		}
1173 		if (cep->sock) {
1174 			siw_socket_disassoc(cep->sock);
1175 			sock_release(cep->sock);
1176 			cep->sock = NULL;
1177 		}
1178 		if (cep->cm_id) {
1179 			cep->cm_id->rem_ref(cep->cm_id);
1180 			cep->cm_id = NULL;
1181 			siw_cep_put(cep);
1182 		}
1183 	}
1184 	siw_cep_set_free(cep);
1185 	siw_put_work(work);
1186 	siw_cep_put(cep);
1187 }
1188 
1189 static struct workqueue_struct *siw_cm_wq;
1190 
1191 int siw_cm_queue_work(struct siw_cep *cep, enum siw_work_type type)
1192 {
1193 	struct siw_cm_work *work = siw_get_work(cep);
1194 	unsigned long delay = 0;
1195 
1196 	if (!work) {
1197 		siw_dbg_cep(cep, "failed with no work available\n");
1198 		return -ENOMEM;
1199 	}
1200 	work->type = type;
1201 	work->cep = cep;
1202 
1203 	siw_cep_get(cep);
1204 
1205 	INIT_DELAYED_WORK(&work->work, siw_cm_work_handler);
1206 
1207 	if (type == SIW_CM_WORK_MPATIMEOUT) {
1208 		cep->mpa_timer = work;
1209 
1210 		if (cep->state == SIW_EPSTATE_AWAIT_MPAREP)
1211 			delay = MPAREQ_TIMEOUT;
1212 		else
1213 			delay = MPAREP_TIMEOUT;
1214 	}
1215 	siw_dbg_cep(cep, "[QP %u]: work type: %d, work 0x%p, timeout %lu\n",
1216 		    cep->qp ? qp_id(cep->qp) : -1, type, work, delay);
1217 
1218 	queue_delayed_work(siw_cm_wq, &work->work, delay);
1219 
1220 	return 0;
1221 }
1222 
1223 static void siw_cm_llp_data_ready(struct sock *sk)
1224 {
1225 	struct siw_cep *cep;
1226 
1227 	read_lock(&sk->sk_callback_lock);
1228 
1229 	cep = sk_to_cep(sk);
1230 	if (!cep) {
1231 		WARN_ON(1);
1232 		goto out;
1233 	}
1234 	siw_dbg_cep(cep, "state: %d\n", cep->state);
1235 
1236 	switch (cep->state) {
1237 	case SIW_EPSTATE_RDMA_MODE:
1238 		/* fall through */
1239 	case SIW_EPSTATE_LISTENING:
1240 		break;
1241 
1242 	case SIW_EPSTATE_AWAIT_MPAREQ:
1243 		/* fall through */
1244 	case SIW_EPSTATE_AWAIT_MPAREP:
1245 		siw_cm_queue_work(cep, SIW_CM_WORK_READ_MPAHDR);
1246 		break;
1247 
1248 	default:
1249 		siw_dbg_cep(cep, "unexpected data, state %d\n", cep->state);
1250 		break;
1251 	}
1252 out:
1253 	read_unlock(&sk->sk_callback_lock);
1254 }
1255 
1256 static void siw_cm_llp_write_space(struct sock *sk)
1257 {
1258 	struct siw_cep *cep = sk_to_cep(sk);
1259 
1260 	if (cep)
1261 		siw_dbg_cep(cep, "state: %d\n", cep->state);
1262 }
1263 
1264 static void siw_cm_llp_error_report(struct sock *sk)
1265 {
1266 	struct siw_cep *cep = sk_to_cep(sk);
1267 
1268 	if (cep) {
1269 		siw_dbg_cep(cep, "error %d, socket state: %d, cep state: %d\n",
1270 			    sk->sk_err, sk->sk_state, cep->state);
1271 		cep->sk_error_report(sk);
1272 	}
1273 }
1274 
1275 static void siw_cm_llp_state_change(struct sock *sk)
1276 {
1277 	struct siw_cep *cep;
1278 	void (*orig_state_change)(struct sock *s);
1279 
1280 	read_lock(&sk->sk_callback_lock);
1281 
1282 	cep = sk_to_cep(sk);
1283 	if (!cep) {
1284 		/* endpoint already disassociated */
1285 		read_unlock(&sk->sk_callback_lock);
1286 		return;
1287 	}
1288 	orig_state_change = cep->sk_state_change;
1289 
1290 	siw_dbg_cep(cep, "state: %d\n", cep->state);
1291 
1292 	switch (sk->sk_state) {
1293 	case TCP_ESTABLISHED:
1294 		/*
1295 		 * handle accepting socket as special case where only
1296 		 * new connection is possible
1297 		 */
1298 		siw_cm_queue_work(cep, SIW_CM_WORK_ACCEPT);
1299 		break;
1300 
1301 	case TCP_CLOSE:
1302 	case TCP_CLOSE_WAIT:
1303 		if (cep->qp)
1304 			cep->qp->tx_ctx.tx_suspend = 1;
1305 		siw_cm_queue_work(cep, SIW_CM_WORK_PEER_CLOSE);
1306 		break;
1307 
1308 	default:
1309 		siw_dbg_cep(cep, "unexpected socket state %d\n", sk->sk_state);
1310 	}
1311 	read_unlock(&sk->sk_callback_lock);
1312 	orig_state_change(sk);
1313 }
1314 
1315 static int kernel_bindconnect(struct socket *s, struct sockaddr *laddr,
1316 			      struct sockaddr *raddr)
1317 {
1318 	int rv, flags = 0, s_val = 1;
1319 	size_t size = laddr->sa_family == AF_INET ?
1320 		sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6);
1321 
1322 	/*
1323 	 * Make address available again asap.
1324 	 */
1325 	rv = kernel_setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (char *)&s_val,
1326 			       sizeof(s_val));
1327 	if (rv < 0)
1328 		return rv;
1329 
1330 	rv = s->ops->bind(s, laddr, size);
1331 	if (rv < 0)
1332 		return rv;
1333 
1334 	rv = s->ops->connect(s, raddr, size, flags);
1335 
1336 	return rv < 0 ? rv : 0;
1337 }
1338 
1339 int siw_connect(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1340 {
1341 	struct siw_device *sdev = to_siw_dev(id->device);
1342 	struct siw_qp *qp;
1343 	struct siw_cep *cep = NULL;
1344 	struct socket *s = NULL;
1345 	struct sockaddr *laddr = (struct sockaddr *)&id->local_addr,
1346 			*raddr = (struct sockaddr *)&id->remote_addr;
1347 	bool p2p_mode = peer_to_peer, v4 = true;
1348 	u16 pd_len = params->private_data_len;
1349 	int version = mpa_version, rv;
1350 
1351 	if (pd_len > MPA_MAX_PRIVDATA)
1352 		return -EINVAL;
1353 
1354 	if (params->ird > sdev->attrs.max_ird ||
1355 	    params->ord > sdev->attrs.max_ord)
1356 		return -ENOMEM;
1357 
1358 	if (laddr->sa_family == AF_INET6)
1359 		v4 = false;
1360 	else if (laddr->sa_family != AF_INET)
1361 		return -EAFNOSUPPORT;
1362 
1363 	/*
1364 	 * Respect any iwarp port mapping: Use mapped remote address
1365 	 * if valid. Local address must not be mapped, since siw
1366 	 * uses kernel TCP stack.
1367 	 */
1368 	if ((v4 && to_sockaddr_in(id->remote_addr).sin_port != 0) ||
1369 	     to_sockaddr_in6(id->remote_addr).sin6_port != 0)
1370 		raddr = (struct sockaddr *)&id->m_remote_addr;
1371 
1372 	qp = siw_qp_id2obj(sdev, params->qpn);
1373 	if (!qp) {
1374 		WARN(1, "[QP %u] does not exist\n", params->qpn);
1375 		rv = -EINVAL;
1376 		goto error;
1377 	}
1378 	if (v4)
1379 		siw_dbg_qp(qp,
1380 			   "id 0x%p, pd_len %d, laddr %pI4 %d, raddr %pI4 %d\n",
1381 			   id, pd_len,
1382 			   &((struct sockaddr_in *)(laddr))->sin_addr,
1383 			   ntohs(((struct sockaddr_in *)(laddr))->sin_port),
1384 			   &((struct sockaddr_in *)(raddr))->sin_addr,
1385 			   ntohs(((struct sockaddr_in *)(raddr))->sin_port));
1386 	else
1387 		siw_dbg_qp(qp,
1388 			   "id 0x%p, pd_len %d, laddr %pI6 %d, raddr %pI6 %d\n",
1389 			   id, pd_len,
1390 			   &((struct sockaddr_in6 *)(laddr))->sin6_addr,
1391 			   ntohs(((struct sockaddr_in6 *)(laddr))->sin6_port),
1392 			   &((struct sockaddr_in6 *)(raddr))->sin6_addr,
1393 			   ntohs(((struct sockaddr_in6 *)(raddr))->sin6_port));
1394 
1395 	rv = sock_create(v4 ? AF_INET : AF_INET6, SOCK_STREAM, IPPROTO_TCP, &s);
1396 	if (rv < 0)
1397 		goto error;
1398 
1399 	/*
1400 	 * NOTE: For simplification, connect() is called in blocking
1401 	 * mode. Might be reconsidered for async connection setup at
1402 	 * TCP level.
1403 	 */
1404 	rv = kernel_bindconnect(s, laddr, raddr);
1405 	if (rv != 0) {
1406 		siw_dbg_qp(qp, "kernel_bindconnect: error %d\n", rv);
1407 		goto error;
1408 	}
1409 	if (siw_tcp_nagle == false) {
1410 		int val = 1;
1411 
1412 		rv = kernel_setsockopt(s, SOL_TCP, TCP_NODELAY, (char *)&val,
1413 				       sizeof(val));
1414 		if (rv) {
1415 			siw_dbg_qp(qp, "setsockopt NODELAY error: %d\n", rv);
1416 			goto error;
1417 		}
1418 	}
1419 	cep = siw_cep_alloc(sdev);
1420 	if (!cep) {
1421 		rv = -ENOMEM;
1422 		goto error;
1423 	}
1424 	siw_cep_set_inuse(cep);
1425 
1426 	/* Associate QP with CEP */
1427 	siw_cep_get(cep);
1428 	qp->cep = cep;
1429 
1430 	/* siw_qp_get(qp) already done by QP lookup */
1431 	cep->qp = qp;
1432 
1433 	id->add_ref(id);
1434 	cep->cm_id = id;
1435 
1436 	/*
1437 	 * 4: Allocate a sufficient number of work elements
1438 	 * to allow concurrent handling of local + peer close
1439 	 * events, MPA header processing + MPA timeout.
1440 	 */
1441 	rv = siw_cm_alloc_work(cep, 4);
1442 	if (rv != 0) {
1443 		rv = -ENOMEM;
1444 		goto error;
1445 	}
1446 	cep->ird = params->ird;
1447 	cep->ord = params->ord;
1448 
1449 	if (p2p_mode && cep->ord == 0)
1450 		cep->ord = 1;
1451 
1452 	cep->state = SIW_EPSTATE_CONNECTING;
1453 
1454 	/*
1455 	 * Associate CEP with socket
1456 	 */
1457 	siw_cep_socket_assoc(cep, s);
1458 
1459 	cep->state = SIW_EPSTATE_AWAIT_MPAREP;
1460 
1461 	/*
1462 	 * Set MPA Request bits: CRC if required, no MPA Markers,
1463 	 * MPA Rev. according to module parameter 'mpa_version', Key 'Request'.
1464 	 */
1465 	cep->mpa.hdr.params.bits = 0;
1466 	if (version > MPA_REVISION_2) {
1467 		pr_warn("Setting MPA version to %u\n", MPA_REVISION_2);
1468 		version = MPA_REVISION_2;
1469 		/* Adjust also module parameter */
1470 		mpa_version = MPA_REVISION_2;
1471 	}
1472 	__mpa_rr_set_revision(&cep->mpa.hdr.params.bits, version);
1473 
1474 	if (try_gso)
1475 		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_GSO_EXP;
1476 
1477 	if (mpa_crc_required)
1478 		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_CRC;
1479 
1480 	/*
1481 	 * If MPA version == 2:
1482 	 * o Include ORD and IRD.
1483 	 * o Indicate peer-to-peer mode, if required by module
1484 	 *   parameter 'peer_to_peer'.
1485 	 */
1486 	if (version == MPA_REVISION_2) {
1487 		cep->enhanced_rdma_conn_est = true;
1488 		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_ENHANCED;
1489 
1490 		cep->mpa.v2_ctrl.ird = htons(cep->ird);
1491 		cep->mpa.v2_ctrl.ord = htons(cep->ord);
1492 
1493 		if (p2p_mode) {
1494 			cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
1495 			cep->mpa.v2_ctrl.ord |= rtr_type;
1496 		}
1497 		/* Remember own P2P mode requested */
1498 		cep->mpa.v2_ctrl_req.ird = cep->mpa.v2_ctrl.ird;
1499 		cep->mpa.v2_ctrl_req.ord = cep->mpa.v2_ctrl.ord;
1500 	}
1501 	memcpy(cep->mpa.hdr.key, MPA_KEY_REQ, 16);
1502 
1503 	rv = siw_send_mpareqrep(cep, params->private_data, pd_len);
1504 	/*
1505 	 * Reset private data.
1506 	 */
1507 	cep->mpa.hdr.params.pd_len = 0;
1508 
1509 	if (rv >= 0) {
1510 		rv = siw_cm_queue_work(cep, SIW_CM_WORK_MPATIMEOUT);
1511 		if (!rv) {
1512 			siw_dbg_cep(cep, "id 0x%p, [QP %u]: exit\n", id,
1513 				    qp_id(qp));
1514 			siw_cep_set_free(cep);
1515 			return 0;
1516 		}
1517 	}
1518 error:
1519 	siw_dbg_qp(qp, "failed: %d\n", rv);
1520 
1521 	if (cep) {
1522 		siw_socket_disassoc(s);
1523 		sock_release(s);
1524 		cep->sock = NULL;
1525 
1526 		cep->qp = NULL;
1527 
1528 		cep->cm_id = NULL;
1529 		id->rem_ref(id);
1530 		siw_cep_put(cep);
1531 
1532 		qp->cep = NULL;
1533 		siw_cep_put(cep);
1534 
1535 		cep->state = SIW_EPSTATE_CLOSED;
1536 
1537 		siw_cep_set_free(cep);
1538 
1539 		siw_cep_put(cep);
1540 
1541 	} else if (s) {
1542 		sock_release(s);
1543 	}
1544 	siw_qp_put(qp);
1545 
1546 	return rv;
1547 }
1548 
1549 /*
1550  * siw_accept - Let SoftiWARP accept an RDMA connection request
1551  *
1552  * @id:		New connection management id to be used for accepted
1553  *		connection request
1554  * @params:	Connection parameters provided by ULP for accepting connection
1555  *
1556  * Transition QP to RTS state, associate new CM id @id with accepted CEP
1557  * and get prepared for TCP input by installing socket callbacks.
1558  * Then send MPA Reply and generate the "connection established" event.
1559  * Socket callbacks must be installed before sending MPA Reply, because
1560  * the latter may cause a first RDMA message to arrive from the RDMA Initiator
1561  * side very quickly, at which time the socket callbacks must be ready.
1562  */
1563 int siw_accept(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1564 {
1565 	struct siw_device *sdev = to_siw_dev(id->device);
1566 	struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1567 	struct siw_qp *qp;
1568 	struct siw_qp_attrs qp_attrs;
1569 	int rv, max_priv_data = MPA_MAX_PRIVDATA;
1570 	bool wait_for_peer_rts = false;
1571 
1572 	siw_cep_set_inuse(cep);
1573 	siw_cep_put(cep);
1574 
1575 	/* Free lingering inbound private data */
1576 	if (cep->mpa.hdr.params.pd_len) {
1577 		cep->mpa.hdr.params.pd_len = 0;
1578 		kfree(cep->mpa.pdata);
1579 		cep->mpa.pdata = NULL;
1580 	}
1581 	siw_cancel_mpatimer(cep);
1582 
1583 	if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1584 		siw_dbg_cep(cep, "id 0x%p: out of state\n", id);
1585 
1586 		siw_cep_set_free(cep);
1587 		siw_cep_put(cep);
1588 
1589 		return -ECONNRESET;
1590 	}
1591 	qp = siw_qp_id2obj(sdev, params->qpn);
1592 	if (!qp) {
1593 		WARN(1, "[QP %d] does not exist\n", params->qpn);
1594 		siw_cep_set_free(cep);
1595 		siw_cep_put(cep);
1596 
1597 		return -EINVAL;
1598 	}
1599 	down_write(&qp->state_lock);
1600 	if (qp->attrs.state > SIW_QP_STATE_RTR) {
1601 		rv = -EINVAL;
1602 		up_write(&qp->state_lock);
1603 		goto error;
1604 	}
1605 	siw_dbg_cep(cep, "id 0x%p\n", id);
1606 
1607 	if (try_gso && cep->mpa.hdr.params.bits & MPA_RR_FLAG_GSO_EXP) {
1608 		siw_dbg_cep(cep, "peer allows GSO on TX\n");
1609 		qp->tx_ctx.gso_seg_limit = 0;
1610 	}
1611 	if (params->ord > sdev->attrs.max_ord ||
1612 	    params->ird > sdev->attrs.max_ird) {
1613 		siw_dbg_cep(
1614 			cep,
1615 			"id 0x%p, [QP %u]: ord %d (max %d), ird %d (max %d)\n",
1616 			id, qp_id(qp), params->ord, sdev->attrs.max_ord,
1617 			params->ird, sdev->attrs.max_ird);
1618 		rv = -EINVAL;
1619 		up_write(&qp->state_lock);
1620 		goto error;
1621 	}
1622 	if (cep->enhanced_rdma_conn_est)
1623 		max_priv_data -= sizeof(struct mpa_v2_data);
1624 
1625 	if (params->private_data_len > max_priv_data) {
1626 		siw_dbg_cep(
1627 			cep,
1628 			"id 0x%p, [QP %u]: private data length: %d (max %d)\n",
1629 			id, qp_id(qp), params->private_data_len, max_priv_data);
1630 		rv = -EINVAL;
1631 		up_write(&qp->state_lock);
1632 		goto error;
1633 	}
1634 	if (cep->enhanced_rdma_conn_est) {
1635 		if (params->ord > cep->ord) {
1636 			if (relaxed_ird_negotiation) {
1637 				params->ord = cep->ord;
1638 			} else {
1639 				cep->ird = params->ird;
1640 				cep->ord = params->ord;
1641 				rv = -EINVAL;
1642 				up_write(&qp->state_lock);
1643 				goto error;
1644 			}
1645 		}
1646 		if (params->ird < cep->ird) {
1647 			if (relaxed_ird_negotiation &&
1648 			    cep->ird <= sdev->attrs.max_ird)
1649 				params->ird = cep->ird;
1650 			else {
1651 				rv = -ENOMEM;
1652 				up_write(&qp->state_lock);
1653 				goto error;
1654 			}
1655 		}
1656 		if (cep->mpa.v2_ctrl.ord &
1657 		    (MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR))
1658 			wait_for_peer_rts = true;
1659 		/*
1660 		 * Signal back negotiated IRD and ORD values
1661 		 */
1662 		cep->mpa.v2_ctrl.ord =
1663 			htons(params->ord & MPA_IRD_ORD_MASK) |
1664 			(cep->mpa.v2_ctrl.ord & ~MPA_V2_MASK_IRD_ORD);
1665 		cep->mpa.v2_ctrl.ird =
1666 			htons(params->ird & MPA_IRD_ORD_MASK) |
1667 			(cep->mpa.v2_ctrl.ird & ~MPA_V2_MASK_IRD_ORD);
1668 	}
1669 	cep->ird = params->ird;
1670 	cep->ord = params->ord;
1671 
1672 	cep->cm_id = id;
1673 	id->add_ref(id);
1674 
1675 	memset(&qp_attrs, 0, sizeof(qp_attrs));
1676 	qp_attrs.orq_size = cep->ord;
1677 	qp_attrs.irq_size = cep->ird;
1678 	qp_attrs.sk = cep->sock;
1679 	if (cep->mpa.hdr.params.bits & MPA_RR_FLAG_CRC)
1680 		qp_attrs.flags = SIW_MPA_CRC;
1681 	qp_attrs.state = SIW_QP_STATE_RTS;
1682 
1683 	siw_dbg_cep(cep, "id 0x%p, [QP%u]: moving to rts\n", id, qp_id(qp));
1684 
1685 	/* Associate QP with CEP */
1686 	siw_cep_get(cep);
1687 	qp->cep = cep;
1688 
1689 	/* siw_qp_get(qp) already done by QP lookup */
1690 	cep->qp = qp;
1691 
1692 	cep->state = SIW_EPSTATE_RDMA_MODE;
1693 
1694 	/* Move socket RX/TX under QP control */
1695 	rv = siw_qp_modify(qp, &qp_attrs,
1696 			   SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
1697 				   SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD |
1698 				   SIW_QP_ATTR_MPA);
1699 	up_write(&qp->state_lock);
1700 
1701 	if (rv)
1702 		goto error;
1703 
1704 	siw_dbg_cep(cep, "id 0x%p, [QP %u]: send mpa reply, %d byte pdata\n",
1705 		    id, qp_id(qp), params->private_data_len);
1706 
1707 	rv = siw_send_mpareqrep(cep, params->private_data,
1708 				params->private_data_len);
1709 	if (rv != 0)
1710 		goto error;
1711 
1712 	if (wait_for_peer_rts) {
1713 		siw_sk_assign_rtr_upcalls(cep);
1714 	} else {
1715 		siw_qp_socket_assoc(cep, qp);
1716 		rv = siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
1717 		if (rv)
1718 			goto error;
1719 	}
1720 	siw_cep_set_free(cep);
1721 
1722 	return 0;
1723 error:
1724 	siw_socket_disassoc(cep->sock);
1725 	sock_release(cep->sock);
1726 	cep->sock = NULL;
1727 
1728 	cep->state = SIW_EPSTATE_CLOSED;
1729 
1730 	if (cep->cm_id) {
1731 		cep->cm_id->rem_ref(id);
1732 		cep->cm_id = NULL;
1733 	}
1734 	if (qp->cep) {
1735 		siw_cep_put(cep);
1736 		qp->cep = NULL;
1737 	}
1738 	cep->qp = NULL;
1739 	siw_qp_put(qp);
1740 
1741 	siw_cep_set_free(cep);
1742 	siw_cep_put(cep);
1743 
1744 	return rv;
1745 }
1746 
1747 /*
1748  * siw_reject()
1749  *
1750  * Local connection reject case. Send private data back to peer,
1751  * close connection and dereference connection id.
1752  */
1753 int siw_reject(struct iw_cm_id *id, const void *pdata, u8 pd_len)
1754 {
1755 	struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1756 
1757 	siw_cep_set_inuse(cep);
1758 	siw_cep_put(cep);
1759 
1760 	siw_cancel_mpatimer(cep);
1761 
1762 	if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1763 		siw_dbg_cep(cep, "id 0x%p: out of state\n", id);
1764 
1765 		siw_cep_set_free(cep);
1766 		siw_cep_put(cep); /* put last reference */
1767 
1768 		return -ECONNRESET;
1769 	}
1770 	siw_dbg_cep(cep, "id 0x%p, cep->state %d, pd_len %d\n", id, cep->state,
1771 		    pd_len);
1772 
1773 	if (__mpa_rr_revision(cep->mpa.hdr.params.bits) >= MPA_REVISION_1) {
1774 		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_REJECT; /* reject */
1775 		siw_send_mpareqrep(cep, pdata, pd_len);
1776 	}
1777 	siw_socket_disassoc(cep->sock);
1778 	sock_release(cep->sock);
1779 	cep->sock = NULL;
1780 
1781 	cep->state = SIW_EPSTATE_CLOSED;
1782 
1783 	siw_cep_set_free(cep);
1784 	siw_cep_put(cep);
1785 
1786 	return 0;
1787 }
1788 
1789 static int siw_listen_address(struct iw_cm_id *id, int backlog,
1790 			      struct sockaddr *laddr, int addr_family)
1791 {
1792 	struct socket *s;
1793 	struct siw_cep *cep = NULL;
1794 	struct siw_device *sdev = to_siw_dev(id->device);
1795 	int rv = 0, s_val;
1796 
1797 	rv = sock_create(addr_family, SOCK_STREAM, IPPROTO_TCP, &s);
1798 	if (rv < 0)
1799 		return rv;
1800 
1801 	/*
1802 	 * Allow binding local port when still in TIME_WAIT from last close.
1803 	 */
1804 	s_val = 1;
1805 	rv = kernel_setsockopt(s, SOL_SOCKET, SO_REUSEADDR, (char *)&s_val,
1806 			       sizeof(s_val));
1807 	if (rv) {
1808 		siw_dbg(id->device, "id 0x%p: setsockopt error: %d\n", id, rv);
1809 		goto error;
1810 	}
1811 	rv = s->ops->bind(s, laddr, addr_family == AF_INET ?
1812 				    sizeof(struct sockaddr_in) :
1813 				    sizeof(struct sockaddr_in6));
1814 	if (rv) {
1815 		siw_dbg(id->device, "id 0x%p: socket bind error: %d\n", id, rv);
1816 		goto error;
1817 	}
1818 	cep = siw_cep_alloc(sdev);
1819 	if (!cep) {
1820 		rv = -ENOMEM;
1821 		goto error;
1822 	}
1823 	siw_cep_socket_assoc(cep, s);
1824 
1825 	rv = siw_cm_alloc_work(cep, backlog);
1826 	if (rv) {
1827 		siw_dbg(id->device,
1828 			"id 0x%p: alloc_work error %d, backlog %d\n", id,
1829 			rv, backlog);
1830 		goto error;
1831 	}
1832 	rv = s->ops->listen(s, backlog);
1833 	if (rv) {
1834 		siw_dbg(id->device, "id 0x%p: listen error %d\n", id, rv);
1835 		goto error;
1836 	}
1837 	cep->cm_id = id;
1838 	id->add_ref(id);
1839 
1840 	/*
1841 	 * In case of a wildcard rdma_listen on a multi-homed device,
1842 	 * a listener's IWCM id is associated with more than one listening CEP.
1843 	 *
1844 	 * We currently use id->provider_data in three different ways:
1845 	 *
1846 	 * o For a listener's IWCM id, id->provider_data points to
1847 	 *   the list_head of the list of listening CEPs.
1848 	 *   Uses: siw_create_listen(), siw_destroy_listen()
1849 	 *
1850 	 * o For each accepted passive-side IWCM id, id->provider_data
1851 	 *   points to the CEP itself. This is a consequence of
1852 	 *   - siw_cm_upcall() setting event.provider_data = cep and
1853 	 *   - the IWCM's cm_conn_req_handler() setting provider_data of the
1854 	 *     new passive-side IWCM id equal to event.provider_data
1855 	 *   Uses: siw_accept(), siw_reject()
1856 	 *
1857 	 * o For an active-side IWCM id, id->provider_data is not used at all.
1858 	 *
1859 	 */
1860 	if (!id->provider_data) {
1861 		id->provider_data =
1862 			kmalloc(sizeof(struct list_head), GFP_KERNEL);
1863 		if (!id->provider_data) {
1864 			rv = -ENOMEM;
1865 			goto error;
1866 		}
1867 		INIT_LIST_HEAD((struct list_head *)id->provider_data);
1868 	}
1869 	list_add_tail(&cep->listenq, (struct list_head *)id->provider_data);
1870 	cep->state = SIW_EPSTATE_LISTENING;
1871 
1872 	if (addr_family == AF_INET)
1873 		siw_dbg(id->device, "Listen at laddr %pI4 %u\n",
1874 			&(((struct sockaddr_in *)laddr)->sin_addr),
1875 			((struct sockaddr_in *)laddr)->sin_port);
1876 	else
1877 		siw_dbg(id->device, "Listen at laddr %pI6 %u\n",
1878 			&(((struct sockaddr_in6 *)laddr)->sin6_addr),
1879 			((struct sockaddr_in6 *)laddr)->sin6_port);
1880 
1881 	return 0;
1882 
1883 error:
1884 	siw_dbg(id->device, "failed: %d\n", rv);
1885 
1886 	if (cep) {
1887 		siw_cep_set_inuse(cep);
1888 
1889 		if (cep->cm_id) {
1890 			cep->cm_id->rem_ref(cep->cm_id);
1891 			cep->cm_id = NULL;
1892 		}
1893 		cep->sock = NULL;
1894 		siw_socket_disassoc(s);
1895 		cep->state = SIW_EPSTATE_CLOSED;
1896 
1897 		siw_cep_set_free(cep);
1898 		siw_cep_put(cep);
1899 	}
1900 	sock_release(s);
1901 
1902 	return rv;
1903 }
1904 
1905 static void siw_drop_listeners(struct iw_cm_id *id)
1906 {
1907 	struct list_head *p, *tmp;
1908 
1909 	/*
1910 	 * In case of a wildcard rdma_listen on a multi-homed device,
1911 	 * a listener's IWCM id is associated with more than one listening CEP.
1912 	 */
1913 	list_for_each_safe(p, tmp, (struct list_head *)id->provider_data) {
1914 		struct siw_cep *cep = list_entry(p, struct siw_cep, listenq);
1915 
1916 		list_del(p);
1917 
1918 		siw_dbg_cep(cep, "id 0x%p: drop cep, state %d\n", id,
1919 			    cep->state);
1920 
1921 		siw_cep_set_inuse(cep);
1922 
1923 		if (cep->cm_id) {
1924 			cep->cm_id->rem_ref(cep->cm_id);
1925 			cep->cm_id = NULL;
1926 		}
1927 		if (cep->sock) {
1928 			siw_socket_disassoc(cep->sock);
1929 			sock_release(cep->sock);
1930 			cep->sock = NULL;
1931 		}
1932 		cep->state = SIW_EPSTATE_CLOSED;
1933 		siw_cep_set_free(cep);
1934 		siw_cep_put(cep);
1935 	}
1936 }
1937 
1938 /*
1939  * siw_create_listen - Create resources for a listener's IWCM ID @id
1940  *
1941  * Listens on the socket addresses id->local_addr and id->remote_addr.
1942  *
1943  * If the listener's @id provides a specific local IP address, at most one
1944  * listening socket is created and associated with @id.
1945  *
1946  * If the listener's @id provides the wildcard (zero) local IP address,
1947  * a separate listen is performed for each local IP address of the device
1948  * by creating a listening socket and binding to that local IP address.
1949  *
1950  */
1951 int siw_create_listen(struct iw_cm_id *id, int backlog)
1952 {
1953 	struct net_device *dev = to_siw_dev(id->device)->netdev;
1954 	int rv = 0, listeners = 0;
1955 
1956 	siw_dbg(id->device, "id 0x%p: backlog %d\n", id, backlog);
1957 
1958 	/*
1959 	 * For each attached address of the interface, create a
1960 	 * listening socket, if id->local_addr is the wildcard
1961 	 * IP address or matches the IP address.
1962 	 */
1963 	if (id->local_addr.ss_family == AF_INET) {
1964 		struct in_device *in_dev = in_dev_get(dev);
1965 		struct sockaddr_in s_laddr, *s_raddr;
1966 		const struct in_ifaddr *ifa;
1967 
1968 		memcpy(&s_laddr, &id->local_addr, sizeof(s_laddr));
1969 		s_raddr = (struct sockaddr_in *)&id->remote_addr;
1970 
1971 		siw_dbg(id->device,
1972 			"id 0x%p: laddr %pI4:%d, raddr %pI4:%d\n",
1973 			id, &s_laddr.sin_addr, ntohs(s_laddr.sin_port),
1974 			&s_raddr->sin_addr, ntohs(s_raddr->sin_port));
1975 
1976 		rtnl_lock();
1977 		in_dev_for_each_ifa_rtnl(ifa, in_dev) {
1978 			if (ipv4_is_zeronet(s_laddr.sin_addr.s_addr) ||
1979 			    s_laddr.sin_addr.s_addr == ifa->ifa_address) {
1980 				s_laddr.sin_addr.s_addr = ifa->ifa_address;
1981 
1982 				rv = siw_listen_address(id, backlog,
1983 						(struct sockaddr *)&s_laddr,
1984 						AF_INET);
1985 				if (!rv)
1986 					listeners++;
1987 			}
1988 		}
1989 		rtnl_unlock();
1990 		in_dev_put(in_dev);
1991 	} else if (id->local_addr.ss_family == AF_INET6) {
1992 		struct inet6_dev *in6_dev = in6_dev_get(dev);
1993 		struct inet6_ifaddr *ifp;
1994 		struct sockaddr_in6 *s_laddr = &to_sockaddr_in6(id->local_addr),
1995 			*s_raddr = &to_sockaddr_in6(id->remote_addr);
1996 
1997 		siw_dbg(id->device,
1998 			"id 0x%p: laddr %pI6:%d, raddr %pI6:%d\n",
1999 			id, &s_laddr->sin6_addr, ntohs(s_laddr->sin6_port),
2000 			&s_raddr->sin6_addr, ntohs(s_raddr->sin6_port));
2001 
2002 		read_lock_bh(&in6_dev->lock);
2003 		list_for_each_entry(ifp, &in6_dev->addr_list, if_list) {
2004 			struct sockaddr_in6 bind_addr;
2005 
2006 			if (ipv6_addr_any(&s_laddr->sin6_addr) ||
2007 			    ipv6_addr_equal(&s_laddr->sin6_addr, &ifp->addr)) {
2008 				bind_addr.sin6_family = AF_INET6;
2009 				bind_addr.sin6_port = s_laddr->sin6_port;
2010 				bind_addr.sin6_flowinfo = 0;
2011 				bind_addr.sin6_addr = ifp->addr;
2012 				bind_addr.sin6_scope_id = dev->ifindex;
2013 
2014 				rv = siw_listen_address(id, backlog,
2015 						(struct sockaddr *)&bind_addr,
2016 						AF_INET6);
2017 				if (!rv)
2018 					listeners++;
2019 			}
2020 		}
2021 		read_unlock_bh(&in6_dev->lock);
2022 
2023 		in6_dev_put(in6_dev);
2024 	} else {
2025 		return -EAFNOSUPPORT;
2026 	}
2027 	if (listeners)
2028 		rv = 0;
2029 	else if (!rv)
2030 		rv = -EINVAL;
2031 
2032 	siw_dbg(id->device, "id 0x%p: %s\n", id, rv ? "FAIL" : "OK");
2033 
2034 	return rv;
2035 }
2036 
2037 int siw_destroy_listen(struct iw_cm_id *id)
2038 {
2039 	siw_dbg(id->device, "id 0x%p\n", id);
2040 
2041 	if (!id->provider_data) {
2042 		siw_dbg(id->device, "id 0x%p: no cep(s)\n", id);
2043 		return 0;
2044 	}
2045 	siw_drop_listeners(id);
2046 	kfree(id->provider_data);
2047 	id->provider_data = NULL;
2048 
2049 	return 0;
2050 }
2051 
2052 int siw_cm_init(void)
2053 {
2054 	/*
2055 	 * create_single_workqueue for strict ordering
2056 	 */
2057 	siw_cm_wq = create_singlethread_workqueue("siw_cm_wq");
2058 	if (!siw_cm_wq)
2059 		return -ENOMEM;
2060 
2061 	return 0;
2062 }
2063 
2064 void siw_cm_exit(void)
2065 {
2066 	if (siw_cm_wq) {
2067 		flush_workqueue(siw_cm_wq);
2068 		destroy_workqueue(siw_cm_wq);
2069 	}
2070 }
2071