cm.c

   1// SPDX-License-Identifier: GPL-2.0 OR Linux-OpenIB
   2/* Copyright (c) 2015 - 2021 Intel Corporation */
   3#include "main.h"
   4#include "trace.h"
   5
   6static void irdma_cm_post_event(struct irdma_cm_event *event);
   7static void irdma_disconnect_worker(struct work_struct *work);
   8
   9/**
  10 * irdma_free_sqbuf - put back puda buffer if refcount is 0
  11 * @vsi: The VSI structure of the device
  12 * @bufp: puda buffer to free
  13 */
  14void irdma_free_sqbuf(struct irdma_sc_vsi *vsi, void *bufp)
  15{
  16	struct irdma_puda_buf *buf = bufp;
  17	struct irdma_puda_rsrc *ilq = vsi->ilq;
  18
  19	if (refcount_dec_and_test(&buf->refcount))
  20		irdma_puda_ret_bufpool(ilq, buf);
  21}
  22
  23/**
  24 * irdma_record_ird_ord - Record IRD/ORD passed in
  25 * @cm_node: connection's node
  26 * @conn_ird: connection IRD
  27 * @conn_ord: connection ORD
  28 */
  29static void irdma_record_ird_ord(struct irdma_cm_node *cm_node, u32 conn_ird,
  30				 u32 conn_ord)
  31{
  32	if (conn_ird > cm_node->dev->hw_attrs.max_hw_ird)
  33		conn_ird = cm_node->dev->hw_attrs.max_hw_ird;
  34
  35	if (conn_ord > cm_node->dev->hw_attrs.max_hw_ord)
  36		conn_ord = cm_node->dev->hw_attrs.max_hw_ord;
  37	else if (!conn_ord && cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO)
  38		conn_ord = 1;
  39	cm_node->ird_size = conn_ird;
  40	cm_node->ord_size = conn_ord;
  41}
  42
  43/**
  44 * irdma_copy_ip_ntohl - copy IP address from  network to host
  45 * @dst: IP address in host order
  46 * @src: IP address in network order (big endian)
  47 */
  48void irdma_copy_ip_ntohl(u32 *dst, __be32 *src)
  49{
  50	*dst++ = ntohl(*src++);
  51	*dst++ = ntohl(*src++);
  52	*dst++ = ntohl(*src++);
  53	*dst = ntohl(*src);
  54}
  55
  56/**
  57 * irdma_copy_ip_htonl - copy IP address from host to network order
  58 * @dst: IP address in network order (big endian)
  59 * @src: IP address in host order
  60 */
  61void irdma_copy_ip_htonl(__be32 *dst, u32 *src)
  62{
  63	*dst++ = htonl(*src++);
  64	*dst++ = htonl(*src++);
  65	*dst++ = htonl(*src++);
  66	*dst = htonl(*src);
  67}
  68
  69/**
  70 * irdma_get_addr_info
  71 * @cm_node: contains ip/tcp info
  72 * @cm_info: to get a copy of the cm_node ip/tcp info
  73 */
  74static void irdma_get_addr_info(struct irdma_cm_node *cm_node,
  75				struct irdma_cm_info *cm_info)
  76{
  77	memset(cm_info, 0, sizeof(*cm_info));
  78	cm_info->ipv4 = cm_node->ipv4;
  79	cm_info->vlan_id = cm_node->vlan_id;
  80	memcpy(cm_info->loc_addr, cm_node->loc_addr, sizeof(cm_info->loc_addr));
  81	memcpy(cm_info->rem_addr, cm_node->rem_addr, sizeof(cm_info->rem_addr));
  82	cm_info->loc_port = cm_node->loc_port;
  83	cm_info->rem_port = cm_node->rem_port;
  84}
  85
  86/**
  87 * irdma_fill_sockaddr4 - fill in addr info for IPv4 connection
  88 * @cm_node: connection's node
  89 * @event: upper layer's cm event
  90 */
  91static inline void irdma_fill_sockaddr4(struct irdma_cm_node *cm_node,
  92					struct iw_cm_event *event)
  93{
  94	struct sockaddr_in *laddr = (struct sockaddr_in *)&event->local_addr;
  95	struct sockaddr_in *raddr = (struct sockaddr_in *)&event->remote_addr;
  96
  97	laddr->sin_family = AF_INET;
  98	raddr->sin_family = AF_INET;
  99
 100	laddr->sin_port = htons(cm_node->loc_port);
 101	raddr->sin_port = htons(cm_node->rem_port);
 102
 103	laddr->sin_addr.s_addr = htonl(cm_node->loc_addr[0]);
 104	raddr->sin_addr.s_addr = htonl(cm_node->rem_addr[0]);
 105}
 106
 107/**
 108 * irdma_fill_sockaddr6 - fill in addr info for IPv6 connection
 109 * @cm_node: connection's node
 110 * @event: upper layer's cm event
 111 */
 112static inline void irdma_fill_sockaddr6(struct irdma_cm_node *cm_node,
 113					struct iw_cm_event *event)
 114{
 115	struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&event->local_addr;
 116	struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)&event->remote_addr;
 117
 118	laddr6->sin6_family = AF_INET6;
 119	raddr6->sin6_family = AF_INET6;
 120
 121	laddr6->sin6_port = htons(cm_node->loc_port);
 122	raddr6->sin6_port = htons(cm_node->rem_port);
 123
 124	irdma_copy_ip_htonl(laddr6->sin6_addr.in6_u.u6_addr32,
 125			    cm_node->loc_addr);
 126	irdma_copy_ip_htonl(raddr6->sin6_addr.in6_u.u6_addr32,
 127			    cm_node->rem_addr);
 128}
 129
 130/**
 131 * irdma_get_cmevent_info - for cm event upcall
 132 * @cm_node: connection's node
 133 * @cm_id: upper layers cm struct for the event
 134 * @event: upper layer's cm event
 135 */
 136static inline void irdma_get_cmevent_info(struct irdma_cm_node *cm_node,
 137					  struct iw_cm_id *cm_id,
 138					  struct iw_cm_event *event)
 139{
 140	memcpy(&event->local_addr, &cm_id->m_local_addr,
 141	       sizeof(event->local_addr));
 142	memcpy(&event->remote_addr, &cm_id->m_remote_addr,
 143	       sizeof(event->remote_addr));
 144	if (cm_node) {
 145		event->private_data = cm_node->pdata_buf;
 146		event->private_data_len = (u8)cm_node->pdata.size;
 147		event->ird = cm_node->ird_size;
 148		event->ord = cm_node->ord_size;
 149	}
 150}
 151
 152/**
 153 * irdma_send_cm_event - upcall cm's event handler
 154 * @cm_node: connection's node
 155 * @cm_id: upper layer's cm info struct
 156 * @type: Event type to indicate
 157 * @status: status for the event type
 158 */
 159static int irdma_send_cm_event(struct irdma_cm_node *cm_node,
 160			       struct iw_cm_id *cm_id,
 161			       enum iw_cm_event_type type, int status)
 162{
 163	struct iw_cm_event event = {};
 164
 165	event.event = type;
 166	event.status = status;
 167	trace_irdma_send_cm_event(cm_node, cm_id, type, status,
 168				  __builtin_return_address(0));
 169
 170	ibdev_dbg(&cm_node->iwdev->ibdev,
 171		  "CM: cm_node %p cm_id=%p state=%d accel=%d event_type=%d status=%d\n",
 172		  cm_node, cm_id, cm_node->accelerated, cm_node->state, type,
 173		  status);
 174
 175	switch (type) {
 176	case IW_CM_EVENT_CONNECT_REQUEST:
 177		if (cm_node->ipv4)
 178			irdma_fill_sockaddr4(cm_node, &event);
 179		else
 180			irdma_fill_sockaddr6(cm_node, &event);
 181		event.provider_data = cm_node;
 182		event.private_data = cm_node->pdata_buf;
 183		event.private_data_len = (u8)cm_node->pdata.size;
 184		event.ird = cm_node->ird_size;
 185		break;
 186	case IW_CM_EVENT_CONNECT_REPLY:
 187		irdma_get_cmevent_info(cm_node, cm_id, &event);
 188		break;
 189	case IW_CM_EVENT_ESTABLISHED:
 190		event.ird = cm_node->ird_size;
 191		event.ord = cm_node->ord_size;
 192		break;
 193	case IW_CM_EVENT_DISCONNECT:
 194	case IW_CM_EVENT_CLOSE:
 195		/* Wait if we are in RTS but havent issued the iwcm event upcall */
 196		if (!cm_node->accelerated)
 197			wait_for_completion(&cm_node->establish_comp);
 198		break;
 199	default:
 200		return -EINVAL;
 201	}
 202
 203	return cm_id->event_handler(cm_id, &event);
 204}
 205
 206/**
 207 * irdma_timer_list_prep - add connection nodes to a list to perform timer tasks
 208 * @cm_core: cm's core
 209 * @timer_list: a timer list to which cm_node will be selected
 210 */
 211static void irdma_timer_list_prep(struct irdma_cm_core *cm_core,
 212				  struct list_head *timer_list)
 213{
 214	struct irdma_cm_node *cm_node;
 215	int bkt;
 216
 217	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
 218		if ((cm_node->close_entry || cm_node->send_entry) &&
 219		    refcount_inc_not_zero(&cm_node->refcnt))
 220			list_add(&cm_node->timer_entry, timer_list);
 221	}
 222}
 223
 224/**
 225 * irdma_create_event - create cm event
 226 * @cm_node: connection's node
 227 * @type: Event type to generate
 228 */
 229static struct irdma_cm_event *irdma_create_event(struct irdma_cm_node *cm_node,
 230						 enum irdma_cm_event_type type)
 231{
 232	struct irdma_cm_event *event;
 233
 234	if (!cm_node->cm_id)
 235		return NULL;
 236
 237	event = kzalloc(sizeof(*event), GFP_ATOMIC);
 238
 239	if (!event)
 240		return NULL;
 241
 242	event->type = type;
 243	event->cm_node = cm_node;
 244	memcpy(event->cm_info.rem_addr, cm_node->rem_addr,
 245	       sizeof(event->cm_info.rem_addr));
 246	memcpy(event->cm_info.loc_addr, cm_node->loc_addr,
 247	       sizeof(event->cm_info.loc_addr));
 248	event->cm_info.rem_port = cm_node->rem_port;
 249	event->cm_info.loc_port = cm_node->loc_port;
 250	event->cm_info.cm_id = cm_node->cm_id;
 251	ibdev_dbg(&cm_node->iwdev->ibdev,
 252		  "CM: node=%p event=%p type=%u dst=%pI4 src=%pI4\n", cm_node,
 253		  event, type, event->cm_info.loc_addr,
 254		  event->cm_info.rem_addr);
 255	trace_irdma_create_event(cm_node, type, __builtin_return_address(0));
 256	irdma_cm_post_event(event);
 257
 258	return event;
 259}
 260
 261/**
 262 * irdma_free_retrans_entry - free send entry
 263 * @cm_node: connection's node
 264 */
 265static void irdma_free_retrans_entry(struct irdma_cm_node *cm_node)
 266{
 267	struct irdma_device *iwdev = cm_node->iwdev;
 268	struct irdma_timer_entry *send_entry;
 269
 270	send_entry = cm_node->send_entry;
 271	if (!send_entry)
 272		return;
 273
 274	cm_node->send_entry = NULL;
 275	irdma_free_sqbuf(&iwdev->vsi, send_entry->sqbuf);
 276	kfree(send_entry);
 277	refcount_dec(&cm_node->refcnt);
 278}
 279
 280/**
 281 * irdma_cleanup_retrans_entry - free send entry with lock
 282 * @cm_node: connection's node
 283 */
 284static void irdma_cleanup_retrans_entry(struct irdma_cm_node *cm_node)
 285{
 286	unsigned long flags;
 287
 288	spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 289	irdma_free_retrans_entry(cm_node);
 290	spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 291}
 292
 293/**
 294 * irdma_form_ah_cm_frame - get a free packet and build frame with address handle
 295 * @cm_node: connection's node ionfo to use in frame
 296 * @options: pointer to options info
 297 * @hdr: pointer mpa header
 298 * @pdata: pointer to private data
 299 * @flags:  indicates FIN or ACK
 300 */
 301static struct irdma_puda_buf *irdma_form_ah_cm_frame(struct irdma_cm_node *cm_node,
 302						     struct irdma_kmem_info *options,
 303						     struct irdma_kmem_info *hdr,
 304						     struct irdma_mpa_priv_info *pdata,
 305						     u8 flags)
 306{
 307	struct irdma_puda_buf *sqbuf;
 308	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
 309	u8 *buf;
 310	struct tcphdr *tcph;
 311	u16 pktsize;
 312	u32 opts_len = 0;
 313	u32 pd_len = 0;
 314	u32 hdr_len = 0;
 315
 316	if (!cm_node->ah || !cm_node->ah->ah_info.ah_valid) {
 317		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: AH invalid\n");
 318		return NULL;
 319	}
 320
 321	sqbuf = irdma_puda_get_bufpool(vsi->ilq);
 322	if (!sqbuf) {
 323		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: SQ buf NULL\n");
 324		return NULL;
 325	}
 326
 327	sqbuf->ah_id = cm_node->ah->ah_info.ah_idx;
 328	buf = sqbuf->mem.va;
 329	if (options)
 330		opts_len = (u32)options->size;
 331
 332	if (hdr)
 333		hdr_len = hdr->size;
 334
 335	if (pdata)
 336		pd_len = pdata->size;
 337
 338	pktsize = sizeof(*tcph) + opts_len + hdr_len + pd_len;
 339
 340	memset(buf, 0, sizeof(*tcph));
 341
 342	sqbuf->totallen = pktsize;
 343	sqbuf->tcphlen = sizeof(*tcph) + opts_len;
 344	sqbuf->scratch = cm_node;
 345
 346	tcph = (struct tcphdr *)buf;
 347	buf += sizeof(*tcph);
 348
 349	tcph->source = htons(cm_node->loc_port);
 350	tcph->dest = htons(cm_node->rem_port);
 351	tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
 352
 353	if (flags & SET_ACK) {
 354		cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
 355		tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
 356		tcph->ack = 1;
 357	} else {
 358		tcph->ack_seq = 0;
 359	}
 360
 361	if (flags & SET_SYN) {
 362		cm_node->tcp_cntxt.loc_seq_num++;
 363		tcph->syn = 1;
 364	} else {
 365		cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
 366	}
 367
 368	if (flags & SET_FIN) {
 369		cm_node->tcp_cntxt.loc_seq_num++;
 370		tcph->fin = 1;
 371	}
 372
 373	if (flags & SET_RST)
 374		tcph->rst = 1;
 375
 376	tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
 377	sqbuf->tcphlen = tcph->doff << 2;
 378	tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd);
 379	tcph->urg_ptr = 0;
 380
 381	if (opts_len) {
 382		memcpy(buf, options->addr, opts_len);
 383		buf += opts_len;
 384	}
 385
 386	if (hdr_len) {
 387		memcpy(buf, hdr->addr, hdr_len);
 388		buf += hdr_len;
 389	}
 390
 391	if (pdata && pdata->addr)
 392		memcpy(buf, pdata->addr, pdata->size);
 393
 394	refcount_set(&sqbuf->refcount, 1);
 395
 396	print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET,
 397			     16, 8, sqbuf->mem.va, sqbuf->totallen, false);
 398
 399	return sqbuf;
 400}
 401
 402/**
 403 * irdma_form_uda_cm_frame - get a free packet and build frame full tcpip packet
 404 * @cm_node: connection's node ionfo to use in frame
 405 * @options: pointer to options info
 406 * @hdr: pointer mpa header
 407 * @pdata: pointer to private data
 408 * @flags:  indicates FIN or ACK
 409 */
 410static struct irdma_puda_buf *irdma_form_uda_cm_frame(struct irdma_cm_node *cm_node,
 411						      struct irdma_kmem_info *options,
 412						      struct irdma_kmem_info *hdr,
 413						      struct irdma_mpa_priv_info *pdata,
 414						      u8 flags)
 415{
 416	struct irdma_puda_buf *sqbuf;
 417	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
 418	u8 *buf;
 419
 420	struct tcphdr *tcph;
 421	struct iphdr *iph;
 422	struct ipv6hdr *ip6h;
 423	struct ethhdr *ethh;
 424	u16 pktsize;
 425	u16 eth_hlen = ETH_HLEN;
 426	u32 opts_len = 0;
 427	u32 pd_len = 0;
 428	u32 hdr_len = 0;
 429
 430	u16 vtag;
 431
 432	sqbuf = irdma_puda_get_bufpool(vsi->ilq);
 433	if (!sqbuf)
 434		return NULL;
 435
 436	buf = sqbuf->mem.va;
 437
 438	if (options)
 439		opts_len = (u32)options->size;
 440
 441	if (hdr)
 442		hdr_len = hdr->size;
 443
 444	if (pdata)
 445		pd_len = pdata->size;
 446
 447	if (cm_node->vlan_id < VLAN_N_VID)
 448		eth_hlen += 4;
 449
 450	if (cm_node->ipv4)
 451		pktsize = sizeof(*iph) + sizeof(*tcph);
 452	else
 453		pktsize = sizeof(*ip6h) + sizeof(*tcph);
 454	pktsize += opts_len + hdr_len + pd_len;
 455
 456	memset(buf, 0, eth_hlen + pktsize);
 457
 458	sqbuf->totallen = pktsize + eth_hlen;
 459	sqbuf->maclen = eth_hlen;
 460	sqbuf->tcphlen = sizeof(*tcph) + opts_len;
 461	sqbuf->scratch = cm_node;
 462
 463	ethh = (struct ethhdr *)buf;
 464	buf += eth_hlen;
 465
 466	if (cm_node->do_lpb)
 467		sqbuf->do_lpb = true;
 468
 469	if (cm_node->ipv4) {
 470		sqbuf->ipv4 = true;
 471
 472		iph = (struct iphdr *)buf;
 473		buf += sizeof(*iph);
 474		tcph = (struct tcphdr *)buf;
 475		buf += sizeof(*tcph);
 476
 477		ether_addr_copy(ethh->h_dest, cm_node->rem_mac);
 478		ether_addr_copy(ethh->h_source, cm_node->loc_mac);
 479		if (cm_node->vlan_id < VLAN_N_VID) {
 480			((struct vlan_ethhdr *)ethh)->h_vlan_proto =
 481				htons(ETH_P_8021Q);
 482			vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
 483			       cm_node->vlan_id;
 484			((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag);
 485
 486			((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto =
 487				htons(ETH_P_IP);
 488		} else {
 489			ethh->h_proto = htons(ETH_P_IP);
 490		}
 491
 492		iph->version = IPVERSION;
 493		iph->ihl = 5; /* 5 * 4Byte words, IP headr len */
 494		iph->tos = cm_node->tos;
 495		iph->tot_len = htons(pktsize);
 496		iph->id = htons(++cm_node->tcp_cntxt.loc_id);
 497
 498		iph->frag_off = htons(0x4000);
 499		iph->ttl = 0x40;
 500		iph->protocol = IPPROTO_TCP;
 501		iph->saddr = htonl(cm_node->loc_addr[0]);
 502		iph->daddr = htonl(cm_node->rem_addr[0]);
 503	} else {
 504		sqbuf->ipv4 = false;
 505		ip6h = (struct ipv6hdr *)buf;
 506		buf += sizeof(*ip6h);
 507		tcph = (struct tcphdr *)buf;
 508		buf += sizeof(*tcph);
 509
 510		ether_addr_copy(ethh->h_dest, cm_node->rem_mac);
 511		ether_addr_copy(ethh->h_source, cm_node->loc_mac);
 512		if (cm_node->vlan_id < VLAN_N_VID) {
 513			((struct vlan_ethhdr *)ethh)->h_vlan_proto =
 514				htons(ETH_P_8021Q);
 515			vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
 516			       cm_node->vlan_id;
 517			((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag);
 518			((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto =
 519				htons(ETH_P_IPV6);
 520		} else {
 521			ethh->h_proto = htons(ETH_P_IPV6);
 522		}
 523		ip6h->version = 6;
 524		ip6h->priority = cm_node->tos >> 4;
 525		ip6h->flow_lbl[0] = cm_node->tos << 4;
 526		ip6h->flow_lbl[1] = 0;
 527		ip6h->flow_lbl[2] = 0;
 528		ip6h->payload_len = htons(pktsize - sizeof(*ip6h));
 529		ip6h->nexthdr = 6;
 530		ip6h->hop_limit = 128;
 531		irdma_copy_ip_htonl(ip6h->saddr.in6_u.u6_addr32,
 532				    cm_node->loc_addr);
 533		irdma_copy_ip_htonl(ip6h->daddr.in6_u.u6_addr32,
 534				    cm_node->rem_addr);
 535	}
 536
 537	tcph->source = htons(cm_node->loc_port);
 538	tcph->dest = htons(cm_node->rem_port);
 539	tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
 540
 541	if (flags & SET_ACK) {
 542		cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
 543		tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num);
 544		tcph->ack = 1;
 545	} else {
 546		tcph->ack_seq = 0;
 547	}
 548
 549	if (flags & SET_SYN) {
 550		cm_node->tcp_cntxt.loc_seq_num++;
 551		tcph->syn = 1;
 552	} else {
 553		cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
 554	}
 555
 556	if (flags & SET_FIN) {
 557		cm_node->tcp_cntxt.loc_seq_num++;
 558		tcph->fin = 1;
 559	}
 560
 561	if (flags & SET_RST)
 562		tcph->rst = 1;
 563
 564	tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
 565	sqbuf->tcphlen = tcph->doff << 2;
 566	tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd);
 567	tcph->urg_ptr = 0;
 568
 569	if (opts_len) {
 570		memcpy(buf, options->addr, opts_len);
 571		buf += opts_len;
 572	}
 573
 574	if (hdr_len) {
 575		memcpy(buf, hdr->addr, hdr_len);
 576		buf += hdr_len;
 577	}
 578
 579	if (pdata && pdata->addr)
 580		memcpy(buf, pdata->addr, pdata->size);
 581
 582	refcount_set(&sqbuf->refcount, 1);
 583
 584	print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET,
 585			     16, 8, sqbuf->mem.va, sqbuf->totallen, false);
 586	return sqbuf;
 587}
 588
 589/**
 590 * irdma_send_reset - Send RST packet
 591 * @cm_node: connection's node
 592 */
 593int irdma_send_reset(struct irdma_cm_node *cm_node)
 594{
 595	struct irdma_puda_buf *sqbuf;
 596	int flags = SET_RST | SET_ACK;
 597
 598	trace_irdma_send_reset(cm_node, 0, __builtin_return_address(0));
 599	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
 600						flags);
 601	if (!sqbuf)
 602		return -ENOMEM;
 603
 604	ibdev_dbg(&cm_node->iwdev->ibdev,
 605		  "CM: caller: %pS cm_node %p cm_id=%p accel=%d state=%d rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n",
 606		  __builtin_return_address(0), cm_node, cm_node->cm_id,
 607		  cm_node->accelerated, cm_node->state, cm_node->rem_port,
 608		  cm_node->loc_port, cm_node->rem_addr, cm_node->loc_addr);
 609
 610	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 0,
 611				       1);
 612}
 613
 614/**
 615 * irdma_active_open_err - send event for active side cm error
 616 * @cm_node: connection's node
 617 * @reset: Flag to send reset or not
 618 */
 619static void irdma_active_open_err(struct irdma_cm_node *cm_node, bool reset)
 620{
 621	trace_irdma_active_open_err(cm_node, reset,
 622				    __builtin_return_address(0));
 623	irdma_cleanup_retrans_entry(cm_node);
 624	cm_node->cm_core->stats_connect_errs++;
 625	if (reset) {
 626		ibdev_dbg(&cm_node->iwdev->ibdev,
 627			  "CM: cm_node=%p state=%d\n", cm_node,
 628			  cm_node->state);
 629		refcount_inc(&cm_node->refcnt);
 630		irdma_send_reset(cm_node);
 631	}
 632
 633	cm_node->state = IRDMA_CM_STATE_CLOSED;
 634	irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
 635}
 636
 637/**
 638 * irdma_passive_open_err - handle passive side cm error
 639 * @cm_node: connection's node
 640 * @reset: send reset or just free cm_node
 641 */
 642static void irdma_passive_open_err(struct irdma_cm_node *cm_node, bool reset)
 643{
 644	irdma_cleanup_retrans_entry(cm_node);
 645	cm_node->cm_core->stats_passive_errs++;
 646	cm_node->state = IRDMA_CM_STATE_CLOSED;
 647	ibdev_dbg(&cm_node->iwdev->ibdev, "CM: cm_node=%p state =%d\n",
 648		  cm_node, cm_node->state);
 649	trace_irdma_passive_open_err(cm_node, reset,
 650				     __builtin_return_address(0));
 651	if (reset)
 652		irdma_send_reset(cm_node);
 653	else
 654		irdma_rem_ref_cm_node(cm_node);
 655}
 656
 657/**
 658 * irdma_event_connect_error - to create connect error event
 659 * @event: cm information for connect event
 660 */
 661static void irdma_event_connect_error(struct irdma_cm_event *event)
 662{
 663	struct irdma_qp *iwqp;
 664	struct iw_cm_id *cm_id;
 665
 666	cm_id = event->cm_node->cm_id;
 667	if (!cm_id)
 668		return;
 669
 670	iwqp = cm_id->provider_data;
 671
 672	if (!iwqp || !iwqp->iwdev)
 673		return;
 674
 675	iwqp->cm_id = NULL;
 676	cm_id->provider_data = NULL;
 677	irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
 678			    -ECONNRESET);
 679	irdma_rem_ref_cm_node(event->cm_node);
 680}
 681
 682/**
 683 * irdma_process_options - process options from TCP header
 684 * @cm_node: connection's node
 685 * @optionsloc: point to start of options
 686 * @optionsize: size of all options
 687 * @syn_pkt: flag if syn packet
 688 */
 689static int irdma_process_options(struct irdma_cm_node *cm_node, u8 *optionsloc,
 690				 u32 optionsize, u32 syn_pkt)
 691{
 692	u32 tmp;
 693	u32 offset = 0;
 694	union all_known_options *all_options;
 695	char got_mss_option = 0;
 696
 697	while (offset < optionsize) {
 698		all_options = (union all_known_options *)(optionsloc + offset);
 699		switch (all_options->base.optionnum) {
 700		case OPTION_NUM_EOL:
 701			offset = optionsize;
 702			break;
 703		case OPTION_NUM_NONE:
 704			offset += 1;
 705			continue;
 706		case OPTION_NUM_MSS:
 707			ibdev_dbg(&cm_node->iwdev->ibdev,
 708				  "CM: MSS Length: %d Offset: %d Size: %d\n",
 709				  all_options->mss.len, offset, optionsize);
 710			got_mss_option = 1;
 711			if (all_options->mss.len != 4)
 712				return -EINVAL;
 713			tmp = ntohs(all_options->mss.mss);
 714			if ((cm_node->ipv4 &&
 715			     (tmp + IRDMA_MTU_TO_MSS_IPV4) < IRDMA_MIN_MTU_IPV4) ||
 716			    (!cm_node->ipv4 &&
 717			     (tmp + IRDMA_MTU_TO_MSS_IPV6) < IRDMA_MIN_MTU_IPV6))
 718				return -EINVAL;
 719			if (tmp < cm_node->tcp_cntxt.mss)
 720				cm_node->tcp_cntxt.mss = tmp;
 721			break;
 722		case OPTION_NUM_WINDOW_SCALE:
 723			cm_node->tcp_cntxt.snd_wscale =
 724				all_options->windowscale.shiftcount;
 725			break;
 726		default:
 727			ibdev_dbg(&cm_node->iwdev->ibdev,
 728				  "CM: Unsupported TCP Option: %x\n",
 729				  all_options->base.optionnum);
 730			break;
 731		}
 732		offset += all_options->base.len;
 733	}
 734	if (!got_mss_option && syn_pkt)
 735		cm_node->tcp_cntxt.mss = IRDMA_CM_DEFAULT_MSS;
 736
 737	return 0;
 738}
 739
 740/**
 741 * irdma_handle_tcp_options - setup TCP context info after parsing TCP options
 742 * @cm_node: connection's node
 743 * @tcph: pointer tcp header
 744 * @optionsize: size of options rcvd
 745 * @passive: active or passive flag
 746 */
 747static int irdma_handle_tcp_options(struct irdma_cm_node *cm_node,
 748				    struct tcphdr *tcph, int optionsize,
 749				    int passive)
 750{
 751	u8 *optionsloc = (u8 *)&tcph[1];
 752	int ret;
 753
 754	if (optionsize) {
 755		ret = irdma_process_options(cm_node, optionsloc, optionsize,
 756					    (u32)tcph->syn);
 757		if (ret) {
 758			ibdev_dbg(&cm_node->iwdev->ibdev,
 759				  "CM: Node %p, Sending Reset\n", cm_node);
 760			if (passive)
 761				irdma_passive_open_err(cm_node, true);
 762			else
 763				irdma_active_open_err(cm_node, true);
 764			return ret;
 765		}
 766	}
 767
 768	cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window)
 769				     << cm_node->tcp_cntxt.snd_wscale;
 770
 771	if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd)
 772		cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
 773
 774	return 0;
 775}
 776
 777/**
 778 * irdma_build_mpa_v1 - build a MPA V1 frame
 779 * @cm_node: connection's node
 780 * @start_addr: address where to build frame
 781 * @mpa_key: to do read0 or write0
 782 */
 783static void irdma_build_mpa_v1(struct irdma_cm_node *cm_node, void *start_addr,
 784			       u8 mpa_key)
 785{
 786	struct ietf_mpa_v1 *mpa_frame = start_addr;
 787
 788	switch (mpa_key) {
 789	case MPA_KEY_REQUEST:
 790		memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
 791		break;
 792	case MPA_KEY_REPLY:
 793		memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
 794		break;
 795	default:
 796		break;
 797	}
 798	mpa_frame->flags = IETF_MPA_FLAGS_CRC;
 799	mpa_frame->rev = cm_node->mpa_frame_rev;
 800	mpa_frame->priv_data_len = htons(cm_node->pdata.size);
 801}
 802
 803/**
 804 * irdma_build_mpa_v2 - build a MPA V2 frame
 805 * @cm_node: connection's node
 806 * @start_addr: buffer start address
 807 * @mpa_key: to do read0 or write0
 808 */
 809static void irdma_build_mpa_v2(struct irdma_cm_node *cm_node, void *start_addr,
 810			       u8 mpa_key)
 811{
 812	struct ietf_mpa_v2 *mpa_frame = start_addr;
 813	struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg;
 814	u16 ctrl_ird, ctrl_ord;
 815
 816	/* initialize the upper 5 bytes of the frame */
 817	irdma_build_mpa_v1(cm_node, start_addr, mpa_key);
 818	mpa_frame->flags |= IETF_MPA_V2_FLAG;
 819	if (cm_node->iwdev->iw_ooo) {
 820		mpa_frame->flags |= IETF_MPA_FLAGS_MARKERS;
 821		cm_node->rcv_mark_en = true;
 822	}
 823	mpa_frame->priv_data_len = cpu_to_be16(be16_to_cpu(mpa_frame->priv_data_len) +
 824					       IETF_RTR_MSG_SIZE);
 825
 826	/* initialize RTR msg */
 827	if (cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) {
 828		ctrl_ird = IETF_NO_IRD_ORD;
 829		ctrl_ord = IETF_NO_IRD_ORD;
 830	} else {
 831		ctrl_ird = (cm_node->ird_size > IETF_NO_IRD_ORD) ?
 832				   IETF_NO_IRD_ORD :
 833				   cm_node->ird_size;
 834		ctrl_ord = (cm_node->ord_size > IETF_NO_IRD_ORD) ?
 835				   IETF_NO_IRD_ORD :
 836				   cm_node->ord_size;
 837	}
 838	ctrl_ird |= IETF_PEER_TO_PEER;
 839
 840	switch (mpa_key) {
 841	case MPA_KEY_REQUEST:
 842		ctrl_ord |= IETF_RDMA0_WRITE;
 843		ctrl_ord |= IETF_RDMA0_READ;
 844		break;
 845	case MPA_KEY_REPLY:
 846		switch (cm_node->send_rdma0_op) {
 847		case SEND_RDMA_WRITE_ZERO:
 848			ctrl_ord |= IETF_RDMA0_WRITE;
 849			break;
 850		case SEND_RDMA_READ_ZERO:
 851			ctrl_ord |= IETF_RDMA0_READ;
 852			break;
 853		}
 854		break;
 855	default:
 856		break;
 857	}
 858	rtr_msg->ctrl_ird = htons(ctrl_ird);
 859	rtr_msg->ctrl_ord = htons(ctrl_ord);
 860}
 861
 862/**
 863 * irdma_cm_build_mpa_frame - build mpa frame for mpa version 1 or version 2
 864 * @cm_node: connection's node
 865 * @mpa: mpa: data buffer
 866 * @mpa_key: to do read0 or write0
 867 */
 868static int irdma_cm_build_mpa_frame(struct irdma_cm_node *cm_node,
 869				    struct irdma_kmem_info *mpa, u8 mpa_key)
 870{
 871	int hdr_len = 0;
 872
 873	switch (cm_node->mpa_frame_rev) {
 874	case IETF_MPA_V1:
 875		hdr_len = sizeof(struct ietf_mpa_v1);
 876		irdma_build_mpa_v1(cm_node, mpa->addr, mpa_key);
 877		break;
 878	case IETF_MPA_V2:
 879		hdr_len = sizeof(struct ietf_mpa_v2);
 880		irdma_build_mpa_v2(cm_node, mpa->addr, mpa_key);
 881		break;
 882	default:
 883		break;
 884	}
 885
 886	return hdr_len;
 887}
 888
 889/**
 890 * irdma_send_mpa_request - active node send mpa request to passive node
 891 * @cm_node: connection's node
 892 */
 893static int irdma_send_mpa_request(struct irdma_cm_node *cm_node)
 894{
 895	struct irdma_puda_buf *sqbuf;
 896
 897	cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
 898	cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
 899							 &cm_node->mpa_hdr,
 900							 MPA_KEY_REQUEST);
 901	if (!cm_node->mpa_hdr.size) {
 902		ibdev_dbg(&cm_node->iwdev->ibdev,
 903			  "CM: mpa size = %d\n", cm_node->mpa_hdr.size);
 904		return -EINVAL;
 905	}
 906
 907	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
 908						&cm_node->mpa_hdr,
 909						&cm_node->pdata, SET_ACK);
 910	if (!sqbuf)
 911		return -ENOMEM;
 912
 913	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
 914				       0);
 915}
 916
 917/**
 918 * irdma_send_mpa_reject -
 919 * @cm_node: connection's node
 920 * @pdata: reject data for connection
 921 * @plen: length of reject data
 922 */
 923static int irdma_send_mpa_reject(struct irdma_cm_node *cm_node,
 924				 const void *pdata, u8 plen)
 925{
 926	struct irdma_puda_buf *sqbuf;
 927	struct irdma_mpa_priv_info priv_info;
 928
 929	cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
 930	cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
 931							 &cm_node->mpa_hdr,
 932							 MPA_KEY_REPLY);
 933
 934	cm_node->mpa_frame.flags |= IETF_MPA_FLAGS_REJECT;
 935	priv_info.addr = pdata;
 936	priv_info.size = plen;
 937
 938	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
 939						&cm_node->mpa_hdr, &priv_info,
 940						SET_ACK | SET_FIN);
 941	if (!sqbuf)
 942		return -ENOMEM;
 943
 944	cm_node->state = IRDMA_CM_STATE_FIN_WAIT1;
 945
 946	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
 947				       0);
 948}
 949
 950/**
 951 * irdma_negotiate_mpa_v2_ird_ord - negotiate MPAv2 IRD/ORD
 952 * @cm_node: connection's node
 953 * @buf: Data pointer
 954 */
 955static int irdma_negotiate_mpa_v2_ird_ord(struct irdma_cm_node *cm_node,
 956					  u8 *buf)
 957{
 958	struct ietf_mpa_v2 *mpa_v2_frame;
 959	struct ietf_rtr_msg *rtr_msg;
 960	u16 ird_size;
 961	u16 ord_size;
 962	u16 ctrl_ord;
 963	u16 ctrl_ird;
 964
 965	mpa_v2_frame = (struct ietf_mpa_v2 *)buf;
 966	rtr_msg = &mpa_v2_frame->rtr_msg;
 967
 968	/* parse rtr message */
 969	ctrl_ord = ntohs(rtr_msg->ctrl_ord);
 970	ctrl_ird = ntohs(rtr_msg->ctrl_ird);
 971	ird_size = ctrl_ird & IETF_NO_IRD_ORD;
 972	ord_size = ctrl_ord & IETF_NO_IRD_ORD;
 973
 974	if (!(ctrl_ird & IETF_PEER_TO_PEER))
 975		return -EOPNOTSUPP;
 976
 977	if (ird_size == IETF_NO_IRD_ORD || ord_size == IETF_NO_IRD_ORD) {
 978		cm_node->mpav2_ird_ord = IETF_NO_IRD_ORD;
 979		goto negotiate_done;
 980	}
 981
 982	if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
 983		/* responder */
 984		if (!ord_size && (ctrl_ord & IETF_RDMA0_READ))
 985			cm_node->ird_size = 1;
 986		if (cm_node->ord_size > ird_size)
 987			cm_node->ord_size = ird_size;
 988	} else {
 989		/* initiator */
 990		if (!ird_size && (ctrl_ord & IETF_RDMA0_READ))
 991			/* Remote peer doesn't support RDMA0_READ */
 992			return -EOPNOTSUPP;
 993
 994		if (cm_node->ord_size > ird_size)
 995			cm_node->ord_size = ird_size;
 996
 997		if (cm_node->ird_size < ord_size)
 998		/* no resources available */
 999			return -EINVAL;
1000	}
1001
1002negotiate_done:
1003	if (ctrl_ord & IETF_RDMA0_READ)
1004		cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
1005	else if (ctrl_ord & IETF_RDMA0_WRITE)
1006		cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO;
1007	else
1008		/* Not supported RDMA0 operation */
1009		return -EOPNOTSUPP;
1010
1011	ibdev_dbg(&cm_node->iwdev->ibdev,
1012		  "CM: MPAV2 Negotiated ORD: %d, IRD: %d\n",
1013		  cm_node->ord_size, cm_node->ird_size);
1014	trace_irdma_negotiate_mpa_v2(cm_node);
1015	return 0;
1016}
1017
1018/**
1019 * irdma_parse_mpa - process an IETF MPA frame
1020 * @cm_node: connection's node
1021 * @buf: Data pointer
1022 * @type: to return accept or reject
1023 * @len: Len of mpa buffer
1024 */
1025static int irdma_parse_mpa(struct irdma_cm_node *cm_node, u8 *buf, u32 *type,
1026			   u32 len)
1027{
1028	struct ietf_mpa_v1 *mpa_frame;
1029	int mpa_hdr_len, priv_data_len, ret;
1030
1031	*type = IRDMA_MPA_REQUEST_ACCEPT;
1032
1033	if (len < sizeof(struct ietf_mpa_v1)) {
1034		ibdev_dbg(&cm_node->iwdev->ibdev,
1035			  "CM: ietf buffer small (%x)\n", len);
1036		return -EINVAL;
1037	}
1038
1039	mpa_frame = (struct ietf_mpa_v1 *)buf;
1040	mpa_hdr_len = sizeof(struct ietf_mpa_v1);
1041	priv_data_len = ntohs(mpa_frame->priv_data_len);
1042
1043	if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) {
1044		ibdev_dbg(&cm_node->iwdev->ibdev,
1045			  "CM: private_data too big %d\n", priv_data_len);
1046		return -EOVERFLOW;
1047	}
1048
1049	if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) {
1050		ibdev_dbg(&cm_node->iwdev->ibdev,
1051			  "CM: unsupported mpa rev = %d\n", mpa_frame->rev);
1052		return -EINVAL;
1053	}
1054
1055	if (mpa_frame->rev > cm_node->mpa_frame_rev) {
1056		ibdev_dbg(&cm_node->iwdev->ibdev, "CM: rev %d\n",
1057			  mpa_frame->rev);
1058		return -EINVAL;
1059	}
1060
1061	cm_node->mpa_frame_rev = mpa_frame->rev;
1062	if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
1063		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ,
1064			   IETF_MPA_KEY_SIZE)) {
1065			ibdev_dbg(&cm_node->iwdev->ibdev,
1066				  "CM: Unexpected MPA Key received\n");
1067			return -EINVAL;
1068		}
1069	} else {
1070		if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REP,
1071			   IETF_MPA_KEY_SIZE)) {
1072			ibdev_dbg(&cm_node->iwdev->ibdev,
1073				  "CM: Unexpected MPA Key received\n");
1074			return -EINVAL;
1075		}
1076	}
1077
1078	if (priv_data_len + mpa_hdr_len > len) {
1079		ibdev_dbg(&cm_node->iwdev->ibdev,
1080			  "CM: ietf buffer len(%x + %x != %x)\n",
1081			  priv_data_len, mpa_hdr_len, len);
1082		return -EOVERFLOW;
1083	}
1084
1085	if (len > IRDMA_MAX_CM_BUF) {
1086		ibdev_dbg(&cm_node->iwdev->ibdev,
1087			  "CM: ietf buffer large len = %d\n", len);
1088		return -EOVERFLOW;
1089	}
1090
1091	switch (mpa_frame->rev) {
1092	case IETF_MPA_V2:
1093		mpa_hdr_len += IETF_RTR_MSG_SIZE;
1094		ret = irdma_negotiate_mpa_v2_ird_ord(cm_node, buf);
1095		if (ret)
1096			return ret;
1097		break;
1098	case IETF_MPA_V1:
1099	default:
1100		break;
1101	}
1102
1103	memcpy(cm_node->pdata_buf, buf + mpa_hdr_len, priv_data_len);
1104	cm_node->pdata.size = priv_data_len;
1105
1106	if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT)
1107		*type = IRDMA_MPA_REQUEST_REJECT;
1108
1109	if (mpa_frame->flags & IETF_MPA_FLAGS_MARKERS)
1110		cm_node->snd_mark_en = true;
1111
1112	return 0;
1113}
1114
1115/**
1116 * irdma_schedule_cm_timer
1117 * @cm_node: connection's node
1118 * @sqbuf: buffer to send
1119 * @type: if it is send or close
1120 * @send_retrans: if rexmits to be done
1121 * @close_when_complete: is cm_node to be removed
1122 *
1123 * note - cm_node needs to be protected before calling this. Encase in:
1124 *		irdma_rem_ref_cm_node(cm_core, cm_node);
1125 *		irdma_schedule_cm_timer(...)
1126 *		refcount_inc(&cm_node->refcnt);
1127 */
1128int irdma_schedule_cm_timer(struct irdma_cm_node *cm_node,
1129			    struct irdma_puda_buf *sqbuf,
1130			    enum irdma_timer_type type, int send_retrans,
1131			    int close_when_complete)
1132{
1133	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
1134	struct irdma_cm_core *cm_core = cm_node->cm_core;
1135	struct irdma_timer_entry *new_send;
1136	u32 was_timer_set;
1137	unsigned long flags;
1138
1139	new_send = kzalloc(sizeof(*new_send), GFP_ATOMIC);
1140	if (!new_send) {
1141		if (type != IRDMA_TIMER_TYPE_CLOSE)
1142			irdma_free_sqbuf(vsi, sqbuf);
1143		return -ENOMEM;
1144	}
1145
1146	new_send->retrycount = IRDMA_DEFAULT_RETRYS;
1147	new_send->retranscount = IRDMA_DEFAULT_RETRANS;
1148	new_send->sqbuf = sqbuf;
1149	new_send->timetosend = jiffies;
1150	new_send->type = type;
1151	new_send->send_retrans = send_retrans;
1152	new_send->close_when_complete = close_when_complete;
1153
1154	if (type == IRDMA_TIMER_TYPE_CLOSE) {
1155		new_send->timetosend += (HZ / 10);
1156		if (cm_node->close_entry) {
1157			kfree(new_send);
1158			ibdev_dbg(&cm_node->iwdev->ibdev,
1159				  "CM: already close entry\n");
1160			return -EINVAL;
1161		}
1162
1163		cm_node->close_entry = new_send;
1164	} else { /* type == IRDMA_TIMER_TYPE_SEND */
1165		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1166		cm_node->send_entry = new_send;
1167		refcount_inc(&cm_node->refcnt);
1168		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1169		new_send->timetosend = jiffies + IRDMA_RETRY_TIMEOUT;
1170
1171		refcount_inc(&sqbuf->refcount);
1172		irdma_puda_send_buf(vsi->ilq, sqbuf);
1173		if (!send_retrans) {
1174			irdma_cleanup_retrans_entry(cm_node);
1175			if (close_when_complete)
1176				irdma_rem_ref_cm_node(cm_node);
1177			return 0;
1178		}
1179	}
1180
1181	spin_lock_irqsave(&cm_core->ht_lock, flags);
1182	was_timer_set = timer_pending(&cm_core->tcp_timer);
1183
1184	if (!was_timer_set) {
1185		cm_core->tcp_timer.expires = new_send->timetosend;
1186		add_timer(&cm_core->tcp_timer);
1187	}
1188	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
1189
1190	return 0;
1191}
1192
1193/**
1194 * irdma_retrans_expired - Could not rexmit the packet
1195 * @cm_node: connection's node
1196 */
1197static void irdma_retrans_expired(struct irdma_cm_node *cm_node)
1198{
1199	enum irdma_cm_node_state state = cm_node->state;
1200
1201	cm_node->state = IRDMA_CM_STATE_CLOSED;
1202	switch (state) {
1203	case IRDMA_CM_STATE_SYN_RCVD:
1204	case IRDMA_CM_STATE_CLOSING:
1205		irdma_rem_ref_cm_node(cm_node);
1206		break;
1207	case IRDMA_CM_STATE_FIN_WAIT1:
1208	case IRDMA_CM_STATE_LAST_ACK:
1209		irdma_send_reset(cm_node);
1210		break;
1211	default:
1212		refcount_inc(&cm_node->refcnt);
1213		irdma_send_reset(cm_node);
1214		irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
1215		break;
1216	}
1217}
1218
1219/**
1220 * irdma_handle_close_entry - for handling retry/timeouts
1221 * @cm_node: connection's node
1222 * @rem_node: flag for remove cm_node
1223 */
1224static void irdma_handle_close_entry(struct irdma_cm_node *cm_node,
1225				     u32 rem_node)
1226{
1227	struct irdma_timer_entry *close_entry = cm_node->close_entry;
1228	struct irdma_qp *iwqp;
1229	unsigned long flags;
1230
1231	if (!close_entry)
1232		return;
1233	iwqp = (struct irdma_qp *)close_entry->sqbuf;
1234	if (iwqp) {
1235		spin_lock_irqsave(&iwqp->lock, flags);
1236		if (iwqp->cm_id) {
1237			iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
1238			iwqp->hw_iwarp_state = IRDMA_QP_STATE_ERROR;
1239			iwqp->last_aeq = IRDMA_AE_RESET_SENT;
1240			iwqp->ibqp_state = IB_QPS_ERR;
1241			spin_unlock_irqrestore(&iwqp->lock, flags);
1242			irdma_cm_disconn(iwqp);
1243		} else {
1244			spin_unlock_irqrestore(&iwqp->lock, flags);
1245		}
1246	} else if (rem_node) {
1247		/* TIME_WAIT state */
1248		irdma_rem_ref_cm_node(cm_node);
1249	}
1250
1251	kfree(close_entry);
1252	cm_node->close_entry = NULL;
1253}
1254
1255/**
1256 * irdma_cm_timer_tick - system's timer expired callback
1257 * @t: Pointer to timer_list
1258 */
1259static void irdma_cm_timer_tick(struct timer_list *t)
1260{
1261	unsigned long nexttimeout = jiffies + IRDMA_LONG_TIME;
1262	struct irdma_cm_node *cm_node;
1263	struct irdma_timer_entry *send_entry, *close_entry;
1264	struct list_head *list_core_temp;
1265	struct list_head *list_node;
1266	struct irdma_cm_core *cm_core = timer_container_of(cm_core, t,
1267							   tcp_timer);
1268	struct irdma_sc_vsi *vsi;
1269	u32 settimer = 0;
1270	unsigned long timetosend;
1271	unsigned long flags;
1272	struct list_head timer_list;
1273
1274	INIT_LIST_HEAD(&timer_list);
1275
1276	rcu_read_lock();
1277	irdma_timer_list_prep(cm_core, &timer_list);
1278	rcu_read_unlock();
1279
1280	list_for_each_safe (list_node, list_core_temp, &timer_list) {
1281		cm_node = container_of(list_node, struct irdma_cm_node,
1282				       timer_entry);
1283		close_entry = cm_node->close_entry;
1284
1285		if (close_entry) {
1286			if (time_after(close_entry->timetosend, jiffies)) {
1287				if (nexttimeout > close_entry->timetosend ||
1288				    !settimer) {
1289					nexttimeout = close_entry->timetosend;
1290					settimer = 1;
1291				}
1292			} else {
1293				irdma_handle_close_entry(cm_node, 1);
1294			}
1295		}
1296
1297		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1298
1299		send_entry = cm_node->send_entry;
1300		if (!send_entry)
1301			goto done;
1302		if (time_after(send_entry->timetosend, jiffies)) {
1303			if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
1304				if (nexttimeout > send_entry->timetosend ||
1305				    !settimer) {
1306					nexttimeout = send_entry->timetosend;
1307					settimer = 1;
1308				}
1309			} else {
1310				irdma_free_retrans_entry(cm_node);
1311			}
1312			goto done;
1313		}
1314
1315		if (cm_node->state == IRDMA_CM_STATE_OFFLOADED ||
1316		    cm_node->state == IRDMA_CM_STATE_CLOSED) {
1317			irdma_free_retrans_entry(cm_node);
1318			goto done;
1319		}
1320
1321		if (!send_entry->retranscount || !send_entry->retrycount) {
1322			irdma_free_retrans_entry(cm_node);
1323
1324			spin_unlock_irqrestore(&cm_node->retrans_list_lock,
1325					       flags);
1326			irdma_retrans_expired(cm_node);
1327			cm_node->state = IRDMA_CM_STATE_CLOSED;
1328			spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1329			goto done;
1330		}
1331		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1332
1333		vsi = &cm_node->iwdev->vsi;
1334		if (!cm_node->ack_rcvd) {
1335			refcount_inc(&send_entry->sqbuf->refcount);
1336			irdma_puda_send_buf(vsi->ilq, send_entry->sqbuf);
1337			cm_node->cm_core->stats_pkt_retrans++;
1338		}
1339
1340		spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
1341		if (send_entry->send_retrans) {
1342			send_entry->retranscount--;
1343			timetosend = (IRDMA_RETRY_TIMEOUT <<
1344				      (IRDMA_DEFAULT_RETRANS -
1345				       send_entry->retranscount));
1346
1347			send_entry->timetosend = jiffies +
1348			    min(timetosend, IRDMA_MAX_TIMEOUT);
1349			if (nexttimeout > send_entry->timetosend || !settimer) {
1350				nexttimeout = send_entry->timetosend;
1351				settimer = 1;
1352			}
1353		} else {
1354			int close_when_complete;
1355
1356			close_when_complete = send_entry->close_when_complete;
1357			irdma_free_retrans_entry(cm_node);
1358			if (close_when_complete)
1359				irdma_rem_ref_cm_node(cm_node);
1360		}
1361done:
1362		spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
1363		irdma_rem_ref_cm_node(cm_node);
1364	}
1365
1366	if (settimer) {
1367		spin_lock_irqsave(&cm_core->ht_lock, flags);
1368		if (!timer_pending(&cm_core->tcp_timer)) {
1369			cm_core->tcp_timer.expires = nexttimeout;
1370			add_timer(&cm_core->tcp_timer);
1371		}
1372		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
1373	}
1374}
1375
1376/**
1377 * irdma_send_syn - send SYN packet
1378 * @cm_node: connection's node
1379 * @sendack: flag to set ACK bit or not
1380 */
1381int irdma_send_syn(struct irdma_cm_node *cm_node, u32 sendack)
1382{
1383	struct irdma_puda_buf *sqbuf;
1384	int flags = SET_SYN;
1385	char optionsbuf[sizeof(struct option_mss) +
1386			sizeof(struct option_windowscale) +
1387			sizeof(struct option_base) + TCP_OPTIONS_PADDING];
1388	struct irdma_kmem_info opts;
1389	int optionssize = 0;
1390	/* Sending MSS option */
1391	union all_known_options *options;
1392
1393	opts.addr = optionsbuf;
1394	if (!cm_node)
1395		return -EINVAL;
1396
1397	options = (union all_known_options *)&optionsbuf[optionssize];
1398	options->mss.optionnum = OPTION_NUM_MSS;
1399	options->mss.len = sizeof(struct option_mss);
1400	options->mss.mss = htons(cm_node->tcp_cntxt.mss);
1401	optionssize += sizeof(struct option_mss);
1402
1403	options = (union all_known_options *)&optionsbuf[optionssize];
1404	options->windowscale.optionnum = OPTION_NUM_WINDOW_SCALE;
1405	options->windowscale.len = sizeof(struct option_windowscale);
1406	options->windowscale.shiftcount = cm_node->tcp_cntxt.rcv_wscale;
1407	optionssize += sizeof(struct option_windowscale);
1408	options = (union all_known_options *)&optionsbuf[optionssize];
1409	options->eol = OPTION_NUM_EOL;
1410	optionssize += 1;
1411
1412	if (sendack)
1413		flags |= SET_ACK;
1414
1415	opts.size = optionssize;
1416
1417	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, &opts, NULL, NULL,
1418						flags);
1419	if (!sqbuf)
1420		return -ENOMEM;
1421
1422	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
1423				       0);
1424}
1425
1426/**
1427 * irdma_send_ack - Send ACK packet
1428 * @cm_node: connection's node
1429 */
1430void irdma_send_ack(struct irdma_cm_node *cm_node)
1431{
1432	struct irdma_puda_buf *sqbuf;
1433	struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
1434
1435	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
1436						SET_ACK);
1437	if (sqbuf)
1438		irdma_puda_send_buf(vsi->ilq, sqbuf);
1439}
1440
1441/**
1442 * irdma_send_fin - Send FIN pkt
1443 * @cm_node: connection's node
1444 */
1445static int irdma_send_fin(struct irdma_cm_node *cm_node)
1446{
1447	struct irdma_puda_buf *sqbuf;
1448
1449	sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
1450						SET_ACK | SET_FIN);
1451	if (!sqbuf)
1452		return -ENOMEM;
1453
1454	return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
1455				       0);
1456}
1457
1458/**
1459 * irdma_find_listener - find a cm node listening on this addr-port pair
1460 * @cm_core: cm's core
1461 * @dst_addr: listener ip addr
1462 * @ipv4: flag indicating IPv4 when true
1463 * @dst_port: listener tcp port num
1464 * @vlan_id: virtual LAN ID
1465 * @listener_state: state to match with listen node's
1466 */
1467static struct irdma_cm_listener *
1468irdma_find_listener(struct irdma_cm_core *cm_core, u32 *dst_addr, bool ipv4,
1469		    u16 dst_port, u16 vlan_id,
1470		    enum irdma_cm_listener_state listener_state)
1471{
1472	struct irdma_cm_listener *listen_node;
1473	static const u32 ip_zero[4] = { 0, 0, 0, 0 };
1474	u32 listen_addr[4];
1475	u16 listen_port;
1476	unsigned long flags;
1477
1478	/* walk list and find cm_node associated with this session ID */
1479	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
1480	list_for_each_entry (listen_node, &cm_core->listen_list, list) {
1481		memcpy(listen_addr, listen_node->loc_addr, sizeof(listen_addr));
1482		listen_port = listen_node->loc_port;
1483		if (listen_node->ipv4 != ipv4 || listen_port != dst_port ||
1484		    !(listener_state & listen_node->listener_state))
1485			continue;
1486		/* compare node pair, return node handle if a match */
1487		if (!memcmp(listen_addr, ip_zero, sizeof(listen_addr)) ||
1488		    (!memcmp(listen_addr, dst_addr, sizeof(listen_addr)) &&
1489		     vlan_id == listen_node->vlan_id)) {
1490			refcount_inc(&listen_node->refcnt);
1491			spin_unlock_irqrestore(&cm_core->listen_list_lock,
1492					       flags);
1493			trace_irdma_find_listener(listen_node);
1494			return listen_node;
1495		}
1496	}
1497	spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
1498
1499	return NULL;
1500}
1501
1502/**
1503 * irdma_del_multiple_qhash - Remove qhash and child listens
1504 * @iwdev: iWarp device
1505 * @cm_info: CM info for parent listen node
1506 * @cm_parent_listen_node: The parent listen node
1507 */
1508static int irdma_del_multiple_qhash(struct irdma_device *iwdev,
1509				    struct irdma_cm_info *cm_info,
1510				    struct irdma_cm_listener *cm_parent_listen_node)
1511{
1512	struct irdma_cm_listener *child_listen_node;
1513	struct list_head *pos, *tpos;
1514	unsigned long flags;
1515	int ret = -EINVAL;
1516
1517	spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
1518	list_for_each_safe (pos, tpos,
1519			    &cm_parent_listen_node->child_listen_list) {
1520		child_listen_node = list_entry(pos, struct irdma_cm_listener,
1521					       child_listen_list);
1522		if (child_listen_node->ipv4)
1523			ibdev_dbg(&iwdev->ibdev,
1524				  "CM: removing child listen for IP=%pI4, port=%d, vlan=%d\n",
1525				  child_listen_node->loc_addr,
1526				  child_listen_node->loc_port,
1527				  child_listen_node->vlan_id);
1528		else
1529			ibdev_dbg(&iwdev->ibdev,
1530				  "CM: removing child listen for IP=%pI6, port=%d, vlan=%d\n",
1531				  child_listen_node->loc_addr,
1532				  child_listen_node->loc_port,
1533				  child_listen_node->vlan_id);
1534		trace_irdma_del_multiple_qhash(child_listen_node);
1535		list_del(pos);
1536		memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1537		       sizeof(cm_info->loc_addr));
1538		cm_info->vlan_id = child_listen_node->vlan_id;
1539		if (child_listen_node->qhash_set) {
1540			ret = irdma_manage_qhash(iwdev, cm_info,
1541						 IRDMA_QHASH_TYPE_TCP_SYN,
1542						 IRDMA_QHASH_MANAGE_TYPE_DELETE,
1543						 NULL, false);
1544			child_listen_node->qhash_set = false;
1545		} else {
1546			ret = 0;
1547		}
1548		ibdev_dbg(&iwdev->ibdev,
1549			  "CM: Child listen node freed = %p\n",
1550			  child_listen_node);
1551		kfree(child_listen_node);
1552		cm_parent_listen_node->cm_core->stats_listen_nodes_destroyed++;
1553	}
1554	spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1555
1556	return ret;
1557}
1558
1559static u8 irdma_iw_get_vlan_prio(u32 *loc_addr, u8 prio, bool ipv4)
1560{
1561	struct net_device *ndev = NULL;
1562
1563	rcu_read_lock();
1564	if (ipv4) {
1565		ndev = ip_dev_find(&init_net, htonl(loc_addr[0]));
1566	} else if (IS_ENABLED(CONFIG_IPV6)) {
1567		struct net_device *ip_dev;
1568		struct in6_addr laddr6;
1569
1570		irdma_copy_ip_htonl(laddr6.in6_u.u6_addr32, loc_addr);
1571
1572		for_each_netdev_rcu (&init_net, ip_dev) {
1573			if (ipv6_chk_addr(&init_net, &laddr6, ip_dev, 1)) {
1574				ndev = ip_dev;
1575				break;
1576			}
1577		}
1578	}
1579
1580	if (!ndev)
1581		goto done;
1582	if (is_vlan_dev(ndev))
1583		prio = (vlan_dev_get_egress_qos_mask(ndev, prio) & VLAN_PRIO_MASK)
1584			>> VLAN_PRIO_SHIFT;
1585	if (ipv4)
1586		dev_put(ndev);
1587
1588done:
1589	rcu_read_unlock();
1590
1591	return prio;
1592}
1593
1594/**
1595 * irdma_get_vlan_mac_ipv6 - Gets the vlan and mac
1596 * @addr: local IPv6 address
1597 * @vlan_id: vlan id for the given IPv6 address
1598 * @mac: mac address for the given IPv6 address
1599 *
1600 * Returns the vlan id and mac for an IPv6 address.
1601 */
1602void irdma_get_vlan_mac_ipv6(u32 *addr, u16 *vlan_id, u8 *mac)
1603{
1604	struct net_device *ip_dev = NULL;
1605	struct in6_addr laddr6;
1606
1607	if (!IS_ENABLED(CONFIG_IPV6))
1608		return;
1609
1610	irdma_copy_ip_htonl(laddr6.in6_u.u6_addr32, addr);
1611	if (vlan_id)
1612		*vlan_id = 0xFFFF;	/* Match rdma_vlan_dev_vlan_id() */
1613	if (mac)
1614		eth_zero_addr(mac);
1615
1616	rcu_read_lock();
1617	for_each_netdev_rcu (&init_net, ip_dev) {
1618		if (ipv6_chk_addr(&init_net, &laddr6, ip_dev, 1)) {
1619			if (vlan_id)
1620				*vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1621			if (ip_dev->dev_addr && mac)
1622				ether_addr_copy(mac, ip_dev->dev_addr);
1623			break;
1624		}
1625	}
1626	rcu_read_unlock();
1627}
1628
1629/**
1630 * irdma_get_vlan_ipv4 - Returns the vlan_id for IPv4 address
1631 * @addr: local IPv4 address
1632 */
1633u16 irdma_get_vlan_ipv4(u32 *addr)
1634{
1635	struct net_device *netdev;
1636	u16 vlan_id = 0xFFFF;
1637
1638	netdev = ip_dev_find(&init_net, htonl(addr[0]));
1639	if (netdev) {
1640		vlan_id = rdma_vlan_dev_vlan_id(netdev);
1641		dev_put(netdev);
1642	}
1643
1644	return vlan_id;
1645}
1646
1647/**
1648 * irdma_add_mqh_6 - Adds multiple qhashes for IPv6
1649 * @iwdev: iWarp device
1650 * @cm_info: CM info for parent listen node
1651 * @cm_parent_listen_node: The parent listen node
1652 *
1653 * Adds a qhash and a child listen node for every IPv6 address
1654 * on the adapter and adds the associated qhash filter
1655 */
1656static int irdma_add_mqh_6(struct irdma_device *iwdev,
1657			   struct irdma_cm_info *cm_info,
1658			   struct irdma_cm_listener *cm_parent_listen_node)
1659{
1660	struct net_device *ip_dev;
1661	struct inet6_dev *idev;
1662	struct inet6_ifaddr *ifp, *tmp;
1663	struct irdma_cm_listener *child_listen_node;
1664	unsigned long flags;
1665	int ret = 0;
1666
1667	rtnl_lock();
1668	for_each_netdev(&init_net, ip_dev) {
1669		if (!(ip_dev->flags & IFF_UP))
1670			continue;
1671
1672		if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
1673		     (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
1674		    ip_dev != iwdev->netdev)
1675			continue;
1676
1677		idev = __in6_dev_get(ip_dev);
1678		if (!idev) {
1679			ibdev_dbg(&iwdev->ibdev, "CM: idev == NULL\n");
1680			break;
1681		}
1682		list_for_each_entry_safe (ifp, tmp, &idev->addr_list, if_list) {
1683			ibdev_dbg(&iwdev->ibdev, "CM: IP=%pI6, vlan_id=%d, MAC=%pM\n",
1684				  &ifp->addr, rdma_vlan_dev_vlan_id(ip_dev),
1685				  ip_dev->dev_addr);
1686			child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
1687			ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n",
1688				  child_listen_node);
1689			if (!child_listen_node) {
1690				ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n");
1691				ret = -ENOMEM;
1692				goto exit;
1693			}
1694
1695			cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1696			cm_parent_listen_node->vlan_id = cm_info->vlan_id;
1697			memcpy(child_listen_node, cm_parent_listen_node,
1698			       sizeof(*child_listen_node));
1699			irdma_copy_ip_ntohl(child_listen_node->loc_addr,
1700					    ifp->addr.in6_u.u6_addr32);
1701			memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1702			       sizeof(cm_info->loc_addr));
1703			if (!iwdev->vsi.dscp_mode)
1704				cm_info->user_pri =
1705				irdma_iw_get_vlan_prio(child_listen_node->loc_addr,
1706						       cm_info->user_pri,
1707						       false);
1708
1709			ret = irdma_manage_qhash(iwdev, cm_info,
1710						 IRDMA_QHASH_TYPE_TCP_SYN,
1711						 IRDMA_QHASH_MANAGE_TYPE_ADD,
1712						 NULL, true);
1713			if (ret) {
1714				kfree(child_listen_node);
1715				continue;
1716			}
1717
1718			trace_irdma_add_mqh_6(iwdev, child_listen_node,
1719					      ip_dev->dev_addr);
1720
1721			child_listen_node->qhash_set = true;
1722			spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
1723			list_add(&child_listen_node->child_listen_list,
1724				 &cm_parent_listen_node->child_listen_list);
1725			spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1726			cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
1727		}
1728	}
1729exit:
1730	rtnl_unlock();
1731
1732	return ret;
1733}
1734
1735/**
1736 * irdma_add_mqh_4 - Adds multiple qhashes for IPv4
1737 * @iwdev: iWarp device
1738 * @cm_info: CM info for parent listen node
1739 * @cm_parent_listen_node: The parent listen node
1740 *
1741 * Adds a qhash and a child listen node for every IPv4 address
1742 * on the adapter and adds the associated qhash filter
1743 */
1744static int irdma_add_mqh_4(struct irdma_device *iwdev,
1745			   struct irdma_cm_info *cm_info,
1746			   struct irdma_cm_listener *cm_parent_listen_node)
1747{
1748	struct net_device *ip_dev;
1749	struct in_device *idev;
1750	struct irdma_cm_listener *child_listen_node;
1751	unsigned long flags;
1752	const struct in_ifaddr *ifa;
1753	int ret = 0;
1754
1755	rtnl_lock();
1756	for_each_netdev(&init_net, ip_dev) {
1757		if (!(ip_dev->flags & IFF_UP))
1758			continue;
1759
1760		if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
1761		     (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
1762		    ip_dev != iwdev->netdev)
1763			continue;
1764
1765		idev = in_dev_get(ip_dev);
1766		if (!idev)
1767			continue;
1768
1769		in_dev_for_each_ifa_rtnl(ifa, idev) {
1770			ibdev_dbg(&iwdev->ibdev,
1771				  "CM: Allocating child CM Listener forIP=%pI4, vlan_id=%d, MAC=%pM\n",
1772				  &ifa->ifa_address, rdma_vlan_dev_vlan_id(ip_dev),
1773				  ip_dev->dev_addr);
1774			child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
1775			cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
1776			ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n",
1777				  child_listen_node);
1778			if (!child_listen_node) {
1779				ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n");
1780				in_dev_put(idev);
1781				ret = -ENOMEM;
1782				goto exit;
1783			}
1784
1785			cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
1786			cm_parent_listen_node->vlan_id = cm_info->vlan_id;
1787			memcpy(child_listen_node, cm_parent_listen_node,
1788			       sizeof(*child_listen_node));
1789			child_listen_node->loc_addr[0] =
1790				ntohl(ifa->ifa_address);
1791			memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
1792			       sizeof(cm_info->loc_addr));
1793			if (!iwdev->vsi.dscp_mode)
1794				cm_info->user_pri =
1795				irdma_iw_get_vlan_prio(child_listen_node->loc_addr,
1796						       cm_info->user_pri,
1797						       true);
1798			ret = irdma_manage_qhash(iwdev, cm_info,
1799						 IRDMA_QHASH_TYPE_TCP_SYN,
1800						 IRDMA_QHASH_MANAGE_TYPE_ADD,
1801						 NULL, true);
1802			if (ret) {
1803				kfree(child_listen_node);
1804				cm_parent_listen_node->cm_core
1805					->stats_listen_nodes_created--;
1806				continue;
1807			}
1808
1809			trace_irdma_add_mqh_4(iwdev, child_listen_node,
1810					      ip_dev->dev_addr);
1811
1812			child_listen_node->qhash_set = true;
1813			spin_lock_irqsave(&iwdev->cm_core.listen_list_lock,
1814					  flags);
1815			list_add(&child_listen_node->child_listen_list,
1816				 &cm_parent_listen_node->child_listen_list);
1817			spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
1818		}
1819		in_dev_put(idev);
1820	}
1821exit:
1822	rtnl_unlock();
1823
1824	return ret;
1825}
1826
1827/**
1828 * irdma_add_mqh - Adds multiple qhashes
1829 * @iwdev: iWarp device
1830 * @cm_info: CM info for parent listen node
1831 * @cm_listen_node: The parent listen node
1832 */
1833static int irdma_add_mqh(struct irdma_device *iwdev,
1834			 struct irdma_cm_info *cm_info,
1835			 struct irdma_cm_listener *cm_listen_node)
1836{
1837	if (cm_info->ipv4)
1838		return irdma_add_mqh_4(iwdev, cm_info, cm_listen_node);
1839	else
1840		return irdma_add_mqh_6(iwdev, cm_info, cm_listen_node);
1841}
1842
1843/**
1844 * irdma_reset_list_prep - add connection nodes slated for reset to list
1845 * @cm_core: cm's core
1846 * @listener: pointer to listener node
1847 * @reset_list: a list to which cm_node will be selected
1848 */
1849static void irdma_reset_list_prep(struct irdma_cm_core *cm_core,
1850				  struct irdma_cm_listener *listener,
1851				  struct list_head *reset_list)
1852{
1853	struct irdma_cm_node *cm_node;
1854	int bkt;
1855
1856	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
1857		if (cm_node->listener == listener &&
1858		    !cm_node->accelerated &&
1859		    refcount_inc_not_zero(&cm_node->refcnt))
1860			list_add(&cm_node->reset_entry, reset_list);
1861	}
1862}
1863
1864/**
1865 * irdma_dec_refcnt_listen - delete listener and associated cm nodes
1866 * @cm_core: cm's core
1867 * @listener: pointer to listener node
1868 * @free_hanging_nodes: to free associated cm_nodes
1869 * @apbvt_del: flag to delete the apbvt
1870 */
1871static int irdma_dec_refcnt_listen(struct irdma_cm_core *cm_core,
1872				   struct irdma_cm_listener *listener,
1873				   int free_hanging_nodes, bool apbvt_del)
1874{
1875	int err;
1876	struct list_head *list_pos;
1877	struct list_head *list_temp;
1878	struct irdma_cm_node *cm_node;
1879	struct list_head reset_list;
1880	struct irdma_cm_info nfo;
1881	enum irdma_cm_node_state old_state;
1882	unsigned long flags;
1883
1884	trace_irdma_dec_refcnt_listen(listener, __builtin_return_address(0));
1885	/* free non-accelerated child nodes for this listener */
1886	INIT_LIST_HEAD(&reset_list);
1887	if (free_hanging_nodes) {
1888		rcu_read_lock();
1889		irdma_reset_list_prep(cm_core, listener, &reset_list);
1890		rcu_read_unlock();
1891	}
1892
1893	list_for_each_safe (list_pos, list_temp, &reset_list) {
1894		cm_node = container_of(list_pos, struct irdma_cm_node,
1895				       reset_entry);
1896		if (cm_node->state >= IRDMA_CM_STATE_FIN_WAIT1) {
1897			irdma_rem_ref_cm_node(cm_node);
1898			continue;
1899		}
1900
1901		irdma_cleanup_retrans_entry(cm_node);
1902		err = irdma_send_reset(cm_node);
1903		if (err) {
1904			cm_node->state = IRDMA_CM_STATE_CLOSED;
1905			ibdev_dbg(&cm_node->iwdev->ibdev,
1906				  "CM: send reset failed\n");
1907		} else {
1908			old_state = cm_node->state;
1909			cm_node->state = IRDMA_CM_STATE_LISTENER_DESTROYED;
1910			if (old_state != IRDMA_CM_STATE_MPAREQ_RCVD)
1911				irdma_rem_ref_cm_node(cm_node);
1912		}
1913	}
1914
1915	if (refcount_dec_and_test(&listener->refcnt)) {
1916		spin_lock_irqsave(&cm_core->listen_list_lock, flags);
1917		list_del(&listener->list);
1918		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
1919
1920		if (apbvt_del)
1921			irdma_del_apbvt(listener->iwdev,
1922					listener->apbvt_entry);
1923		memcpy(nfo.loc_addr, listener->loc_addr, sizeof(nfo.loc_addr));
1924		nfo.loc_port = listener->loc_port;
1925		nfo.ipv4 = listener->ipv4;
1926		nfo.vlan_id = listener->vlan_id;
1927		nfo.user_pri = listener->user_pri;
1928		nfo.qh_qpid = listener->iwdev->vsi.ilq->qp_id;
1929
1930		if (!list_empty(&listener->child_listen_list)) {
1931			irdma_del_multiple_qhash(listener->iwdev, &nfo,
1932						 listener);
1933		} else {
1934			if (listener->qhash_set)
1935				irdma_manage_qhash(listener->iwdev,
1936						   &nfo,
1937						   IRDMA_QHASH_TYPE_TCP_SYN,
1938						   IRDMA_QHASH_MANAGE_TYPE_DELETE,
1939						   NULL, false);
1940		}
1941
1942		cm_core->stats_listen_destroyed++;
1943		cm_core->stats_listen_nodes_destroyed++;
1944		ibdev_dbg(&listener->iwdev->ibdev,
1945			  "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d apbvt_del=%d\n",
1946			  listener->loc_port, listener->loc_addr, listener,
1947			  listener->cm_id, listener->qhash_set,
1948			  listener->vlan_id, apbvt_del);
1949		kfree(listener);
1950		listener = NULL;
1951		return 0;
1952	}
1953
1954	return -EINVAL;
1955}
1956
1957/**
1958 * irdma_cm_del_listen - delete a listener
1959 * @cm_core: cm's core
1960 * @listener: passive connection's listener
1961 * @apbvt_del: flag to delete apbvt
1962 */
1963static int irdma_cm_del_listen(struct irdma_cm_core *cm_core,
1964			       struct irdma_cm_listener *listener,
1965			       bool apbvt_del)
1966{
1967	listener->listener_state = IRDMA_CM_LISTENER_PASSIVE_STATE;
1968	listener->cm_id = NULL;
1969
1970	return irdma_dec_refcnt_listen(cm_core, listener, 1, apbvt_del);
1971}
1972
1973/**
1974 * irdma_addr_resolve_neigh - resolve neighbor address
1975 * @iwdev: iwarp device structure
1976 * @src_ip: local ip address
1977 * @dst_ip: remote ip address
1978 * @arpindex: if there is an arp entry
1979 */
1980static int irdma_addr_resolve_neigh(struct irdma_device *iwdev, u32 src_ip,
1981				    u32 dst_ip, int arpindex)
1982{
1983	struct rtable *rt;
1984	struct neighbour *neigh;
1985	int rc = arpindex;
1986	__be32 dst_ipaddr = htonl(dst_ip);
1987	__be32 src_ipaddr = htonl(src_ip);
1988
1989	rt = ip_route_output(&init_net, dst_ipaddr, src_ipaddr, 0, 0,
1990			     RT_SCOPE_UNIVERSE);
1991	if (IS_ERR(rt)) {
1992		ibdev_dbg(&iwdev->ibdev, "CM: ip_route_output fail\n");
1993		return -EINVAL;
1994	}
1995
1996	neigh = dst_neigh_lookup(&rt->dst, &dst_ipaddr);
1997	if (!neigh)
1998		goto exit;
1999
2000	if (neigh->nud_state & NUD_VALID)
2001		rc = irdma_add_arp(iwdev->rf, &dst_ip, true, neigh->ha);
2002	else
2003		neigh_event_send(neigh, NULL);
2004	if (neigh)
2005		neigh_release(neigh);
2006exit:
2007	ip_rt_put(rt);
2008
2009	return rc;
2010}
2011
2012/**
2013 * irdma_get_dst_ipv6 - get destination cache entry via ipv6 lookup
2014 * @src_addr: local ipv6 sock address
2015 * @dst_addr: destination ipv6 sock address
2016 */
2017static struct dst_entry *irdma_get_dst_ipv6(struct sockaddr_in6 *src_addr,
2018					    struct sockaddr_in6 *dst_addr)
2019{
2020	struct dst_entry *dst = NULL;
2021
2022	if ((IS_ENABLED(CONFIG_IPV6))) {
2023		struct flowi6 fl6 = {};
2024
2025		fl6.daddr = dst_addr->sin6_addr;
2026		fl6.saddr = src_addr->sin6_addr;
2027		if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL)
2028			fl6.flowi6_oif = dst_addr->sin6_scope_id;
2029
2030		dst = ip6_route_output(&init_net, NULL, &fl6);
2031	}
2032
2033	return dst;
2034}
2035
2036/**
2037 * irdma_addr_resolve_neigh_ipv6 - resolve neighbor ipv6 address
2038 * @iwdev: iwarp device structure
2039 * @src: local ip address
2040 * @dest: remote ip address
2041 * @arpindex: if there is an arp entry
2042 */
2043static int irdma_addr_resolve_neigh_ipv6(struct irdma_device *iwdev, u32 *src,
2044					 u32 *dest, int arpindex)
2045{
2046	struct neighbour *neigh;
2047	int rc = arpindex;
2048	struct dst_entry *dst;
2049	struct sockaddr_in6 dst_addr = {};
2050	struct sockaddr_in6 src_addr = {};
2051
2052	dst_addr.sin6_family = AF_INET6;
2053	irdma_copy_ip_htonl(dst_addr.sin6_addr.in6_u.u6_addr32, dest);
2054	src_addr.sin6_family = AF_INET6;
2055	irdma_copy_ip_htonl(src_addr.sin6_addr.in6_u.u6_addr32, src);
2056	dst = irdma_get_dst_ipv6(&src_addr, &dst_addr);
2057	if (!dst || dst->error) {
2058		if (dst) {
2059			dst_release(dst);
2060			ibdev_dbg(&iwdev->ibdev,
2061				  "CM: ip6_route_output returned dst->error = %d\n",
2062				  dst->error);
2063		}
2064		return -EINVAL;
2065	}
2066
2067	neigh = dst_neigh_lookup(dst, dst_addr.sin6_addr.in6_u.u6_addr32);
2068	if (!neigh)
2069		goto exit;
2070
2071	ibdev_dbg(&iwdev->ibdev, "CM: dst_neigh_lookup MAC=%pM\n",
2072		  neigh->ha);
2073
2074	trace_irdma_addr_resolve(iwdev, neigh->ha);
2075
2076	if (neigh->nud_state & NUD_VALID)
2077		rc = irdma_add_arp(iwdev->rf, dest, false, neigh->ha);
2078	else
2079		neigh_event_send(neigh, NULL);
2080	if (neigh)
2081		neigh_release(neigh);
2082exit:
2083	dst_release(dst);
2084
2085	return rc;
2086}
2087
2088/**
2089 * irdma_find_node - find a cm node that matches the reference cm node
2090 * @cm_core: cm's core
2091 * @rem_port: remote tcp port num
2092 * @rem_addr: remote ip addr
2093 * @loc_port: local tcp port num
2094 * @loc_addr: local ip addr
2095 * @vlan_id: local VLAN ID
2096 */
2097struct irdma_cm_node *irdma_find_node(struct irdma_cm_core *cm_core,
2098				      u16 rem_port, u32 *rem_addr, u16 loc_port,
2099				      u32 *loc_addr, u16 vlan_id)
2100{
2101	struct irdma_cm_node *cm_node;
2102	u32 key = (rem_port << 16) | loc_port;
2103
2104	rcu_read_lock();
2105	hash_for_each_possible_rcu(cm_core->cm_hash_tbl, cm_node, list, key) {
2106		if (cm_node->vlan_id == vlan_id &&
2107		    cm_node->loc_port == loc_port && cm_node->rem_port == rem_port &&
2108		    !memcmp(cm_node->loc_addr, loc_addr, sizeof(cm_node->loc_addr)) &&
2109		    !memcmp(cm_node->rem_addr, rem_addr, sizeof(cm_node->rem_addr))) {
2110			if (!refcount_inc_not_zero(&cm_node->refcnt))
2111				goto exit;
2112			rcu_read_unlock();
2113			trace_irdma_find_node(cm_node, 0, NULL);
2114			return cm_node;
2115		}
2116	}
2117
2118exit:
2119	rcu_read_unlock();
2120
2121	/* no owner node */
2122	return NULL;
2123}
2124
2125/**
2126 * irdma_add_hte_node - add a cm node to the hash table
2127 * @cm_core: cm's core
2128 * @cm_node: connection's node
2129 */
2130static void irdma_add_hte_node(struct irdma_cm_core *cm_core,
2131			       struct irdma_cm_node *cm_node)
2132{
2133	unsigned long flags;
2134	u32 key = (cm_node->rem_port << 16) | cm_node->loc_port;
2135
2136	spin_lock_irqsave(&cm_core->ht_lock, flags);
2137	hash_add_rcu(cm_core->cm_hash_tbl, &cm_node->list, key);
2138	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2139}
2140
2141/**
2142 * irdma_ipv4_is_lpb - check if loopback
2143 * @loc_addr: local addr to compare
2144 * @rem_addr: remote address
2145 */
2146bool irdma_ipv4_is_lpb(u32 loc_addr, u32 rem_addr)
2147{
2148	return ipv4_is_loopback(htonl(rem_addr)) || (loc_addr == rem_addr);
2149}
2150
2151/**
2152 * irdma_ipv6_is_lpb - check if loopback
2153 * @loc_addr: local addr to compare
2154 * @rem_addr: remote address
2155 */
2156bool irdma_ipv6_is_lpb(u32 *loc_addr, u32 *rem_addr)
2157{
2158	struct in6_addr raddr6;
2159
2160	irdma_copy_ip_htonl(raddr6.in6_u.u6_addr32, rem_addr);
2161
2162	return !memcmp(loc_addr, rem_addr, 16) || ipv6_addr_loopback(&raddr6);
2163}
2164
2165/**
2166 * irdma_cm_create_ah - create a cm address handle
2167 * @cm_node: The connection manager node to create AH for
2168 * @wait: Provides option to wait for ah creation or not
2169 */
2170static int irdma_cm_create_ah(struct irdma_cm_node *cm_node, bool wait)
2171{
2172	struct irdma_ah_info ah_info = {};
2173	struct irdma_device *iwdev = cm_node->iwdev;
2174
2175	ether_addr_copy(ah_info.mac_addr, iwdev->netdev->dev_addr);
2176
2177	ah_info.hop_ttl = 0x40;
2178	ah_info.tc_tos = cm_node->tos;
2179	ah_info.vsi = &iwdev->vsi;
2180
2181	if (cm_node->ipv4) {
2182		ah_info.ipv4_valid = true;
2183		ah_info.dest_ip_addr[0] = cm_node->rem_addr[0];
2184		ah_info.src_ip_addr[0] = cm_node->loc_addr[0];
2185		ah_info.do_lpbk = irdma_ipv4_is_lpb(ah_info.src_ip_addr[0],
2186						    ah_info.dest_ip_addr[0]);
2187	} else {
2188		memcpy(ah_info.dest_ip_addr, cm_node->rem_addr,
2189		       sizeof(ah_info.dest_ip_addr));
2190		memcpy(ah_info.src_ip_addr, cm_node->loc_addr,
2191		       sizeof(ah_info.src_ip_addr));
2192		ah_info.do_lpbk = irdma_ipv6_is_lpb(ah_info.src_ip_addr,
2193						    ah_info.dest_ip_addr);
2194	}
2195
2196	ah_info.vlan_tag = cm_node->vlan_id;
2197	if (cm_node->vlan_id < VLAN_N_VID) {
2198		ah_info.insert_vlan_tag = 1;
2199		ah_info.vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
2200	}
2201
2202	ah_info.dst_arpindex =
2203		irdma_arp_table(iwdev->rf, ah_info.dest_ip_addr,
2204				ah_info.ipv4_valid, NULL, IRDMA_ARP_RESOLVE);
2205
2206	if (irdma_puda_create_ah(&iwdev->rf->sc_dev, &ah_info, wait,
2207				 IRDMA_PUDA_RSRC_TYPE_ILQ, cm_node,
2208				 &cm_node->ah))
2209		return -ENOMEM;
2210
2211	trace_irdma_create_ah(cm_node);
2212	return 0;
2213}
2214
2215/**
2216 * irdma_cm_free_ah - free a cm address handle
2217 * @cm_node: The connection manager node to create AH for
2218 */
2219static void irdma_cm_free_ah(struct irdma_cm_node *cm_node)
2220{
2221	struct irdma_device *iwdev = cm_node->iwdev;
2222
2223	trace_irdma_cm_free_ah(cm_node);
2224	irdma_puda_free_ah(&iwdev->rf->sc_dev, cm_node->ah);
2225	cm_node->ah = NULL;
2226}
2227
2228/**
2229 * irdma_make_cm_node - create a new instance of a cm node
2230 * @cm_core: cm's core
2231 * @iwdev: iwarp device structure
2232 * @cm_info: quad info for connection
2233 * @listener: passive connection's listener
2234 */
2235static struct irdma_cm_node *
2236irdma_make_cm_node(struct irdma_cm_core *cm_core, struct irdma_device *iwdev,
2237		   struct irdma_cm_info *cm_info,
2238		   struct irdma_cm_listener *listener)
2239{
2240	struct irdma_cm_node *cm_node;
2241	int oldarpindex;
2242	int arpindex;
2243	struct net_device *netdev = iwdev->netdev;
2244	int ret;
2245
2246	/* create an hte and cm_node for this instance */
2247	cm_node = kzalloc(sizeof(*cm_node), GFP_ATOMIC);
2248	if (!cm_node)
2249		return ERR_PTR(-ENOMEM);
2250
2251	/* set our node specific transport info */
2252	cm_node->ipv4 = cm_info->ipv4;
2253	cm_node->vlan_id = cm_info->vlan_id;
2254	if (cm_node->vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
2255		cm_node->vlan_id = 0;
2256	cm_node->tos = cm_info->tos;
2257	cm_node->user_pri = cm_info->user_pri;
2258	if (listener) {
2259		if (listener->tos != cm_info->tos)
2260			ibdev_warn(&iwdev->ibdev,
2261				   "application TOS[%d] and remote client TOS[%d] mismatch\n",
2262				   listener->tos, cm_info->tos);
2263		if (iwdev->vsi.dscp_mode) {
2264			cm_node->user_pri = listener->user_pri;
2265		} else {
2266			cm_node->tos = max(listener->tos, cm_info->tos);
2267			cm_node->user_pri = rt_tos2priority(cm_node->tos);
2268			cm_node->user_pri =
2269				irdma_iw_get_vlan_prio(cm_info->loc_addr,
2270						       cm_node->user_pri,
2271						       cm_info->ipv4);
2272		}
2273		ibdev_dbg(&iwdev->ibdev,
2274			  "DCB: listener: TOS:[%d] UP:[%d]\n", cm_node->tos,
2275			  cm_node->user_pri);
2276		trace_irdma_listener_tos(iwdev, cm_node->tos,
2277					 cm_node->user_pri);
2278	}
2279	memcpy(cm_node->loc_addr, cm_info->loc_addr, sizeof(cm_node->loc_addr));
2280	memcpy(cm_node->rem_addr, cm_info->rem_addr, sizeof(cm_node->rem_addr));
2281	cm_node->loc_port = cm_info->loc_port;
2282	cm_node->rem_port = cm_info->rem_port;
2283
2284	cm_node->mpa_frame_rev = IRDMA_CM_DEFAULT_MPA_VER;
2285	cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
2286	cm_node->iwdev = iwdev;
2287	cm_node->dev = &iwdev->rf->sc_dev;
2288
2289	cm_node->ird_size = cm_node->dev->hw_attrs.max_hw_ird;
2290	cm_node->ord_size = cm_node->dev->hw_attrs.max_hw_ord;
2291
2292	cm_node->listener = listener;
2293	cm_node->cm_id = cm_info->cm_id;
2294	ether_addr_copy(cm_node->loc_mac, netdev->dev_addr);
2295	spin_lock_init(&cm_node->retrans_list_lock);
2296	cm_node->ack_rcvd = false;
2297
2298	init_completion(&cm_node->establish_comp);
2299	refcount_set(&cm_node->refcnt, 1);
2300	/* associate our parent CM core */
2301	cm_node->cm_core = cm_core;
2302	cm_node->tcp_cntxt.loc_id = IRDMA_CM_DEFAULT_LOCAL_ID;
2303	cm_node->tcp_cntxt.rcv_wscale = iwdev->rcv_wscale;
2304	cm_node->tcp_cntxt.rcv_wnd = iwdev->rcv_wnd >> cm_node->tcp_cntxt.rcv_wscale;
2305	if (cm_node->ipv4) {
2306		cm_node->tcp_cntxt.loc_seq_num = secure_tcp_seq(htonl(cm_node->loc_addr[0]),
2307								htonl(cm_node->rem_addr[0]),
2308								htons(cm_node->loc_port),
2309								htons(cm_node->rem_port));
2310		cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV4;
2311	} else if (IS_ENABLED(CONFIG_IPV6)) {
2312		__be32 loc[4] = {
2313			htonl(cm_node->loc_addr[0]), htonl(cm_node->loc_addr[1]),
2314			htonl(cm_node->loc_addr[2]), htonl(cm_node->loc_addr[3])
2315		};
2316		__be32 rem[4] = {
2317			htonl(cm_node->rem_addr[0]), htonl(cm_node->rem_addr[1]),
2318			htonl(cm_node->rem_addr[2]), htonl(cm_node->rem_addr[3])
2319		};
2320		cm_node->tcp_cntxt.loc_seq_num = secure_tcpv6_seq(loc, rem,
2321								  htons(cm_node->loc_port),
2322								  htons(cm_node->rem_port));
2323		cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV6;
2324	}
2325
2326	if ((cm_node->ipv4 &&
2327	     irdma_ipv4_is_lpb(cm_node->loc_addr[0], cm_node->rem_addr[0])) ||
2328	    (!cm_node->ipv4 &&
2329	     irdma_ipv6_is_lpb(cm_node->loc_addr, cm_node->rem_addr))) {
2330		cm_node->do_lpb = true;
2331		arpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr,
2332					   cm_node->ipv4, NULL,
2333					   IRDMA_ARP_RESOLVE);
2334	} else {
2335		oldarpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr,
2336					      cm_node->ipv4, NULL,
2337					      IRDMA_ARP_RESOLVE);
2338		if (cm_node->ipv4)
2339			arpindex = irdma_addr_resolve_neigh(iwdev,
2340							    cm_info->loc_addr[0],
2341							    cm_info->rem_addr[0],
2342							    oldarpindex);
2343		else if (IS_ENABLED(CONFIG_IPV6))
2344			arpindex = irdma_addr_resolve_neigh_ipv6(iwdev,
2345								 cm_info->loc_addr,
2346								 cm_info->rem_addr,
2347								 oldarpindex);
2348		else
2349			arpindex = -EINVAL;
2350	}
2351
2352	if (arpindex < 0) {
2353		ret = -EINVAL;
2354		goto err;
2355	}
2356
2357	ether_addr_copy(cm_node->rem_mac,
2358			iwdev->rf->arp_table[arpindex].mac_addr);
2359	irdma_add_hte_node(cm_core, cm_node);
2360	cm_core->stats_nodes_created++;
2361	return cm_node;
2362
2363err:
2364	kfree(cm_node);
2365
2366	return ERR_PTR(ret);
2367}
2368
2369static void irdma_destroy_connection(struct irdma_cm_node *cm_node)
2370{
2371	struct irdma_cm_core *cm_core = cm_node->cm_core;
2372	struct irdma_qp *iwqp;
2373	struct irdma_cm_info nfo;
2374
2375	/* if the node is destroyed before connection was accelerated */
2376	if (!cm_node->accelerated && cm_node->accept_pend) {
2377		ibdev_dbg(&cm_node->iwdev->ibdev,
2378			  "CM: node destroyed before established\n");
2379		atomic_dec(&cm_node->listener->pend_accepts_cnt);
2380	}
2381	if (cm_node->close_entry)
2382		irdma_handle_close_entry(cm_node, 0);
2383	if (cm_node->listener) {
2384		irdma_dec_refcnt_listen(cm_core, cm_node->listener, 0, true);
2385	} else {
2386		if (cm_node->apbvt_set) {
2387			irdma_del_apbvt(cm_node->iwdev, cm_node->apbvt_entry);
2388			cm_node->apbvt_set = 0;
2389		}
2390		irdma_get_addr_info(cm_node, &nfo);
2391		if (cm_node->qhash_set) {
2392			nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2393			irdma_manage_qhash(cm_node->iwdev, &nfo,
2394					   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2395					   IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL,
2396					   false);
2397			cm_node->qhash_set = 0;
2398		}
2399	}
2400
2401	iwqp = cm_node->iwqp;
2402	if (iwqp) {
2403		cm_node->cm_id->rem_ref(cm_node->cm_id);
2404		cm_node->cm_id = NULL;
2405		iwqp->cm_id = NULL;
2406		irdma_qp_rem_ref(&iwqp->ibqp);
2407		cm_node->iwqp = NULL;
2408	} else if (cm_node->qhash_set) {
2409		irdma_get_addr_info(cm_node, &nfo);
2410		nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2411		irdma_manage_qhash(cm_node->iwdev, &nfo,
2412				   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2413				   IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL, false);
2414		cm_node->qhash_set = 0;
2415	}
2416
2417	cm_core->cm_free_ah(cm_node);
2418}
2419
2420/**
2421 * irdma_rem_ref_cm_node - destroy an instance of a cm node
2422 * @cm_node: connection's node
2423 */
2424void irdma_rem_ref_cm_node(struct irdma_cm_node *cm_node)
2425{
2426	struct irdma_cm_core *cm_core = cm_node->cm_core;
2427	unsigned long flags;
2428
2429	trace_irdma_rem_ref_cm_node(cm_node, 0, __builtin_return_address(0));
2430	spin_lock_irqsave(&cm_core->ht_lock, flags);
2431
2432	if (!refcount_dec_and_test(&cm_node->refcnt)) {
2433		spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2434		return;
2435	}
2436	if (cm_node->iwqp) {
2437		cm_node->iwqp->cm_node = NULL;
2438		cm_node->iwqp->cm_id = NULL;
2439	}
2440	hash_del_rcu(&cm_node->list);
2441	cm_node->cm_core->stats_nodes_destroyed++;
2442
2443	spin_unlock_irqrestore(&cm_core->ht_lock, flags);
2444
2445	irdma_destroy_connection(cm_node);
2446
2447	kfree_rcu(cm_node, rcu_head);
2448}
2449
2450/**
2451 * irdma_handle_fin_pkt - FIN packet received
2452 * @cm_node: connection's node
2453 */
2454static void irdma_handle_fin_pkt(struct irdma_cm_node *cm_node)
2455{
2456	switch (cm_node->state) {
2457	case IRDMA_CM_STATE_SYN_RCVD:
2458	case IRDMA_CM_STATE_SYN_SENT:
2459	case IRDMA_CM_STATE_ESTABLISHED:
2460	case IRDMA_CM_STATE_MPAREJ_RCVD:
2461		cm_node->tcp_cntxt.rcv_nxt++;
2462		irdma_cleanup_retrans_entry(cm_node);
2463		cm_node->state = IRDMA_CM_STATE_LAST_ACK;
2464		irdma_send_fin(cm_node);
2465		break;
2466	case IRDMA_CM_STATE_MPAREQ_SENT:
2467		irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
2468		cm_node->tcp_cntxt.rcv_nxt++;
2469		irdma_cleanup_retrans_entry(cm_node);
2470		cm_node->state = IRDMA_CM_STATE_CLOSED;
2471		refcount_inc(&cm_node->refcnt);
2472		irdma_send_reset(cm_node);
2473		break;
2474	case IRDMA_CM_STATE_FIN_WAIT1:
2475		cm_node->tcp_cntxt.rcv_nxt++;
2476		irdma_cleanup_retrans_entry(cm_node);
2477		cm_node->state = IRDMA_CM_STATE_CLOSING;
2478		irdma_send_ack(cm_node);
2479		/*
2480		 * Wait for ACK as this is simultaneous close.
2481		 * After we receive ACK, do not send anything.
2482		 * Just rm the node.
2483		 */
2484		break;
2485	case IRDMA_CM_STATE_FIN_WAIT2:
2486		cm_node->tcp_cntxt.rcv_nxt++;
2487		irdma_cleanup_retrans_entry(cm_node);
2488		cm_node->state = IRDMA_CM_STATE_TIME_WAIT;
2489		irdma_send_ack(cm_node);
2490		irdma_schedule_cm_timer(cm_node, NULL, IRDMA_TIMER_TYPE_CLOSE,
2491					1, 0);
2492		break;
2493	case IRDMA_CM_STATE_TIME_WAIT:
2494		cm_node->tcp_cntxt.rcv_nxt++;
2495		irdma_cleanup_retrans_entry(cm_node);
2496		cm_node->state = IRDMA_CM_STATE_CLOSED;
2497		irdma_rem_ref_cm_node(cm_node);
2498		break;
2499	case IRDMA_CM_STATE_OFFLOADED:
2500	default:
2501		ibdev_dbg(&cm_node->iwdev->ibdev,
2502			  "CM: bad state node state = %d\n", cm_node->state);
2503		break;
2504	}
2505}
2506
2507/**
2508 * irdma_handle_rst_pkt - process received RST packet
2509 * @cm_node: connection's node
2510 * @rbuf: receive buffer
2511 */
2512static void irdma_handle_rst_pkt(struct irdma_cm_node *cm_node,
2513				 struct irdma_puda_buf *rbuf)
2514{
2515	ibdev_dbg(&cm_node->iwdev->ibdev,
2516		  "CM: caller: %pS cm_node=%p state=%d rem_port=0x%04x loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n",
2517		  __builtin_return_address(0), cm_node, cm_node->state,
2518		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
2519		  cm_node->loc_addr);
2520
2521	irdma_cleanup_retrans_entry(cm_node);
2522	switch (cm_node->state) {
2523	case IRDMA_CM_STATE_SYN_SENT:
2524	case IRDMA_CM_STATE_MPAREQ_SENT:
2525		switch (cm_node->mpa_frame_rev) {
2526		case IETF_MPA_V2:
2527			/* Drop down to MPA_V1*/
2528			cm_node->mpa_frame_rev = IETF_MPA_V1;
2529			/* send a syn and goto syn sent state */
2530			cm_node->state = IRDMA_CM_STATE_SYN_SENT;
2531			if (irdma_send_syn(cm_node, 0))
2532				irdma_active_open_err(cm_node, false);
2533			break;
2534		case IETF_MPA_V1:
2535		default:
2536			irdma_active_open_err(cm_node, false);
2537			break;
2538		}
2539		break;
2540	case IRDMA_CM_STATE_MPAREQ_RCVD:
2541		atomic_inc(&cm_node->passive_state);
2542		break;
2543	case IRDMA_CM_STATE_ESTABLISHED:
2544	case IRDMA_CM_STATE_SYN_RCVD:
2545	case IRDMA_CM_STATE_LISTENING:
2546		irdma_passive_open_err(cm_node, false);
2547		break;
2548	case IRDMA_CM_STATE_OFFLOADED:
2549		irdma_active_open_err(cm_node, false);
2550		break;
2551	case IRDMA_CM_STATE_CLOSED:
2552		break;
2553	case IRDMA_CM_STATE_FIN_WAIT2:
2554	case IRDMA_CM_STATE_FIN_WAIT1:
2555	case IRDMA_CM_STATE_LAST_ACK:
2556	case IRDMA_CM_STATE_TIME_WAIT:
2557		cm_node->state = IRDMA_CM_STATE_CLOSED;
2558		irdma_rem_ref_cm_node(cm_node);
2559		break;
2560	default:
2561		break;
2562	}
2563}
2564
2565/**
2566 * irdma_handle_rcv_mpa - Process a recv'd mpa buffer
2567 * @cm_node: connection's node
2568 * @rbuf: receive buffer
2569 */
2570static void irdma_handle_rcv_mpa(struct irdma_cm_node *cm_node,
2571				 struct irdma_puda_buf *rbuf)
2572{
2573	int err;
2574	int datasize = rbuf->datalen;
2575	u8 *dataloc = rbuf->data;
2576
2577	enum irdma_cm_event_type type = IRDMA_CM_EVENT_UNKNOWN;
2578	u32 res_type;
2579
2580	err = irdma_parse_mpa(cm_node, dataloc, &res_type, datasize);
2581	if (err) {
2582		if (cm_node->state == IRDMA_CM_STATE_MPAREQ_SENT)
2583			irdma_active_open_err(cm_node, true);
2584		else
2585			irdma_passive_open_err(cm_node, true);
2586		return;
2587	}
2588
2589	switch (cm_node->state) {
2590	case IRDMA_CM_STATE_ESTABLISHED:
2591		if (res_type == IRDMA_MPA_REQUEST_REJECT)
2592			ibdev_dbg(&cm_node->iwdev->ibdev,
2593				  "CM: state for reject\n");
2594		cm_node->state = IRDMA_CM_STATE_MPAREQ_RCVD;
2595		type = IRDMA_CM_EVENT_MPA_REQ;
2596		irdma_send_ack(cm_node); /* ACK received MPA request */
2597		atomic_set(&cm_node->passive_state,
2598			   IRDMA_PASSIVE_STATE_INDICATED);
2599		break;
2600	case IRDMA_CM_STATE_MPAREQ_SENT:
2601		irdma_cleanup_retrans_entry(cm_node);
2602		if (res_type == IRDMA_MPA_REQUEST_REJECT) {
2603			type = IRDMA_CM_EVENT_MPA_REJECT;
2604			cm_node->state = IRDMA_CM_STATE_MPAREJ_RCVD;
2605		} else {
2606			type = IRDMA_CM_EVENT_CONNECTED;
2607			cm_node->state = IRDMA_CM_STATE_OFFLOADED;
2608		}
2609		irdma_send_ack(cm_node);
2610		break;
2611	default:
2612		ibdev_dbg(&cm_node->iwdev->ibdev,
2613			  "CM: wrong cm_node state =%d\n", cm_node->state);
2614		break;
2615	}
2616	irdma_create_event(cm_node, type);
2617}
2618
2619/**
2620 * irdma_check_syn - Check for error on received syn ack
2621 * @cm_node: connection's node
2622 * @tcph: pointer tcp header
2623 */
2624static int irdma_check_syn(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
2625{
2626	if (ntohl(tcph->ack_seq) != cm_node->tcp_cntxt.loc_seq_num) {
2627		irdma_active_open_err(cm_node, true);
2628		return 1;
2629	}
2630
2631	return 0;
2632}
2633
2634/**
2635 * irdma_check_seq - check seq numbers if OK
2636 * @cm_node: connection's node
2637 * @tcph: pointer tcp header
2638 */
2639static int irdma_check_seq(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
2640{
2641	u32 seq;
2642	u32 ack_seq;
2643	u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num;
2644	u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
2645	u32 rcv_wnd;
2646	int err = 0;
2647
2648	seq = ntohl(tcph->seq);
2649	ack_seq = ntohl(tcph->ack_seq);
2650	rcv_wnd = cm_node->tcp_cntxt.rcv_wnd;
2651	if (ack_seq != loc_seq_num ||
2652	    !between(seq, rcv_nxt, (rcv_nxt + rcv_wnd)))
2653		err = -1;
2654	if (err)
2655		ibdev_dbg(&cm_node->iwdev->ibdev,
2656			  "CM: seq number err\n");
2657
2658	return err;
2659}
2660
2661void irdma_add_conn_est_qh(struct irdma_cm_node *cm_node)
2662{
2663	struct irdma_cm_info nfo;
2664
2665	irdma_get_addr_info(cm_node, &nfo);
2666	nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
2667	irdma_manage_qhash(cm_node->iwdev, &nfo,
2668			   IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
2669			   IRDMA_QHASH_MANAGE_TYPE_ADD,
2670			   cm_node, false);
2671	cm_node->qhash_set = true;
2672}
2673
2674/**
2675 * irdma_handle_syn_pkt - is for Passive node
2676 * @cm_node: connection's node
2677 * @rbuf: receive buffer
2678 */
2679static void irdma_handle_syn_pkt(struct irdma_cm_node *cm_node,
2680				 struct irdma_puda_buf *rbuf)
2681{
2682	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2683	int err;
2684	u32 inc_sequence;
2685	int optionsize;
2686
2687	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2688	inc_sequence = ntohl(tcph->seq);
2689
2690	switch (cm_node->state) {
2691	case IRDMA_CM_STATE_SYN_SENT:
2692	case IRDMA_CM_STATE_MPAREQ_SENT:
2693		/* Rcvd syn on active open connection */
2694		irdma_active_open_err(cm_node, 1);
2695		break;
2696	case IRDMA_CM_STATE_LISTENING:
2697		/* Passive OPEN */
2698		if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
2699		    cm_node->listener->backlog) {
2700			cm_node->cm_core->stats_backlog_drops++;
2701			irdma_passive_open_err(cm_node, false);
2702			break;
2703		}
2704		err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
2705		if (err) {
2706			irdma_passive_open_err(cm_node, false);
2707			/* drop pkt */
2708			break;
2709		}
2710		err = cm_node->cm_core->cm_create_ah(cm_node, false);
2711		if (err) {
2712			irdma_passive_open_err(cm_node, false);
2713			/* drop pkt */
2714			break;
2715		}
2716		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
2717		cm_node->accept_pend = 1;
2718		atomic_inc(&cm_node->listener->pend_accepts_cnt);
2719
2720		cm_node->state = IRDMA_CM_STATE_SYN_RCVD;
2721		break;
2722	case IRDMA_CM_STATE_CLOSED:
2723		irdma_cleanup_retrans_entry(cm_node);
2724		refcount_inc(&cm_node->refcnt);
2725		irdma_send_reset(cm_node);
2726		break;
2727	case IRDMA_CM_STATE_OFFLOADED:
2728	case IRDMA_CM_STATE_ESTABLISHED:
2729	case IRDMA_CM_STATE_FIN_WAIT1:
2730	case IRDMA_CM_STATE_FIN_WAIT2:
2731	case IRDMA_CM_STATE_MPAREQ_RCVD:
2732	case IRDMA_CM_STATE_LAST_ACK:
2733	case IRDMA_CM_STATE_CLOSING:
2734	case IRDMA_CM_STATE_UNKNOWN:
2735	default:
2736		break;
2737	}
2738}
2739
2740/**
2741 * irdma_handle_synack_pkt - Process SYN+ACK packet (active side)
2742 * @cm_node: connection's node
2743 * @rbuf: receive buffer
2744 */
2745static void irdma_handle_synack_pkt(struct irdma_cm_node *cm_node,
2746				    struct irdma_puda_buf *rbuf)
2747{
2748	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2749	int err;
2750	u32 inc_sequence;
2751	int optionsize;
2752
2753	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2754	inc_sequence = ntohl(tcph->seq);
2755	switch (cm_node->state) {
2756	case IRDMA_CM_STATE_SYN_SENT:
2757		irdma_cleanup_retrans_entry(cm_node);
2758		/* active open */
2759		if (irdma_check_syn(cm_node, tcph)) {
2760			ibdev_dbg(&cm_node->iwdev->ibdev,
2761				  "CM: check syn fail\n");
2762			return;
2763		}
2764		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2765		/* setup options */
2766		err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 0);
2767		if (err) {
2768			ibdev_dbg(&cm_node->iwdev->ibdev,
2769				  "CM: cm_node=%p tcp_options failed\n",
2770				  cm_node);
2771			break;
2772		}
2773		irdma_cleanup_retrans_entry(cm_node);
2774		cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
2775		irdma_send_ack(cm_node); /* ACK  for the syn_ack */
2776		err = irdma_send_mpa_request(cm_node);
2777		if (err) {
2778			ibdev_dbg(&cm_node->iwdev->ibdev,
2779				  "CM: cm_node=%p irdma_send_mpa_request failed\n",
2780				  cm_node);
2781			break;
2782		}
2783		cm_node->state = IRDMA_CM_STATE_MPAREQ_SENT;
2784		break;
2785	case IRDMA_CM_STATE_MPAREQ_RCVD:
2786		irdma_passive_open_err(cm_node, true);
2787		break;
2788	case IRDMA_CM_STATE_LISTENING:
2789		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
2790		irdma_cleanup_retrans_entry(cm_node);
2791		cm_node->state = IRDMA_CM_STATE_CLOSED;
2792		irdma_send_reset(cm_node);
2793		break;
2794	case IRDMA_CM_STATE_CLOSED:
2795		cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq);
2796		irdma_cleanup_retrans_entry(cm_node);
2797		refcount_inc(&cm_node->refcnt);
2798		irdma_send_reset(cm_node);
2799		break;
2800	case IRDMA_CM_STATE_ESTABLISHED:
2801	case IRDMA_CM_STATE_FIN_WAIT1:
2802	case IRDMA_CM_STATE_FIN_WAIT2:
2803	case IRDMA_CM_STATE_LAST_ACK:
2804	case IRDMA_CM_STATE_OFFLOADED:
2805	case IRDMA_CM_STATE_CLOSING:
2806	case IRDMA_CM_STATE_UNKNOWN:
2807	case IRDMA_CM_STATE_MPAREQ_SENT:
2808	default:
2809		break;
2810	}
2811}
2812
2813/**
2814 * irdma_handle_ack_pkt - process packet with ACK
2815 * @cm_node: connection's node
2816 * @rbuf: receive buffer
2817 */
2818static int irdma_handle_ack_pkt(struct irdma_cm_node *cm_node,
2819				struct irdma_puda_buf *rbuf)
2820{
2821	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2822	u32 inc_sequence;
2823	int ret;
2824	int optionsize;
2825	u32 datasize = rbuf->datalen;
2826
2827	optionsize = (tcph->doff << 2) - sizeof(struct tcphdr);
2828
2829	if (irdma_check_seq(cm_node, tcph))
2830		return -EINVAL;
2831
2832	inc_sequence = ntohl(tcph->seq);
2833	switch (cm_node->state) {
2834	case IRDMA_CM_STATE_SYN_RCVD:
2835		irdma_cleanup_retrans_entry(cm_node);
2836		ret = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
2837		if (ret)
2838			return ret;
2839		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2840		cm_node->state = IRDMA_CM_STATE_ESTABLISHED;
2841		if (datasize) {
2842			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2843			irdma_handle_rcv_mpa(cm_node, rbuf);
2844		}
2845		break;
2846	case IRDMA_CM_STATE_ESTABLISHED:
2847		irdma_cleanup_retrans_entry(cm_node);
2848		if (datasize) {
2849			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2850			irdma_handle_rcv_mpa(cm_node, rbuf);
2851		}
2852		break;
2853	case IRDMA_CM_STATE_MPAREQ_SENT:
2854		cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq);
2855		if (datasize) {
2856			cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
2857			cm_node->ack_rcvd = false;
2858			irdma_handle_rcv_mpa(cm_node, rbuf);
2859		} else {
2860			cm_node->ack_rcvd = true;
2861		}
2862		break;
2863	case IRDMA_CM_STATE_LISTENING:
2864		irdma_cleanup_retrans_entry(cm_node);
2865		cm_node->state = IRDMA_CM_STATE_CLOSED;
2866		irdma_send_reset(cm_node);
2867		break;
2868	case IRDMA_CM_STATE_CLOSED:
2869		irdma_cleanup_retrans_entry(cm_node);
2870		refcount_inc(&cm_node->refcnt);
2871		irdma_send_reset(cm_node);
2872		break;
2873	case IRDMA_CM_STATE_LAST_ACK:
2874	case IRDMA_CM_STATE_CLOSING:
2875		irdma_cleanup_retrans_entry(cm_node);
2876		cm_node->state = IRDMA_CM_STATE_CLOSED;
2877		irdma_rem_ref_cm_node(cm_node);
2878		break;
2879	case IRDMA_CM_STATE_FIN_WAIT1:
2880		irdma_cleanup_retrans_entry(cm_node);
2881		cm_node->state = IRDMA_CM_STATE_FIN_WAIT2;
2882		break;
2883	case IRDMA_CM_STATE_SYN_SENT:
2884	case IRDMA_CM_STATE_FIN_WAIT2:
2885	case IRDMA_CM_STATE_OFFLOADED:
2886	case IRDMA_CM_STATE_MPAREQ_RCVD:
2887	case IRDMA_CM_STATE_UNKNOWN:
2888	default:
2889		irdma_cleanup_retrans_entry(cm_node);
2890		break;
2891	}
2892
2893	return 0;
2894}
2895
2896/**
2897 * irdma_process_pkt - process cm packet
2898 * @cm_node: connection's node
2899 * @rbuf: receive buffer
2900 */
2901static void irdma_process_pkt(struct irdma_cm_node *cm_node,
2902			      struct irdma_puda_buf *rbuf)
2903{
2904	enum irdma_tcpip_pkt_type pkt_type = IRDMA_PKT_TYPE_UNKNOWN;
2905	struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
2906	u32 fin_set = 0;
2907	int err;
2908
2909	if (tcph->rst) {
2910		pkt_type = IRDMA_PKT_TYPE_RST;
2911	} else if (tcph->syn) {
2912		pkt_type = IRDMA_PKT_TYPE_SYN;
2913		if (tcph->ack)
2914			pkt_type = IRDMA_PKT_TYPE_SYNACK;
2915	} else if (tcph->ack) {
2916		pkt_type = IRDMA_PKT_TYPE_ACK;
2917	}
2918	if (tcph->fin)
2919		fin_set = 1;
2920
2921	switch (pkt_type) {
2922	case IRDMA_PKT_TYPE_SYN:
2923		irdma_handle_syn_pkt(cm_node, rbuf);
2924		break;
2925	case IRDMA_PKT_TYPE_SYNACK:
2926		irdma_handle_synack_pkt(cm_node, rbuf);
2927		break;
2928	case IRDMA_PKT_TYPE_ACK:
2929		err = irdma_handle_ack_pkt(cm_node, rbuf);
2930		if (fin_set && !err)
2931			irdma_handle_fin_pkt(cm_node);
2932		break;
2933	case IRDMA_PKT_TYPE_RST:
2934		irdma_handle_rst_pkt(cm_node, rbuf);
2935		break;
2936	default:
2937		if (fin_set &&
2938		    (!irdma_check_seq(cm_node, (struct tcphdr *)rbuf->tcph)))
2939			irdma_handle_fin_pkt(cm_node);
2940		break;
2941	}
2942}
2943
2944/**
2945 * irdma_make_listen_node - create a listen node with params
2946 * @cm_core: cm's core
2947 * @iwdev: iwarp device structure
2948 * @cm_info: quad info for connection
2949 */
2950static struct irdma_cm_listener *
2951irdma_make_listen_node(struct irdma_cm_core *cm_core,
2952		       struct irdma_device *iwdev,
2953		       struct irdma_cm_info *cm_info)
2954{
2955	struct irdma_cm_listener *listener;
2956	unsigned long flags;
2957
2958	/* cannot have multiple matching listeners */
2959	listener =
2960		irdma_find_listener(cm_core, cm_info->loc_addr, cm_info->ipv4,
2961				    cm_info->loc_port, cm_info->vlan_id,
2962				    IRDMA_CM_LISTENER_EITHER_STATE);
2963	if (listener &&
2964	    listener->listener_state == IRDMA_CM_LISTENER_ACTIVE_STATE) {
2965		refcount_dec(&listener->refcnt);
2966		return NULL;
2967	}
2968
2969	if (!listener) {
2970		/* create a CM listen node
2971		 * 1/2 node to compare incoming traffic to
2972		 */
2973		listener = kzalloc(sizeof(*listener), GFP_KERNEL);
2974		if (!listener)
2975			return NULL;
2976		cm_core->stats_listen_nodes_created++;
2977		memcpy(listener->loc_addr, cm_info->loc_addr,
2978		       sizeof(listener->loc_addr));
2979		listener->loc_port = cm_info->loc_port;
2980
2981		INIT_LIST_HEAD(&listener->child_listen_list);
2982
2983		refcount_set(&listener->refcnt, 1);
2984	} else {
2985		listener->reused_node = 1;
2986	}
2987
2988	listener->cm_id = cm_info->cm_id;
2989	listener->ipv4 = cm_info->ipv4;
2990	listener->vlan_id = cm_info->vlan_id;
2991	atomic_set(&listener->pend_accepts_cnt, 0);
2992	listener->cm_core = cm_core;
2993	listener->iwdev = iwdev;
2994
2995	listener->backlog = cm_info->backlog;
2996	listener->listener_state = IRDMA_CM_LISTENER_ACTIVE_STATE;
2997
2998	if (!listener->reused_node) {
2999		spin_lock_irqsave(&cm_core->listen_list_lock, flags);
3000		list_add(&listener->list, &cm_core->listen_list);
3001		spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
3002	}
3003
3004	return listener;
3005}
3006
3007/**
3008 * irdma_create_cm_node - make a connection node with params
3009 * @cm_core: cm's core
3010 * @iwdev: iwarp device structure
3011 * @conn_param: connection parameters
3012 * @cm_info: quad info for connection
3013 * @caller_cm_node: pointer to cm_node structure to return
3014 */
3015static int irdma_create_cm_node(struct irdma_cm_core *cm_core,
3016				struct irdma_device *iwdev,
3017				struct iw_cm_conn_param *conn_param,
3018				struct irdma_cm_info *cm_info,
3019				struct irdma_cm_node **caller_cm_node)
3020{
3021	struct irdma_cm_node *cm_node;
3022	u16 private_data_len = conn_param->private_data_len;
3023	const void *private_data = conn_param->private_data;
3024
3025	/* create a CM connection node */
3026	cm_node = irdma_make_cm_node(cm_core, iwdev, cm_info, NULL);
3027	if (IS_ERR(cm_node))
3028		return PTR_ERR(cm_node);
3029
3030	/* set our node side to client (active) side */
3031	cm_node->tcp_cntxt.client = 1;
3032	cm_node->tcp_cntxt.rcv_wscale = IRDMA_CM_DEFAULT_RCV_WND_SCALE;
3033
3034	irdma_record_ird_ord(cm_node, conn_param->ird, conn_param->ord);
3035
3036	cm_node->pdata.size = private_data_len;
3037	cm_node->pdata.addr = cm_node->pdata_buf;
3038
3039	memcpy(cm_node->pdata_buf, private_data, private_data_len);
3040	*caller_cm_node = cm_node;
3041
3042	return 0;
3043}
3044
3045/**
3046 * irdma_cm_reject - reject and teardown a connection
3047 * @cm_node: connection's node
3048 * @pdata: ptr to private data for reject
3049 * @plen: size of private data
3050 */
3051static int irdma_cm_reject(struct irdma_cm_node *cm_node, const void *pdata,
3052			   u8 plen)
3053{
3054	int ret;
3055	int passive_state;
3056
3057	if (cm_node->tcp_cntxt.client)
3058		return 0;
3059
3060	irdma_cleanup_retrans_entry(cm_node);
3061
3062	passive_state = atomic_add_return(1, &cm_node->passive_state);
3063	if (passive_state == IRDMA_SEND_RESET_EVENT) {
3064		cm_node->state = IRDMA_CM_STATE_CLOSED;
3065		irdma_rem_ref_cm_node(cm_node);
3066		return 0;
3067	}
3068
3069	if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
3070		irdma_rem_ref_cm_node(cm_node);
3071		return 0;
3072	}
3073
3074	ret = irdma_send_mpa_reject(cm_node, pdata, plen);
3075	if (!ret)
3076		return 0;
3077
3078	cm_node->state = IRDMA_CM_STATE_CLOSED;
3079	if (irdma_send_reset(cm_node))
3080		ibdev_dbg(&cm_node->iwdev->ibdev,
3081			  "CM: send reset failed\n");
3082
3083	return ret;
3084}
3085
3086/**
3087 * irdma_cm_close - close of cm connection
3088 * @cm_node: connection's node
3089 */
3090static int irdma_cm_close(struct irdma_cm_node *cm_node)
3091{
3092	switch (cm_node->state) {
3093	case IRDMA_CM_STATE_SYN_RCVD:
3094	case IRDMA_CM_STATE_SYN_SENT:
3095	case IRDMA_CM_STATE_ONE_SIDE_ESTABLISHED:
3096	case IRDMA_CM_STATE_ESTABLISHED:
3097	case IRDMA_CM_STATE_ACCEPTING:
3098	case IRDMA_CM_STATE_MPAREQ_SENT:
3099	case IRDMA_CM_STATE_MPAREQ_RCVD:
3100		irdma_cleanup_retrans_entry(cm_node);
3101		irdma_send_reset(cm_node);
3102		break;
3103	case IRDMA_CM_STATE_CLOSE_WAIT:
3104		cm_node->state = IRDMA_CM_STATE_LAST_ACK;
3105		irdma_send_fin(cm_node);
3106		break;
3107	case IRDMA_CM_STATE_FIN_WAIT1:
3108	case IRDMA_CM_STATE_FIN_WAIT2:
3109	case IRDMA_CM_STATE_LAST_ACK:
3110	case IRDMA_CM_STATE_TIME_WAIT:
3111	case IRDMA_CM_STATE_CLOSING:
3112		return -EINVAL;
3113	case IRDMA_CM_STATE_LISTENING:
3114		irdma_cleanup_retrans_entry(cm_node);
3115		irdma_send_reset(cm_node);
3116		break;
3117	case IRDMA_CM_STATE_MPAREJ_RCVD:
3118	case IRDMA_CM_STATE_UNKNOWN:
3119	case IRDMA_CM_STATE_INITED:
3120	case IRDMA_CM_STATE_CLOSED:
3121	case IRDMA_CM_STATE_LISTENER_DESTROYED:
3122		irdma_rem_ref_cm_node(cm_node);
3123		break;
3124	case IRDMA_CM_STATE_OFFLOADED:
3125		if (cm_node->send_entry)
3126			ibdev_dbg(&cm_node->iwdev->ibdev,
3127				  "CM: CM send_entry in OFFLOADED state\n");
3128		irdma_rem_ref_cm_node(cm_node);
3129		break;
3130	}
3131
3132	return 0;
3133}
3134
3135/**
3136 * irdma_receive_ilq - recv an ETHERNET packet, and process it
3137 * through CM
3138 * @vsi: VSI structure of dev
3139 * @rbuf: receive buffer
3140 */
3141void irdma_receive_ilq(struct irdma_sc_vsi *vsi, struct irdma_puda_buf *rbuf)
3142{
3143	struct irdma_cm_node *cm_node;
3144	struct irdma_cm_listener *listener;
3145	struct iphdr *iph;
3146	struct ipv6hdr *ip6h;
3147	struct tcphdr *tcph;
3148	struct irdma_cm_info cm_info = {};
3149	struct irdma_device *iwdev = vsi->back_vsi;
3150	struct irdma_cm_core *cm_core = &iwdev->cm_core;
3151	struct vlan_ethhdr *ethh;
3152	u16 vtag;
3153
3154	/* if vlan, then maclen = 18 else 14 */
3155	iph = (struct iphdr *)rbuf->iph;
3156	print_hex_dump_debug("ILQ: RECEIVE ILQ BUFFER", DUMP_PREFIX_OFFSET,
3157			     16, 8, rbuf->mem.va, rbuf->totallen, false);
3158	if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
3159		if (rbuf->vlan_valid) {
3160			vtag = rbuf->vlan_id;
3161			cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >>
3162					   VLAN_PRIO_SHIFT;
3163			cm_info.vlan_id = vtag & VLAN_VID_MASK;
3164		} else {
3165			cm_info.vlan_id = 0xFFFF;
3166		}
3167	} else {
3168		ethh = rbuf->mem.va;
3169
3170		if (ethh->h_vlan_proto == htons(ETH_P_8021Q)) {
3171			vtag = ntohs(ethh->h_vlan_TCI);
3172			cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >>
3173					   VLAN_PRIO_SHIFT;
3174			cm_info.vlan_id = vtag & VLAN_VID_MASK;
3175			ibdev_dbg(&cm_core->iwdev->ibdev,
3176				  "CM: vlan_id=%d\n", cm_info.vlan_id);
3177		} else {
3178			cm_info.vlan_id = 0xFFFF;
3179		}
3180	}
3181	tcph = (struct tcphdr *)rbuf->tcph;
3182
3183	if (rbuf->ipv4) {
3184		cm_info.loc_addr[0] = ntohl(iph->daddr);
3185		cm_info.rem_addr[0] = ntohl(iph->saddr);
3186		cm_info.ipv4 = true;
3187		cm_info.tos = iph->tos;
3188	} else {
3189		ip6h = (struct ipv6hdr *)rbuf->iph;
3190		irdma_copy_ip_ntohl(cm_info.loc_addr,
3191				    ip6h->daddr.in6_u.u6_addr32);
3192		irdma_copy_ip_ntohl(cm_info.rem_addr,
3193				    ip6h->saddr.in6_u.u6_addr32);
3194		cm_info.ipv4 = false;
3195		cm_info.tos = (ip6h->priority << 4) | (ip6h->flow_lbl[0] >> 4);
3196	}
3197	cm_info.loc_port = ntohs(tcph->dest);
3198	cm_info.rem_port = ntohs(tcph->source);
3199	cm_node = irdma_find_node(cm_core, cm_info.rem_port, cm_info.rem_addr,
3200				  cm_info.loc_port, cm_info.loc_addr, cm_info.vlan_id);
3201
3202	if (!cm_node) {
3203		/* Only type of packet accepted are for the
3204		 * PASSIVE open (syn only)
3205		 */
3206		if (!tcph->syn || tcph->ack)
3207			return;
3208
3209		listener = irdma_find_listener(cm_core,
3210					       cm_info.loc_addr,
3211					       cm_info.ipv4,
3212					       cm_info.loc_port,
3213					       cm_info.vlan_id,
3214					       IRDMA_CM_LISTENER_ACTIVE_STATE);
3215		if (!listener) {
3216			cm_info.cm_id = NULL;
3217			ibdev_dbg(&cm_core->iwdev->ibdev,
3218				  "CM: no listener found\n");
3219			return;
3220		}
3221
3222		cm_info.cm_id = listener->cm_id;
3223		cm_node = irdma_make_cm_node(cm_core, iwdev, &cm_info,
3224					     listener);
3225		if (IS_ERR(cm_node)) {
3226			ibdev_dbg(&cm_core->iwdev->ibdev,
3227				  "CM: allocate node failed ret=%ld\n", PTR_ERR(cm_node));
3228			refcount_dec(&listener->refcnt);
3229			return;
3230		}
3231
3232		if (!tcph->rst && !tcph->fin) {
3233			cm_node->state = IRDMA_CM_STATE_LISTENING;
3234		} else {
3235			irdma_rem_ref_cm_node(cm_node);
3236			return;
3237		}
3238
3239		refcount_inc(&cm_node->refcnt);
3240	} else if (cm_node->state == IRDMA_CM_STATE_OFFLOADED) {
3241		irdma_rem_ref_cm_node(cm_node);
3242		return;
3243	}
3244
3245	irdma_process_pkt(cm_node, rbuf);
3246	irdma_rem_ref_cm_node(cm_node);
3247}
3248
3249static int irdma_add_qh(struct irdma_cm_node *cm_node, bool active)
3250{
3251	if (!active)
3252		irdma_add_conn_est_qh(cm_node);
3253	return 0;
3254}
3255
3256static void irdma_cm_free_ah_nop(struct irdma_cm_node *cm_node)
3257{
3258}
3259
3260/**
3261 * irdma_setup_cm_core - setup top level instance of a cm core
3262 * @iwdev: iwarp device structure
3263 * @rdma_ver: HW version
3264 */
3265int irdma_setup_cm_core(struct irdma_device *iwdev, u8 rdma_ver)
3266{
3267	struct irdma_cm_core *cm_core = &iwdev->cm_core;
3268
3269	cm_core->iwdev = iwdev;
3270	cm_core->dev = &iwdev->rf->sc_dev;
3271
3272	/* Handles CM event work items send to Iwarp core */
3273	cm_core->event_wq = alloc_ordered_workqueue("iwarp-event-wq", 0);
3274	if (!cm_core->event_wq)
3275		return -ENOMEM;
3276
3277	INIT_LIST_HEAD(&cm_core->listen_list);
3278
3279	timer_setup(&cm_core->tcp_timer, irdma_cm_timer_tick, 0);
3280
3281	spin_lock_init(&cm_core->ht_lock);
3282	spin_lock_init(&cm_core->listen_list_lock);
3283	spin_lock_init(&cm_core->apbvt_lock);
3284	switch (rdma_ver) {
3285	case IRDMA_GEN_1:
3286		cm_core->form_cm_frame = irdma_form_uda_cm_frame;
3287		cm_core->cm_create_ah = irdma_add_qh;
3288		cm_core->cm_free_ah = irdma_cm_free_ah_nop;
3289		break;
3290	case IRDMA_GEN_2:
3291	default:
3292		cm_core->form_cm_frame = irdma_form_ah_cm_frame;
3293		cm_core->cm_create_ah = irdma_cm_create_ah;
3294		cm_core->cm_free_ah = irdma_cm_free_ah;
3295	}
3296
3297	return 0;
3298}
3299
3300/**
3301 * irdma_cleanup_cm_core - deallocate a top level instance of a
3302 * cm core
3303 * @cm_core: cm's core
3304 */
3305void irdma_cleanup_cm_core(struct irdma_cm_core *cm_core)
3306{
3307	if (!cm_core)
3308		return;
3309
3310	timer_delete_sync(&cm_core->tcp_timer);
3311
3312	destroy_workqueue(cm_core->event_wq);
3313	cm_core->dev->ws_reset(&cm_core->iwdev->vsi);
3314}
3315
3316/**
3317 * irdma_init_tcp_ctx - setup qp context
3318 * @cm_node: connection's node
3319 * @tcp_info: offload info for tcp
3320 * @iwqp: associate qp for the connection
3321 */
3322static void irdma_init_tcp_ctx(struct irdma_cm_node *cm_node,
3323			       struct irdma_tcp_offload_info *tcp_info,
3324			       struct irdma_qp *iwqp)
3325{
3326	tcp_info->ipv4 = cm_node->ipv4;
3327	tcp_info->drop_ooo_seg = !iwqp->iwdev->iw_ooo;
3328	tcp_info->wscale = true;
3329	tcp_info->ignore_tcp_opt = true;
3330	tcp_info->ignore_tcp_uns_opt = true;
3331	tcp_info->no_nagle = false;
3332
3333	tcp_info->ttl = IRDMA_DEFAULT_TTL;
3334	tcp_info->rtt_var = IRDMA_DEFAULT_RTT_VAR;
3335	tcp_info->ss_thresh = IRDMA_DEFAULT_SS_THRESH;
3336	tcp_info->rexmit_thresh = IRDMA_DEFAULT_REXMIT_THRESH;
3337
3338	tcp_info->tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
3339	tcp_info->snd_wscale = cm_node->tcp_cntxt.snd_wscale;
3340	tcp_info->rcv_wscale = cm_node->tcp_cntxt.rcv_wscale;
3341
3342	tcp_info->snd_nxt = cm_node->tcp_cntxt.loc_seq_num;
3343	tcp_info->snd_wnd = cm_node->tcp_cntxt.snd_wnd;
3344	tcp_info->rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
3345	tcp_info->snd_max = cm_node->tcp_cntxt.loc_seq_num;
3346
3347	tcp_info->snd_una = cm_node->tcp_cntxt.loc_seq_num;
3348	tcp_info->cwnd = 2 * cm_node->tcp_cntxt.mss;
3349	tcp_info->snd_wl1 = cm_node->tcp_cntxt.rcv_nxt;
3350	tcp_info->snd_wl2 = cm_node->tcp_cntxt.loc_seq_num;
3351	tcp_info->max_snd_window = cm_node->tcp_cntxt.max_snd_wnd;
3352	tcp_info->rcv_wnd = cm_node->tcp_cntxt.rcv_wnd
3353			    << cm_node->tcp_cntxt.rcv_wscale;
3354
3355	tcp_info->flow_label = 0;
3356	tcp_info->snd_mss = (u32)cm_node->tcp_cntxt.mss;
3357	tcp_info->tos = cm_node->tos;
3358	if (cm_node->vlan_id < VLAN_N_VID) {
3359		tcp_info->insert_vlan_tag = true;
3360		tcp_info->vlan_tag = cm_node->vlan_id;
3361		tcp_info->vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
3362	}
3363	if (cm_node->ipv4) {
3364		tcp_info->src_port = cm_node->loc_port;
3365		tcp_info->dst_port = cm_node->rem_port;
3366
3367		tcp_info->dest_ip_addr[3] = cm_node->rem_addr[0];
3368		tcp_info->local_ipaddr[3] = cm_node->loc_addr[0];
3369		tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf,
3370							 &tcp_info->dest_ip_addr[3],
3371							 true, NULL,
3372							 IRDMA_ARP_RESOLVE);
3373	} else {
3374		tcp_info->src_port = cm_node->loc_port;
3375		tcp_info->dst_port = cm_node->rem_port;
3376		memcpy(tcp_info->dest_ip_addr, cm_node->rem_addr,
3377		       sizeof(tcp_info->dest_ip_addr));
3378		memcpy(tcp_info->local_ipaddr, cm_node->loc_addr,
3379		       sizeof(tcp_info->local_ipaddr));
3380
3381		tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf,
3382							 &tcp_info->dest_ip_addr[0],
3383							 false, NULL,
3384							 IRDMA_ARP_RESOLVE);
3385	}
3386}
3387
3388/**
3389 * irdma_cm_init_tsa_conn - setup qp for RTS
3390 * @iwqp: associate qp for the connection
3391 * @cm_node: connection's node
3392 */
3393static void irdma_cm_init_tsa_conn(struct irdma_qp *iwqp,
3394				   struct irdma_cm_node *cm_node)
3395{
3396	struct irdma_iwarp_offload_info *iwarp_info;
3397	struct irdma_qp_host_ctx_info *ctx_info;
3398
3399	iwarp_info = &iwqp->iwarp_info;
3400	ctx_info = &iwqp->ctx_info;
3401
3402	ctx_info->tcp_info = &iwqp->tcp_info;
3403	ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
3404	ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
3405
3406	iwarp_info->ord_size = cm_node->ord_size;
3407	iwarp_info->ird_size = cm_node->ird_size;
3408	iwarp_info->rd_en = true;
3409	iwarp_info->rdmap_ver = 1;
3410	iwarp_info->ddp_ver = 1;
3411	iwarp_info->pd_id = iwqp->iwpd->sc_pd.pd_id;
3412
3413	ctx_info->tcp_info_valid = true;
3414	ctx_info->iwarp_info_valid = true;
3415	ctx_info->user_pri = cm_node->user_pri;
3416
3417	irdma_init_tcp_ctx(cm_node, &iwqp->tcp_info, iwqp);
3418	if (cm_node->snd_mark_en) {
3419		iwarp_info->snd_mark_en = true;
3420		iwarp_info->snd_mark_offset = (iwqp->tcp_info.snd_nxt & SNDMARKER_SEQNMASK) +
3421					       cm_node->lsmm_size;
3422	}
3423
3424	cm_node->state = IRDMA_CM_STATE_OFFLOADED;
3425	iwqp->tcp_info.tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
3426	iwqp->tcp_info.src_mac_addr_idx = iwqp->iwdev->mac_ip_table_idx;
3427
3428	if (cm_node->rcv_mark_en) {
3429		iwarp_info->rcv_mark_en = true;
3430		iwarp_info->align_hdrs = true;
3431	}
3432
3433	irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
3434
3435	/* once tcp_info is set, no need to do it again */
3436	ctx_info->tcp_info_valid = false;
3437	ctx_info->iwarp_info_valid = false;
3438}
3439
3440/**
3441 * irdma_cm_disconn - when a connection is being closed
3442 * @iwqp: associated qp for the connection
3443 */
3444void irdma_cm_disconn(struct irdma_qp *iwqp)
3445{
3446	struct irdma_device *iwdev = iwqp->iwdev;
3447	struct disconn_work *work;
3448	unsigned long flags;
3449
3450	work = kzalloc(sizeof(*work), GFP_ATOMIC);
3451	if (!work)
3452		return;
3453
3454	spin_lock_irqsave(&iwdev->rf->qptable_lock, flags);
3455	if (!iwdev->rf->qp_table[iwqp->ibqp.qp_num]) {
3456		spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
3457		ibdev_dbg(&iwdev->ibdev,
3458			  "CM: qp_id %d is already freed\n",
3459			  iwqp->ibqp.qp_num);
3460		kfree(work);
3461		return;
3462	}
3463	irdma_qp_add_ref(&iwqp->ibqp);
3464	spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
3465
3466	work->iwqp = iwqp;
3467	INIT_WORK(&work->work, irdma_disconnect_worker);
3468	queue_work(iwdev->cleanup_wq, &work->work);
3469}
3470
3471/**
3472 * irdma_qp_disconnect - free qp and close cm
3473 * @iwqp: associate qp for the connection
3474 */
3475static void irdma_qp_disconnect(struct irdma_qp *iwqp)
3476{
3477	struct irdma_device *iwdev = iwqp->iwdev;
3478
3479	iwqp->active_conn = 0;
3480	/* close the CM node down if it is still active */
3481	ibdev_dbg(&iwdev->ibdev, "CM: Call close API\n");
3482	irdma_cm_close(iwqp->cm_node);
3483}
3484
3485/**
3486 * irdma_cm_disconn_true - called by worker thread to disconnect qp
3487 * @iwqp: associate qp for the connection
3488 */
3489static void irdma_cm_disconn_true(struct irdma_qp *iwqp)
3490{
3491	struct iw_cm_id *cm_id;
3492	struct irdma_device *iwdev;
3493	struct irdma_sc_qp *qp = &iwqp->sc_qp;
3494	u16 last_ae;
3495	u8 original_hw_tcp_state;
3496	u8 original_ibqp_state;
3497	int disconn_status = 0;
3498	int issue_disconn = 0;
3499	int issue_close = 0;
3500	int issue_flush = 0;
3501	unsigned long flags;
3502	int err;
3503
3504	iwdev = iwqp->iwdev;
3505	spin_lock_irqsave(&iwqp->lock, flags);
3506	if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
3507		struct ib_qp_attr attr;
3508
3509		if (iwqp->flush_issued || iwqp->sc_qp.qp_uk.destroy_pending) {
3510			spin_unlock_irqrestore(&iwqp->lock, flags);
3511			return;
3512		}
3513
3514		spin_unlock_irqrestore(&iwqp->lock, flags);
3515
3516		attr.qp_state = IB_QPS_ERR;
3517		irdma_modify_qp_roce(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
3518		irdma_ib_qp_event(iwqp, qp->event_type);
3519		return;
3520	}
3521
3522	cm_id = iwqp->cm_id;
3523	original_hw_tcp_state = iwqp->hw_tcp_state;
3524	original_ibqp_state = iwqp->ibqp_state;
3525	last_ae = iwqp->last_aeq;
3526
3527	if (qp->term_flags) {
3528		issue_disconn = 1;
3529		issue_close = 1;
3530		iwqp->cm_id = NULL;
3531		irdma_terminate_del_timer(qp);
3532		if (!iwqp->flush_issued) {
3533			iwqp->flush_issued = 1;
3534			issue_flush = 1;
3535		}
3536	} else if ((original_hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT) ||
3537		   ((original_ibqp_state == IB_QPS_RTS) &&
3538		    (last_ae == IRDMA_AE_LLP_CONNECTION_RESET))) {
3539		issue_disconn = 1;
3540		if (last_ae == IRDMA_AE_LLP_CONNECTION_RESET)
3541			disconn_status = -ECONNRESET;
3542	}
3543
3544	if (original_hw_tcp_state == IRDMA_TCP_STATE_CLOSED ||
3545	    original_hw_tcp_state == IRDMA_TCP_STATE_TIME_WAIT ||
3546	    last_ae == IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE ||
3547	    last_ae == IRDMA_AE_BAD_CLOSE ||
3548	    last_ae == IRDMA_AE_LLP_CONNECTION_RESET || iwdev->rf->reset || !cm_id) {
3549		issue_close = 1;
3550		iwqp->cm_id = NULL;
3551		qp->term_flags = 0;
3552		if (!iwqp->flush_issued) {
3553			iwqp->flush_issued = 1;
3554			issue_flush = 1;
3555		}
3556	}
3557
3558	spin_unlock_irqrestore(&iwqp->lock, flags);
3559	if (issue_flush && !iwqp->sc_qp.qp_uk.destroy_pending) {
3560		irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_FLUSH_RQ |
3561				 IRDMA_FLUSH_WAIT);
3562
3563		if (qp->term_flags)
3564			irdma_ib_qp_event(iwqp, qp->event_type);
3565	}
3566
3567	if (!cm_id || !cm_id->event_handler)
3568		return;
3569
3570	spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
3571	if (!iwqp->cm_node) {
3572		spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
3573		return;
3574	}
3575	refcount_inc(&iwqp->cm_node->refcnt);
3576
3577	spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
3578
3579	if (issue_disconn) {
3580		err = irdma_send_cm_event(iwqp->cm_node, cm_id,
3581					  IW_CM_EVENT_DISCONNECT,
3582					  disconn_status);
3583		if (err)
3584			ibdev_dbg(&iwdev->ibdev,
3585				  "CM: disconnect event failed: - cm_id = %p\n",
3586				  cm_id);
3587	}
3588	if (issue_close) {
3589		cm_id->provider_data = iwqp;
3590		err = irdma_send_cm_event(iwqp->cm_node, cm_id,
3591					  IW_CM_EVENT_CLOSE, 0);
3592		if (err)
3593			ibdev_dbg(&iwdev->ibdev,
3594				  "CM: close event failed: - cm_id = %p\n",
3595				  cm_id);
3596		irdma_qp_disconnect(iwqp);
3597	}
3598	irdma_rem_ref_cm_node(iwqp->cm_node);
3599}
3600
3601/**
3602 * irdma_disconnect_worker - worker for connection close
3603 * @work: points or disconn structure
3604 */
3605static void irdma_disconnect_worker(struct work_struct *work)
3606{
3607	struct disconn_work *dwork = container_of(work, struct disconn_work, work);
3608	struct irdma_qp *iwqp = dwork->iwqp;
3609
3610	kfree(dwork);
3611	irdma_cm_disconn_true(iwqp);
3612	irdma_qp_rem_ref(&iwqp->ibqp);
3613}
3614
3615/**
3616 * irdma_free_lsmm_rsrc - free lsmm memory and deregister
3617 * @iwqp: associate qp for the connection
3618 */
3619void irdma_free_lsmm_rsrc(struct irdma_qp *iwqp)
3620{
3621	struct irdma_device *iwdev;
3622
3623	iwdev = iwqp->iwdev;
3624
3625	if (iwqp->ietf_mem.va) {
3626		if (iwqp->lsmm_mr)
3627			iwdev->ibdev.ops.dereg_mr(iwqp->lsmm_mr, NULL);
3628		dma_free_coherent(iwdev->rf->sc_dev.hw->device,
3629				  iwqp->ietf_mem.size, iwqp->ietf_mem.va,
3630				  iwqp->ietf_mem.pa);
3631		iwqp->ietf_mem.va = NULL;
3632	}
3633}
3634
3635/**
3636 * irdma_accept - registered call for connection to be accepted
3637 * @cm_id: cm information for passive connection
3638 * @conn_param: accept parameters
3639 */
3640int irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3641{
3642	struct ib_qp *ibqp;
3643	struct irdma_qp *iwqp;
3644	struct irdma_device *iwdev;
3645	struct irdma_sc_dev *dev;
3646	struct irdma_cm_node *cm_node;
3647	struct ib_qp_attr attr = {};
3648	int passive_state;
3649	struct ib_mr *ibmr;
3650	struct irdma_pd *iwpd;
3651	u16 buf_len = 0;
3652	struct irdma_kmem_info accept;
3653	u64 tagged_offset;
3654	int wait_ret;
3655	int ret = 0;
3656
3657	ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
3658	if (!ibqp)
3659		return -EINVAL;
3660
3661	iwqp = to_iwqp(ibqp);
3662	iwdev = iwqp->iwdev;
3663	dev = &iwdev->rf->sc_dev;
3664	cm_node = cm_id->provider_data;
3665
3666	if (((struct sockaddr_in *)&cm_id->local_addr)->sin_family == AF_INET) {
3667		cm_node->ipv4 = true;
3668		cm_node->vlan_id = irdma_get_vlan_ipv4(cm_node->loc_addr);
3669	} else {
3670		cm_node->ipv4 = false;
3671		irdma_get_vlan_mac_ipv6(cm_node->loc_addr, &cm_node->vlan_id,
3672					NULL);
3673	}
3674	ibdev_dbg(&iwdev->ibdev, "CM: Accept vlan_id=%d\n",
3675		  cm_node->vlan_id);
3676
3677	trace_irdma_accept(cm_node, 0, NULL);
3678
3679	if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
3680		ret = -EINVAL;
3681		goto error;
3682	}
3683
3684	passive_state = atomic_add_return(1, &cm_node->passive_state);
3685	if (passive_state == IRDMA_SEND_RESET_EVENT) {
3686		ret = -ECONNRESET;
3687		goto error;
3688	}
3689
3690	buf_len = conn_param->private_data_len + IRDMA_MAX_IETF_SIZE;
3691	iwqp->ietf_mem.size = ALIGN(buf_len, 1);
3692	iwqp->ietf_mem.va = dma_alloc_coherent(dev->hw->device,
3693					       iwqp->ietf_mem.size,
3694					       &iwqp->ietf_mem.pa, GFP_KERNEL);
3695	if (!iwqp->ietf_mem.va) {
3696		ret = -ENOMEM;
3697		goto error;
3698	}
3699
3700	cm_node->pdata.size = conn_param->private_data_len;
3701	accept.addr = iwqp->ietf_mem.va;
3702	accept.size = irdma_cm_build_mpa_frame(cm_node, &accept, MPA_KEY_REPLY);
3703	memcpy((u8 *)accept.addr + accept.size, conn_param->private_data,
3704	       conn_param->private_data_len);
3705
3706	if (cm_node->dev->ws_add(iwqp->sc_qp.vsi, cm_node->user_pri)) {
3707		ret = -ENOMEM;
3708		goto error;
3709	}
3710	iwqp->sc_qp.user_pri = cm_node->user_pri;
3711	irdma_qp_add_qos(&iwqp->sc_qp);
3712	/* setup our first outgoing iWarp send WQE (the IETF frame response) */
3713	iwpd = iwqp->iwpd;
3714	tagged_offset = (uintptr_t)iwqp->ietf_mem.va;
3715	ibmr = irdma_reg_phys_mr(&iwpd->ibpd, iwqp->ietf_mem.pa, buf_len,
3716				 IB_ACCESS_LOCAL_WRITE, &tagged_offset, false);
3717	if (IS_ERR(ibmr)) {
3718		ret = -ENOMEM;
3719		goto error;
3720	}
3721
3722	ibmr->pd = &iwpd->ibpd;
3723	ibmr->device = iwpd->ibpd.device;
3724	iwqp->lsmm_mr = ibmr;
3725	if (iwqp->page)
3726		iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
3727
3728	cm_node->lsmm_size = accept.size + conn_param->private_data_len;
3729	irdma_sc_send_lsmm(&iwqp->sc_qp, iwqp->ietf_mem.va, cm_node->lsmm_size,
3730			   ibmr->lkey);
3731
3732	if (iwqp->page)
3733		kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
3734
3735	iwqp->cm_id = cm_id;
3736	cm_node->cm_id = cm_id;
3737
3738	cm_id->provider_data = iwqp;
3739	iwqp->active_conn = 0;
3740	iwqp->cm_node = cm_node;
3741	cm_node->iwqp = iwqp;
3742	irdma_cm_init_tsa_conn(iwqp, cm_node);
3743	irdma_qp_add_ref(&iwqp->ibqp);
3744	cm_id->add_ref(cm_id);
3745
3746	attr.qp_state = IB_QPS_RTS;
3747	cm_node->qhash_set = false;
3748	cm_node->cm_core->cm_free_ah(cm_node);
3749
3750	irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
3751	if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
3752		wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
3753							    iwqp->rts_ae_rcvd,
3754							    IRDMA_MAX_TIMEOUT);
3755		if (!wait_ret) {
3756			ibdev_dbg(&iwdev->ibdev,
3757				  "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
3758				  cm_node, cm_node->loc_port,
3759				  cm_node->rem_port, cm_node->cm_id);
3760			ret = -ECONNRESET;
3761			goto error;
3762		}
3763	}
3764
3765	irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_ESTABLISHED, 0);
3766	cm_node->accelerated = true;
3767	complete(&cm_node->establish_comp);
3768
3769	if (cm_node->accept_pend) {
3770		atomic_dec(&cm_node->listener->pend_accepts_cnt);
3771		cm_node->accept_pend = 0;
3772	}
3773
3774	ibdev_dbg(&iwdev->ibdev,
3775		  "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n",
3776		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
3777		  cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
3778	cm_node->cm_core->stats_accepts++;
3779
3780	return 0;
3781error:
3782	irdma_free_lsmm_rsrc(iwqp);
3783	irdma_rem_ref_cm_node(cm_node);
3784
3785	return ret;
3786}
3787
3788/**
3789 * irdma_reject - registered call for connection to be rejected
3790 * @cm_id: cm information for passive connection
3791 * @pdata: private data to be sent
3792 * @pdata_len: private data length
3793 */
3794int irdma_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
3795{
3796	struct irdma_device *iwdev;
3797	struct irdma_cm_node *cm_node;
3798
3799	cm_node = cm_id->provider_data;
3800	cm_node->pdata.size = pdata_len;
3801
3802	trace_irdma_reject(cm_node, 0, NULL);
3803
3804	iwdev = to_iwdev(cm_id->device);
3805	if (!iwdev)
3806		return -EINVAL;
3807
3808	cm_node->cm_core->stats_rejects++;
3809
3810	if (pdata_len + sizeof(struct ietf_mpa_v2) > IRDMA_MAX_CM_BUF)
3811		return -EINVAL;
3812
3813	return irdma_cm_reject(cm_node, pdata, pdata_len);
3814}
3815
3816/**
3817 * irdma_connect - registered call for connection to be established
3818 * @cm_id: cm information for passive connection
3819 * @conn_param: Information about the connection
3820 */
3821int irdma_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
3822{
3823	struct ib_qp *ibqp;
3824	struct irdma_qp *iwqp;
3825	struct irdma_device *iwdev;
3826	struct irdma_cm_node *cm_node;
3827	struct irdma_cm_info cm_info;
3828	struct sockaddr_in *laddr;
3829	struct sockaddr_in *raddr;
3830	struct sockaddr_in6 *laddr6;
3831	struct sockaddr_in6 *raddr6;
3832	int ret = 0;
3833
3834	ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
3835	if (!ibqp)
3836		return -EINVAL;
3837	iwqp = to_iwqp(ibqp);
3838	if (!iwqp)
3839		return -EINVAL;
3840	iwdev = iwqp->iwdev;
3841	if (!iwdev)
3842		return -EINVAL;
3843
3844	laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3845	raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
3846	laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3847	raddr6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
3848
3849	if (!(laddr->sin_port) || !(raddr->sin_port))
3850		return -EINVAL;
3851
3852	iwqp->active_conn = 1;
3853	iwqp->cm_id = NULL;
3854	cm_id->provider_data = iwqp;
3855
3856	/* set up the connection params for the node */
3857	if (cm_id->remote_addr.ss_family == AF_INET) {
3858		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
3859			return -EINVAL;
3860
3861		cm_info.ipv4 = true;
3862		memset(cm_info.loc_addr, 0, sizeof(cm_info.loc_addr));
3863		memset(cm_info.rem_addr, 0, sizeof(cm_info.rem_addr));
3864		cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
3865		cm_info.rem_addr[0] = ntohl(raddr->sin_addr.s_addr);
3866		cm_info.loc_port = ntohs(laddr->sin_port);
3867		cm_info.rem_port = ntohs(raddr->sin_port);
3868		cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
3869	} else {
3870		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
3871			return -EINVAL;
3872
3873		cm_info.ipv4 = false;
3874		irdma_copy_ip_ntohl(cm_info.loc_addr,
3875				    laddr6->sin6_addr.in6_u.u6_addr32);
3876		irdma_copy_ip_ntohl(cm_info.rem_addr,
3877				    raddr6->sin6_addr.in6_u.u6_addr32);
3878		cm_info.loc_port = ntohs(laddr6->sin6_port);
3879		cm_info.rem_port = ntohs(raddr6->sin6_port);
3880		irdma_get_vlan_mac_ipv6(cm_info.loc_addr, &cm_info.vlan_id,
3881					NULL);
3882	}
3883	cm_info.cm_id = cm_id;
3884	cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
3885	cm_info.tos = cm_id->tos;
3886	if (iwdev->vsi.dscp_mode) {
3887		cm_info.user_pri =
3888			iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(cm_info.tos)];
3889	} else {
3890		cm_info.user_pri = rt_tos2priority(cm_id->tos);
3891		cm_info.user_pri = irdma_iw_get_vlan_prio(cm_info.loc_addr,
3892							  cm_info.user_pri,
3893							  cm_info.ipv4);
3894	}
3895
3896	if (iwqp->sc_qp.dev->ws_add(iwqp->sc_qp.vsi, cm_info.user_pri))
3897		return -ENOMEM;
3898	iwqp->sc_qp.user_pri = cm_info.user_pri;
3899	irdma_qp_add_qos(&iwqp->sc_qp);
3900	ibdev_dbg(&iwdev->ibdev, "DCB: TOS:[%d] UP:[%d]\n", cm_id->tos,
3901		  cm_info.user_pri);
3902
3903	trace_irdma_dcb_tos(iwdev, cm_id->tos, cm_info.user_pri);
3904
3905	ret = irdma_create_cm_node(&iwdev->cm_core, iwdev, conn_param, &cm_info,
3906				   &cm_node);
3907	if (ret)
3908		return ret;
3909	ret = cm_node->cm_core->cm_create_ah(cm_node, true);
3910	if (ret)
3911		goto err;
3912	if (irdma_manage_qhash(iwdev, &cm_info,
3913			       IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
3914			       IRDMA_QHASH_MANAGE_TYPE_ADD, NULL, true)) {
3915		ret = -EINVAL;
3916		goto err;
3917	}
3918	cm_node->qhash_set = true;
3919
3920	cm_node->apbvt_entry = irdma_add_apbvt(iwdev, cm_info.loc_port);
3921	if (!cm_node->apbvt_entry) {
3922		ret = -EINVAL;
3923		goto err;
3924	}
3925
3926	cm_node->apbvt_set = true;
3927	iwqp->cm_node = cm_node;
3928	cm_node->iwqp = iwqp;
3929	iwqp->cm_id = cm_id;
3930	irdma_qp_add_ref(&iwqp->ibqp);
3931	cm_id->add_ref(cm_id);
3932
3933	if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
3934		cm_node->state = IRDMA_CM_STATE_SYN_SENT;
3935		ret = irdma_send_syn(cm_node, 0);
3936		if (ret)
3937			goto err;
3938	}
3939
3940	ibdev_dbg(&iwdev->ibdev,
3941		  "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n",
3942		  cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
3943		  cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
3944
3945	trace_irdma_connect(cm_node, 0, NULL);
3946
3947	return 0;
3948
3949err:
3950	if (cm_info.ipv4)
3951		ibdev_dbg(&iwdev->ibdev,
3952			  "CM: connect() FAILED: dest addr=%pI4",
3953			  cm_info.rem_addr);
3954	else
3955		ibdev_dbg(&iwdev->ibdev,
3956			  "CM: connect() FAILED: dest addr=%pI6",
3957			  cm_info.rem_addr);
3958	irdma_rem_ref_cm_node(cm_node);
3959	iwdev->cm_core.stats_connect_errs++;
3960
3961	return ret;
3962}
3963
3964/**
3965 * irdma_create_listen - registered call creating listener
3966 * @cm_id: cm information for passive connection
3967 * @backlog: to max accept pending count
3968 */
3969int irdma_create_listen(struct iw_cm_id *cm_id, int backlog)
3970{
3971	struct irdma_device *iwdev;
3972	struct irdma_cm_listener *cm_listen_node;
3973	struct irdma_cm_info cm_info = {};
3974	struct sockaddr_in *laddr;
3975	struct sockaddr_in6 *laddr6;
3976	bool wildcard = false;
3977	int err;
3978
3979	iwdev = to_iwdev(cm_id->device);
3980	if (!iwdev)
3981		return -EINVAL;
3982
3983	laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
3984	laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
3985	cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
3986
3987	if (laddr->sin_family == AF_INET) {
3988		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
3989			return -EINVAL;
3990
3991		cm_info.ipv4 = true;
3992		cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
3993		cm_info.loc_port = ntohs(laddr->sin_port);
3994
3995		if (laddr->sin_addr.s_addr != htonl(INADDR_ANY)) {
3996			cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
3997		} else {
3998			cm_info.vlan_id = 0xFFFF;
3999			wildcard = true;
4000		}
4001	} else {
4002		if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
4003			return -EINVAL;
4004
4005		cm_info.ipv4 = false;
4006		irdma_copy_ip_ntohl(cm_info.loc_addr,
4007				    laddr6->sin6_addr.in6_u.u6_addr32);
4008		cm_info.loc_port = ntohs(laddr6->sin6_port);
4009		if (ipv6_addr_type(&laddr6->sin6_addr) != IPV6_ADDR_ANY) {
4010			irdma_get_vlan_mac_ipv6(cm_info.loc_addr,
4011						&cm_info.vlan_id, NULL);
4012		} else {
4013			cm_info.vlan_id = 0xFFFF;
4014			wildcard = true;
4015		}
4016	}
4017
4018	if (cm_info.vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
4019		cm_info.vlan_id = 0;
4020	cm_info.backlog = backlog;
4021	cm_info.cm_id = cm_id;
4022
4023	trace_irdma_create_listen(iwdev, &cm_info);
4024
4025	cm_listen_node = irdma_make_listen_node(&iwdev->cm_core, iwdev,
4026						&cm_info);
4027	if (!cm_listen_node) {
4028		ibdev_dbg(&iwdev->ibdev,
4029			  "CM: cm_listen_node == NULL\n");
4030		return -ENOMEM;
4031	}
4032
4033	cm_id->provider_data = cm_listen_node;
4034
4035	cm_listen_node->tos = cm_id->tos;
4036	if (iwdev->vsi.dscp_mode)
4037		cm_listen_node->user_pri =
4038		iwdev->vsi.dscp_map[irdma_tos2dscp(cm_id->tos)];
4039	else
4040		cm_listen_node->user_pri = rt_tos2priority(cm_id->tos);
4041	cm_info.user_pri = cm_listen_node->user_pri;
4042	if (!cm_listen_node->reused_node) {
4043		if (wildcard) {
4044			err = irdma_add_mqh(iwdev, &cm_info, cm_listen_node);
4045			if (err)
4046				goto error;
4047		} else {
4048			if (!iwdev->vsi.dscp_mode)
4049				cm_listen_node->user_pri =
4050				irdma_iw_get_vlan_prio(cm_info.loc_addr,
4051						       cm_info.user_pri,
4052						       cm_info.ipv4);
4053			cm_info.user_pri = cm_listen_node->user_pri;
4054			err = irdma_manage_qhash(iwdev, &cm_info,
4055						 IRDMA_QHASH_TYPE_TCP_SYN,
4056						 IRDMA_QHASH_MANAGE_TYPE_ADD,
4057						 NULL, true);
4058			if (err)
4059				goto error;
4060
4061			cm_listen_node->qhash_set = true;
4062		}
4063
4064		cm_listen_node->apbvt_entry = irdma_add_apbvt(iwdev,
4065							      cm_info.loc_port);
4066		if (!cm_listen_node->apbvt_entry)
4067			goto error;
4068	}
4069	cm_id->add_ref(cm_id);
4070	cm_listen_node->cm_core->stats_listen_created++;
4071	ibdev_dbg(&iwdev->ibdev,
4072		  "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d\n",
4073		  cm_listen_node->loc_port, cm_listen_node->loc_addr,
4074		  cm_listen_node, cm_listen_node->cm_id,
4075		  cm_listen_node->qhash_set, cm_listen_node->vlan_id);
4076
4077	return 0;
4078
4079error:
4080
4081	irdma_cm_del_listen(&iwdev->cm_core, cm_listen_node, false);
4082
4083	return -EINVAL;
4084}
4085
4086/**
4087 * irdma_destroy_listen - registered call to destroy listener
4088 * @cm_id: cm information for passive connection
4089 */
4090int irdma_destroy_listen(struct iw_cm_id *cm_id)
4091{
4092	struct irdma_device *iwdev;
4093
4094	iwdev = to_iwdev(cm_id->device);
4095	if (cm_id->provider_data)
4096		irdma_cm_del_listen(&iwdev->cm_core, cm_id->provider_data,
4097				    true);
4098	else
4099		ibdev_dbg(&iwdev->ibdev,
4100			  "CM: cm_id->provider_data was NULL\n");
4101
4102	cm_id->rem_ref(cm_id);
4103
4104	return 0;
4105}
4106
4107/**
4108 * irdma_teardown_list_prep - add conn nodes slated for tear down to list
4109 * @cm_core: cm's core
4110 * @teardown_list: a list to which cm_node will be selected
4111 * @ipaddr: pointer to ip address
4112 * @nfo: pointer to cm_info structure instance
4113 * @disconnect_all: flag indicating disconnect all QPs
4114 */
4115static void irdma_teardown_list_prep(struct irdma_cm_core *cm_core,
4116				     struct list_head *teardown_list,
4117				     u32 *ipaddr,
4118				     struct irdma_cm_info *nfo,
4119				     bool disconnect_all)
4120{
4121	struct irdma_cm_node *cm_node;
4122	int bkt;
4123
4124	hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) {
4125		if ((disconnect_all ||
4126		     (nfo->vlan_id == cm_node->vlan_id &&
4127		      !memcmp(cm_node->loc_addr, ipaddr, nfo->ipv4 ? 4 : 16))) &&
4128		    refcount_inc_not_zero(&cm_node->refcnt))
4129			list_add(&cm_node->teardown_entry, teardown_list);
4130	}
4131}
4132
4133/**
4134 * irdma_cm_event_connected - handle connected active node
4135 * @event: the info for cm_node of connection
4136 */
4137static void irdma_cm_event_connected(struct irdma_cm_event *event)
4138{
4139	struct irdma_qp *iwqp;
4140	struct irdma_device *iwdev;
4141	struct irdma_cm_node *cm_node;
4142	struct irdma_sc_dev *dev;
4143	struct ib_qp_attr attr = {};
4144	struct iw_cm_id *cm_id;
4145	int status;
4146	bool read0;
4147	int wait_ret = 0;
4148
4149	cm_node = event->cm_node;
4150	cm_id = cm_node->cm_id;
4151	iwqp = cm_id->provider_data;
4152	iwdev = iwqp->iwdev;
4153	dev = &iwdev->rf->sc_dev;
4154	if (iwqp->sc_qp.qp_uk.destroy_pending) {
4155		status = -ETIMEDOUT;
4156		goto error;
4157	}
4158
4159	irdma_cm_init_tsa_conn(iwqp, cm_node);
4160	read0 = (cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO);
4161	if (iwqp->page)
4162		iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
4163	irdma_sc_send_rtt(&iwqp->sc_qp, read0);
4164	if (iwqp->page)
4165		kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
4166
4167	attr.qp_state = IB_QPS_RTS;
4168	cm_node->qhash_set = false;
4169	irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
4170	if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
4171		wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
4172							    iwqp->rts_ae_rcvd,
4173							    IRDMA_MAX_TIMEOUT);
4174		if (!wait_ret)
4175			ibdev_dbg(&iwdev->ibdev,
4176				  "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
4177				  cm_node, cm_node->loc_port,
4178				  cm_node->rem_port, cm_node->cm_id);
4179	}
4180
4181	irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 0);
4182	cm_node->accelerated = true;
4183	complete(&cm_node->establish_comp);
4184	cm_node->cm_core->cm_free_ah(cm_node);
4185	return;
4186
4187error:
4188	iwqp->cm_id = NULL;
4189	cm_id->provider_data = NULL;
4190	irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
4191			    status);
4192	irdma_rem_ref_cm_node(event->cm_node);
4193}
4194
4195/**
4196 * irdma_cm_event_reset - handle reset
4197 * @event: the info for cm_node of connection
4198 */
4199static void irdma_cm_event_reset(struct irdma_cm_event *event)
4200{
4201	struct irdma_cm_node *cm_node = event->cm_node;
4202	struct iw_cm_id *cm_id = cm_node->cm_id;
4203	struct irdma_qp *iwqp;
4204
4205	if (!cm_id)
4206		return;
4207
4208	iwqp = cm_id->provider_data;
4209	if (!iwqp)
4210		return;
4211
4212	ibdev_dbg(&cm_node->iwdev->ibdev,
4213		  "CM: reset event %p - cm_id = %p\n", event->cm_node, cm_id);
4214	iwqp->cm_id = NULL;
4215
4216	irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_DISCONNECT,
4217			    -ECONNRESET);
4218	irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_CLOSE, 0);
4219}
4220
4221/**
4222 * irdma_cm_event_handler - send event to cm upper layer
4223 * @work: pointer of cm event info.
4224 */
4225static void irdma_cm_event_handler(struct work_struct *work)
4226{
4227	struct irdma_cm_event *event = container_of(work, struct irdma_cm_event, event_work);
4228	struct irdma_cm_node *cm_node;
4229
4230	if (!event || !event->cm_node || !event->cm_node->cm_core)
4231		return;
4232
4233	cm_node = event->cm_node;
4234	trace_irdma_cm_event_handler(cm_node, event->type, NULL);
4235
4236	switch (event->type) {
4237	case IRDMA_CM_EVENT_MPA_REQ:
4238		irdma_send_cm_event(cm_node, cm_node->cm_id,
4239				    IW_CM_EVENT_CONNECT_REQUEST, 0);
4240		break;
4241	case IRDMA_CM_EVENT_RESET:
4242		irdma_cm_event_reset(event);
4243		break;
4244	case IRDMA_CM_EVENT_CONNECTED:
4245		if (!cm_node->cm_id ||
4246		    cm_node->state != IRDMA_CM_STATE_OFFLOADED)
4247			break;
4248		irdma_cm_event_connected(event);
4249		break;
4250	case IRDMA_CM_EVENT_MPA_REJECT:
4251		if (!cm_node->cm_id ||
4252		    cm_node->state == IRDMA_CM_STATE_OFFLOADED)
4253			break;
4254		irdma_send_cm_event(cm_node, cm_node->cm_id,
4255				    IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
4256		break;
4257	case IRDMA_CM_EVENT_ABORTED:
4258		if (!cm_node->cm_id ||
4259		    cm_node->state == IRDMA_CM_STATE_OFFLOADED)
4260			break;
4261		irdma_event_connect_error(event);
4262		break;
4263	default:
4264		ibdev_dbg(&cm_node->iwdev->ibdev,
4265			  "CM: bad event type = %d\n", event->type);
4266		break;
4267	}
4268
4269	irdma_rem_ref_cm_node(cm_node);
4270	kfree(event);
4271}
4272
4273/**
4274 * irdma_cm_post_event - queue event request for worker thread
4275 * @event: cm node's info for up event call
4276 */
4277static void irdma_cm_post_event(struct irdma_cm_event *event)
4278{
4279	refcount_inc(&event->cm_node->refcnt);
4280	INIT_WORK(&event->event_work, irdma_cm_event_handler);
4281	queue_work(event->cm_node->cm_core->event_wq, &event->event_work);
4282}
4283
4284/**
4285 * irdma_cm_teardown_connections - teardown QPs
4286 * @iwdev: device pointer
4287 * @ipaddr: Pointer to IPv4 or IPv6 address
4288 * @nfo: Connection info
4289 * @disconnect_all: flag indicating disconnect all QPs
4290 *
4291 * teardown QPs where source or destination addr matches ip addr
4292 */
4293void irdma_cm_teardown_connections(struct irdma_device *iwdev, u32 *ipaddr,
4294				   struct irdma_cm_info *nfo,
4295				   bool disconnect_all)
4296{
4297	struct irdma_cm_core *cm_core = &iwdev->cm_core;
4298	struct list_head *list_core_temp;
4299	struct list_head *list_node;
4300	struct irdma_cm_node *cm_node;
4301	struct list_head teardown_list;
4302	struct ib_qp_attr attr;
4303
4304	INIT_LIST_HEAD(&teardown_list);
4305
4306	rcu_read_lock();
4307	irdma_teardown_list_prep(cm_core, &teardown_list, ipaddr, nfo, disconnect_all);
4308	rcu_read_unlock();
4309
4310	list_for_each_safe (list_node, list_core_temp, &teardown_list) {
4311		cm_node = container_of(list_node, struct irdma_cm_node,
4312				       teardown_entry);
4313		attr.qp_state = IB_QPS_ERR;
4314		irdma_modify_qp(&cm_node->iwqp->ibqp, &attr, IB_QP_STATE, NULL);
4315		if (iwdev->rf->reset)
4316			irdma_cm_disconn(cm_node->iwqp);
4317		irdma_rem_ref_cm_node(cm_node);
4318	}
4319}
4320
4321/**
4322 * irdma_qhash_ctrl - enable/disable qhash for list
4323 * @iwdev: device pointer
4324 * @parent_listen_node: parent listen node
4325 * @nfo: cm info node
4326 * @ipaddr: Pointer to IPv4 or IPv6 address
4327 * @ipv4: flag indicating IPv4 when true
4328 * @ifup: flag indicating interface up when true
4329 *
4330 * Enables or disables the qhash for the node in the child
4331 * listen list that matches ipaddr. If no matching IP was found
4332 * it will allocate and add a new child listen node to the
4333 * parent listen node. The listen_list_lock is assumed to be
4334 * held when called.
4335 */
4336static void irdma_qhash_ctrl(struct irdma_device *iwdev,
4337			     struct irdma_cm_listener *parent_listen_node,
4338			     struct irdma_cm_info *nfo, u32 *ipaddr, bool ipv4,
4339			     bool ifup)
4340{
4341	struct list_head *child_listen_list = &parent_listen_node->child_listen_list;
4342	struct irdma_cm_listener *child_listen_node;
4343	struct list_head *pos, *tpos;
4344	bool node_allocated = false;
4345	enum irdma_quad_hash_manage_type op = ifup ?
4346					      IRDMA_QHASH_MANAGE_TYPE_ADD :
4347					      IRDMA_QHASH_MANAGE_TYPE_DELETE;
4348	int err;
4349
4350	list_for_each_safe (pos, tpos, child_listen_list) {
4351		child_listen_node = list_entry(pos, struct irdma_cm_listener,
4352					       child_listen_list);
4353		if (!memcmp(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16))
4354			goto set_qhash;
4355	}
4356
4357	/* if not found then add a child listener if interface is going up */
4358	if (!ifup)
4359		return;
4360	child_listen_node = kmemdup(parent_listen_node,
4361				    sizeof(*child_listen_node), GFP_ATOMIC);
4362	if (!child_listen_node)
4363		return;
4364
4365	node_allocated = true;
4366	memcpy(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16);
4367
4368set_qhash:
4369	memcpy(nfo->loc_addr, child_listen_node->loc_addr,
4370	       sizeof(nfo->loc_addr));
4371	nfo->vlan_id = child_listen_node->vlan_id;
4372	err = irdma_manage_qhash(iwdev, nfo, IRDMA_QHASH_TYPE_TCP_SYN, op, NULL,
4373				 false);
4374	if (!err) {
4375		child_listen_node->qhash_set = ifup;
4376		if (node_allocated)
4377			list_add(&child_listen_node->child_listen_list,
4378				 &parent_listen_node->child_listen_list);
4379	} else if (node_allocated) {
4380		kfree(child_listen_node);
4381	}
4382}
4383
4384/**
4385 * irdma_if_notify - process an ifdown on an interface
4386 * @iwdev: device pointer
4387 * @netdev: network device structure
4388 * @ipaddr: Pointer to IPv4 or IPv6 address
4389 * @ipv4: flag indicating IPv4 when true
4390 * @ifup: flag indicating interface up when true
4391 */
4392void irdma_if_notify(struct irdma_device *iwdev, struct net_device *netdev,
4393		     u32 *ipaddr, bool ipv4, bool ifup)
4394{
4395	struct irdma_cm_core *cm_core = &iwdev->cm_core;
4396	unsigned long flags;
4397	struct irdma_cm_listener *listen_node;
4398	static const u32 ip_zero[4] = { 0, 0, 0, 0 };
4399	struct irdma_cm_info nfo = {};
4400	u16 vlan_id = rdma_vlan_dev_vlan_id(netdev);
4401	enum irdma_quad_hash_manage_type op = ifup ?
4402					      IRDMA_QHASH_MANAGE_TYPE_ADD :
4403					      IRDMA_QHASH_MANAGE_TYPE_DELETE;
4404
4405	nfo.vlan_id = vlan_id;
4406	nfo.ipv4 = ipv4;
4407	nfo.qh_qpid = 1;
4408
4409	/* Disable or enable qhash for listeners */
4410	spin_lock_irqsave(&cm_core->listen_list_lock, flags);
4411	list_for_each_entry (listen_node, &cm_core->listen_list, list) {
4412		if (vlan_id != listen_node->vlan_id ||
4413		    (memcmp(listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16) &&
4414		     memcmp(listen_node->loc_addr, ip_zero, ipv4 ? 4 : 16)))
4415			continue;
4416
4417		memcpy(nfo.loc_addr, listen_node->loc_addr,
4418		       sizeof(nfo.loc_addr));
4419		nfo.loc_port = listen_node->loc_port;
4420		nfo.user_pri = listen_node->user_pri;
4421		if (!list_empty(&listen_node->child_listen_list)) {
4422			irdma_qhash_ctrl(iwdev, listen_node, &nfo, ipaddr, ipv4,
4423					 ifup);
4424		} else if (memcmp(listen_node->loc_addr, ip_zero,
4425				  ipv4 ? 4 : 16)) {
4426			if (!irdma_manage_qhash(iwdev, &nfo,
4427						IRDMA_QHASH_TYPE_TCP_SYN, op,
4428						NULL, false))
4429				listen_node->qhash_set = ifup;
4430		}
4431	}
4432	spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
4433
4434	/* disconnect any connected qp's on ifdown */
4435	if (!ifup)
4436		irdma_cm_teardown_connections(iwdev, ipaddr, &nfo, false);
4437}