siw_cm.c

   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#include <trace/events/sock.h>
  20
  21#include <rdma/iw_cm.h>
  22#include <rdma/ib_verbs.h>
  23#include <rdma/ib_user_verbs.h>
  24
  25#include "siw.h"
  26#include "siw_cm.h"
  27
  28/*
  29 * Set to any combination of
  30 * MPA_V2_RDMA_NO_RTR, MPA_V2_RDMA_READ_RTR, MPA_V2_RDMA_WRITE_RTR
  31 */
  32static __be16 rtr_type = MPA_V2_RDMA_READ_RTR | MPA_V2_RDMA_WRITE_RTR;
  33static const bool relaxed_ird_negotiation = true;
  34
  35static void siw_cm_llp_state_change(struct sock *s);
  36static void siw_cm_llp_data_ready(struct sock *s);
  37static void siw_cm_llp_write_space(struct sock *s);
  38static void siw_cm_llp_error_report(struct sock *s);
  39static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
  40			 int status);
  41
  42
  43#ifdef CONFIG_DEBUG_LOCK_ALLOC
  44/*
  45 * lockdep can detect false positive circular dependencies
  46 * when there are user-space socket API users or in kernel
  47 * users switching between a tcp and rdma transport.
  48 * Maybe also switching between siw and rxe may cause
  49 * problems as per default sockets are only classified
  50 * by family and not by ip protocol. And there might
  51 * be different locks used between the application
  52 * and the low level sockets.
  53 *
  54 * Problems were seen with ksmbd.ko and cifs.ko,
  55 * switching transports, use git blame to find
  56 * more details.
  57 */
  58static struct lock_class_key siw_sk_key[2];
  59static struct lock_class_key siw_slock_key[2];
  60#endif /* CONFIG_DEBUG_LOCK_ALLOC */
  61
  62static inline void siw_reclassify_socket(struct socket *sock)
  63{
  64#ifdef CONFIG_DEBUG_LOCK_ALLOC
  65	struct sock *sk = sock->sk;
  66
  67	if (WARN_ON_ONCE(!sock_allow_reclassification(sk)))
  68		return;
  69
  70	switch (sk->sk_family) {
  71	case AF_INET:
  72		sock_lock_init_class_and_name(sk,
  73					      "slock-AF_INET-RDMA-SIW",
  74					      &siw_slock_key[0],
  75					      "sk_lock-AF_INET-RDMA-SIW",
  76					      &siw_sk_key[0]);
  77		break;
  78	case AF_INET6:
  79		sock_lock_init_class_and_name(sk,
  80					      "slock-AF_INET6-RDMA-SIW",
  81					      &siw_slock_key[1],
  82					      "sk_lock-AF_INET6-RDMA-SIW",
  83					      &siw_sk_key[1]);
  84		break;
  85	default:
  86		WARN_ON_ONCE(1);
  87	}
  88#endif /* CONFIG_DEBUG_LOCK_ALLOC */
  89}
  90
  91static void siw_sk_assign_cm_upcalls(struct sock *sk)
  92{
  93	struct siw_cep *cep = sk_to_cep(sk);
  94
  95	write_lock_bh(&sk->sk_callback_lock);
  96	cep->sk_state_change = sk->sk_state_change;
  97	cep->sk_data_ready = sk->sk_data_ready;
  98	cep->sk_write_space = sk->sk_write_space;
  99	cep->sk_error_report = sk->sk_error_report;
 100
 101	sk->sk_state_change = siw_cm_llp_state_change;
 102	sk->sk_data_ready = siw_cm_llp_data_ready;
 103	sk->sk_write_space = siw_cm_llp_write_space;
 104	sk->sk_error_report = siw_cm_llp_error_report;
 105	write_unlock_bh(&sk->sk_callback_lock);
 106}
 107
 108static void siw_sk_restore_upcalls(struct sock *sk, struct siw_cep *cep)
 109{
 110	sk->sk_state_change = cep->sk_state_change;
 111	sk->sk_data_ready = cep->sk_data_ready;
 112	sk->sk_write_space = cep->sk_write_space;
 113	sk->sk_error_report = cep->sk_error_report;
 114	sk->sk_user_data = NULL;
 115}
 116
 117static void siw_qp_socket_assoc(struct siw_cep *cep, struct siw_qp *qp)
 118{
 119	struct socket *s = cep->sock;
 120	struct sock *sk = s->sk;
 121
 122	write_lock_bh(&sk->sk_callback_lock);
 123
 124	qp->attrs.sk = s;
 125	sk->sk_data_ready = siw_qp_llp_data_ready;
 126	sk->sk_write_space = siw_qp_llp_write_space;
 127
 128	write_unlock_bh(&sk->sk_callback_lock);
 129}
 130
 131static void siw_socket_disassoc(struct socket *s)
 132{
 133	struct sock *sk = s->sk;
 134	struct siw_cep *cep;
 135
 136	if (sk) {
 137		write_lock_bh(&sk->sk_callback_lock);
 138		cep = sk_to_cep(sk);
 139		if (cep) {
 140			siw_sk_restore_upcalls(sk, cep);
 141			siw_cep_put(cep);
 142		} else {
 143			pr_warn("siw: cannot restore sk callbacks: no ep\n");
 144		}
 145		write_unlock_bh(&sk->sk_callback_lock);
 146	} else {
 147		pr_warn("siw: cannot restore sk callbacks: no sk\n");
 148	}
 149}
 150
 151static void siw_rtr_data_ready(struct sock *sk)
 152{
 153	struct siw_cep *cep;
 154	struct siw_qp *qp = NULL;
 155	read_descriptor_t rd_desc;
 156
 157	trace_sk_data_ready(sk);
 158
 159	read_lock(&sk->sk_callback_lock);
 160
 161	cep = sk_to_cep(sk);
 162	if (!cep) {
 163		WARN(1, "No connection endpoint\n");
 164		goto out;
 165	}
 166	qp = sk_to_qp(sk);
 167
 168	memset(&rd_desc, 0, sizeof(rd_desc));
 169	rd_desc.arg.data = qp;
 170	rd_desc.count = 1;
 171
 172	tcp_read_sock(sk, &rd_desc, siw_tcp_rx_data);
 173	/*
 174	 * Check if first frame was successfully processed.
 175	 * Signal connection full establishment if yes.
 176	 * Failed data processing would have already scheduled
 177	 * connection drop.
 178	 */
 179	if (!qp->rx_stream.rx_suspend)
 180		siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
 181out:
 182	read_unlock(&sk->sk_callback_lock);
 183	if (qp)
 184		siw_qp_socket_assoc(cep, qp);
 185}
 186
 187static void siw_sk_assign_rtr_upcalls(struct siw_cep *cep)
 188{
 189	struct sock *sk = cep->sock->sk;
 190
 191	write_lock_bh(&sk->sk_callback_lock);
 192	sk->sk_data_ready = siw_rtr_data_ready;
 193	sk->sk_write_space = siw_qp_llp_write_space;
 194	write_unlock_bh(&sk->sk_callback_lock);
 195}
 196
 197static void siw_cep_socket_assoc(struct siw_cep *cep, struct socket *s)
 198{
 199	cep->sock = s;
 200	siw_cep_get(cep);
 201	s->sk->sk_user_data = cep;
 202
 203	siw_sk_assign_cm_upcalls(s->sk);
 204}
 205
 206static struct siw_cep *siw_cep_alloc(struct siw_device *sdev)
 207{
 208	struct siw_cep *cep = kzalloc(sizeof(*cep), GFP_KERNEL);
 209	unsigned long flags;
 210
 211	if (!cep)
 212		return NULL;
 213
 214	INIT_LIST_HEAD(&cep->listenq);
 215	INIT_LIST_HEAD(&cep->devq);
 216	INIT_LIST_HEAD(&cep->work_freelist);
 217
 218	kref_init(&cep->ref);
 219	cep->state = SIW_EPSTATE_IDLE;
 220	init_waitqueue_head(&cep->waitq);
 221	spin_lock_init(&cep->lock);
 222	cep->sdev = sdev;
 223	cep->enhanced_rdma_conn_est = false;
 224
 225	spin_lock_irqsave(&sdev->lock, flags);
 226	list_add_tail(&cep->devq, &sdev->cep_list);
 227	spin_unlock_irqrestore(&sdev->lock, flags);
 228
 229	siw_dbg_cep(cep, "new endpoint\n");
 230	return cep;
 231}
 232
 233static void siw_cm_free_work(struct siw_cep *cep)
 234{
 235	struct list_head *w, *tmp;
 236	struct siw_cm_work *work;
 237
 238	list_for_each_safe(w, tmp, &cep->work_freelist) {
 239		work = list_entry(w, struct siw_cm_work, list);
 240		list_del(&work->list);
 241		kfree(work);
 242	}
 243}
 244
 245static void siw_cancel_mpatimer(struct siw_cep *cep)
 246{
 247	spin_lock_bh(&cep->lock);
 248	if (cep->mpa_timer) {
 249		if (cancel_delayed_work(&cep->mpa_timer->work)) {
 250			siw_cep_put(cep);
 251			kfree(cep->mpa_timer); /* not needed again */
 252		}
 253		cep->mpa_timer = NULL;
 254	}
 255	spin_unlock_bh(&cep->lock);
 256}
 257
 258static void siw_put_work(struct siw_cm_work *work)
 259{
 260	INIT_LIST_HEAD(&work->list);
 261	spin_lock_bh(&work->cep->lock);
 262	list_add(&work->list, &work->cep->work_freelist);
 263	spin_unlock_bh(&work->cep->lock);
 264}
 265
 266static void siw_cep_set_inuse(struct siw_cep *cep)
 267{
 268	unsigned long flags;
 269retry:
 270	spin_lock_irqsave(&cep->lock, flags);
 271
 272	if (cep->in_use) {
 273		spin_unlock_irqrestore(&cep->lock, flags);
 274		wait_event_interruptible(cep->waitq, !cep->in_use);
 275		if (signal_pending(current))
 276			flush_signals(current);
 277		goto retry;
 278	} else {
 279		cep->in_use = 1;
 280		spin_unlock_irqrestore(&cep->lock, flags);
 281	}
 282}
 283
 284static void siw_cep_set_free(struct siw_cep *cep)
 285{
 286	unsigned long flags;
 287
 288	spin_lock_irqsave(&cep->lock, flags);
 289	cep->in_use = 0;
 290	spin_unlock_irqrestore(&cep->lock, flags);
 291
 292	wake_up(&cep->waitq);
 293}
 294
 295static void __siw_cep_dealloc(struct kref *ref)
 296{
 297	struct siw_cep *cep = container_of(ref, struct siw_cep, ref);
 298	struct siw_device *sdev = cep->sdev;
 299	unsigned long flags;
 300
 301	WARN_ON(cep->listen_cep);
 302
 303	/* kfree(NULL) is safe */
 304	kfree(cep->mpa.pdata);
 305	spin_lock_bh(&cep->lock);
 306	if (!list_empty(&cep->work_freelist))
 307		siw_cm_free_work(cep);
 308	spin_unlock_bh(&cep->lock);
 309
 310	spin_lock_irqsave(&sdev->lock, flags);
 311	list_del(&cep->devq);
 312	spin_unlock_irqrestore(&sdev->lock, flags);
 313
 314	siw_dbg_cep(cep, "free endpoint\n");
 315	kfree(cep);
 316}
 317
 318static struct siw_cm_work *siw_get_work(struct siw_cep *cep)
 319{
 320	struct siw_cm_work *work = NULL;
 321
 322	spin_lock_bh(&cep->lock);
 323	if (!list_empty(&cep->work_freelist)) {
 324		work = list_entry(cep->work_freelist.next, struct siw_cm_work,
 325				  list);
 326		list_del_init(&work->list);
 327	}
 328	spin_unlock_bh(&cep->lock);
 329	return work;
 330}
 331
 332static int siw_cm_alloc_work(struct siw_cep *cep, int num)
 333{
 334	struct siw_cm_work *work;
 335
 336	while (num--) {
 337		work = kmalloc(sizeof(*work), GFP_KERNEL);
 338		if (!work) {
 339			if (!(list_empty(&cep->work_freelist)))
 340				siw_cm_free_work(cep);
 341			return -ENOMEM;
 342		}
 343		work->cep = cep;
 344		INIT_LIST_HEAD(&work->list);
 345		list_add(&work->list, &cep->work_freelist);
 346	}
 347	return 0;
 348}
 349
 350/*
 351 * siw_cm_upcall()
 352 *
 353 * Upcall to IWCM to inform about async connection events
 354 */
 355static int siw_cm_upcall(struct siw_cep *cep, enum iw_cm_event_type reason,
 356			 int status)
 357{
 358	struct iw_cm_event event;
 359	struct iw_cm_id *id;
 360
 361	memset(&event, 0, sizeof(event));
 362	event.status = status;
 363	event.event = reason;
 364
 365	if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
 366		event.provider_data = cep;
 367		id = cep->listen_cep->cm_id;
 368	} else {
 369		id = cep->cm_id;
 370	}
 371	/* Signal IRD and ORD */
 372	if (reason == IW_CM_EVENT_ESTABLISHED ||
 373	    reason == IW_CM_EVENT_CONNECT_REPLY) {
 374		/* Signal negotiated IRD/ORD values we will use */
 375		event.ird = cep->ird;
 376		event.ord = cep->ord;
 377	} else if (reason == IW_CM_EVENT_CONNECT_REQUEST) {
 378		event.ird = cep->ord;
 379		event.ord = cep->ird;
 380	}
 381	/* Signal private data and address information */
 382	if (reason == IW_CM_EVENT_CONNECT_REQUEST ||
 383	    reason == IW_CM_EVENT_CONNECT_REPLY) {
 384		u16 pd_len = be16_to_cpu(cep->mpa.hdr.params.pd_len);
 385
 386		if (pd_len) {
 387			/*
 388			 * hand over MPA private data
 389			 */
 390			event.private_data_len = pd_len;
 391			event.private_data = cep->mpa.pdata;
 392
 393			/* Hide MPA V2 IRD/ORD control */
 394			if (cep->enhanced_rdma_conn_est) {
 395				event.private_data_len -=
 396					sizeof(struct mpa_v2_data);
 397				event.private_data +=
 398					sizeof(struct mpa_v2_data);
 399			}
 400		}
 401		getname_local(cep->sock, &event.local_addr);
 402		getname_peer(cep->sock, &event.remote_addr);
 403	}
 404	siw_dbg_cep(cep, "[QP %u]: reason=%d, status=%d\n",
 405		    cep->qp ? qp_id(cep->qp) : UINT_MAX, reason, status);
 406
 407	return id->event_handler(id, &event);
 408}
 409
 410static void siw_free_cm_id(struct siw_cep *cep)
 411{
 412	if (!cep->cm_id)
 413		return;
 414
 415	cep->cm_id->rem_ref(cep->cm_id);
 416	cep->cm_id = NULL;
 417}
 418
 419static void siw_destroy_cep_sock(struct siw_cep *cep)
 420{
 421	if (cep->sock) {
 422		siw_socket_disassoc(cep->sock);
 423		sock_release(cep->sock);
 424		cep->sock = NULL;
 425	}
 426}
 427
 428/*
 429 * siw_qp_cm_drop()
 430 *
 431 * Drops established LLP connection if present and not already
 432 * scheduled for dropping. Called from user context, SQ workqueue
 433 * or receive IRQ. Caller signals if socket can be immediately
 434 * closed (basically, if not in IRQ).
 435 */
 436void siw_qp_cm_drop(struct siw_qp *qp, int schedule)
 437{
 438	struct siw_cep *cep = qp->cep;
 439
 440	qp->rx_stream.rx_suspend = 1;
 441	qp->tx_ctx.tx_suspend = 1;
 442
 443	if (!qp->cep)
 444		return;
 445
 446	if (schedule) {
 447		siw_cm_queue_work(cep, SIW_CM_WORK_CLOSE_LLP);
 448	} else {
 449		siw_cep_set_inuse(cep);
 450
 451		if (cep->state == SIW_EPSTATE_CLOSED) {
 452			siw_dbg_cep(cep, "already closed\n");
 453			goto out;
 454		}
 455		siw_dbg_cep(cep, "immediate close, state %d\n", cep->state);
 456
 457		siw_send_terminate(qp);
 458
 459		if (cep->cm_id) {
 460			switch (cep->state) {
 461			case SIW_EPSTATE_AWAIT_MPAREP:
 462				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
 463					      -EINVAL);
 464				break;
 465
 466			case SIW_EPSTATE_RDMA_MODE:
 467				siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
 468				break;
 469
 470			case SIW_EPSTATE_IDLE:
 471			case SIW_EPSTATE_LISTENING:
 472			case SIW_EPSTATE_CONNECTING:
 473			case SIW_EPSTATE_AWAIT_MPAREQ:
 474			case SIW_EPSTATE_RECVD_MPAREQ:
 475			case SIW_EPSTATE_CLOSED:
 476			default:
 477				break;
 478			}
 479			siw_free_cm_id(cep);
 480			siw_cep_put(cep);
 481		}
 482		cep->state = SIW_EPSTATE_CLOSED;
 483
 484		siw_destroy_cep_sock(cep);
 485		if (cep->qp) {
 486			cep->qp = NULL;
 487			siw_qp_put(qp);
 488		}
 489out:
 490		siw_cep_set_free(cep);
 491	}
 492}
 493
 494void siw_cep_put(struct siw_cep *cep)
 495{
 496	WARN_ON(kref_read(&cep->ref) < 1);
 497	kref_put(&cep->ref, __siw_cep_dealloc);
 498}
 499
 500static void siw_cep_set_free_and_put(struct siw_cep *cep)
 501{
 502	siw_cep_set_free(cep);
 503	siw_cep_put(cep);
 504}
 505
 506void siw_cep_get(struct siw_cep *cep)
 507{
 508	kref_get(&cep->ref);
 509}
 510
 511/*
 512 * Expects params->pd_len in host byte order
 513 */
 514static int siw_send_mpareqrep(struct siw_cep *cep, const void *pdata, u8 pd_len)
 515{
 516	struct socket *s = cep->sock;
 517	struct mpa_rr *rr = &cep->mpa.hdr;
 518	struct kvec iov[3];
 519	struct msghdr msg;
 520	int rv;
 521	int iovec_num = 0;
 522	int mpa_len;
 523
 524	memset(&msg, 0, sizeof(msg));
 525
 526	iov[iovec_num].iov_base = rr;
 527	iov[iovec_num].iov_len = sizeof(*rr);
 528	mpa_len = sizeof(*rr);
 529
 530	if (cep->enhanced_rdma_conn_est) {
 531		iovec_num++;
 532		iov[iovec_num].iov_base = &cep->mpa.v2_ctrl;
 533		iov[iovec_num].iov_len = sizeof(cep->mpa.v2_ctrl);
 534		mpa_len += sizeof(cep->mpa.v2_ctrl);
 535	}
 536	if (pd_len) {
 537		iovec_num++;
 538		iov[iovec_num].iov_base = (char *)pdata;
 539		iov[iovec_num].iov_len = pd_len;
 540		mpa_len += pd_len;
 541	}
 542	if (cep->enhanced_rdma_conn_est)
 543		pd_len += sizeof(cep->mpa.v2_ctrl);
 544
 545	rr->params.pd_len = cpu_to_be16(pd_len);
 546
 547	rv = kernel_sendmsg(s, &msg, iov, iovec_num + 1, mpa_len);
 548
 549	return rv < 0 ? rv : 0;
 550}
 551
 552/*
 553 * Receive MPA Request/Reply header.
 554 *
 555 * Returns 0 if complete MPA Request/Reply header including
 556 * eventual private data was received. Returns -EAGAIN if
 557 * header was partially received or negative error code otherwise.
 558 *
 559 * Context: May be called in process context only
 560 */
 561static int siw_recv_mpa_rr(struct siw_cep *cep)
 562{
 563	struct mpa_rr *hdr = &cep->mpa.hdr;
 564	struct socket *s = cep->sock;
 565	u16 pd_len;
 566	int rcvd, to_rcv;
 567
 568	if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr)) {
 569		rcvd = ksock_recv(s, (char *)hdr + cep->mpa.bytes_rcvd,
 570				  sizeof(struct mpa_rr) - cep->mpa.bytes_rcvd,
 571				  0);
 572		if (rcvd <= 0)
 573			return -ECONNABORTED;
 574
 575		cep->mpa.bytes_rcvd += rcvd;
 576
 577		if (cep->mpa.bytes_rcvd < sizeof(struct mpa_rr))
 578			return -EAGAIN;
 579
 580		if (be16_to_cpu(hdr->params.pd_len) > MPA_MAX_PRIVDATA)
 581			return -EPROTO;
 582	}
 583	pd_len = be16_to_cpu(hdr->params.pd_len);
 584
 585	/*
 586	 * At least the MPA Request/Reply header (frame not including
 587	 * private data) has been received.
 588	 * Receive (or continue receiving) any private data.
 589	 */
 590	to_rcv = pd_len - (cep->mpa.bytes_rcvd - sizeof(struct mpa_rr));
 591
 592	if (!to_rcv) {
 593		/*
 594		 * We must have hdr->params.pd_len == 0 and thus received a
 595		 * complete MPA Request/Reply frame.
 596		 * Check against peer protocol violation.
 597		 */
 598		u32 word;
 599
 600		rcvd = ksock_recv(s, (char *)&word, sizeof(word), MSG_DONTWAIT);
 601		if (rcvd == -EAGAIN)
 602			return 0;
 603
 604		if (rcvd == 0) {
 605			siw_dbg_cep(cep, "peer EOF\n");
 606			return -EPIPE;
 607		}
 608		if (rcvd < 0) {
 609			siw_dbg_cep(cep, "error: %d\n", rcvd);
 610			return rcvd;
 611		}
 612		siw_dbg_cep(cep, "peer sent extra data: %d\n", rcvd);
 613
 614		return -EPROTO;
 615	}
 616
 617	/*
 618	 * At this point, we must have hdr->params.pd_len != 0.
 619	 * A private data buffer gets allocated if hdr->params.pd_len != 0.
 620	 */
 621	if (!cep->mpa.pdata) {
 622		cep->mpa.pdata = kmalloc(pd_len + 4, GFP_KERNEL);
 623		if (!cep->mpa.pdata)
 624			return -ENOMEM;
 625	}
 626	rcvd = ksock_recv(
 627		s, cep->mpa.pdata + cep->mpa.bytes_rcvd - sizeof(struct mpa_rr),
 628		to_rcv + 4, MSG_DONTWAIT);
 629
 630	if (rcvd < 0)
 631		return rcvd;
 632
 633	if (rcvd > to_rcv)
 634		return -EPROTO;
 635
 636	cep->mpa.bytes_rcvd += rcvd;
 637
 638	if (to_rcv == rcvd) {
 639		siw_dbg_cep(cep, "%d bytes private data received\n", pd_len);
 640		return 0;
 641	}
 642	return -EAGAIN;
 643}
 644
 645/*
 646 * siw_proc_mpareq()
 647 *
 648 * Read MPA Request from socket and signal new connection to IWCM
 649 * if success. Caller must hold lock on corresponding listening CEP.
 650 */
 651static int siw_proc_mpareq(struct siw_cep *cep)
 652{
 653	struct mpa_rr *req;
 654	int version, rv;
 655	u16 pd_len;
 656
 657	rv = siw_recv_mpa_rr(cep);
 658	if (rv)
 659		return rv;
 660
 661	req = &cep->mpa.hdr;
 662
 663	version = __mpa_rr_revision(req->params.bits);
 664	pd_len = be16_to_cpu(req->params.pd_len);
 665
 666	if (version > MPA_REVISION_2)
 667		/* allow for 0, 1, and 2 only */
 668		return -EPROTO;
 669
 670	if (memcmp(req->key, MPA_KEY_REQ, 16))
 671		return -EPROTO;
 672
 673	/* Prepare for sending MPA reply */
 674	memcpy(req->key, MPA_KEY_REP, 16);
 675
 676	if (version == MPA_REVISION_2 &&
 677	    (req->params.bits & MPA_RR_FLAG_ENHANCED)) {
 678		/*
 679		 * MPA version 2 must signal IRD/ORD values and P2P mode
 680		 * in private data if header flag MPA_RR_FLAG_ENHANCED
 681		 * is set.
 682		 */
 683		if (pd_len < sizeof(struct mpa_v2_data))
 684			goto reject_conn;
 685
 686		cep->enhanced_rdma_conn_est = true;
 687	}
 688
 689	/* MPA Markers: currently not supported. Marker TX to be added. */
 690	if (req->params.bits & MPA_RR_FLAG_MARKERS)
 691		goto reject_conn;
 692
 693	if (req->params.bits & MPA_RR_FLAG_CRC) {
 694		/*
 695		 * RFC 5044, page 27: CRC MUST be used if peer requests it.
 696		 * siw specific: 'mpa_crc_strict' parameter to reject
 697		 * connection with CRC if local CRC off enforced by
 698		 * 'mpa_crc_strict' module parameter.
 699		 */
 700		if (!mpa_crc_required && mpa_crc_strict)
 701			goto reject_conn;
 702
 703		/* Enable CRC if requested by module parameter */
 704		if (mpa_crc_required)
 705			req->params.bits |= MPA_RR_FLAG_CRC;
 706	}
 707	if (cep->enhanced_rdma_conn_est) {
 708		struct mpa_v2_data *v2 = (struct mpa_v2_data *)cep->mpa.pdata;
 709
 710		/*
 711		 * Peer requested ORD becomes requested local IRD,
 712		 * peer requested IRD becomes requested local ORD.
 713		 * IRD and ORD get limited by global maximum values.
 714		 */
 715		cep->ord = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
 716		cep->ord = min(cep->ord, SIW_MAX_ORD_QP);
 717		cep->ird = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
 718		cep->ird = min(cep->ird, SIW_MAX_IRD_QP);
 719
 720		/* May get overwritten by locally negotiated values */
 721		cep->mpa.v2_ctrl.ird = htons(cep->ird);
 722		cep->mpa.v2_ctrl.ord = htons(cep->ord);
 723
 724		/*
 725		 * Support for peer sent zero length Write or Read to
 726		 * let local side enter RTS. Writes are preferred.
 727		 * Sends would require pre-posting a Receive and are
 728		 * not supported.
 729		 * Propose zero length Write if none of Read and Write
 730		 * is indicated.
 731		 */
 732		if (v2->ird & MPA_V2_PEER_TO_PEER) {
 733			cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
 734
 735			if (v2->ord & MPA_V2_RDMA_WRITE_RTR)
 736				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
 737			else if (v2->ord & MPA_V2_RDMA_READ_RTR)
 738				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_READ_RTR;
 739			else
 740				cep->mpa.v2_ctrl.ord |= MPA_V2_RDMA_WRITE_RTR;
 741		}
 742	}
 743
 744	cep->state = SIW_EPSTATE_RECVD_MPAREQ;
 745
 746	/* Keep reference until IWCM accepts/rejects */
 747	siw_cep_get(cep);
 748	rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REQUEST, 0);
 749	if (rv)
 750		siw_cep_put(cep);
 751
 752	return rv;
 753
 754reject_conn:
 755	siw_dbg_cep(cep, "reject: crc %d:%d:%d, m %d:%d\n",
 756		    req->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
 757		    mpa_crc_required, mpa_crc_strict,
 758		    req->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
 759
 760	req->params.bits &= ~MPA_RR_FLAG_MARKERS;
 761	req->params.bits |= MPA_RR_FLAG_REJECT;
 762
 763	if (!mpa_crc_required && mpa_crc_strict)
 764		req->params.bits &= ~MPA_RR_FLAG_CRC;
 765
 766	if (pd_len)
 767		kfree(cep->mpa.pdata);
 768
 769	cep->mpa.pdata = NULL;
 770
 771	siw_send_mpareqrep(cep, NULL, 0);
 772
 773	return -EOPNOTSUPP;
 774}
 775
 776static int siw_proc_mpareply(struct siw_cep *cep)
 777{
 778	struct siw_qp_attrs qp_attrs;
 779	enum siw_qp_attr_mask qp_attr_mask;
 780	struct siw_qp *qp = cep->qp;
 781	struct mpa_rr *rep;
 782	int rv;
 783	u16 rep_ord;
 784	u16 rep_ird;
 785	bool ird_insufficient = false;
 786	enum mpa_v2_ctrl mpa_p2p_mode = MPA_V2_RDMA_NO_RTR;
 787
 788	rv = siw_recv_mpa_rr(cep);
 789	if (rv)
 790		goto out_err;
 791
 792	siw_cancel_mpatimer(cep);
 793
 794	rep = &cep->mpa.hdr;
 795
 796	if (__mpa_rr_revision(rep->params.bits) > MPA_REVISION_2) {
 797		/* allow for 0, 1,  and 2 only */
 798		rv = -EPROTO;
 799		goto out_err;
 800	}
 801	if (memcmp(rep->key, MPA_KEY_REP, 16)) {
 802		siw_init_terminate(qp, TERM_ERROR_LAYER_LLP, LLP_ETYPE_MPA,
 803				   LLP_ECODE_INVALID_REQ_RESP, 0);
 804		siw_send_terminate(qp);
 805		rv = -EPROTO;
 806		goto out_err;
 807	}
 808	if (rep->params.bits & MPA_RR_FLAG_REJECT) {
 809		siw_dbg_cep(cep, "got mpa reject\n");
 810		siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNRESET);
 811
 812		return -ECONNRESET;
 813	}
 814	if (try_gso && rep->params.bits & MPA_RR_FLAG_GSO_EXP) {
 815		siw_dbg_cep(cep, "peer allows GSO on TX\n");
 816		qp->tx_ctx.gso_seg_limit = 0;
 817	}
 818	if ((rep->params.bits & MPA_RR_FLAG_MARKERS) ||
 819	    (mpa_crc_required && !(rep->params.bits & MPA_RR_FLAG_CRC)) ||
 820	    (mpa_crc_strict && !mpa_crc_required &&
 821	     (rep->params.bits & MPA_RR_FLAG_CRC))) {
 822		siw_dbg_cep(cep, "reply unsupp: crc %d:%d:%d, m %d:%d\n",
 823			    rep->params.bits & MPA_RR_FLAG_CRC ? 1 : 0,
 824			    mpa_crc_required, mpa_crc_strict,
 825			    rep->params.bits & MPA_RR_FLAG_MARKERS ? 1 : 0, 0);
 826
 827		siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
 828
 829		return -EINVAL;
 830	}
 831	if (cep->enhanced_rdma_conn_est) {
 832		struct mpa_v2_data *v2;
 833
 834		if (__mpa_rr_revision(rep->params.bits) < MPA_REVISION_2 ||
 835		    !(rep->params.bits & MPA_RR_FLAG_ENHANCED)) {
 836			/*
 837			 * Protocol failure: The responder MUST reply with
 838			 * MPA version 2 and MUST set MPA_RR_FLAG_ENHANCED.
 839			 */
 840			siw_dbg_cep(cep, "mpa reply error: vers %d, enhcd %d\n",
 841				    __mpa_rr_revision(rep->params.bits),
 842				    rep->params.bits & MPA_RR_FLAG_ENHANCED ?
 843					    1 :
 844					    0);
 845
 846			siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
 847				      -ECONNRESET);
 848			return -EINVAL;
 849		}
 850		v2 = (struct mpa_v2_data *)cep->mpa.pdata;
 851		rep_ird = ntohs(v2->ird) & MPA_IRD_ORD_MASK;
 852		rep_ord = ntohs(v2->ord) & MPA_IRD_ORD_MASK;
 853
 854		if (cep->ird < rep_ord &&
 855		    (relaxed_ird_negotiation == false ||
 856		     rep_ord > cep->sdev->attrs.max_ird)) {
 857			siw_dbg_cep(cep, "ird %d, rep_ord %d, max_ord %d\n",
 858				    cep->ird, rep_ord,
 859				    cep->sdev->attrs.max_ord);
 860			ird_insufficient = true;
 861		}
 862		if (cep->ord > rep_ird && relaxed_ird_negotiation == false) {
 863			siw_dbg_cep(cep, "ord %d, rep_ird %d\n", cep->ord,
 864				    rep_ird);
 865			ird_insufficient = true;
 866		}
 867		/*
 868		 * Always report negotiated peer values to user,
 869		 * even if IRD/ORD negotiation failed
 870		 */
 871		cep->ird = rep_ord;
 872		cep->ord = rep_ird;
 873
 874		if (ird_insufficient) {
 875			/*
 876			 * If the initiator IRD is insuffient for the
 877			 * responder ORD, send a TERM.
 878			 */
 879			siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
 880					   LLP_ETYPE_MPA,
 881					   LLP_ECODE_INSUFFICIENT_IRD, 0);
 882			siw_send_terminate(qp);
 883			rv = -ENOMEM;
 884			goto out_err;
 885		}
 886		if (cep->mpa.v2_ctrl_req.ird & MPA_V2_PEER_TO_PEER)
 887			mpa_p2p_mode =
 888				cep->mpa.v2_ctrl_req.ord &
 889				(MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR);
 890
 891		/*
 892		 * Check if we requested P2P mode, and if peer agrees
 893		 */
 894		if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
 895			if ((mpa_p2p_mode & v2->ord) == 0) {
 896				/*
 897				 * We requested RTR mode(s), but the peer
 898				 * did not pick any mode we support.
 899				 */
 900				siw_dbg_cep(cep,
 901					    "rtr mode:  req %2x, got %2x\n",
 902					    mpa_p2p_mode,
 903					    v2->ord & (MPA_V2_RDMA_WRITE_RTR |
 904						       MPA_V2_RDMA_READ_RTR));
 905
 906				siw_init_terminate(qp, TERM_ERROR_LAYER_LLP,
 907						   LLP_ETYPE_MPA,
 908						   LLP_ECODE_NO_MATCHING_RTR,
 909						   0);
 910				siw_send_terminate(qp);
 911				rv = -EPROTO;
 912				goto out_err;
 913			}
 914			mpa_p2p_mode = v2->ord & (MPA_V2_RDMA_WRITE_RTR |
 915						  MPA_V2_RDMA_READ_RTR);
 916		}
 917	}
 918	memset(&qp_attrs, 0, sizeof(qp_attrs));
 919
 920	if (rep->params.bits & MPA_RR_FLAG_CRC)
 921		qp_attrs.flags = SIW_MPA_CRC;
 922
 923	qp_attrs.irq_size = cep->ird;
 924	qp_attrs.orq_size = cep->ord;
 925	qp_attrs.sk = cep->sock;
 926	qp_attrs.state = SIW_QP_STATE_RTS;
 927
 928	qp_attr_mask = SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
 929		       SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD | SIW_QP_ATTR_MPA;
 930
 931	/* Move socket RX/TX under QP control */
 932	down_write(&qp->state_lock);
 933	if (qp->attrs.state > SIW_QP_STATE_RTR) {
 934		rv = -EINVAL;
 935		up_write(&qp->state_lock);
 936		goto out_err;
 937	}
 938	rv = siw_qp_modify(qp, &qp_attrs, qp_attr_mask);
 939
 940	siw_qp_socket_assoc(cep, qp);
 941
 942	up_write(&qp->state_lock);
 943
 944	/* Send extra RDMA frame to trigger peer RTS if negotiated */
 945	if (mpa_p2p_mode != MPA_V2_RDMA_NO_RTR) {
 946		rv = siw_qp_mpa_rts(qp, mpa_p2p_mode);
 947		if (rv)
 948			goto out_err;
 949	}
 950	if (!rv) {
 951		rv = siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, 0);
 952		if (!rv)
 953			cep->state = SIW_EPSTATE_RDMA_MODE;
 954
 955		return 0;
 956	}
 957
 958out_err:
 959	if (rv != -EAGAIN)
 960		siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY, -EINVAL);
 961
 962	return rv;
 963}
 964
 965/*
 966 * siw_accept_newconn - accept an incoming pending connection
 967 *
 968 */
 969static void siw_accept_newconn(struct siw_cep *cep)
 970{
 971	struct socket *s = cep->sock;
 972	struct socket *new_s = NULL;
 973	struct siw_cep *new_cep = NULL;
 974	int rv = 0; /* debug only. should disappear */
 975
 976	if (cep->state != SIW_EPSTATE_LISTENING)
 977		goto error;
 978
 979	new_cep = siw_cep_alloc(cep->sdev);
 980	if (!new_cep)
 981		goto error;
 982
 983	/*
 984	 * 4: Allocate a sufficient number of work elements
 985	 * to allow concurrent handling of local + peer close
 986	 * events, MPA header processing + MPA timeout.
 987	 */
 988	if (siw_cm_alloc_work(new_cep, 4) != 0)
 989		goto error;
 990
 991	/*
 992	 * Copy saved socket callbacks from listening CEP
 993	 * and assign new socket with new CEP
 994	 */
 995	new_cep->sk_state_change = cep->sk_state_change;
 996	new_cep->sk_data_ready = cep->sk_data_ready;
 997	new_cep->sk_write_space = cep->sk_write_space;
 998	new_cep->sk_error_report = cep->sk_error_report;
 999
1000	rv = kernel_accept(s, &new_s, O_NONBLOCK);
1001	if (rv != 0) {
1002		/*
1003		 * Connection already aborted by peer..?
1004		 */
1005		siw_dbg_cep(cep, "kernel_accept() error: %d\n", rv);
1006		goto error;
1007	}
1008	new_cep->sock = new_s;
1009	siw_cep_get(new_cep);
1010	new_s->sk->sk_user_data = new_cep;
1011
1012	if (siw_tcp_nagle == false)
1013		tcp_sock_set_nodelay(new_s->sk);
1014	new_cep->state = SIW_EPSTATE_AWAIT_MPAREQ;
1015
1016	rv = siw_cm_queue_work(new_cep, SIW_CM_WORK_MPATIMEOUT);
1017	if (rv)
1018		goto error;
1019	/*
1020	 * See siw_proc_mpareq() etc. for the use of new_cep->listen_cep.
1021	 */
1022	new_cep->listen_cep = cep;
1023	siw_cep_get(cep);
1024
1025	if (atomic_read(&new_s->sk->sk_rmem_alloc)) {
1026		/*
1027		 * MPA REQ already queued
1028		 */
1029		siw_dbg_cep(cep, "immediate mpa request\n");
1030
1031		siw_cep_set_inuse(new_cep);
1032		rv = siw_proc_mpareq(new_cep);
1033		if (rv != -EAGAIN) {
1034			siw_cep_put(cep);
1035			new_cep->listen_cep = NULL;
1036			if (rv) {
1037				siw_cancel_mpatimer(new_cep);
1038				siw_cep_set_free(new_cep);
1039				goto error;
1040			}
1041		}
1042		siw_cep_set_free(new_cep);
1043	}
1044	return;
1045
1046error:
1047	if (new_cep)
1048		siw_cep_put(new_cep);
1049
1050	if (new_s) {
1051		siw_socket_disassoc(new_s);
1052		sock_release(new_s);
1053		new_cep->sock = NULL;
1054	}
1055	siw_dbg_cep(cep, "error %d\n", rv);
1056}
1057
1058static void siw_cm_work_handler(struct work_struct *w)
1059{
1060	struct siw_cm_work *work;
1061	struct siw_cep *cep;
1062	int release_cep = 0, rv = 0;
1063
1064	work = container_of(w, struct siw_cm_work, work.work);
1065	cep = work->cep;
1066
1067	siw_dbg_cep(cep, "[QP %u]: work type: %d, state %d\n",
1068		    cep->qp ? qp_id(cep->qp) : UINT_MAX,
1069		    work->type, cep->state);
1070
1071	siw_cep_set_inuse(cep);
1072
1073	switch (work->type) {
1074	case SIW_CM_WORK_ACCEPT:
1075		siw_accept_newconn(cep);
1076		break;
1077
1078	case SIW_CM_WORK_READ_MPAHDR:
1079		if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1080			if (cep->listen_cep) {
1081				siw_cep_set_inuse(cep->listen_cep);
1082
1083				if (cep->listen_cep->state ==
1084				    SIW_EPSTATE_LISTENING)
1085					rv = siw_proc_mpareq(cep);
1086				else
1087					rv = -EFAULT;
1088
1089				siw_cep_set_free(cep->listen_cep);
1090
1091				if (rv != -EAGAIN) {
1092					siw_cep_put(cep->listen_cep);
1093					cep->listen_cep = NULL;
1094					if (rv)
1095						siw_cep_put(cep);
1096				}
1097			}
1098		} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1099			rv = siw_proc_mpareply(cep);
1100		} else {
1101			/*
1102			 * CEP already moved out of MPA handshake.
1103			 * any connection management already done.
1104			 * silently ignore the mpa packet.
1105			 */
1106			if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1107				cep->sock->sk->sk_data_ready(cep->sock->sk);
1108				siw_dbg_cep(cep, "already in RDMA mode");
1109			} else {
1110				siw_dbg_cep(cep, "out of state: %d\n",
1111					    cep->state);
1112			}
1113		}
1114		if (rv && rv != -EAGAIN)
1115			release_cep = 1;
1116		break;
1117
1118	case SIW_CM_WORK_CLOSE_LLP:
1119		/*
1120		 * QP scheduled LLP close
1121		 */
1122		if (cep->qp)
1123			siw_send_terminate(cep->qp);
1124
1125		if (cep->cm_id)
1126			siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1127
1128		release_cep = 1;
1129		break;
1130
1131	case SIW_CM_WORK_PEER_CLOSE:
1132		if (cep->cm_id) {
1133			if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1134				/*
1135				 * MPA reply not received, but connection drop
1136				 */
1137				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1138					      -ECONNRESET);
1139			} else if (cep->state == SIW_EPSTATE_RDMA_MODE) {
1140				/*
1141				 * NOTE: IW_CM_EVENT_DISCONNECT is given just
1142				 *       to transition IWCM into CLOSING.
1143				 */
1144				siw_cm_upcall(cep, IW_CM_EVENT_DISCONNECT, 0);
1145				siw_cm_upcall(cep, IW_CM_EVENT_CLOSE, 0);
1146			}
1147			/*
1148			 * for other states there is no connection
1149			 * known to the IWCM.
1150			 */
1151		} else {
1152			if (cep->state == SIW_EPSTATE_RECVD_MPAREQ) {
1153				/*
1154				 * Wait for the ulp/CM to call accept/reject
1155				 */
1156				siw_dbg_cep(cep,
1157					    "mpa req recvd, wait for ULP\n");
1158			} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1159				/*
1160				 * Socket close before MPA request received.
1161				 */
1162				if (cep->listen_cep) {
1163					siw_dbg_cep(cep,
1164						"no mpareq: drop listener\n");
1165					siw_cep_put(cep->listen_cep);
1166					cep->listen_cep = NULL;
1167				}
1168			}
1169		}
1170		release_cep = 1;
1171		break;
1172
1173	case SIW_CM_WORK_MPATIMEOUT:
1174		cep->mpa_timer = NULL;
1175
1176		if (cep->state == SIW_EPSTATE_AWAIT_MPAREP) {
1177			/*
1178			 * MPA request timed out:
1179			 * Hide any partially received private data and signal
1180			 * timeout
1181			 */
1182			cep->mpa.hdr.params.pd_len = 0;
1183
1184			if (cep->cm_id)
1185				siw_cm_upcall(cep, IW_CM_EVENT_CONNECT_REPLY,
1186					      -ETIMEDOUT);
1187			release_cep = 1;
1188
1189		} else if (cep->state == SIW_EPSTATE_AWAIT_MPAREQ) {
1190			/*
1191			 * No MPA request received after peer TCP stream setup.
1192			 */
1193			if (cep->listen_cep) {
1194				siw_cep_put(cep->listen_cep);
1195				cep->listen_cep = NULL;
1196			}
1197			release_cep = 1;
1198		}
1199		break;
1200
1201	default:
1202		WARN(1, "Undefined CM work type: %d\n", work->type);
1203	}
1204	if (release_cep) {
1205		siw_dbg_cep(cep,
1206			    "release: timer=%s, QP[%u]\n",
1207			    cep->mpa_timer ? "y" : "n",
1208			    cep->qp ? qp_id(cep->qp) : UINT_MAX);
1209
1210		siw_cancel_mpatimer(cep);
1211
1212		cep->state = SIW_EPSTATE_CLOSED;
1213
1214		if (cep->qp) {
1215			struct siw_qp *qp = cep->qp;
1216			/*
1217			 * Serialize a potential race with application
1218			 * closing the QP and calling siw_qp_cm_drop()
1219			 */
1220			siw_qp_get(qp);
1221			siw_cep_set_free(cep);
1222
1223			siw_qp_llp_close(qp);
1224			siw_qp_put(qp);
1225
1226			siw_cep_set_inuse(cep);
1227			cep->qp = NULL;
1228			siw_qp_put(qp);
1229		}
1230		if (cep->sock) {
1231			siw_socket_disassoc(cep->sock);
1232			sock_release(cep->sock);
1233			cep->sock = NULL;
1234		}
1235		if (cep->cm_id) {
1236			siw_free_cm_id(cep);
1237			siw_cep_put(cep);
1238		}
1239	}
1240	siw_cep_set_free(cep);
1241	siw_put_work(work);
1242	siw_cep_put(cep);
1243}
1244
1245static struct workqueue_struct *siw_cm_wq;
1246
1247int siw_cm_queue_work(struct siw_cep *cep, enum siw_work_type type)
1248{
1249	struct siw_cm_work *work = siw_get_work(cep);
1250	unsigned long delay = 0;
1251
1252	if (!work) {
1253		siw_dbg_cep(cep, "failed with no work available\n");
1254		return -ENOMEM;
1255	}
1256	work->type = type;
1257	work->cep = cep;
1258
1259	siw_cep_get(cep);
1260
1261	INIT_DELAYED_WORK(&work->work, siw_cm_work_handler);
1262
1263	if (type == SIW_CM_WORK_MPATIMEOUT) {
1264		cep->mpa_timer = work;
1265
1266		if (cep->state == SIW_EPSTATE_AWAIT_MPAREP)
1267			delay = MPAREQ_TIMEOUT;
1268		else
1269			delay = MPAREP_TIMEOUT;
1270	}
1271	siw_dbg_cep(cep, "[QP %u]: work type: %d, timeout %lu\n",
1272		    cep->qp ? qp_id(cep->qp) : -1, type, delay);
1273
1274	queue_delayed_work(siw_cm_wq, &work->work, delay);
1275
1276	return 0;
1277}
1278
1279static void siw_cm_llp_data_ready(struct sock *sk)
1280{
1281	struct siw_cep *cep;
1282
1283	trace_sk_data_ready(sk);
1284
1285	read_lock(&sk->sk_callback_lock);
1286
1287	cep = sk_to_cep(sk);
1288	if (!cep)
1289		goto out;
1290
1291	siw_dbg_cep(cep, "cep state: %d, socket state %d\n",
1292		    cep->state, sk->sk_state);
1293
1294	if (sk->sk_state != TCP_ESTABLISHED)
1295		goto out;
1296
1297	switch (cep->state) {
1298	case SIW_EPSTATE_RDMA_MODE:
1299	case SIW_EPSTATE_LISTENING:
1300		break;
1301
1302	case SIW_EPSTATE_AWAIT_MPAREQ:
1303	case SIW_EPSTATE_AWAIT_MPAREP:
1304		siw_cm_queue_work(cep, SIW_CM_WORK_READ_MPAHDR);
1305		break;
1306
1307	default:
1308		siw_dbg_cep(cep, "unexpected data, state %d\n", cep->state);
1309		break;
1310	}
1311out:
1312	read_unlock(&sk->sk_callback_lock);
1313}
1314
1315static void siw_cm_llp_write_space(struct sock *sk)
1316{
1317	struct siw_cep *cep = sk_to_cep(sk);
1318
1319	if (cep)
1320		siw_dbg_cep(cep, "state: %d\n", cep->state);
1321}
1322
1323static void siw_cm_llp_error_report(struct sock *sk)
1324{
1325	struct siw_cep *cep = sk_to_cep(sk);
1326
1327	if (cep) {
1328		siw_dbg_cep(cep, "error %d, socket state: %d, cep state: %d\n",
1329			    sk->sk_err, sk->sk_state, cep->state);
1330		cep->sk_error_report(sk);
1331	}
1332}
1333
1334static void siw_cm_llp_state_change(struct sock *sk)
1335{
1336	struct siw_cep *cep;
1337	void (*orig_state_change)(struct sock *s);
1338
1339	read_lock(&sk->sk_callback_lock);
1340
1341	cep = sk_to_cep(sk);
1342	if (!cep) {
1343		/* endpoint already disassociated */
1344		read_unlock(&sk->sk_callback_lock);
1345		return;
1346	}
1347	orig_state_change = cep->sk_state_change;
1348
1349	siw_dbg_cep(cep, "state: %d\n", cep->state);
1350
1351	switch (sk->sk_state) {
1352	case TCP_ESTABLISHED:
1353		/*
1354		 * handle accepting socket as special case where only
1355		 * new connection is possible
1356		 */
1357		siw_cm_queue_work(cep, SIW_CM_WORK_ACCEPT);
1358		break;
1359
1360	case TCP_CLOSE:
1361	case TCP_CLOSE_WAIT:
1362		if (cep->qp)
1363			cep->qp->tx_ctx.tx_suspend = 1;
1364		siw_cm_queue_work(cep, SIW_CM_WORK_PEER_CLOSE);
1365		break;
1366
1367	default:
1368		siw_dbg_cep(cep, "unexpected socket state %d\n", sk->sk_state);
1369	}
1370	read_unlock(&sk->sk_callback_lock);
1371	orig_state_change(sk);
1372}
1373
1374static int kernel_bindconnect(struct socket *s, struct sockaddr *laddr,
1375			      struct sockaddr *raddr, bool afonly)
1376{
1377	int rv, flags = 0;
1378	size_t size = laddr->sa_family == AF_INET ?
1379		sizeof(struct sockaddr_in) : sizeof(struct sockaddr_in6);
1380
1381	/*
1382	 * Make address available again asap.
1383	 */
1384	sock_set_reuseaddr(s->sk);
1385
1386	if (afonly) {
1387		rv = ip6_sock_set_v6only(s->sk);
1388		if (rv)
1389			return rv;
1390	}
1391
1392	rv = s->ops->bind(s, (struct sockaddr_unsized *)laddr, size);
1393	if (rv < 0)
1394		return rv;
1395
1396	rv = s->ops->connect(s, (struct sockaddr_unsized *)raddr, size, flags);
1397
1398	return rv < 0 ? rv : 0;
1399}
1400
1401int siw_connect(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1402{
1403	struct siw_device *sdev = to_siw_dev(id->device);
1404	struct siw_qp *qp;
1405	struct siw_cep *cep = NULL;
1406	struct socket *s = NULL;
1407	struct sockaddr *laddr = (struct sockaddr *)&id->local_addr,
1408			*raddr = (struct sockaddr *)&id->remote_addr;
1409	bool p2p_mode = peer_to_peer, v4 = true;
1410	u16 pd_len = params->private_data_len;
1411	int version = mpa_version, rv;
1412
1413	if (pd_len > MPA_MAX_PRIVDATA)
1414		return -EINVAL;
1415
1416	if (params->ird > sdev->attrs.max_ird ||
1417	    params->ord > sdev->attrs.max_ord)
1418		return -ENOMEM;
1419
1420	if (laddr->sa_family == AF_INET6)
1421		v4 = false;
1422	else if (laddr->sa_family != AF_INET)
1423		return -EAFNOSUPPORT;
1424
1425	/*
1426	 * Respect any iwarp port mapping: Use mapped remote address
1427	 * if valid. Local address must not be mapped, since siw
1428	 * uses kernel TCP stack.
1429	 */
1430	if ((v4 && to_sockaddr_in(id->remote_addr).sin_port != 0) ||
1431	     to_sockaddr_in6(id->remote_addr).sin6_port != 0)
1432		raddr = (struct sockaddr *)&id->m_remote_addr;
1433
1434	qp = siw_qp_id2obj(sdev, params->qpn);
1435	if (!qp) {
1436		WARN(1, "[QP %u] does not exist\n", params->qpn);
1437		rv = -EINVAL;
1438		goto error;
1439	}
1440	siw_dbg_qp(qp, "pd_len %d, laddr %pISp, raddr %pISp\n", pd_len, laddr,
1441		   raddr);
1442
1443	rv = sock_create(v4 ? AF_INET : AF_INET6, SOCK_STREAM, IPPROTO_TCP, &s);
1444	if (rv < 0)
1445		goto error;
1446	siw_reclassify_socket(s);
1447
1448	/*
1449	 * NOTE: For simplification, connect() is called in blocking
1450	 * mode. Might be reconsidered for async connection setup at
1451	 * TCP level.
1452	 */
1453	rv = kernel_bindconnect(s, laddr, raddr, id->afonly);
1454	if (rv != 0) {
1455		siw_dbg_qp(qp, "kernel_bindconnect: error %d\n", rv);
1456		goto error;
1457	}
1458	if (siw_tcp_nagle == false)
1459		tcp_sock_set_nodelay(s->sk);
1460	cep = siw_cep_alloc(sdev);
1461	if (!cep) {
1462		rv = -ENOMEM;
1463		goto error;
1464	}
1465	siw_cep_set_inuse(cep);
1466
1467	/* Associate QP with CEP */
1468	siw_cep_get(cep);
1469	qp->cep = cep;
1470
1471	/* siw_qp_get(qp) already done by QP lookup */
1472	cep->qp = qp;
1473
1474	id->add_ref(id);
1475	cep->cm_id = id;
1476
1477	/*
1478	 * 4: Allocate a sufficient number of work elements
1479	 * to allow concurrent handling of local + peer close
1480	 * events, MPA header processing + MPA timeout.
1481	 */
1482	rv = siw_cm_alloc_work(cep, 4);
1483	if (rv != 0) {
1484		rv = -ENOMEM;
1485		goto error;
1486	}
1487	cep->ird = params->ird;
1488	cep->ord = params->ord;
1489
1490	if (p2p_mode && cep->ord == 0)
1491		cep->ord = 1;
1492
1493	cep->state = SIW_EPSTATE_CONNECTING;
1494
1495	/*
1496	 * Associate CEP with socket
1497	 */
1498	siw_cep_socket_assoc(cep, s);
1499
1500	cep->state = SIW_EPSTATE_AWAIT_MPAREP;
1501
1502	/*
1503	 * Set MPA Request bits: CRC if required, no MPA Markers,
1504	 * MPA Rev. according to module parameter 'mpa_version', Key 'Request'.
1505	 */
1506	cep->mpa.hdr.params.bits = 0;
1507	if (version > MPA_REVISION_2) {
1508		pr_warn("Setting MPA version to %u\n", MPA_REVISION_2);
1509		version = MPA_REVISION_2;
1510		/* Adjust also module parameter */
1511		mpa_version = MPA_REVISION_2;
1512	}
1513	__mpa_rr_set_revision(&cep->mpa.hdr.params.bits, version);
1514
1515	if (try_gso)
1516		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_GSO_EXP;
1517
1518	if (mpa_crc_required)
1519		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_CRC;
1520
1521	/*
1522	 * If MPA version == 2:
1523	 * o Include ORD and IRD.
1524	 * o Indicate peer-to-peer mode, if required by module
1525	 *   parameter 'peer_to_peer'.
1526	 */
1527	if (version == MPA_REVISION_2) {
1528		cep->enhanced_rdma_conn_est = true;
1529		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_ENHANCED;
1530
1531		cep->mpa.v2_ctrl.ird = htons(cep->ird);
1532		cep->mpa.v2_ctrl.ord = htons(cep->ord);
1533
1534		if (p2p_mode) {
1535			cep->mpa.v2_ctrl.ird |= MPA_V2_PEER_TO_PEER;
1536			cep->mpa.v2_ctrl.ord |= rtr_type;
1537		}
1538		/* Remember own P2P mode requested */
1539		cep->mpa.v2_ctrl_req.ird = cep->mpa.v2_ctrl.ird;
1540		cep->mpa.v2_ctrl_req.ord = cep->mpa.v2_ctrl.ord;
1541	}
1542	memcpy(cep->mpa.hdr.key, MPA_KEY_REQ, 16);
1543
1544	rv = siw_send_mpareqrep(cep, params->private_data, pd_len);
1545	/*
1546	 * Reset private data.
1547	 */
1548	cep->mpa.hdr.params.pd_len = 0;
1549
1550	if (rv >= 0) {
1551		rv = siw_cm_queue_work(cep, SIW_CM_WORK_MPATIMEOUT);
1552		if (!rv) {
1553			siw_dbg_cep(cep, "[QP %u]: exit\n", qp_id(qp));
1554			siw_cep_set_free(cep);
1555			return 0;
1556		}
1557	}
1558error:
1559	siw_dbg(id->device, "failed: %d\n", rv);
1560
1561	if (cep) {
1562		siw_socket_disassoc(s);
1563		sock_release(s);
1564		cep->sock = NULL;
1565
1566		cep->qp = NULL;
1567
1568		cep->cm_id = NULL;
1569		id->rem_ref(id);
1570
1571		qp->cep = NULL;
1572		siw_cep_put(cep);
1573
1574		cep->state = SIW_EPSTATE_CLOSED;
1575
1576		siw_cep_set_free_and_put(cep);
1577
1578	} else if (s) {
1579		sock_release(s);
1580	}
1581	if (qp)
1582		siw_qp_put(qp);
1583
1584	return rv;
1585}
1586
1587/*
1588 * siw_accept - Let SoftiWARP accept an RDMA connection request
1589 *
1590 * @id:		New connection management id to be used for accepted
1591 *		connection request
1592 * @params:	Connection parameters provided by ULP for accepting connection
1593 *
1594 * Transition QP to RTS state, associate new CM id @id with accepted CEP
1595 * and get prepared for TCP input by installing socket callbacks.
1596 * Then send MPA Reply and generate the "connection established" event.
1597 * Socket callbacks must be installed before sending MPA Reply, because
1598 * the latter may cause a first RDMA message to arrive from the RDMA Initiator
1599 * side very quickly, at which time the socket callbacks must be ready.
1600 */
1601int siw_accept(struct iw_cm_id *id, struct iw_cm_conn_param *params)
1602{
1603	struct siw_device *sdev = to_siw_dev(id->device);
1604	struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1605	struct siw_qp *qp;
1606	struct siw_qp_attrs qp_attrs;
1607	int rv = -EINVAL, max_priv_data = MPA_MAX_PRIVDATA;
1608	bool wait_for_peer_rts = false;
1609
1610	siw_cep_set_inuse(cep);
1611	siw_cep_put(cep);
1612
1613	/* Free lingering inbound private data */
1614	if (cep->mpa.hdr.params.pd_len) {
1615		cep->mpa.hdr.params.pd_len = 0;
1616		kfree(cep->mpa.pdata);
1617		cep->mpa.pdata = NULL;
1618	}
1619	siw_cancel_mpatimer(cep);
1620
1621	if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1622		siw_dbg_cep(cep, "out of state\n");
1623		rv = -ECONNRESET;
1624		goto free_cep;
1625	}
1626	qp = siw_qp_id2obj(sdev, params->qpn);
1627	if (!qp) {
1628		WARN(1, "[QP %d] does not exist\n", params->qpn);
1629		goto free_cep;
1630	}
1631	down_write(&qp->state_lock);
1632	if (qp->attrs.state > SIW_QP_STATE_RTR)
1633		goto error_unlock;
1634	siw_dbg_cep(cep, "[QP %d]\n", params->qpn);
1635
1636	if (try_gso && cep->mpa.hdr.params.bits & MPA_RR_FLAG_GSO_EXP) {
1637		siw_dbg_cep(cep, "peer allows GSO on TX\n");
1638		qp->tx_ctx.gso_seg_limit = 0;
1639	}
1640	if (params->ord > sdev->attrs.max_ord ||
1641	    params->ird > sdev->attrs.max_ird) {
1642		siw_dbg_cep(
1643			cep,
1644			"[QP %u]: ord %d (max %d), ird %d (max %d)\n",
1645			qp_id(qp), params->ord, sdev->attrs.max_ord,
1646			params->ird, sdev->attrs.max_ird);
1647		goto error_unlock;
1648	}
1649	if (cep->enhanced_rdma_conn_est)
1650		max_priv_data -= sizeof(struct mpa_v2_data);
1651
1652	if (params->private_data_len > max_priv_data) {
1653		siw_dbg_cep(
1654			cep,
1655			"[QP %u]: private data length: %d (max %d)\n",
1656			qp_id(qp), params->private_data_len, max_priv_data);
1657		goto error_unlock;
1658	}
1659	if (cep->enhanced_rdma_conn_est) {
1660		if (params->ord > cep->ord) {
1661			if (relaxed_ird_negotiation) {
1662				params->ord = cep->ord;
1663			} else {
1664				cep->ird = params->ird;
1665				cep->ord = params->ord;
1666				goto error_unlock;
1667			}
1668		}
1669		if (params->ird < cep->ird) {
1670			if (relaxed_ird_negotiation &&
1671			    cep->ird <= sdev->attrs.max_ird)
1672				params->ird = cep->ird;
1673			else {
1674				rv = -ENOMEM;
1675				goto error_unlock;
1676			}
1677		}
1678		if (cep->mpa.v2_ctrl.ord &
1679		    (MPA_V2_RDMA_WRITE_RTR | MPA_V2_RDMA_READ_RTR))
1680			wait_for_peer_rts = true;
1681		/*
1682		 * Signal back negotiated IRD and ORD values
1683		 */
1684		cep->mpa.v2_ctrl.ord =
1685			htons(params->ord & MPA_IRD_ORD_MASK) |
1686			(cep->mpa.v2_ctrl.ord & ~MPA_V2_MASK_IRD_ORD);
1687		cep->mpa.v2_ctrl.ird =
1688			htons(params->ird & MPA_IRD_ORD_MASK) |
1689			(cep->mpa.v2_ctrl.ird & ~MPA_V2_MASK_IRD_ORD);
1690	}
1691	cep->ird = params->ird;
1692	cep->ord = params->ord;
1693
1694	cep->cm_id = id;
1695	id->add_ref(id);
1696
1697	memset(&qp_attrs, 0, sizeof(qp_attrs));
1698	qp_attrs.orq_size = cep->ord;
1699	qp_attrs.irq_size = cep->ird;
1700	qp_attrs.sk = cep->sock;
1701	if (cep->mpa.hdr.params.bits & MPA_RR_FLAG_CRC)
1702		qp_attrs.flags = SIW_MPA_CRC;
1703	qp_attrs.state = SIW_QP_STATE_RTS;
1704
1705	siw_dbg_cep(cep, "[QP%u]: moving to rts\n", qp_id(qp));
1706
1707	/* Associate QP with CEP */
1708	siw_cep_get(cep);
1709	qp->cep = cep;
1710
1711	/* siw_qp_get(qp) already done by QP lookup */
1712	cep->qp = qp;
1713
1714	cep->state = SIW_EPSTATE_RDMA_MODE;
1715
1716	/* Move socket RX/TX under QP control */
1717	rv = siw_qp_modify(qp, &qp_attrs,
1718			   SIW_QP_ATTR_STATE | SIW_QP_ATTR_LLP_HANDLE |
1719				   SIW_QP_ATTR_ORD | SIW_QP_ATTR_IRD |
1720				   SIW_QP_ATTR_MPA);
1721	up_write(&qp->state_lock);
1722	if (rv)
1723		goto error;
1724
1725	siw_dbg_cep(cep, "[QP %u]: send mpa reply, %d byte pdata\n",
1726		    qp_id(qp), params->private_data_len);
1727
1728	rv = siw_send_mpareqrep(cep, params->private_data,
1729				params->private_data_len);
1730	if (rv != 0)
1731		goto error;
1732
1733	if (wait_for_peer_rts) {
1734		siw_sk_assign_rtr_upcalls(cep);
1735	} else {
1736		siw_qp_socket_assoc(cep, qp);
1737		rv = siw_cm_upcall(cep, IW_CM_EVENT_ESTABLISHED, 0);
1738		if (rv)
1739			goto error;
1740	}
1741	siw_cep_set_free(cep);
1742
1743	return 0;
1744
1745error_unlock:
1746	up_write(&qp->state_lock);
1747error:
1748	siw_destroy_cep_sock(cep);
1749
1750	cep->state = SIW_EPSTATE_CLOSED;
1751
1752	siw_free_cm_id(cep);
1753	if (qp->cep) {
1754		siw_cep_put(cep);
1755		qp->cep = NULL;
1756	}
1757	cep->qp = NULL;
1758	siw_qp_put(qp);
1759free_cep:
1760	siw_cep_set_free_and_put(cep);
1761	return rv;
1762}
1763
1764/*
1765 * siw_reject()
1766 *
1767 * Local connection reject case. Send private data back to peer,
1768 * close connection and dereference connection id.
1769 */
1770int siw_reject(struct iw_cm_id *id, const void *pdata, u8 pd_len)
1771{
1772	struct siw_cep *cep = (struct siw_cep *)id->provider_data;
1773
1774	siw_cep_set_inuse(cep);
1775	siw_cep_put(cep);
1776
1777	siw_cancel_mpatimer(cep);
1778
1779	if (cep->state != SIW_EPSTATE_RECVD_MPAREQ) {
1780		siw_dbg_cep(cep, "out of state\n");
1781
1782		siw_cep_set_free_and_put(cep); /* put last reference */
1783
1784		return -ECONNRESET;
1785	}
1786	siw_dbg_cep(cep, "cep->state %d, pd_len %d\n", cep->state,
1787		    pd_len);
1788
1789	if (__mpa_rr_revision(cep->mpa.hdr.params.bits) >= MPA_REVISION_1) {
1790		cep->mpa.hdr.params.bits |= MPA_RR_FLAG_REJECT; /* reject */
1791		siw_send_mpareqrep(cep, pdata, pd_len);
1792	}
1793	siw_destroy_cep_sock(cep);
1794
1795	cep->state = SIW_EPSTATE_CLOSED;
1796
1797	siw_cep_set_free_and_put(cep);
1798
1799	return 0;
1800}
1801
1802/*
1803 * siw_create_listen - Create resources for a listener's IWCM ID @id
1804 *
1805 * Starts listen on the socket address id->local_addr.
1806 *
1807 */
1808int siw_create_listen(struct iw_cm_id *id, int backlog)
1809{
1810	struct socket *s;
1811	struct siw_cep *cep = NULL;
1812	struct net_device *ndev = NULL;
1813	struct siw_device *sdev = to_siw_dev(id->device);
1814	int addr_family = id->local_addr.ss_family;
1815	int rv = 0;
1816
1817	if (addr_family != AF_INET && addr_family != AF_INET6)
1818		return -EAFNOSUPPORT;
1819
1820	rv = sock_create(addr_family, SOCK_STREAM, IPPROTO_TCP, &s);
1821	if (rv < 0)
1822		return rv;
1823	siw_reclassify_socket(s);
1824
1825	/*
1826	 * Allow binding local port when still in TIME_WAIT from last close.
1827	 */
1828	sock_set_reuseaddr(s->sk);
1829
1830	if (addr_family == AF_INET) {
1831		struct sockaddr_in *laddr = &to_sockaddr_in(id->local_addr);
1832
1833		/* For wildcard addr, limit binding to current device only */
1834		if (ipv4_is_zeronet(laddr->sin_addr.s_addr)) {
1835			ndev = ib_device_get_netdev(id->device, SIW_PORT);
1836			if (ndev) {
1837				s->sk->sk_bound_dev_if = ndev->ifindex;
1838			} else {
1839				rv = -ENODEV;
1840				goto error;
1841			}
1842		}
1843		rv = s->ops->bind(s, (struct sockaddr_unsized *)laddr,
1844				  sizeof(struct sockaddr_in));
1845	} else {
1846		struct sockaddr_in6 *laddr = &to_sockaddr_in6(id->local_addr);
1847
1848		if (id->afonly) {
1849			rv = ip6_sock_set_v6only(s->sk);
1850			if (rv) {
1851				siw_dbg(id->device,
1852					"ip6_sock_set_v6only erro: %d\n", rv);
1853				goto error;
1854			}
1855		}
1856
1857		/* For wildcard addr, limit binding to current device only */
1858		if (ipv6_addr_any(&laddr->sin6_addr)) {
1859			ndev = ib_device_get_netdev(id->device, SIW_PORT);
1860			if (ndev) {
1861				s->sk->sk_bound_dev_if = ndev->ifindex;
1862			} else {
1863				rv = -ENODEV;
1864				goto error;
1865			}
1866		}
1867		rv = s->ops->bind(s, (struct sockaddr_unsized *)laddr,
1868				  sizeof(struct sockaddr_in6));
1869	}
1870	if (rv) {
1871		siw_dbg(id->device, "socket bind error: %d\n", rv);
1872		goto error;
1873	}
1874	cep = siw_cep_alloc(sdev);
1875	if (!cep) {
1876		rv = -ENOMEM;
1877		goto error;
1878	}
1879	siw_cep_socket_assoc(cep, s);
1880
1881	rv = siw_cm_alloc_work(cep, backlog);
1882	if (rv) {
1883		siw_dbg(id->device,
1884			"alloc_work error %d, backlog %d\n",
1885			rv, backlog);
1886		goto error;
1887	}
1888	rv = s->ops->listen(s, backlog);
1889	if (rv) {
1890		siw_dbg(id->device, "listen error %d\n", rv);
1891		goto error;
1892	}
1893	cep->cm_id = id;
1894	id->add_ref(id);
1895
1896	/*
1897	 * In case of a wildcard rdma_listen on a multi-homed device,
1898	 * a listener's IWCM id is associated with more than one listening CEP.
1899	 *
1900	 * We currently use id->provider_data in three different ways:
1901	 *
1902	 * o For a listener's IWCM id, id->provider_data points to
1903	 *   the list_head of the list of listening CEPs.
1904	 *   Uses: siw_create_listen(), siw_destroy_listen()
1905	 *
1906	 * o For each accepted passive-side IWCM id, id->provider_data
1907	 *   points to the CEP itself. This is a consequence of
1908	 *   - siw_cm_upcall() setting event.provider_data = cep and
1909	 *   - the IWCM's cm_conn_req_handler() setting provider_data of the
1910	 *     new passive-side IWCM id equal to event.provider_data
1911	 *   Uses: siw_accept(), siw_reject()
1912	 *
1913	 * o For an active-side IWCM id, id->provider_data is not used at all.
1914	 *
1915	 */
1916	if (!id->provider_data) {
1917		id->provider_data =
1918			kmalloc(sizeof(struct list_head), GFP_KERNEL);
1919		if (!id->provider_data) {
1920			rv = -ENOMEM;
1921			goto error;
1922		}
1923		INIT_LIST_HEAD((struct list_head *)id->provider_data);
1924	}
1925	list_add_tail(&cep->listenq, (struct list_head *)id->provider_data);
1926	cep->state = SIW_EPSTATE_LISTENING;
1927	dev_put(ndev);
1928
1929	siw_dbg(id->device, "Listen at laddr %pISp\n", &id->local_addr);
1930
1931	return 0;
1932
1933error:
1934	siw_dbg(id->device, "failed: %d\n", rv);
1935
1936	if (cep) {
1937		siw_cep_set_inuse(cep);
1938
1939		siw_free_cm_id(cep);
1940		cep->sock = NULL;
1941		siw_socket_disassoc(s);
1942		cep->state = SIW_EPSTATE_CLOSED;
1943
1944		siw_cep_set_free_and_put(cep);
1945	}
1946	sock_release(s);
1947	dev_put(ndev);
1948
1949	return rv;
1950}
1951
1952static void siw_drop_listeners(struct iw_cm_id *id)
1953{
1954	struct list_head *p, *tmp;
1955
1956	/*
1957	 * In case of a wildcard rdma_listen on a multi-homed device,
1958	 * a listener's IWCM id is associated with more than one listening CEP.
1959	 */
1960	list_for_each_safe(p, tmp, (struct list_head *)id->provider_data) {
1961		struct siw_cep *cep = list_entry(p, struct siw_cep, listenq);
1962
1963		list_del(p);
1964
1965		siw_dbg_cep(cep, "drop cep, state %d\n", cep->state);
1966
1967		siw_cep_set_inuse(cep);
1968
1969		siw_free_cm_id(cep);
1970		if (cep->sock) {
1971			siw_socket_disassoc(cep->sock);
1972			sock_release(cep->sock);
1973			cep->sock = NULL;
1974		}
1975		cep->state = SIW_EPSTATE_CLOSED;
1976		siw_cep_set_free_and_put(cep);
1977	}
1978}
1979
1980int siw_destroy_listen(struct iw_cm_id *id)
1981{
1982	if (!id->provider_data) {
1983		siw_dbg(id->device, "no cep(s)\n");
1984		return 0;
1985	}
1986	siw_drop_listeners(id);
1987	kfree(id->provider_data);
1988	id->provider_data = NULL;
1989
1990	return 0;
1991}
1992
1993int siw_cm_init(void)
1994{
1995	/*
1996	 * create_single_workqueue for strict ordering
1997	 */
1998	siw_cm_wq = create_singlethread_workqueue("siw_cm_wq");
1999	if (!siw_cm_wq)
2000		return -ENOMEM;
2001
2002	return 0;
2003}
2004
2005void siw_cm_exit(void)
2006{
2007	if (siw_cm_wq)
2008		destroy_workqueue(siw_cm_wq);
2009}