af_smc.c

   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 *  Shared Memory Communications over RDMA (SMC-R) and RoCE
   4 *
   5 *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
   6 *  applies to SOCK_STREAM sockets only
   7 *  offers an alternative communication option for TCP-protocol sockets
   8 *  applicable with RoCE-cards only
   9 *
  10 *  Initial restrictions:
  11 *    - support for alternate links postponed
  12 *
  13 *  Copyright IBM Corp. 2016, 2018
  14 *
  15 *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
  16 *              based on prototype from Frank Blaschka
  17 */
  18
  19#define pr_fmt(fmt) "smc: " fmt
  20
  21#include <linux/module.h>
  22#include <linux/socket.h>
  23#include <linux/workqueue.h>
  24#include <linux/in.h>
  25#include <linux/sched/signal.h>
  26#include <linux/if_vlan.h>
  27#include <linux/rcupdate_wait.h>
  28#include <linux/ctype.h>
  29#include <linux/splice.h>
  30
  31#include <net/sock.h>
  32#include <net/inet_common.h>
  33#if IS_ENABLED(CONFIG_IPV6)
  34#include <net/ipv6.h>
  35#endif
  36#include <net/tcp.h>
  37#include <net/smc.h>
  38#include <asm/ioctls.h>
  39
  40#include <net/net_namespace.h>
  41#include <net/netns/generic.h>
  42#include "smc_netns.h"
  43
  44#include "smc.h"
  45#include "smc_clc.h"
  46#include "smc_llc.h"
  47#include "smc_cdc.h"
  48#include "smc_core.h"
  49#include "smc_ib.h"
  50#include "smc_ism.h"
  51#include "smc_pnet.h"
  52#include "smc_netlink.h"
  53#include "smc_tx.h"
  54#include "smc_rx.h"
  55#include "smc_close.h"
  56#include "smc_stats.h"
  57#include "smc_tracepoint.h"
  58#include "smc_sysctl.h"
  59#include "smc_inet.h"
  60#include "smc_hs_bpf.h"
  61
  62static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
  63						 * creation on server
  64						 */
  65static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
  66						 * creation on client
  67						 */
  68
  69static struct workqueue_struct	*smc_tcp_ls_wq;	/* wq for tcp listen work */
  70struct workqueue_struct	*smc_hs_wq;	/* wq for handshake work */
  71struct workqueue_struct	*smc_close_wq;	/* wq for close work */
  72
  73static void smc_tcp_listen_work(struct work_struct *);
  74static void smc_connect_work(struct work_struct *);
  75
  76int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb)
  77{
  78	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
  79	void *hdr;
  80
  81	if (cb_ctx->pos[0])
  82		goto out;
  83
  84	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
  85			  &smc_gen_nl_family, NLM_F_MULTI,
  86			  SMC_NETLINK_DUMP_HS_LIMITATION);
  87	if (!hdr)
  88		return -ENOMEM;
  89
  90	if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED,
  91		       sock_net(skb->sk)->smc.limit_smc_hs))
  92		goto err;
  93
  94	genlmsg_end(skb, hdr);
  95	cb_ctx->pos[0] = 1;
  96out:
  97	return skb->len;
  98err:
  99	genlmsg_cancel(skb, hdr);
 100	return -EMSGSIZE;
 101}
 102
 103int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
 104{
 105	sock_net(skb->sk)->smc.limit_smc_hs = true;
 106	return 0;
 107}
 108
 109int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
 110{
 111	sock_net(skb->sk)->smc.limit_smc_hs = false;
 112	return 0;
 113}
 114
 115static void smc_set_keepalive(struct sock *sk, int val)
 116{
 117	struct smc_sock *smc = smc_sk(sk);
 118
 119	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
 120}
 121
 122static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk,
 123					  struct sk_buff *skb,
 124					  struct request_sock *req,
 125					  struct dst_entry *dst,
 126					  struct request_sock *req_unhash,
 127					  bool *own_req,
 128					  void (*opt_child_init)(struct sock *newsk,
 129								 const struct sock *sk))
 130{
 131	struct smc_sock *smc;
 132	struct sock *child;
 133
 134	rcu_read_lock();
 135	smc = smc_clcsock_user_data_rcu(sk);
 136	if (!smc || !refcount_inc_not_zero(&smc->sk.sk_refcnt)) {
 137		rcu_read_unlock();
 138		smc = NULL;
 139		goto drop;
 140	}
 141	rcu_read_unlock();
 142
 143	if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) >
 144				sk->sk_max_ack_backlog)
 145		goto drop;
 146
 147	if (sk_acceptq_is_full(&smc->sk)) {
 148		NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
 149		goto drop;
 150	}
 151
 152	/* passthrough to original syn recv sock fct */
 153	child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash,
 154					       own_req, opt_child_init);
 155	/* child must not inherit smc or its ops */
 156	if (child) {
 157		rcu_assign_sk_user_data(child, NULL);
 158
 159		/* v4-mapped sockets don't inherit parent ops. Don't restore. */
 160		if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops)
 161			inet_csk(child)->icsk_af_ops = smc->ori_af_ops;
 162	}
 163	sock_put(&smc->sk);
 164	return child;
 165
 166drop:
 167	dst_release(dst);
 168	tcp_listendrop(sk);
 169	if (smc)
 170		sock_put(&smc->sk);
 171	return NULL;
 172}
 173
 174static bool smc_hs_congested(const struct sock *sk)
 175{
 176	const struct smc_sock *smc;
 177
 178	smc = smc_clcsock_user_data(sk);
 179
 180	if (!smc)
 181		return true;
 182
 183	if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq))
 184		return true;
 185
 186	return false;
 187}
 188
 189struct smc_hashinfo smc_v4_hashinfo = {
 190	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
 191};
 192
 193struct smc_hashinfo smc_v6_hashinfo = {
 194	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
 195};
 196
 197int smc_hash_sk(struct sock *sk)
 198{
 199	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
 200	struct hlist_head *head;
 201
 202	head = &h->ht;
 203
 204	write_lock_bh(&h->lock);
 205	sk_add_node(sk, head);
 206	write_unlock_bh(&h->lock);
 207	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
 208
 209	return 0;
 210}
 211
 212void smc_unhash_sk(struct sock *sk)
 213{
 214	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
 215
 216	write_lock_bh(&h->lock);
 217	if (sk_del_node_init(sk))
 218		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
 219	write_unlock_bh(&h->lock);
 220}
 221
 222/* This will be called before user really release sock_lock. So do the
 223 * work which we didn't do because of user hold the sock_lock in the
 224 * BH context
 225 */
 226void smc_release_cb(struct sock *sk)
 227{
 228	struct smc_sock *smc = smc_sk(sk);
 229
 230	if (smc->conn.tx_in_release_sock) {
 231		smc_tx_pending(&smc->conn);
 232		smc->conn.tx_in_release_sock = false;
 233	}
 234}
 235
 236struct proto smc_proto = {
 237	.name		= "SMC",
 238	.owner		= THIS_MODULE,
 239	.keepalive	= smc_set_keepalive,
 240	.hash		= smc_hash_sk,
 241	.unhash		= smc_unhash_sk,
 242	.release_cb	= smc_release_cb,
 243	.obj_size	= sizeof(struct smc_sock),
 244	.h.smc_hash	= &smc_v4_hashinfo,
 245	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
 246};
 247EXPORT_SYMBOL_GPL(smc_proto);
 248
 249struct proto smc_proto6 = {
 250	.name		= "SMC6",
 251	.owner		= THIS_MODULE,
 252	.keepalive	= smc_set_keepalive,
 253	.hash		= smc_hash_sk,
 254	.unhash		= smc_unhash_sk,
 255	.release_cb	= smc_release_cb,
 256	.obj_size	= sizeof(struct smc_sock),
 257	.h.smc_hash	= &smc_v6_hashinfo,
 258	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
 259};
 260EXPORT_SYMBOL_GPL(smc_proto6);
 261
 262static void smc_fback_restore_callbacks(struct smc_sock *smc)
 263{
 264	struct sock *clcsk = smc->clcsock->sk;
 265
 266	write_lock_bh(&clcsk->sk_callback_lock);
 267	rcu_assign_sk_user_data(clcsk, NULL);
 268
 269	smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change);
 270	smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready);
 271	smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space);
 272	smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report);
 273
 274	write_unlock_bh(&clcsk->sk_callback_lock);
 275}
 276
 277static void smc_restore_fallback_changes(struct smc_sock *smc)
 278{
 279	if (smc->clcsock->file) { /* non-accepted sockets have no file yet */
 280		smc->clcsock->file->private_data = smc->sk.sk_socket;
 281		smc->clcsock->file = NULL;
 282		smc_fback_restore_callbacks(smc);
 283	}
 284}
 285
 286static int __smc_release(struct smc_sock *smc)
 287{
 288	struct sock *sk = &smc->sk;
 289	int rc = 0;
 290
 291	if (!smc->use_fallback) {
 292		rc = smc_close_active(smc);
 293		smc_sock_set_flag(sk, SOCK_DEAD);
 294		sk->sk_shutdown |= SHUTDOWN_MASK;
 295	} else {
 296		if (sk->sk_state != SMC_CLOSED) {
 297			if (sk->sk_state != SMC_LISTEN &&
 298			    sk->sk_state != SMC_INIT)
 299				sock_put(sk); /* passive closing */
 300			if (sk->sk_state == SMC_LISTEN) {
 301				/* wake up clcsock accept */
 302				rc = kernel_sock_shutdown(smc->clcsock,
 303							  SHUT_RDWR);
 304			}
 305			sk->sk_state = SMC_CLOSED;
 306			sk->sk_state_change(sk);
 307		}
 308		smc_restore_fallback_changes(smc);
 309	}
 310
 311	sk->sk_prot->unhash(sk);
 312
 313	if (sk->sk_state == SMC_CLOSED) {
 314		if (smc->clcsock) {
 315			release_sock(sk);
 316			smc_clcsock_release(smc);
 317			lock_sock(sk);
 318		}
 319		if (!smc->use_fallback)
 320			smc_conn_free(&smc->conn);
 321	}
 322
 323	return rc;
 324}
 325
 326int smc_release(struct socket *sock)
 327{
 328	struct sock *sk = sock->sk;
 329	struct smc_sock *smc;
 330	int old_state, rc = 0;
 331
 332	if (!sk)
 333		goto out;
 334
 335	sock_hold(sk); /* sock_put below */
 336	smc = smc_sk(sk);
 337
 338	old_state = sk->sk_state;
 339
 340	/* cleanup for a dangling non-blocking connect */
 341	if (smc->connect_nonblock && old_state == SMC_INIT)
 342		tcp_abort(smc->clcsock->sk, ECONNABORTED);
 343
 344	if (cancel_work_sync(&smc->connect_work))
 345		sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */
 346
 347	if (sk->sk_state == SMC_LISTEN)
 348		/* smc_close_non_accepted() is called and acquires
 349		 * sock lock for child sockets again
 350		 */
 351		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 352	else
 353		lock_sock(sk);
 354
 355	if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE &&
 356	    !smc->use_fallback)
 357		smc_close_active_abort(smc);
 358
 359	rc = __smc_release(smc);
 360
 361	/* detach socket */
 362	sock_orphan(sk);
 363	sock->sk = NULL;
 364	release_sock(sk);
 365
 366	sock_put(sk); /* sock_hold above */
 367	sock_put(sk); /* final sock_put */
 368out:
 369	return rc;
 370}
 371
 372static void smc_destruct(struct sock *sk)
 373{
 374	if (sk->sk_state != SMC_CLOSED)
 375		return;
 376	if (!sock_flag(sk, SOCK_DEAD))
 377		return;
 378	switch (sk->sk_family) {
 379	case AF_INET:
 380		inet_sock_destruct(sk);
 381		break;
 382#if IS_ENABLED(CONFIG_IPV6)
 383	case AF_INET6:
 384		inet6_sock_destruct(sk);
 385		break;
 386#endif
 387	}
 388}
 389
 390static struct lock_class_key smc_key;
 391static struct lock_class_key smc_slock_key;
 392
 393void smc_sk_init(struct net *net, struct sock *sk, int protocol)
 394{
 395	struct smc_sock *smc = smc_sk(sk);
 396
 397	sk->sk_state = SMC_INIT;
 398	sk->sk_destruct = smc_destruct;
 399	sk->sk_protocol = protocol;
 400	WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem));
 401	WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem));
 402	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
 403	INIT_WORK(&smc->connect_work, smc_connect_work);
 404	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
 405	INIT_LIST_HEAD(&smc->accept_q);
 406	sock_lock_init_class_and_name(sk, "slock-AF_SMC", &smc_slock_key,
 407				      "sk_lock-AF_SMC", &smc_key);
 408	spin_lock_init(&smc->accept_q_lock);
 409	spin_lock_init(&smc->conn.send_lock);
 410	sk->sk_prot->hash(sk);
 411	mutex_init(&smc->clcsock_release_lock);
 412	smc_init_saved_callbacks(smc);
 413	smc->limit_smc_hs = net->smc.limit_smc_hs;
 414	smc->use_fallback = false; /* assume rdma capability first */
 415	smc->fallback_rsn = 0;
 416	smc_close_init(smc);
 417}
 418
 419static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
 420				   int protocol)
 421{
 422	struct proto *prot;
 423	struct sock *sk;
 424
 425	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
 426	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
 427	if (!sk)
 428		return NULL;
 429
 430	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
 431	smc_sk_init(net, sk, protocol);
 432
 433	return sk;
 434}
 435
 436int smc_bind(struct socket *sock, struct sockaddr_unsized *uaddr,
 437	     int addr_len)
 438{
 439	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 440	struct sock *sk = sock->sk;
 441	struct smc_sock *smc;
 442	int rc;
 443
 444	smc = smc_sk(sk);
 445
 446	/* replicate tests from inet_bind(), to be safe wrt. future changes */
 447	rc = -EINVAL;
 448	if (addr_len < sizeof(struct sockaddr_in))
 449		goto out;
 450
 451	rc = -EAFNOSUPPORT;
 452	if (addr->sin_family != AF_INET &&
 453	    addr->sin_family != AF_INET6 &&
 454	    addr->sin_family != AF_UNSPEC)
 455		goto out;
 456	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
 457	if (addr->sin_family == AF_UNSPEC &&
 458	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
 459		goto out;
 460
 461	lock_sock(sk);
 462
 463	/* Check if socket is already active */
 464	rc = -EINVAL;
 465	if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
 466		goto out_rel;
 467
 468	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
 469	smc->clcsock->sk->sk_reuseport = sk->sk_reuseport;
 470	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
 471
 472out_rel:
 473	release_sock(sk);
 474out:
 475	return rc;
 476}
 477
 478/* copy only relevant settings and flags of SOL_SOCKET level from smc to
 479 * clc socket (since smc is not called for these options from net/core)
 480 */
 481
 482#define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
 483			     (1UL << SOCK_KEEPOPEN) | \
 484			     (1UL << SOCK_LINGER) | \
 485			     (1UL << SOCK_BROADCAST) | \
 486			     (1UL << SOCK_TIMESTAMP) | \
 487			     (1UL << SOCK_DBG) | \
 488			     (1UL << SOCK_RCVTSTAMP) | \
 489			     (1UL << SOCK_RCVTSTAMPNS) | \
 490			     (1UL << SOCK_LOCALROUTE) | \
 491			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
 492			     (1UL << SOCK_RXQ_OVFL) | \
 493			     (1UL << SOCK_WIFI_STATUS) | \
 494			     (1UL << SOCK_NOFCS) | \
 495			     (1UL << SOCK_FILTER_LOCKED) | \
 496			     (1UL << SOCK_TSTAMP_NEW))
 497
 498/* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */
 499static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk,
 500				     unsigned long mask)
 501{
 502	nsk->sk_userlocks = osk->sk_userlocks;
 503	if (osk->sk_userlocks & SOCK_SNDBUF_LOCK)
 504		nsk->sk_sndbuf = osk->sk_sndbuf;
 505	if (osk->sk_userlocks & SOCK_RCVBUF_LOCK)
 506		nsk->sk_rcvbuf = osk->sk_rcvbuf;
 507}
 508
 509static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
 510				   unsigned long mask)
 511{
 512	/* options we don't get control via setsockopt for */
 513	nsk->sk_type = osk->sk_type;
 514	nsk->sk_sndtimeo = READ_ONCE(osk->sk_sndtimeo);
 515	nsk->sk_rcvtimeo = READ_ONCE(osk->sk_rcvtimeo);
 516	nsk->sk_mark = READ_ONCE(osk->sk_mark);
 517	nsk->sk_priority = READ_ONCE(osk->sk_priority);
 518	nsk->sk_rcvlowat = osk->sk_rcvlowat;
 519	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
 520	nsk->sk_err = osk->sk_err;
 521
 522	nsk->sk_flags &= ~mask;
 523	nsk->sk_flags |= osk->sk_flags & mask;
 524
 525	smc_adjust_sock_bufsizes(nsk, osk, mask);
 526}
 527
 528static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
 529{
 530	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
 531}
 532
 533#define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
 534			     (1UL << SOCK_KEEPOPEN) | \
 535			     (1UL << SOCK_LINGER) | \
 536			     (1UL << SOCK_DBG))
 537/* copy only settings and flags relevant for smc from clc to smc socket */
 538static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
 539{
 540	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
 541}
 542
 543/* register the new vzalloced sndbuf on all links */
 544static int smcr_lgr_reg_sndbufs(struct smc_link *link,
 545				struct smc_buf_desc *snd_desc)
 546{
 547	struct smc_link_group *lgr = link->lgr;
 548	int i, rc = 0;
 549
 550	if (!snd_desc->is_vm)
 551		return -EINVAL;
 552
 553	/* protect against parallel smcr_link_reg_buf() */
 554	down_write(&lgr->llc_conf_mutex);
 555	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
 556		if (!smc_link_active(&lgr->lnk[i]))
 557			continue;
 558		rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc);
 559		if (rc)
 560			break;
 561	}
 562	up_write(&lgr->llc_conf_mutex);
 563	return rc;
 564}
 565
 566/* register the new rmb on all links */
 567static int smcr_lgr_reg_rmbs(struct smc_link *link,
 568			     struct smc_buf_desc *rmb_desc)
 569{
 570	struct smc_link_group *lgr = link->lgr;
 571	bool do_slow = false;
 572	int i, rc = 0;
 573
 574	rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
 575	if (rc)
 576		return rc;
 577
 578	down_read(&lgr->llc_conf_mutex);
 579	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
 580		if (!smc_link_active(&lgr->lnk[i]))
 581			continue;
 582		if (!rmb_desc->is_reg_mr[link->link_idx]) {
 583			up_read(&lgr->llc_conf_mutex);
 584			goto slow_path;
 585		}
 586	}
 587	/* mr register already */
 588	goto fast_path;
 589slow_path:
 590	do_slow = true;
 591	/* protect against parallel smc_llc_cli_rkey_exchange() and
 592	 * parallel smcr_link_reg_buf()
 593	 */
 594	down_write(&lgr->llc_conf_mutex);
 595	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
 596		if (!smc_link_active(&lgr->lnk[i]))
 597			continue;
 598		rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc);
 599		if (rc)
 600			goto out;
 601	}
 602fast_path:
 603	/* exchange confirm_rkey msg with peer */
 604	rc = smc_llc_do_confirm_rkey(link, rmb_desc);
 605	if (rc) {
 606		rc = -EFAULT;
 607		goto out;
 608	}
 609	rmb_desc->is_conf_rkey = true;
 610out:
 611	do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex);
 612	smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
 613	return rc;
 614}
 615
 616static int smcr_clnt_conf_first_link(struct smc_sock *smc)
 617{
 618	struct smc_link *link = smc->conn.lnk;
 619	struct smc_llc_qentry *qentry;
 620	int rc;
 621
 622	/* Receive CONFIRM LINK request from server over RoCE fabric.
 623	 * Increasing the client's timeout by twice as much as the server's
 624	 * timeout by default can temporarily avoid decline messages of
 625	 * both sides crossing or colliding
 626	 */
 627	qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME,
 628			      SMC_LLC_CONFIRM_LINK);
 629	if (!qentry) {
 630		struct smc_clc_msg_decline dclc;
 631
 632		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
 633				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
 634		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
 635	}
 636	smc_llc_save_peer_uid(qentry);
 637	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
 638	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
 639	if (rc)
 640		return SMC_CLC_DECL_RMBE_EC;
 641
 642	rc = smc_ib_modify_qp_rts(link);
 643	if (rc)
 644		return SMC_CLC_DECL_ERR_RDYLNK;
 645
 646	smc_wr_remember_qp_attr(link);
 647
 648	/* reg the sndbuf if it was vzalloced */
 649	if (smc->conn.sndbuf_desc->is_vm) {
 650		if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
 651			return SMC_CLC_DECL_ERR_REGBUF;
 652	}
 653
 654	/* reg the rmb */
 655	if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
 656		return SMC_CLC_DECL_ERR_REGBUF;
 657
 658	/* confirm_rkey is implicit on 1st contact */
 659	smc->conn.rmb_desc->is_conf_rkey = true;
 660
 661	/* send CONFIRM LINK response over RoCE fabric */
 662	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
 663	if (rc < 0)
 664		return SMC_CLC_DECL_TIMEOUT_CL;
 665
 666	smc_llc_link_active(link);
 667	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
 668
 669	if (link->lgr->max_links > 1) {
 670		/* optional 2nd link, receive ADD LINK request from server */
 671		qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
 672				      SMC_LLC_ADD_LINK);
 673		if (!qentry) {
 674			struct smc_clc_msg_decline dclc;
 675
 676			rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
 677					      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
 678			if (rc == -EAGAIN)
 679				rc = 0; /* no DECLINE received, go with one link */
 680			return rc;
 681		}
 682		smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
 683		smc_llc_cli_add_link(link, qentry);
 684	}
 685	return 0;
 686}
 687
 688static bool smc_isascii(char *hostname)
 689{
 690	int i;
 691
 692	for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
 693		if (!isascii(hostname[i]))
 694			return false;
 695	return true;
 696}
 697
 698static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
 699					struct smc_clc_msg_accept_confirm *clc)
 700{
 701	struct smc_clc_first_contact_ext *fce;
 702	int clc_v2_len;
 703
 704	if (clc->hdr.version == SMC_V1 ||
 705	    !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
 706		return;
 707
 708	if (smc->conn.lgr->is_smcd) {
 709		memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid,
 710		       SMC_MAX_EID_LEN);
 711		clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1);
 712	} else {
 713		memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid,
 714		       SMC_MAX_EID_LEN);
 715		clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1);
 716	}
 717	fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len);
 718	smc->conn.lgr->peer_os = fce->os_type;
 719	smc->conn.lgr->peer_smc_release = fce->release;
 720	if (smc_isascii(fce->hostname))
 721		memcpy(smc->conn.lgr->peer_hostname, fce->hostname,
 722		       SMC_MAX_HOSTNAME_LEN);
 723}
 724
 725static void smcr_conn_save_peer_info(struct smc_sock *smc,
 726				     struct smc_clc_msg_accept_confirm *clc)
 727{
 728	int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
 729
 730	smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx;
 731	smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token);
 732	smc->conn.peer_rmbe_size = bufsize;
 733	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
 734	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
 735}
 736
 737static void smcd_conn_save_peer_info(struct smc_sock *smc,
 738				     struct smc_clc_msg_accept_confirm *clc)
 739{
 740	int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
 741
 742	smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
 743	smc->conn.peer_token = ntohll(clc->d0.token);
 744	/* msg header takes up space in the buffer */
 745	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
 746	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
 747	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
 748}
 749
 750static void smc_conn_save_peer_info(struct smc_sock *smc,
 751				    struct smc_clc_msg_accept_confirm *clc)
 752{
 753	if (smc->conn.lgr->is_smcd)
 754		smcd_conn_save_peer_info(smc, clc);
 755	else
 756		smcr_conn_save_peer_info(smc, clc);
 757	smc_conn_save_peer_info_fce(smc, clc);
 758}
 759
 760static void smc_link_save_peer_info(struct smc_link *link,
 761				    struct smc_clc_msg_accept_confirm *clc,
 762				    struct smc_init_info *ini)
 763{
 764	link->peer_qpn = ntoh24(clc->r0.qpn);
 765	memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE);
 766	memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac));
 767	link->peer_psn = ntoh24(clc->r0.psn);
 768	link->peer_mtu = clc->r0.qp_mtu;
 769}
 770
 771static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
 772				       struct smc_stats_fback *fback_arr)
 773{
 774	int cnt;
 775
 776	for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) {
 777		if (fback_arr[cnt].fback_code == smc->fallback_rsn) {
 778			fback_arr[cnt].count++;
 779			break;
 780		}
 781		if (!fback_arr[cnt].fback_code) {
 782			fback_arr[cnt].fback_code = smc->fallback_rsn;
 783			fback_arr[cnt].count++;
 784			break;
 785		}
 786	}
 787}
 788
 789static void smc_stat_fallback(struct smc_sock *smc)
 790{
 791	struct net *net = sock_net(&smc->sk);
 792
 793	mutex_lock(&net->smc.mutex_fback_rsn);
 794	if (smc->listen_smc) {
 795		smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv);
 796		net->smc.fback_rsn->srv_fback_cnt++;
 797	} else {
 798		smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
 799		net->smc.fback_rsn->clnt_fback_cnt++;
 800	}
 801	mutex_unlock(&net->smc.mutex_fback_rsn);
 802}
 803
 804/* must be called under rcu read lock */
 805static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
 806{
 807	struct socket_wq *wq;
 808	__poll_t flags;
 809
 810	wq = rcu_dereference(smc->sk.sk_wq);
 811	if (!skwq_has_sleeper(wq))
 812		return;
 813
 814	/* wake up smc sk->sk_wq */
 815	if (!key) {
 816		/* sk_state_change */
 817		wake_up_interruptible_all(&wq->wait);
 818	} else {
 819		flags = key_to_poll(key);
 820		if (flags & (EPOLLIN | EPOLLOUT))
 821			/* sk_data_ready or sk_write_space */
 822			wake_up_interruptible_sync_poll(&wq->wait, flags);
 823		else if (flags & EPOLLERR)
 824			/* sk_error_report */
 825			wake_up_interruptible_poll(&wq->wait, flags);
 826	}
 827}
 828
 829static int smc_fback_mark_woken(wait_queue_entry_t *wait,
 830				unsigned int mode, int sync, void *key)
 831{
 832	struct smc_mark_woken *mark =
 833		container_of(wait, struct smc_mark_woken, wait_entry);
 834
 835	mark->woken = true;
 836	mark->key = key;
 837	return 0;
 838}
 839
 840static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
 841				     void (*clcsock_callback)(struct sock *sk))
 842{
 843	struct smc_mark_woken mark = { .woken = false };
 844	struct socket_wq *wq;
 845
 846	init_waitqueue_func_entry(&mark.wait_entry,
 847				  smc_fback_mark_woken);
 848	rcu_read_lock();
 849	wq = rcu_dereference(clcsk->sk_wq);
 850	if (!wq)
 851		goto out;
 852	add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
 853	clcsock_callback(clcsk);
 854	remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
 855
 856	if (mark.woken)
 857		smc_fback_wakeup_waitqueue(smc, mark.key);
 858out:
 859	rcu_read_unlock();
 860}
 861
 862static void smc_fback_state_change(struct sock *clcsk)
 863{
 864	struct smc_sock *smc;
 865
 866	read_lock_bh(&clcsk->sk_callback_lock);
 867	smc = smc_clcsock_user_data(clcsk);
 868	if (smc)
 869		smc_fback_forward_wakeup(smc, clcsk,
 870					 smc->clcsk_state_change);
 871	read_unlock_bh(&clcsk->sk_callback_lock);
 872}
 873
 874static void smc_fback_data_ready(struct sock *clcsk)
 875{
 876	struct smc_sock *smc;
 877
 878	read_lock_bh(&clcsk->sk_callback_lock);
 879	smc = smc_clcsock_user_data(clcsk);
 880	if (smc)
 881		smc_fback_forward_wakeup(smc, clcsk,
 882					 smc->clcsk_data_ready);
 883	read_unlock_bh(&clcsk->sk_callback_lock);
 884}
 885
 886static void smc_fback_write_space(struct sock *clcsk)
 887{
 888	struct smc_sock *smc;
 889
 890	read_lock_bh(&clcsk->sk_callback_lock);
 891	smc = smc_clcsock_user_data(clcsk);
 892	if (smc)
 893		smc_fback_forward_wakeup(smc, clcsk,
 894					 smc->clcsk_write_space);
 895	read_unlock_bh(&clcsk->sk_callback_lock);
 896}
 897
 898static void smc_fback_error_report(struct sock *clcsk)
 899{
 900	struct smc_sock *smc;
 901
 902	read_lock_bh(&clcsk->sk_callback_lock);
 903	smc = smc_clcsock_user_data(clcsk);
 904	if (smc)
 905		smc_fback_forward_wakeup(smc, clcsk,
 906					 smc->clcsk_error_report);
 907	read_unlock_bh(&clcsk->sk_callback_lock);
 908}
 909
 910static void smc_fback_replace_callbacks(struct smc_sock *smc)
 911{
 912	struct sock *clcsk = smc->clcsock->sk;
 913
 914	write_lock_bh(&clcsk->sk_callback_lock);
 915	__rcu_assign_sk_user_data_with_flags(clcsk, smc, SK_USER_DATA_NOCOPY);
 916
 917	smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change,
 918			       &smc->clcsk_state_change);
 919	smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready,
 920			       &smc->clcsk_data_ready);
 921	smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space,
 922			       &smc->clcsk_write_space);
 923	smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report,
 924			       &smc->clcsk_error_report);
 925
 926	write_unlock_bh(&clcsk->sk_callback_lock);
 927}
 928
 929static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
 930{
 931	int rc = 0;
 932
 933	mutex_lock(&smc->clcsock_release_lock);
 934	if (!smc->clcsock) {
 935		rc = -EBADF;
 936		goto out;
 937	}
 938
 939	smc->use_fallback = true;
 940	smc->fallback_rsn = reason_code;
 941	smc_stat_fallback(smc);
 942	trace_smc_switch_to_fallback(smc, reason_code);
 943	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
 944		smc->clcsock->file = smc->sk.sk_socket->file;
 945		smc->clcsock->file->private_data = smc->clcsock;
 946		smc->clcsock->wq.fasync_list =
 947			smc->sk.sk_socket->wq.fasync_list;
 948		smc->sk.sk_socket->wq.fasync_list = NULL;
 949
 950		/* There might be some wait entries remaining
 951		 * in smc sk->sk_wq and they should be woken up
 952		 * as clcsock's wait queue is woken up.
 953		 */
 954		smc_fback_replace_callbacks(smc);
 955	}
 956out:
 957	mutex_unlock(&smc->clcsock_release_lock);
 958	return rc;
 959}
 960
 961/* fall back during connect */
 962static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
 963{
 964	struct net *net = sock_net(&smc->sk);
 965	int rc = 0;
 966
 967	rc = smc_switch_to_fallback(smc, reason_code);
 968	if (rc) { /* fallback fails */
 969		this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
 970		if (smc->sk.sk_state == SMC_INIT)
 971			sock_put(&smc->sk); /* passive closing */
 972		return rc;
 973	}
 974	smc_copy_sock_settings_to_clc(smc);
 975	smc->connect_nonblock = 0;
 976	if (smc->sk.sk_state == SMC_INIT)
 977		smc->sk.sk_state = SMC_ACTIVE;
 978	return 0;
 979}
 980
 981/* decline and fall back during connect */
 982static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
 983					u8 version)
 984{
 985	struct net *net = sock_net(&smc->sk);
 986	int rc;
 987
 988	if (reason_code < 0) { /* error, fallback is not possible */
 989		this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
 990		if (smc->sk.sk_state == SMC_INIT)
 991			sock_put(&smc->sk); /* passive closing */
 992		return reason_code;
 993	}
 994	if (reason_code != SMC_CLC_DECL_PEERDECL) {
 995		rc = smc_clc_send_decline(smc, reason_code, version);
 996		if (rc < 0) {
 997			this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
 998			if (smc->sk.sk_state == SMC_INIT)
 999				sock_put(&smc->sk); /* passive closing */
1000			return rc;
1001		}
1002	}
1003	return smc_connect_fallback(smc, reason_code);
1004}
1005
1006static void smc_conn_abort(struct smc_sock *smc, int local_first)
1007{
1008	struct smc_connection *conn = &smc->conn;
1009	struct smc_link_group *lgr = conn->lgr;
1010	bool lgr_valid = false;
1011
1012	if (smc_conn_lgr_valid(conn))
1013		lgr_valid = true;
1014
1015	smc_conn_free(conn);
1016	if (local_first && lgr_valid)
1017		smc_lgr_cleanup_early(lgr);
1018}
1019
1020/* check if there is a rdma device available for this connection. */
1021/* called for connect and listen */
1022static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
1023{
1024	/* PNET table look up: search active ib_device and port
1025	 * within same PNETID that also contains the ethernet device
1026	 * used for the internal TCP socket
1027	 */
1028	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
1029	if (!ini->check_smcrv2 && !ini->ib_dev)
1030		return SMC_CLC_DECL_NOSMCRDEV;
1031	if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1032		return SMC_CLC_DECL_NOSMCRDEV;
1033	return 0;
1034}
1035
1036/* check if there is an ISM device available for this connection. */
1037/* called for connect and listen */
1038static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1039{
1040	/* Find ISM device with same PNETID as connecting interface  */
1041	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1042	if (!ini->ism_dev[0])
1043		return SMC_CLC_DECL_NOSMCDDEV;
1044	else
1045		ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1046	return 0;
1047}
1048
1049/* is chid unique for the ism devices that are already determined? */
1050static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1051					   int cnt)
1052{
1053	int i = (!ini->ism_dev[0]) ? 1 : 0;
1054
1055	for (; i < cnt; i++)
1056		if (ini->ism_chid[i] == chid)
1057			return false;
1058	return true;
1059}
1060
1061/* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1062 * PNETID matching net_device)
1063 */
1064static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1065				       struct smc_init_info *ini)
1066{
1067	int rc = SMC_CLC_DECL_NOSMCDDEV;
1068	struct smcd_dev *smcd;
1069	int i = 1, entry = 1;
1070	bool is_emulated;
1071	u16 chid;
1072
1073	if (smcd_indicated(ini->smc_type_v1))
1074		rc = 0;		/* already initialized for V1 */
1075	mutex_lock(&smcd_dev_list.mutex);
1076	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1077		if (smcd->going_away || smcd == ini->ism_dev[0])
1078			continue;
1079		chid = smc_ism_get_chid(smcd);
1080		if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1081			continue;
1082		is_emulated = __smc_ism_is_emulated(chid);
1083		if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1084		    smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1085			if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES)
1086				/* It's the last GID-CHID entry left in CLC
1087				 * Proposal SMC-Dv2 extension, but an Emulated-
1088				 * ISM device will take two entries. So give
1089				 * up it and try the next potential ISM device.
1090				 */
1091				continue;
1092			ini->ism_dev[i] = smcd;
1093			ini->ism_chid[i] = chid;
1094			ini->is_smcd = true;
1095			rc = 0;
1096			i++;
1097			entry = is_emulated ? entry + 2 : entry + 1;
1098			if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES)
1099				break;
1100		}
1101	}
1102	mutex_unlock(&smcd_dev_list.mutex);
1103	ini->ism_offered_cnt = i - 1;
1104	if (!ini->ism_dev[0] && !ini->ism_dev[1])
1105		ini->smcd_version = 0;
1106
1107	return rc;
1108}
1109
1110/* Check for VLAN ID and register it on ISM device just for CLC handshake */
1111static int smc_connect_ism_vlan_setup(struct smc_init_info *ini)
1112{
1113	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1114		return SMC_CLC_DECL_ISMVLANERR;
1115	return 0;
1116}
1117
1118static int smc_find_proposal_devices(struct smc_sock *smc,
1119				     struct smc_init_info *ini)
1120{
1121	int rc = 0;
1122
1123	/* check if there is an ism device available */
1124	if (!(ini->smcd_version & SMC_V1) ||
1125	    smc_find_ism_device(smc, ini) ||
1126	    smc_connect_ism_vlan_setup(ini))
1127		ini->smcd_version &= ~SMC_V1;
1128	/* else ISM V1 is supported for this connection */
1129
1130	/* check if there is an rdma device available */
1131	if (!(ini->smcr_version & SMC_V1) ||
1132	    smc_find_rdma_device(smc, ini))
1133		ini->smcr_version &= ~SMC_V1;
1134	/* else RDMA is supported for this connection */
1135
1136	ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1137					      ini->smcr_version & SMC_V1);
1138
1139	/* check if there is an ism v2 device available */
1140	if (!(ini->smcd_version & SMC_V2) ||
1141	    !smc_ism_is_v2_capable() ||
1142	    smc_find_ism_v2_device_clnt(smc, ini))
1143		ini->smcd_version &= ~SMC_V2;
1144
1145	/* check if there is an rdma v2 device available */
1146	ini->check_smcrv2 = true;
1147	ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1148	if (!(ini->smcr_version & SMC_V2) ||
1149#if IS_ENABLED(CONFIG_IPV6)
1150	    (smc->clcsock->sk->sk_family == AF_INET6 &&
1151	     !ipv6_addr_v4mapped(&smc->clcsock->sk->sk_v6_rcv_saddr)) ||
1152#endif
1153	    !smc_clc_ueid_count() ||
1154	    smc_find_rdma_device(smc, ini))
1155		ini->smcr_version &= ~SMC_V2;
1156	ini->check_smcrv2 = false;
1157
1158	ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1159					      ini->smcr_version & SMC_V2);
1160
1161	/* if neither ISM nor RDMA are supported, fallback */
1162	if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1163		rc = SMC_CLC_DECL_NOSMCDEV;
1164
1165	return rc;
1166}
1167
1168/* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1169 * used, the VLAN ID will be registered again during the connection setup.
1170 */
1171static int smc_connect_ism_vlan_cleanup(struct smc_init_info *ini)
1172{
1173	if (!smcd_indicated(ini->smc_type_v1))
1174		return 0;
1175	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1176		return SMC_CLC_DECL_CNFERR;
1177	return 0;
1178}
1179
1180#define SMC_CLC_MAX_ACCEPT_LEN \
1181	(sizeof(struct smc_clc_msg_accept_confirm) + \
1182	 sizeof(struct smc_clc_first_contact_ext_v2x) + \
1183	 sizeof(struct smc_clc_msg_trail))
1184
1185/* CLC handshake during connect */
1186static int smc_connect_clc(struct smc_sock *smc,
1187			   struct smc_clc_msg_accept_confirm *aclc,
1188			   struct smc_init_info *ini)
1189{
1190	int rc = 0;
1191
1192	/* do inband token exchange */
1193	rc = smc_clc_send_proposal(smc, ini);
1194	if (rc)
1195		return rc;
1196	/* receive SMC Accept CLC message */
1197	return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN,
1198				SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1199}
1200
1201void smc_fill_gid_list(struct smc_link_group *lgr,
1202		       struct smc_gidlist *gidlist,
1203		       struct smc_ib_device *known_dev, u8 *known_gid)
1204{
1205	struct smc_init_info *alt_ini = NULL;
1206
1207	memset(gidlist, 0, sizeof(*gidlist));
1208	memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1209
1210	alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1211	if (!alt_ini)
1212		goto out;
1213
1214	alt_ini->vlan_id = lgr->vlan_id;
1215	alt_ini->check_smcrv2 = true;
1216	alt_ini->smcrv2.saddr = lgr->saddr;
1217	smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1218
1219	if (!alt_ini->smcrv2.ib_dev_v2)
1220		goto out;
1221
1222	memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1223	       SMC_GID_SIZE);
1224
1225out:
1226	kfree(alt_ini);
1227}
1228
1229static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1230				       struct smc_clc_msg_accept_confirm *aclc,
1231				       struct smc_init_info *ini)
1232{
1233	struct smc_clc_first_contact_ext *fce =
1234		smc_get_clc_first_contact_ext(aclc, false);
1235	struct net *net = sock_net(&smc->sk);
1236	int rc;
1237
1238	if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1239		return 0;
1240
1241	if (fce->v2_direct) {
1242		memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1243		ini->smcrv2.uses_gateway = false;
1244	} else {
1245		if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr,
1246				      smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1247				      ini->smcrv2.nexthop_mac,
1248				      &ini->smcrv2.uses_gateway))
1249			return SMC_CLC_DECL_NOROUTE;
1250		if (!ini->smcrv2.uses_gateway) {
1251			/* mismatch: peer claims indirect, but its direct */
1252			return SMC_CLC_DECL_NOINDIRECT;
1253		}
1254	}
1255
1256	ini->release_nr = fce->release;
1257	rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1258	if (rc)
1259		return rc;
1260
1261	return 0;
1262}
1263
1264/* setup for RDMA connection of client */
1265static int smc_connect_rdma(struct smc_sock *smc,
1266			    struct smc_clc_msg_accept_confirm *aclc,
1267			    struct smc_init_info *ini)
1268{
1269	int i, reason_code = 0;
1270	struct smc_link *link;
1271	u8 *eid = NULL;
1272
1273	ini->is_smcd = false;
1274	ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1275	ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1276	memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1277	memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1278	memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1279	ini->max_conns = SMC_CONN_PER_LGR_MAX;
1280	ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1281
1282	reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1283	if (reason_code)
1284		return reason_code;
1285
1286	mutex_lock(&smc_client_lgr_pending);
1287	reason_code = smc_conn_create(smc, ini);
1288	if (reason_code) {
1289		mutex_unlock(&smc_client_lgr_pending);
1290		return reason_code;
1291	}
1292
1293	smc_conn_save_peer_info(smc, aclc);
1294
1295	if (ini->first_contact_local) {
1296		link = smc->conn.lnk;
1297	} else {
1298		/* set link that was assigned by server */
1299		link = NULL;
1300		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1301			struct smc_link *l = &smc->conn.lgr->lnk[i];
1302
1303			if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1304			    !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1305				    SMC_GID_SIZE) &&
1306			    (aclc->hdr.version > SMC_V1 ||
1307			     !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1308				     sizeof(l->peer_mac)))) {
1309				link = l;
1310				break;
1311			}
1312		}
1313		if (!link) {
1314			reason_code = SMC_CLC_DECL_NOSRVLINK;
1315			goto connect_abort;
1316		}
1317		smc_switch_link_and_count(&smc->conn, link);
1318	}
1319
1320	/* create send buffer and rmb */
1321	if (smc_buf_create(smc, false)) {
1322		reason_code = SMC_CLC_DECL_MEM;
1323		goto connect_abort;
1324	}
1325
1326	if (ini->first_contact_local)
1327		smc_link_save_peer_info(link, aclc, ini);
1328
1329	if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1330		reason_code = SMC_CLC_DECL_ERR_RTOK;
1331		goto connect_abort;
1332	}
1333
1334	smc_rx_init(smc);
1335
1336	if (ini->first_contact_local) {
1337		if (smc_ib_ready_link(link)) {
1338			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1339			goto connect_abort;
1340		}
1341	} else {
1342		/* reg sendbufs if they were vzalloced */
1343		if (smc->conn.sndbuf_desc->is_vm) {
1344			if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) {
1345				reason_code = SMC_CLC_DECL_ERR_REGBUF;
1346				goto connect_abort;
1347			}
1348		}
1349		if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1350			reason_code = SMC_CLC_DECL_ERR_REGBUF;
1351			goto connect_abort;
1352		}
1353	}
1354
1355	if (aclc->hdr.version > SMC_V1) {
1356		eid = aclc->r1.eid;
1357		if (ini->first_contact_local)
1358			smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1359					  link->smcibdev, link->gid);
1360	}
1361
1362	reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1363					   aclc->hdr.version, eid, ini);
1364	if (reason_code)
1365		goto connect_abort;
1366
1367	smc_tx_init(smc);
1368
1369	if (ini->first_contact_local) {
1370		/* QP confirmation over RoCE fabric */
1371		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1372		reason_code = smcr_clnt_conf_first_link(smc);
1373		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1374		if (reason_code)
1375			goto connect_abort;
1376	}
1377	mutex_unlock(&smc_client_lgr_pending);
1378
1379	smc_copy_sock_settings_to_clc(smc);
1380	smc->connect_nonblock = 0;
1381	if (smc->sk.sk_state == SMC_INIT)
1382		smc->sk.sk_state = SMC_ACTIVE;
1383
1384	return 0;
1385connect_abort:
1386	smc_conn_abort(smc, ini->first_contact_local);
1387	mutex_unlock(&smc_client_lgr_pending);
1388	smc->connect_nonblock = 0;
1389
1390	return reason_code;
1391}
1392
1393/* The server has chosen one of the proposed ISM devices for the communication.
1394 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1395 */
1396static int
1397smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc,
1398			       struct smc_init_info *ini)
1399{
1400	int i;
1401
1402	for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1403		if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1404			ini->ism_selected = i;
1405			return 0;
1406		}
1407	}
1408
1409	return -EPROTO;
1410}
1411
1412/* setup for ISM connection of client */
1413static int smc_connect_ism(struct smc_sock *smc,
1414			   struct smc_clc_msg_accept_confirm *aclc,
1415			   struct smc_init_info *ini)
1416{
1417	u8 *eid = NULL;
1418	int rc = 0;
1419
1420	ini->is_smcd = true;
1421	ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1422
1423	if (aclc->hdr.version == SMC_V2) {
1424		if (ini->first_contact_peer) {
1425			struct smc_clc_first_contact_ext *fce =
1426				smc_get_clc_first_contact_ext(aclc, true);
1427
1428			ini->release_nr = fce->release;
1429			rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1430			if (rc)
1431				return rc;
1432		}
1433
1434		rc = smc_v2_determine_accepted_chid(aclc, ini);
1435		if (rc)
1436			return rc;
1437
1438		if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected]))
1439			ini->ism_peer_gid[ini->ism_selected].gid_ext =
1440						ntohll(aclc->d1.gid_ext);
1441		/* for non-Emulated-ISM devices, peer gid_ext remains 0. */
1442	}
1443	ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid);
1444
1445	/* there is only one lgr role for SMC-D; use server lock */
1446	mutex_lock(&smc_server_lgr_pending);
1447	rc = smc_conn_create(smc, ini);
1448	if (rc) {
1449		mutex_unlock(&smc_server_lgr_pending);
1450		return rc;
1451	}
1452
1453	/* Create send and receive buffers */
1454	rc = smc_buf_create(smc, true);
1455	if (rc) {
1456		rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1457		goto connect_abort;
1458	}
1459
1460	smc_conn_save_peer_info(smc, aclc);
1461
1462	if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) {
1463		rc = smcd_buf_attach(smc);
1464		if (rc) {
1465			rc = SMC_CLC_DECL_MEM;	/* try to fallback */
1466			goto connect_abort;
1467		}
1468	}
1469	smc_rx_init(smc);
1470	smc_tx_init(smc);
1471
1472	if (aclc->hdr.version > SMC_V1)
1473		eid = aclc->d1.eid;
1474
1475	rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1476				  aclc->hdr.version, eid, ini);
1477	if (rc)
1478		goto connect_abort;
1479	mutex_unlock(&smc_server_lgr_pending);
1480
1481	smc_copy_sock_settings_to_clc(smc);
1482	smc->connect_nonblock = 0;
1483	if (smc->sk.sk_state == SMC_INIT)
1484		smc->sk.sk_state = SMC_ACTIVE;
1485
1486	return 0;
1487connect_abort:
1488	smc_conn_abort(smc, ini->first_contact_local);
1489	mutex_unlock(&smc_server_lgr_pending);
1490	smc->connect_nonblock = 0;
1491
1492	return rc;
1493}
1494
1495/* check if received accept type and version matches a proposed one */
1496static int smc_connect_check_aclc(struct smc_init_info *ini,
1497				  struct smc_clc_msg_accept_confirm *aclc)
1498{
1499	if (aclc->hdr.version >= SMC_V2) {
1500		if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1501		     !smcr_indicated(ini->smc_type_v2)) ||
1502		    (aclc->hdr.typev1 == SMC_TYPE_D &&
1503		     !smcd_indicated(ini->smc_type_v2)))
1504			return SMC_CLC_DECL_MODEUNSUPP;
1505	} else {
1506		if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1507		     !smcr_indicated(ini->smc_type_v1)) ||
1508		    (aclc->hdr.typev1 == SMC_TYPE_D &&
1509		     !smcd_indicated(ini->smc_type_v1)))
1510			return SMC_CLC_DECL_MODEUNSUPP;
1511	}
1512
1513	return 0;
1514}
1515
1516/* perform steps before actually connecting */
1517static int __smc_connect(struct smc_sock *smc)
1518{
1519	u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1520	struct smc_clc_msg_accept_confirm *aclc;
1521	struct smc_init_info *ini = NULL;
1522	u8 *buf = NULL;
1523	int rc = 0;
1524
1525	if (smc->use_fallback)
1526		return smc_connect_fallback(smc, smc->fallback_rsn);
1527
1528	/* if peer has not signalled SMC-capability, fall back */
1529	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1530		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1531
1532	/* IPSec connections opt out of SMC optimizations */
1533	if (using_ipsec(smc))
1534		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1535						    version);
1536
1537	ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1538	if (!ini)
1539		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1540						    version);
1541
1542	ini->smcd_version = SMC_V1 | SMC_V2;
1543	ini->smcr_version = SMC_V1 | SMC_V2;
1544	ini->smc_type_v1 = SMC_TYPE_B;
1545	ini->smc_type_v2 = SMC_TYPE_B;
1546
1547	/* get vlan id from IP device */
1548	if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1549		ini->smcd_version &= ~SMC_V1;
1550		ini->smcr_version = 0;
1551		ini->smc_type_v1 = SMC_TYPE_N;
1552	}
1553
1554	rc = smc_find_proposal_devices(smc, ini);
1555	if (rc)
1556		goto fallback;
1557
1558	buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1559	if (!buf) {
1560		rc = SMC_CLC_DECL_MEM;
1561		goto fallback;
1562	}
1563	aclc = (struct smc_clc_msg_accept_confirm *)buf;
1564
1565	/* perform CLC handshake */
1566	rc = smc_connect_clc(smc, aclc, ini);
1567	if (rc) {
1568		/* -EAGAIN on timeout, see tcp_recvmsg() */
1569		if (rc == -EAGAIN) {
1570			rc = -ETIMEDOUT;
1571			smc->sk.sk_err = ETIMEDOUT;
1572		}
1573		goto vlan_cleanup;
1574	}
1575
1576	/* check if smc modes and versions of CLC proposal and accept match */
1577	rc = smc_connect_check_aclc(ini, aclc);
1578	version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1579	if (rc)
1580		goto vlan_cleanup;
1581
1582	/* depending on previous steps, connect using rdma or ism */
1583	if (aclc->hdr.typev1 == SMC_TYPE_R) {
1584		ini->smcr_version = version;
1585		rc = smc_connect_rdma(smc, aclc, ini);
1586	} else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1587		ini->smcd_version = version;
1588		rc = smc_connect_ism(smc, aclc, ini);
1589	}
1590	if (rc)
1591		goto vlan_cleanup;
1592
1593	SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1594	smc_connect_ism_vlan_cleanup(ini);
1595	kfree(buf);
1596	kfree(ini);
1597	return 0;
1598
1599vlan_cleanup:
1600	smc_connect_ism_vlan_cleanup(ini);
1601	kfree(buf);
1602fallback:
1603	kfree(ini);
1604	return smc_connect_decline_fallback(smc, rc, version);
1605}
1606
1607static void smc_connect_work(struct work_struct *work)
1608{
1609	struct smc_sock *smc = container_of(work, struct smc_sock,
1610					    connect_work);
1611	long timeo = READ_ONCE(smc->sk.sk_sndtimeo);
1612	int rc = 0;
1613
1614	if (!timeo)
1615		timeo = MAX_SCHEDULE_TIMEOUT;
1616	lock_sock(smc->clcsock->sk);
1617	if (smc->clcsock->sk->sk_err) {
1618		smc->sk.sk_err = smc->clcsock->sk->sk_err;
1619	} else if ((1 << smc->clcsock->sk->sk_state) &
1620					(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1621		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1622		if ((rc == -EPIPE) &&
1623		    ((1 << smc->clcsock->sk->sk_state) &
1624					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1625			rc = 0;
1626	}
1627	release_sock(smc->clcsock->sk);
1628	lock_sock(&smc->sk);
1629	if (rc != 0 || smc->sk.sk_err) {
1630		smc->sk.sk_state = SMC_CLOSED;
1631		if (rc == -EPIPE || rc == -EAGAIN)
1632			smc->sk.sk_err = EPIPE;
1633		else if (rc == -ECONNREFUSED)
1634			smc->sk.sk_err = ECONNREFUSED;
1635		else if (signal_pending(current))
1636			smc->sk.sk_err = -sock_intr_errno(timeo);
1637		sock_put(&smc->sk); /* passive closing */
1638		goto out;
1639	}
1640
1641	rc = __smc_connect(smc);
1642	if (rc < 0)
1643		smc->sk.sk_err = -rc;
1644
1645out:
1646	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1647		if (smc->sk.sk_err) {
1648			smc->sk.sk_state_change(&smc->sk);
1649		} else { /* allow polling before and after fallback decision */
1650			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1651			smc->sk.sk_write_space(&smc->sk);
1652		}
1653	}
1654	release_sock(&smc->sk);
1655}
1656
1657int smc_connect(struct socket *sock, struct sockaddr_unsized *addr,
1658		int alen, int flags)
1659{
1660	struct sock *sk = sock->sk;
1661	struct smc_sock *smc;
1662	int rc = -EINVAL;
1663
1664	smc = smc_sk(sk);
1665
1666	/* separate smc parameter checking to be safe */
1667	if (alen < sizeof(addr->sa_family))
1668		goto out_err;
1669	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1670		goto out_err;
1671
1672	lock_sock(sk);
1673	switch (sock->state) {
1674	default:
1675		rc = -EINVAL;
1676		goto out;
1677	case SS_CONNECTED:
1678		rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1679		goto out;
1680	case SS_CONNECTING:
1681		if (sk->sk_state == SMC_ACTIVE)
1682			goto connected;
1683		break;
1684	case SS_UNCONNECTED:
1685		sock->state = SS_CONNECTING;
1686		break;
1687	}
1688
1689	switch (sk->sk_state) {
1690	default:
1691		goto out;
1692	case SMC_CLOSED:
1693		rc = sock_error(sk) ? : -ECONNABORTED;
1694		sock->state = SS_UNCONNECTED;
1695		goto out;
1696	case SMC_ACTIVE:
1697		rc = -EISCONN;
1698		goto out;
1699	case SMC_INIT:
1700		break;
1701	}
1702
1703	smc_copy_sock_settings_to_clc(smc);
1704	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1705	if (smc->connect_nonblock) {
1706		rc = -EALREADY;
1707		goto out;
1708	}
1709	rc = kernel_connect(smc->clcsock, (struct sockaddr_unsized *)addr, alen, flags);
1710	if (rc && rc != -EINPROGRESS)
1711		goto out;
1712
1713	if (smc->use_fallback) {
1714		sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1715		goto out;
1716	}
1717	sock_hold(&smc->sk); /* sock put in passive closing */
1718	if (flags & O_NONBLOCK) {
1719		if (queue_work(smc_hs_wq, &smc->connect_work))
1720			smc->connect_nonblock = 1;
1721		rc = -EINPROGRESS;
1722		goto out;
1723	} else {
1724		rc = __smc_connect(smc);
1725		if (rc < 0)
1726			goto out;
1727	}
1728
1729connected:
1730	rc = 0;
1731	sock->state = SS_CONNECTED;
1732out:
1733	release_sock(sk);
1734out_err:
1735	return rc;
1736}
1737
1738static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1739{
1740	struct socket *new_clcsock = NULL;
1741	struct sock *lsk = &lsmc->sk;
1742	struct sock *new_sk;
1743	int rc = -EINVAL;
1744
1745	release_sock(lsk);
1746	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1747	if (!new_sk) {
1748		rc = -ENOMEM;
1749		lsk->sk_err = ENOMEM;
1750		*new_smc = NULL;
1751		lock_sock(lsk);
1752		goto out;
1753	}
1754	*new_smc = smc_sk(new_sk);
1755
1756	mutex_lock(&lsmc->clcsock_release_lock);
1757	if (lsmc->clcsock)
1758		rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1759	mutex_unlock(&lsmc->clcsock_release_lock);
1760	lock_sock(lsk);
1761	if  (rc < 0 && rc != -EAGAIN)
1762		lsk->sk_err = -rc;
1763	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1764		new_sk->sk_prot->unhash(new_sk);
1765		if (new_clcsock)
1766			sock_release(new_clcsock);
1767		new_sk->sk_state = SMC_CLOSED;
1768		smc_sock_set_flag(new_sk, SOCK_DEAD);
1769		sock_put(new_sk); /* final */
1770		*new_smc = NULL;
1771		goto out;
1772	}
1773
1774	/* new clcsock has inherited the smc listen-specific sk_data_ready
1775	 * function; switch it back to the original sk_data_ready function
1776	 */
1777	new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1778
1779	/* if new clcsock has also inherited the fallback-specific callback
1780	 * functions, switch them back to the original ones.
1781	 */
1782	if (lsmc->use_fallback) {
1783		if (lsmc->clcsk_state_change)
1784			new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1785		if (lsmc->clcsk_write_space)
1786			new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1787		if (lsmc->clcsk_error_report)
1788			new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1789	}
1790
1791	(*new_smc)->clcsock = new_clcsock;
1792out:
1793	return rc;
1794}
1795
1796/* add a just created sock to the accept queue of the listen sock as
1797 * candidate for a following socket accept call from user space
1798 */
1799static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1800{
1801	struct smc_sock *par = smc_sk(parent);
1802
1803	sock_hold(sk); /* sock_put in smc_accept_unlink () */
1804	spin_lock(&par->accept_q_lock);
1805	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1806	spin_unlock(&par->accept_q_lock);
1807	sk_acceptq_added(parent);
1808}
1809
1810/* remove a socket from the accept queue of its parental listening socket */
1811static void smc_accept_unlink(struct sock *sk)
1812{
1813	struct smc_sock *par = smc_sk(sk)->listen_smc;
1814
1815	spin_lock(&par->accept_q_lock);
1816	list_del_init(&smc_sk(sk)->accept_q);
1817	spin_unlock(&par->accept_q_lock);
1818	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1819	sock_put(sk); /* sock_hold in smc_accept_enqueue */
1820}
1821
1822/* remove a sock from the accept queue to bind it to a new socket created
1823 * for a socket accept call from user space
1824 */
1825struct sock *smc_accept_dequeue(struct sock *parent,
1826				struct socket *new_sock)
1827{
1828	struct smc_sock *isk, *n;
1829	struct sock *new_sk;
1830
1831	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1832		new_sk = (struct sock *)isk;
1833
1834		smc_accept_unlink(new_sk);
1835		if (new_sk->sk_state == SMC_CLOSED) {
1836			new_sk->sk_prot->unhash(new_sk);
1837			if (isk->clcsock) {
1838				sock_release(isk->clcsock);
1839				isk->clcsock = NULL;
1840			}
1841			sock_put(new_sk); /* final */
1842			continue;
1843		}
1844		if (new_sock) {
1845			sock_graft(new_sk, new_sock);
1846			new_sock->state = SS_CONNECTED;
1847			if (isk->use_fallback) {
1848				smc_sk(new_sk)->clcsock->file = new_sock->file;
1849				isk->clcsock->file->private_data = isk->clcsock;
1850			}
1851		}
1852		return new_sk;
1853	}
1854	return NULL;
1855}
1856
1857/* clean up for a created but never accepted sock */
1858void smc_close_non_accepted(struct sock *sk)
1859{
1860	struct smc_sock *smc = smc_sk(sk);
1861
1862	sock_hold(sk); /* sock_put below */
1863	lock_sock(sk);
1864	if (!sk->sk_lingertime)
1865		/* wait for peer closing */
1866		WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT);
1867	__smc_release(smc);
1868	release_sock(sk);
1869	sock_put(sk); /* sock_hold above */
1870	sock_put(sk); /* final sock_put */
1871}
1872
1873static int smcr_serv_conf_first_link(struct smc_sock *smc)
1874{
1875	struct smc_link *link = smc->conn.lnk;
1876	struct smc_llc_qentry *qentry;
1877	int rc;
1878
1879	/* reg the sndbuf if it was vzalloced*/
1880	if (smc->conn.sndbuf_desc->is_vm) {
1881		if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1882			return SMC_CLC_DECL_ERR_REGBUF;
1883	}
1884
1885	/* reg the rmb */
1886	if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1887		return SMC_CLC_DECL_ERR_REGBUF;
1888
1889	/* send CONFIRM LINK request to client over the RoCE fabric */
1890	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1891	if (rc < 0)
1892		return SMC_CLC_DECL_TIMEOUT_CL;
1893
1894	/* receive CONFIRM LINK response from client over the RoCE fabric */
1895	qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1896			      SMC_LLC_CONFIRM_LINK);
1897	if (!qentry) {
1898		struct smc_clc_msg_decline dclc;
1899
1900		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1901				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1902		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1903	}
1904	smc_llc_save_peer_uid(qentry);
1905	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1906	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1907	if (rc)
1908		return SMC_CLC_DECL_RMBE_EC;
1909
1910	/* confirm_rkey is implicit on 1st contact */
1911	smc->conn.rmb_desc->is_conf_rkey = true;
1912
1913	smc_llc_link_active(link);
1914	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1915
1916	if (link->lgr->max_links > 1) {
1917		down_write(&link->lgr->llc_conf_mutex);
1918		/* initial contact - try to establish second link */
1919		smc_llc_srv_add_link(link, NULL);
1920		up_write(&link->lgr->llc_conf_mutex);
1921	}
1922	return 0;
1923}
1924
1925/* listen worker: finish */
1926static void smc_listen_out(struct smc_sock *new_smc)
1927{
1928	struct smc_sock *lsmc = new_smc->listen_smc;
1929	struct sock *newsmcsk = &new_smc->sk;
1930
1931	if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1932		atomic_dec(&lsmc->queued_smc_hs);
1933
1934	release_sock(newsmcsk); /* lock in smc_listen_work() */
1935	if (lsmc->sk.sk_state == SMC_LISTEN) {
1936		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1937		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1938		release_sock(&lsmc->sk);
1939	} else { /* no longer listening */
1940		smc_close_non_accepted(newsmcsk);
1941	}
1942
1943	/* Wake up accept */
1944	lsmc->sk.sk_data_ready(&lsmc->sk);
1945	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1946}
1947
1948/* listen worker: finish in state connected */
1949static void smc_listen_out_connected(struct smc_sock *new_smc)
1950{
1951	struct sock *newsmcsk = &new_smc->sk;
1952
1953	if (newsmcsk->sk_state == SMC_INIT)
1954		newsmcsk->sk_state = SMC_ACTIVE;
1955
1956	smc_listen_out(new_smc);
1957}
1958
1959/* listen worker: finish in error state */
1960static void smc_listen_out_err(struct smc_sock *new_smc)
1961{
1962	struct sock *newsmcsk = &new_smc->sk;
1963	struct net *net = sock_net(newsmcsk);
1964
1965	this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1966	if (newsmcsk->sk_state == SMC_INIT)
1967		sock_put(&new_smc->sk); /* passive closing */
1968	newsmcsk->sk_state = SMC_CLOSED;
1969
1970	smc_listen_out(new_smc);
1971}
1972
1973/* listen worker: decline and fall back if possible */
1974static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1975			       int local_first, u8 version)
1976{
1977	/* RDMA setup failed, switch back to TCP */
1978	smc_conn_abort(new_smc, local_first);
1979	if (reason_code < 0 ||
1980	    smc_switch_to_fallback(new_smc, reason_code)) {
1981		/* error, no fallback possible */
1982		smc_listen_out_err(new_smc);
1983		return;
1984	}
1985	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1986		if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1987			smc_listen_out_err(new_smc);
1988			return;
1989		}
1990	}
1991	smc_listen_out_connected(new_smc);
1992}
1993
1994/* listen worker: version checking */
1995static int smc_listen_v2_check(struct smc_sock *new_smc,
1996			       struct smc_clc_msg_proposal *pclc,
1997			       struct smc_init_info *ini)
1998{
1999	struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
2000	struct smc_clc_v2_extension *pclc_v2_ext;
2001	int rc = SMC_CLC_DECL_PEERNOSMC;
2002
2003	ini->smc_type_v1 = pclc->hdr.typev1;
2004	ini->smc_type_v2 = pclc->hdr.typev2;
2005	ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
2006	ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
2007	if (pclc->hdr.version > SMC_V1) {
2008		if (smcd_indicated(ini->smc_type_v2))
2009			ini->smcd_version |= SMC_V2;
2010		if (smcr_indicated(ini->smc_type_v2))
2011			ini->smcr_version |= SMC_V2;
2012	}
2013	if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
2014		rc = SMC_CLC_DECL_PEERNOSMC;
2015		goto out;
2016	}
2017	pclc_v2_ext = smc_get_clc_v2_ext(pclc);
2018	if (!pclc_v2_ext) {
2019		ini->smcd_version &= ~SMC_V2;
2020		ini->smcr_version &= ~SMC_V2;
2021		rc = SMC_CLC_DECL_NOV2EXT;
2022		goto out;
2023	}
2024	pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
2025	if (ini->smcd_version & SMC_V2) {
2026		if (!smc_ism_is_v2_capable()) {
2027			ini->smcd_version &= ~SMC_V2;
2028			rc = SMC_CLC_DECL_NOISM2SUPP;
2029		} else if (!pclc_smcd_v2_ext) {
2030			ini->smcd_version &= ~SMC_V2;
2031			rc = SMC_CLC_DECL_NOV2DEXT;
2032		} else if (!pclc_v2_ext->hdr.eid_cnt &&
2033			   !pclc_v2_ext->hdr.flag.seid) {
2034			ini->smcd_version &= ~SMC_V2;
2035			rc = SMC_CLC_DECL_NOUEID;
2036		}
2037	}
2038	if (ini->smcr_version & SMC_V2) {
2039		if (!pclc_v2_ext->hdr.eid_cnt) {
2040			ini->smcr_version &= ~SMC_V2;
2041			rc = SMC_CLC_DECL_NOUEID;
2042		}
2043	}
2044
2045	ini->release_nr = pclc_v2_ext->hdr.flag.release;
2046	if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE)
2047		ini->release_nr = SMC_RELEASE;
2048
2049out:
2050	if (!ini->smcd_version && !ini->smcr_version)
2051		return rc;
2052
2053	return 0;
2054}
2055
2056/* listen worker: check prefixes */
2057static int smc_listen_prfx_check(struct smc_sock *new_smc,
2058				 struct smc_clc_msg_proposal *pclc)
2059{
2060	struct smc_clc_msg_proposal_prefix *pclc_prfx;
2061	struct socket *newclcsock = new_smc->clcsock;
2062
2063	if (pclc->hdr.typev1 == SMC_TYPE_N)
2064		return 0;
2065	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2066	if (!pclc_prfx)
2067		return -EPROTO;
2068	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2069		return SMC_CLC_DECL_DIFFPREFIX;
2070
2071	return 0;
2072}
2073
2074/* listen worker: initialize connection and buffers */
2075static int smc_listen_rdma_init(struct smc_sock *new_smc,
2076				struct smc_init_info *ini)
2077{
2078	int rc;
2079
2080	/* allocate connection / link group */
2081	rc = smc_conn_create(new_smc, ini);
2082	if (rc)
2083		return rc;
2084
2085	/* create send buffer and rmb */
2086	if (smc_buf_create(new_smc, false)) {
2087		smc_conn_abort(new_smc, ini->first_contact_local);
2088		return SMC_CLC_DECL_MEM;
2089	}
2090
2091	return 0;
2092}
2093
2094/* listen worker: initialize connection and buffers for SMC-D */
2095static int smc_listen_ism_init(struct smc_sock *new_smc,
2096			       struct smc_init_info *ini)
2097{
2098	int rc;
2099
2100	rc = smc_conn_create(new_smc, ini);
2101	if (rc)
2102		return rc;
2103
2104	/* Create send and receive buffers */
2105	rc = smc_buf_create(new_smc, true);
2106	if (rc) {
2107		smc_conn_abort(new_smc, ini->first_contact_local);
2108		return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2109					 SMC_CLC_DECL_MEM;
2110	}
2111
2112	return 0;
2113}
2114
2115static bool smc_is_already_selected(struct smcd_dev *smcd,
2116				    struct smc_init_info *ini,
2117				    int matches)
2118{
2119	int i;
2120
2121	for (i = 0; i < matches; i++)
2122		if (smcd == ini->ism_dev[i])
2123			return true;
2124
2125	return false;
2126}
2127
2128/* check for ISM devices matching proposed ISM devices */
2129static void smc_check_ism_v2_match(struct smc_init_info *ini,
2130				   u16 proposed_chid,
2131				   struct smcd_gid *proposed_gid,
2132				   unsigned int *matches)
2133{
2134	struct smcd_dev *smcd;
2135
2136	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2137		if (smcd->going_away)
2138			continue;
2139		if (smc_is_already_selected(smcd, ini, *matches))
2140			continue;
2141		if (smc_ism_get_chid(smcd) == proposed_chid &&
2142		    !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2143			ini->ism_peer_gid[*matches].gid = proposed_gid->gid;
2144			if (__smc_ism_is_emulated(proposed_chid))
2145				ini->ism_peer_gid[*matches].gid_ext =
2146							proposed_gid->gid_ext;
2147				/* non-Emulated-ISM's peer gid_ext remains 0. */
2148			ini->ism_dev[*matches] = smcd;
2149			(*matches)++;
2150			break;
2151		}
2152	}
2153}
2154
2155static void smc_init_info_store_rc(u32 rc, struct smc_init_info *ini)
2156{
2157	if (!ini->rc)
2158		ini->rc = rc;
2159}
2160
2161static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc,
2162					struct smc_clc_msg_proposal *pclc,
2163					struct smc_init_info *ini)
2164{
2165	struct smc_clc_smcd_v2_extension *smcd_v2_ext;
2166	struct smc_clc_v2_extension *smc_v2_ext;
2167	struct smc_clc_msg_smcd *pclc_smcd;
2168	unsigned int matches = 0;
2169	struct smcd_gid smcd_gid;
2170	u8 smcd_version;
2171	u8 *eid = NULL;
2172	int i, rc;
2173	u16 chid;
2174
2175	if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2176		goto not_found;
2177
2178	pclc_smcd = smc_get_clc_msg_smcd(pclc);
2179	smc_v2_ext = smc_get_clc_v2_ext(pclc);
2180	smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2181	if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext)
2182		goto not_found;
2183
2184	mutex_lock(&smcd_dev_list.mutex);
2185	if (pclc_smcd->ism.chid) {
2186		/* check for ISM device matching proposed native ISM device */
2187		smcd_gid.gid = ntohll(pclc_smcd->ism.gid);
2188		smcd_gid.gid_ext = 0;
2189		smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2190				       &smcd_gid, &matches);
2191	}
2192	for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) {
2193		/* check for ISM devices matching proposed non-native ISM
2194		 * devices
2195		 */
2196		smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid);
2197		smcd_gid.gid_ext = 0;
2198		chid = ntohs(smcd_v2_ext->gidchid[i].chid);
2199		if (__smc_ism_is_emulated(chid)) {
2200			if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt ||
2201			    chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid))
2202				/* each Emulated-ISM device takes two GID-CHID
2203				 * entries and CHID of the second entry repeats
2204				 * that of the first entry.
2205				 *
2206				 * So check if the next GID-CHID entry exists
2207				 * and both two entries' CHIDs are the same.
2208				 */
2209				continue;
2210			smcd_gid.gid_ext =
2211				ntohll(smcd_v2_ext->gidchid[++i].gid);
2212		}
2213		smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches);
2214	}
2215	mutex_unlock(&smcd_dev_list.mutex);
2216
2217	if (!ini->ism_dev[0]) {
2218		smc_init_info_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2219		goto not_found;
2220	}
2221
2222	smc_ism_get_system_eid(&eid);
2223	if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2224			       smcd_v2_ext->system_eid, eid))
2225		goto not_found;
2226
2227	/* separate - outside the smcd_dev_list.lock */
2228	smcd_version = ini->smcd_version;
2229	for (i = 0; i < matches; i++) {
2230		ini->smcd_version = SMC_V2;
2231		ini->is_smcd = true;
2232		ini->ism_selected = i;
2233		rc = smc_listen_ism_init(new_smc, ini);
2234		if (rc) {
2235			smc_init_info_store_rc(rc, ini);
2236			/* try next active ISM device */
2237			continue;
2238		}
2239		return; /* matching and usable V2 ISM device found */
2240	}
2241	/* no V2 ISM device could be initialized */
2242	ini->smcd_version = smcd_version;	/* restore original value */
2243	ini->negotiated_eid[0] = 0;
2244
2245not_found:
2246	ini->smcd_version &= ~SMC_V2;
2247	ini->ism_dev[0] = NULL;
2248	ini->is_smcd = false;
2249}
2250
2251static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2252					struct smc_clc_msg_proposal *pclc,
2253					struct smc_init_info *ini)
2254{
2255	struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2256	int rc = 0;
2257
2258	/* check if ISM V1 is available */
2259	if (!(ini->smcd_version & SMC_V1) ||
2260	    !smcd_indicated(ini->smc_type_v1) ||
2261	    !pclc_smcd)
2262		goto not_found;
2263	ini->is_smcd = true; /* prepare ISM check */
2264	ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid);
2265	ini->ism_peer_gid[0].gid_ext = 0;
2266	rc = smc_find_ism_device(new_smc, ini);
2267	if (rc)
2268		goto not_found;
2269	ini->ism_selected = 0;
2270	rc = smc_listen_ism_init(new_smc, ini);
2271	if (!rc)
2272		return;		/* V1 ISM device found */
2273
2274not_found:
2275	smc_init_info_store_rc(rc, ini);
2276	ini->smcd_version &= ~SMC_V1;
2277	ini->ism_dev[0] = NULL;
2278	ini->is_smcd = false;
2279}
2280
2281/* listen worker: register buffers */
2282static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2283{
2284	struct smc_connection *conn = &new_smc->conn;
2285
2286	if (!local_first) {
2287		/* reg sendbufs if they were vzalloced */
2288		if (conn->sndbuf_desc->is_vm) {
2289			if (smcr_lgr_reg_sndbufs(conn->lnk,
2290						 conn->sndbuf_desc))
2291				return SMC_CLC_DECL_ERR_REGBUF;
2292		}
2293		if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2294			return SMC_CLC_DECL_ERR_REGBUF;
2295	}
2296
2297	return 0;
2298}
2299
2300static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2301					 struct smc_clc_msg_proposal *pclc,
2302					 struct smc_init_info *ini)
2303{
2304	struct smc_clc_v2_extension *smc_v2_ext;
2305	u8 smcr_version;
2306	int rc;
2307
2308	if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2309		goto not_found;
2310
2311	smc_v2_ext = smc_get_clc_v2_ext(pclc);
2312	if (!smc_v2_ext ||
2313	    !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2314		goto not_found;
2315
2316	/* prepare RDMA check */
2317	memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2318	memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2319	memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2320	ini->check_smcrv2 = true;
2321	ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2322	ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2323	ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2324	rc = smc_find_rdma_device(new_smc, ini);
2325	if (rc) {
2326		smc_init_info_store_rc(rc, ini);
2327		goto not_found;
2328	}
2329	if (!ini->smcrv2.uses_gateway)
2330		memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2331
2332	smcr_version = ini->smcr_version;
2333	ini->smcr_version = SMC_V2;
2334	rc = smc_listen_rdma_init(new_smc, ini);
2335	if (!rc) {
2336		rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2337		if (rc)
2338			smc_conn_abort(new_smc, ini->first_contact_local);
2339	}
2340	if (!rc)
2341		return;
2342	ini->smcr_version = smcr_version;
2343	smc_init_info_store_rc(rc, ini);
2344
2345not_found:
2346	ini->smcr_version &= ~SMC_V2;
2347	ini->smcrv2.ib_dev_v2 = NULL;
2348	ini->check_smcrv2 = false;
2349}
2350
2351static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2352					struct smc_clc_msg_proposal *pclc,
2353					struct smc_init_info *ini)
2354{
2355	int rc;
2356
2357	if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2358		return SMC_CLC_DECL_NOSMCDEV;
2359
2360	/* prepare RDMA check */
2361	memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2362	memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2363	memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2364	rc = smc_find_rdma_device(new_smc, ini);
2365	if (rc) {
2366		/* no RDMA device found */
2367		return SMC_CLC_DECL_NOSMCDEV;
2368	}
2369	rc = smc_listen_rdma_init(new_smc, ini);
2370	if (rc)
2371		return rc;
2372	return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2373}
2374
2375/* determine the local device matching to proposal */
2376static int smc_listen_find_device(struct smc_sock *new_smc,
2377				  struct smc_clc_msg_proposal *pclc,
2378				  struct smc_init_info *ini)
2379{
2380	int prfx_rc;
2381
2382	/* check for ISM device matching V2 proposed device */
2383	smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2384	if (ini->ism_dev[0])
2385		return 0;
2386
2387	/* check for matching IP prefix and subnet length (V1) */
2388	prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2389	if (prfx_rc)
2390		smc_init_info_store_rc(prfx_rc, ini);
2391
2392	/* get vlan id from IP device */
2393	if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2394		return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2395
2396	/* check for ISM device matching V1 proposed device */
2397	if (!prfx_rc)
2398		smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2399	if (ini->ism_dev[0])
2400		return 0;
2401
2402	if (!smcr_indicated(pclc->hdr.typev1) &&
2403	    !smcr_indicated(pclc->hdr.typev2))
2404		/* skip RDMA and decline */
2405		return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2406
2407	/* check if RDMA V2 is available */
2408	smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2409	if (ini->smcrv2.ib_dev_v2)
2410		return 0;
2411
2412	/* check if RDMA V1 is available */
2413	if (!prfx_rc) {
2414		int rc;
2415
2416		rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2417		smc_init_info_store_rc(rc, ini);
2418		return (!rc) ? 0 : ini->rc;
2419	}
2420	return prfx_rc;
2421}
2422
2423/* listen worker: finish RDMA setup */
2424static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2425				  struct smc_clc_msg_accept_confirm *cclc,
2426				  bool local_first,
2427				  struct smc_init_info *ini)
2428{
2429	struct smc_link *link = new_smc->conn.lnk;
2430	int reason_code = 0;
2431
2432	if (local_first)
2433		smc_link_save_peer_info(link, cclc, ini);
2434
2435	if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2436		return SMC_CLC_DECL_ERR_RTOK;
2437
2438	if (local_first) {
2439		if (smc_ib_ready_link(link))
2440			return SMC_CLC_DECL_ERR_RDYLNK;
2441		/* QP confirmation over RoCE fabric */
2442		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2443		reason_code = smcr_serv_conf_first_link(new_smc);
2444		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2445	}
2446	return reason_code;
2447}
2448
2449/* setup for connection of server */
2450static void smc_listen_work(struct work_struct *work)
2451{
2452	struct smc_sock *new_smc = container_of(work, struct smc_sock,
2453						smc_listen_work);
2454	struct socket *newclcsock = new_smc->clcsock;
2455	struct smc_clc_msg_accept_confirm *cclc;
2456	struct smc_clc_msg_proposal_area *buf;
2457	struct smc_clc_msg_proposal *pclc;
2458	struct smc_init_info *ini = NULL;
2459	u8 proposal_version = SMC_V1;
2460	u8 accept_version;
2461	int rc = 0;
2462
2463	lock_sock(&new_smc->sk); /* release in smc_listen_out() */
2464	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2465		return smc_listen_out_err(new_smc);
2466
2467	if (new_smc->use_fallback) {
2468		smc_listen_out_connected(new_smc);
2469		return;
2470	}
2471
2472	/* check if peer is smc capable */
2473	if (!tcp_sk(newclcsock->sk)->syn_smc) {
2474		rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2475		if (rc)
2476			smc_listen_out_err(new_smc);
2477		else
2478			smc_listen_out_connected(new_smc);
2479		return;
2480	}
2481
2482	/* do inband token exchange -
2483	 * wait for and receive SMC Proposal CLC message
2484	 */
2485	buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2486	if (!buf) {
2487		rc = SMC_CLC_DECL_MEM;
2488		goto out_decl;
2489	}
2490	pclc = (struct smc_clc_msg_proposal *)buf;
2491	rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2492			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2493	if (rc)
2494		goto out_decl;
2495
2496	if (pclc->hdr.version > SMC_V1)
2497		proposal_version = SMC_V2;
2498
2499	/* IPSec connections opt out of SMC optimizations */
2500	if (using_ipsec(new_smc)) {
2501		rc = SMC_CLC_DECL_IPSEC;
2502		goto out_decl;
2503	}
2504
2505	ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2506	if (!ini) {
2507		rc = SMC_CLC_DECL_MEM;
2508		goto out_decl;
2509	}
2510
2511	/* initial version checking */
2512	rc = smc_listen_v2_check(new_smc, pclc, ini);
2513	if (rc)
2514		goto out_decl;
2515
2516	rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini);
2517	if (rc)
2518		goto out_decl;
2519
2520	mutex_lock(&smc_server_lgr_pending);
2521	smc_rx_init(new_smc);
2522	smc_tx_init(new_smc);
2523
2524	/* determine ISM or RoCE device used for connection */
2525	rc = smc_listen_find_device(new_smc, pclc, ini);
2526	if (rc)
2527		goto out_unlock;
2528
2529	/* send SMC Accept CLC message */
2530	accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2531	rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2532				 accept_version, ini->negotiated_eid, ini);
2533	if (rc)
2534		goto out_unlock;
2535
2536	/* SMC-D does not need this lock any more */
2537	if (ini->is_smcd)
2538		mutex_unlock(&smc_server_lgr_pending);
2539
2540	/* receive SMC Confirm CLC message */
2541	memset(buf, 0, sizeof(*buf));
2542	cclc = (struct smc_clc_msg_accept_confirm *)buf;
2543	rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2544			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2545	if (rc) {
2546		if (!ini->is_smcd)
2547			goto out_unlock;
2548		goto out_decl;
2549	}
2550
2551	rc = smc_clc_v2x_features_confirm_check(cclc, ini);
2552	if (rc) {
2553		if (!ini->is_smcd)
2554			goto out_unlock;
2555		goto out_decl;
2556	}
2557
2558	/* fce smc release version is needed in smc_listen_rdma_finish,
2559	 * so save fce info here.
2560	 */
2561	smc_conn_save_peer_info_fce(new_smc, cclc);
2562
2563	/* finish worker */
2564	if (!ini->is_smcd) {
2565		rc = smc_listen_rdma_finish(new_smc, cclc,
2566					    ini->first_contact_local, ini);
2567		if (rc)
2568			goto out_unlock;
2569		mutex_unlock(&smc_server_lgr_pending);
2570	}
2571	smc_conn_save_peer_info(new_smc, cclc);
2572
2573	if (ini->is_smcd &&
2574	    smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) {
2575		rc = smcd_buf_attach(new_smc);
2576		if (rc)
2577			goto out_decl;
2578	}
2579
2580	SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2581	/* smc_listen_out() will release smcsk */
2582	smc_listen_out_connected(new_smc);
2583	goto out_free;
2584
2585out_unlock:
2586	mutex_unlock(&smc_server_lgr_pending);
2587out_decl:
2588	smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2589			   proposal_version);
2590out_free:
2591	kfree(ini);
2592	kfree(buf);
2593}
2594
2595static void smc_tcp_listen_work(struct work_struct *work)
2596{
2597	struct smc_sock *lsmc = container_of(work, struct smc_sock,
2598					     tcp_listen_work);
2599	struct sock *lsk = &lsmc->sk;
2600	struct smc_sock *new_smc;
2601	int rc = 0;
2602
2603	lock_sock(lsk);
2604	while (lsk->sk_state == SMC_LISTEN) {
2605		rc = smc_clcsock_accept(lsmc, &new_smc);
2606		if (rc) /* clcsock accept queue empty or error */
2607			goto out;
2608		if (!new_smc)
2609			continue;
2610
2611		if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2612			atomic_inc(&lsmc->queued_smc_hs);
2613
2614		new_smc->listen_smc = lsmc;
2615		new_smc->use_fallback = lsmc->use_fallback;
2616		new_smc->fallback_rsn = lsmc->fallback_rsn;
2617		sock_hold(lsk); /* sock_put in smc_listen_work */
2618		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2619		smc_copy_sock_settings_to_smc(new_smc);
2620		sock_hold(&new_smc->sk); /* sock_put in passive closing */
2621		if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2622			sock_put(&new_smc->sk);
2623	}
2624
2625out:
2626	release_sock(lsk);
2627	sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2628}
2629
2630static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2631{
2632	struct smc_sock *lsmc;
2633
2634	read_lock_bh(&listen_clcsock->sk_callback_lock);
2635	lsmc = smc_clcsock_user_data(listen_clcsock);
2636	if (!lsmc)
2637		goto out;
2638	lsmc->clcsk_data_ready(listen_clcsock);
2639	if (lsmc->sk.sk_state == SMC_LISTEN) {
2640		sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2641		if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2642			sock_put(&lsmc->sk);
2643	}
2644out:
2645	read_unlock_bh(&listen_clcsock->sk_callback_lock);
2646}
2647
2648int smc_listen(struct socket *sock, int backlog)
2649{
2650	struct sock *sk = sock->sk;
2651	struct smc_sock *smc;
2652	int rc;
2653
2654	smc = smc_sk(sk);
2655	lock_sock(sk);
2656
2657	rc = -EINVAL;
2658	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2659	    smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2660		goto out;
2661
2662	rc = 0;
2663	if (sk->sk_state == SMC_LISTEN) {
2664		sk->sk_max_ack_backlog = backlog;
2665		goto out;
2666	}
2667	/* some socket options are handled in core, so we could not apply
2668	 * them to the clc socket -- copy smc socket options to clc socket
2669	 */
2670	smc_copy_sock_settings_to_clc(smc);
2671	if (!smc->use_fallback)
2672		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2673
2674	/* save original sk_data_ready function and establish
2675	 * smc-specific sk_data_ready function
2676	 */
2677	write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2678	__rcu_assign_sk_user_data_with_flags(smc->clcsock->sk, smc,
2679					     SK_USER_DATA_NOCOPY);
2680	smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2681			       smc_clcsock_data_ready, &smc->clcsk_data_ready);
2682	write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2683
2684	/* save original ops */
2685	smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2686
2687	smc->af_ops = *smc->ori_af_ops;
2688	smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2689
2690	inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2691
2692	if (smc->limit_smc_hs)
2693		tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2694
2695	rc = kernel_listen(smc->clcsock, backlog);
2696	if (rc) {
2697		write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2698		smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2699				       &smc->clcsk_data_ready);
2700		rcu_assign_sk_user_data(smc->clcsock->sk, NULL);
2701		write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2702		goto out;
2703	}
2704	sock_set_flag(sk, SOCK_RCU_FREE);
2705	sk->sk_max_ack_backlog = backlog;
2706	sk->sk_ack_backlog = 0;
2707	sk->sk_state = SMC_LISTEN;
2708
2709out:
2710	release_sock(sk);
2711	return rc;
2712}
2713
2714int smc_accept(struct socket *sock, struct socket *new_sock,
2715	       struct proto_accept_arg *arg)
2716{
2717	struct sock *sk = sock->sk, *nsk;
2718	DECLARE_WAITQUEUE(wait, current);
2719	struct smc_sock *lsmc;
2720	long timeo;
2721	int rc = 0;
2722
2723	lsmc = smc_sk(sk);
2724	sock_hold(sk); /* sock_put below */
2725	lock_sock(sk);
2726
2727	if (lsmc->sk.sk_state != SMC_LISTEN) {
2728		rc = -EINVAL;
2729		release_sock(sk);
2730		goto out;
2731	}
2732
2733	/* Wait for an incoming connection */
2734	timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
2735	add_wait_queue_exclusive(sk_sleep(sk), &wait);
2736	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2737		set_current_state(TASK_INTERRUPTIBLE);
2738		if (!timeo) {
2739			rc = -EAGAIN;
2740			break;
2741		}
2742		release_sock(sk);
2743		timeo = schedule_timeout(timeo);
2744		/* wakeup by sk_data_ready in smc_listen_work() */
2745		sched_annotate_sleep();
2746		lock_sock(sk);
2747		if (signal_pending(current)) {
2748			rc = sock_intr_errno(timeo);
2749			break;
2750		}
2751	}
2752	set_current_state(TASK_RUNNING);
2753	remove_wait_queue(sk_sleep(sk), &wait);
2754
2755	if (!rc)
2756		rc = sock_error(nsk);
2757	release_sock(sk);
2758	if (rc)
2759		goto out;
2760
2761	if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) {
2762		/* wait till data arrives on the socket */
2763		timeo = secs_to_jiffies(lsmc->sockopt_defer_accept);
2764		if (smc_sk(nsk)->use_fallback) {
2765			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2766
2767			lock_sock(clcsk);
2768			if (skb_queue_empty(&clcsk->sk_receive_queue))
2769				sk_wait_data(clcsk, &timeo, NULL);
2770			release_sock(clcsk);
2771		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2772			lock_sock(nsk);
2773			smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available);
2774			release_sock(nsk);
2775		}
2776	}
2777
2778out:
2779	sock_put(sk); /* sock_hold above */
2780	return rc;
2781}
2782
2783int smc_getname(struct socket *sock, struct sockaddr *addr,
2784		int peer)
2785{
2786	struct smc_sock *smc;
2787
2788	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2789	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2790		return -ENOTCONN;
2791
2792	smc = smc_sk(sock->sk);
2793
2794	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2795}
2796
2797int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2798{
2799	struct sock *sk = sock->sk;
2800	struct smc_sock *smc;
2801	int rc;
2802
2803	smc = smc_sk(sk);
2804	lock_sock(sk);
2805
2806	/* SMC does not support connect with fastopen */
2807	if (msg->msg_flags & MSG_FASTOPEN) {
2808		/* not connected yet, fallback */
2809		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2810			rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2811			if (rc)
2812				goto out;
2813		} else {
2814			rc = -EINVAL;
2815			goto out;
2816		}
2817	} else if ((sk->sk_state != SMC_ACTIVE) &&
2818		   (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2819		   (sk->sk_state != SMC_INIT)) {
2820		rc = -EPIPE;
2821		goto out;
2822	}
2823
2824	if (smc->use_fallback) {
2825		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2826	} else {
2827		rc = smc_tx_sendmsg(smc, msg, len);
2828		SMC_STAT_TX_PAYLOAD(smc, len, rc);
2829	}
2830out:
2831	release_sock(sk);
2832	return rc;
2833}
2834
2835int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2836		int flags)
2837{
2838	struct sock *sk = sock->sk;
2839	struct smc_sock *smc;
2840	int rc = -ENOTCONN;
2841
2842	smc = smc_sk(sk);
2843	lock_sock(sk);
2844	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2845		/* socket was connected before, no more data to read */
2846		rc = 0;
2847		goto out;
2848	}
2849	if ((sk->sk_state == SMC_INIT) ||
2850	    (sk->sk_state == SMC_LISTEN) ||
2851	    (sk->sk_state == SMC_CLOSED))
2852		goto out;
2853
2854	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2855		rc = 0;
2856		goto out;
2857	}
2858
2859	if (smc->use_fallback) {
2860		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2861	} else {
2862		msg->msg_namelen = 0;
2863		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2864		SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2865	}
2866
2867out:
2868	release_sock(sk);
2869	return rc;
2870}
2871
2872static __poll_t smc_accept_poll(struct sock *parent)
2873{
2874	struct smc_sock *isk = smc_sk(parent);
2875	__poll_t mask = 0;
2876
2877	spin_lock(&isk->accept_q_lock);
2878	if (!list_empty(&isk->accept_q))
2879		mask = EPOLLIN | EPOLLRDNORM;
2880	spin_unlock(&isk->accept_q_lock);
2881
2882	return mask;
2883}
2884
2885__poll_t smc_poll(struct file *file, struct socket *sock,
2886		  poll_table *wait)
2887{
2888	struct sock *sk = sock->sk;
2889	struct smc_sock *smc;
2890	__poll_t mask = 0;
2891
2892	if (!sk)
2893		return EPOLLNVAL;
2894
2895	smc = smc_sk(sock->sk);
2896	if (smc->use_fallback) {
2897		/* delegate to CLC child sock */
2898		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2899		sk->sk_err = smc->clcsock->sk->sk_err;
2900	} else {
2901		if (sk->sk_state != SMC_CLOSED)
2902			sock_poll_wait(file, sock, wait);
2903		if (sk->sk_err)
2904			mask |= EPOLLERR;
2905		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2906		    (sk->sk_state == SMC_CLOSED))
2907			mask |= EPOLLHUP;
2908		if (sk->sk_state == SMC_LISTEN) {
2909			/* woken up by sk_data_ready in smc_listen_work() */
2910			mask |= smc_accept_poll(sk);
2911		} else if (smc->use_fallback) { /* as result of connect_work()*/
2912			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2913							   wait);
2914			sk->sk_err = smc->clcsock->sk->sk_err;
2915		} else {
2916			if ((sk->sk_state != SMC_INIT &&
2917			     atomic_read(&smc->conn.sndbuf_space)) ||
2918			    sk->sk_shutdown & SEND_SHUTDOWN) {
2919				mask |= EPOLLOUT | EPOLLWRNORM;
2920			} else {
2921				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2922				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2923
2924				if (sk->sk_state != SMC_INIT) {
2925					/* Race breaker the same way as tcp_poll(). */
2926					smp_mb__after_atomic();
2927					if (atomic_read(&smc->conn.sndbuf_space))
2928						mask |= EPOLLOUT | EPOLLWRNORM;
2929				}
2930			}
2931			if (atomic_read(&smc->conn.bytes_to_rcv))
2932				mask |= EPOLLIN | EPOLLRDNORM;
2933			if (sk->sk_shutdown & RCV_SHUTDOWN)
2934				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2935			if (sk->sk_state == SMC_APPCLOSEWAIT1)
2936				mask |= EPOLLIN;
2937			if (smc->conn.urg_state == SMC_URG_VALID)
2938				mask |= EPOLLPRI;
2939		}
2940	}
2941
2942	return mask;
2943}
2944
2945int smc_shutdown(struct socket *sock, int how)
2946{
2947	struct sock *sk = sock->sk;
2948	bool do_shutdown = true;
2949	struct smc_sock *smc;
2950	int rc = -EINVAL;
2951	int old_state;
2952	int rc1 = 0;
2953
2954	smc = smc_sk(sk);
2955
2956	if ((how < SHUT_RD) || (how > SHUT_RDWR))
2957		return rc;
2958
2959	lock_sock(sk);
2960
2961	if (sock->state == SS_CONNECTING) {
2962		if (sk->sk_state == SMC_ACTIVE)
2963			sock->state = SS_CONNECTED;
2964		else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2965			 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2966			 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2967			 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2968			 sk->sk_state == SMC_APPFINCLOSEWAIT)
2969			sock->state = SS_DISCONNECTING;
2970	}
2971
2972	rc = -ENOTCONN;
2973	if ((sk->sk_state != SMC_ACTIVE) &&
2974	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2975	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2976	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2977	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2978	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
2979		goto out;
2980	if (smc->use_fallback) {
2981		rc = kernel_sock_shutdown(smc->clcsock, how);
2982		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2983		if (sk->sk_shutdown == SHUTDOWN_MASK) {
2984			sk->sk_state = SMC_CLOSED;
2985			sk->sk_socket->state = SS_UNCONNECTED;
2986			sock_put(sk);
2987		}
2988		goto out;
2989	}
2990	switch (how) {
2991	case SHUT_RDWR:		/* shutdown in both directions */
2992		old_state = sk->sk_state;
2993		rc = smc_close_active(smc);
2994		if (old_state == SMC_ACTIVE &&
2995		    sk->sk_state == SMC_PEERCLOSEWAIT1)
2996			do_shutdown = false;
2997		break;
2998	case SHUT_WR:
2999		rc = smc_close_shutdown_write(smc);
3000		break;
3001	case SHUT_RD:
3002		rc = 0;
3003		/* nothing more to do because peer is not involved */
3004		break;
3005	}
3006	if (do_shutdown && smc->clcsock)
3007		rc1 = kernel_sock_shutdown(smc->clcsock, how);
3008	/* map sock_shutdown_cmd constants to sk_shutdown value range */
3009	sk->sk_shutdown |= how + 1;
3010
3011	if (sk->sk_state == SMC_CLOSED)
3012		sock->state = SS_UNCONNECTED;
3013	else
3014		sock->state = SS_DISCONNECTING;
3015out:
3016	release_sock(sk);
3017	return rc ? rc : rc1;
3018}
3019
3020static int __smc_getsockopt(struct socket *sock, int level, int optname,
3021			    char __user *optval, int __user *optlen)
3022{
3023	struct smc_sock *smc;
3024	int val, len;
3025
3026	smc = smc_sk(sock->sk);
3027
3028	if (get_user(len, optlen))
3029		return -EFAULT;
3030
3031	len = min_t(int, len, sizeof(int));
3032
3033	if (len < 0)
3034		return -EINVAL;
3035
3036	switch (optname) {
3037	case SMC_LIMIT_HS:
3038		val = smc->limit_smc_hs;
3039		break;
3040	default:
3041		return -EOPNOTSUPP;
3042	}
3043
3044	if (put_user(len, optlen))
3045		return -EFAULT;
3046	if (copy_to_user(optval, &val, len))
3047		return -EFAULT;
3048
3049	return 0;
3050}
3051
3052static int __smc_setsockopt(struct socket *sock, int level, int optname,
3053			    sockptr_t optval, unsigned int optlen)
3054{
3055	struct sock *sk = sock->sk;
3056	struct smc_sock *smc;
3057	int val, rc;
3058
3059	smc = smc_sk(sk);
3060
3061	lock_sock(sk);
3062	switch (optname) {
3063	case SMC_LIMIT_HS:
3064		if (optlen < sizeof(int)) {
3065			rc = -EINVAL;
3066			break;
3067		}
3068		if (copy_from_sockptr(&val, optval, sizeof(int))) {
3069			rc = -EFAULT;
3070			break;
3071		}
3072
3073		smc->limit_smc_hs = !!val;
3074		rc = 0;
3075		break;
3076	default:
3077		rc = -EOPNOTSUPP;
3078		break;
3079	}
3080	release_sock(sk);
3081
3082	return rc;
3083}
3084
3085int smc_setsockopt(struct socket *sock, int level, int optname,
3086		   sockptr_t optval, unsigned int optlen)
3087{
3088	struct sock *sk = sock->sk;
3089	struct smc_sock *smc;
3090	int val, rc;
3091
3092	if (level == SOL_TCP && optname == TCP_ULP)
3093		return -EOPNOTSUPP;
3094	else if (level == SOL_SMC)
3095		return __smc_setsockopt(sock, level, optname, optval, optlen);
3096
3097	smc = smc_sk(sk);
3098
3099	/* generic setsockopts reaching us here always apply to the
3100	 * CLC socket
3101	 */
3102	mutex_lock(&smc->clcsock_release_lock);
3103	if (!smc->clcsock) {
3104		mutex_unlock(&smc->clcsock_release_lock);
3105		return -EBADF;
3106	}
3107	if (unlikely(!smc->clcsock->ops->setsockopt))
3108		rc = -EOPNOTSUPP;
3109	else
3110		rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
3111						   optval, optlen);
3112	if (smc->clcsock->sk->sk_err) {
3113		sk->sk_err = smc->clcsock->sk->sk_err;
3114		sk_error_report(sk);
3115	}
3116	mutex_unlock(&smc->clcsock_release_lock);
3117
3118	if (optlen < sizeof(int))
3119		return -EINVAL;
3120	if (copy_from_sockptr(&val, optval, sizeof(int)))
3121		return -EFAULT;
3122
3123	lock_sock(sk);
3124	if (rc || smc->use_fallback)
3125		goto out;
3126	switch (optname) {
3127	case TCP_FASTOPEN:
3128	case TCP_FASTOPEN_CONNECT:
3129	case TCP_FASTOPEN_KEY:
3130	case TCP_FASTOPEN_NO_COOKIE:
3131		/* option not supported by SMC */
3132		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3133			rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3134		} else {
3135			rc = -EINVAL;
3136		}
3137		break;
3138	case TCP_NODELAY:
3139		if (sk->sk_state != SMC_INIT &&
3140		    sk->sk_state != SMC_LISTEN &&
3141		    sk->sk_state != SMC_CLOSED) {
3142			if (val) {
3143				SMC_STAT_INC(smc, ndly_cnt);
3144				smc_tx_pending(&smc->conn);
3145				cancel_delayed_work(&smc->conn.tx_work);
3146			}
3147		}
3148		break;
3149	case TCP_CORK:
3150		if (sk->sk_state != SMC_INIT &&
3151		    sk->sk_state != SMC_LISTEN &&
3152		    sk->sk_state != SMC_CLOSED) {
3153			if (!val) {
3154				SMC_STAT_INC(smc, cork_cnt);
3155				smc_tx_pending(&smc->conn);
3156				cancel_delayed_work(&smc->conn.tx_work);
3157			}
3158		}
3159		break;
3160	case TCP_DEFER_ACCEPT:
3161		smc->sockopt_defer_accept = val;
3162		break;
3163	default:
3164		break;
3165	}
3166out:
3167	release_sock(sk);
3168
3169	return rc;
3170}
3171
3172int smc_getsockopt(struct socket *sock, int level, int optname,
3173		   char __user *optval, int __user *optlen)
3174{
3175	struct smc_sock *smc;
3176	int rc;
3177
3178	if (level == SOL_SMC)
3179		return __smc_getsockopt(sock, level, optname, optval, optlen);
3180
3181	smc = smc_sk(sock->sk);
3182	mutex_lock(&smc->clcsock_release_lock);
3183	if (!smc->clcsock) {
3184		mutex_unlock(&smc->clcsock_release_lock);
3185		return -EBADF;
3186	}
3187	/* socket options apply to the CLC socket */
3188	if (unlikely(!smc->clcsock->ops->getsockopt)) {
3189		mutex_unlock(&smc->clcsock_release_lock);
3190		return -EOPNOTSUPP;
3191	}
3192	rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3193					   optval, optlen);
3194	mutex_unlock(&smc->clcsock_release_lock);
3195	return rc;
3196}
3197
3198int smc_ioctl(struct socket *sock, unsigned int cmd,
3199	      unsigned long arg)
3200{
3201	union smc_host_cursor cons, urg;
3202	struct smc_connection *conn;
3203	struct smc_sock *smc;
3204	int answ;
3205
3206	smc = smc_sk(sock->sk);
3207	conn = &smc->conn;
3208	lock_sock(&smc->sk);
3209	if (smc->use_fallback) {
3210		if (!smc->clcsock) {
3211			release_sock(&smc->sk);
3212			return -EBADF;
3213		}
3214		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3215		release_sock(&smc->sk);
3216		return answ;
3217	}
3218	switch (cmd) {
3219	case SIOCINQ: /* same as FIONREAD */
3220		if (smc->sk.sk_state == SMC_LISTEN) {
3221			release_sock(&smc->sk);
3222			return -EINVAL;
3223		}
3224		if (smc->sk.sk_state == SMC_INIT ||
3225		    smc->sk.sk_state == SMC_CLOSED)
3226			answ = 0;
3227		else
3228			answ = atomic_read(&smc->conn.bytes_to_rcv);
3229		break;
3230	case SIOCOUTQ:
3231		/* output queue size (not send + not acked) */
3232		if (smc->sk.sk_state == SMC_LISTEN) {
3233			release_sock(&smc->sk);
3234			return -EINVAL;
3235		}
3236		if (smc->sk.sk_state == SMC_INIT ||
3237		    smc->sk.sk_state == SMC_CLOSED)
3238			answ = 0;
3239		else
3240			answ = smc->conn.sndbuf_desc->len -
3241					atomic_read(&smc->conn.sndbuf_space);
3242		break;
3243	case SIOCOUTQNSD:
3244		/* output queue size (not send only) */
3245		if (smc->sk.sk_state == SMC_LISTEN) {
3246			release_sock(&smc->sk);
3247			return -EINVAL;
3248		}
3249		if (smc->sk.sk_state == SMC_INIT ||
3250		    smc->sk.sk_state == SMC_CLOSED)
3251			answ = 0;
3252		else
3253			answ = smc_tx_prepared_sends(&smc->conn);
3254		break;
3255	case SIOCATMARK:
3256		if (smc->sk.sk_state == SMC_LISTEN) {
3257			release_sock(&smc->sk);
3258			return -EINVAL;
3259		}
3260		if (smc->sk.sk_state == SMC_INIT ||
3261		    smc->sk.sk_state == SMC_CLOSED) {
3262			answ = 0;
3263		} else {
3264			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3265			smc_curs_copy(&urg, &conn->urg_curs, conn);
3266			answ = smc_curs_diff(conn->rmb_desc->len,
3267					     &cons, &urg) == 1;
3268		}
3269		break;
3270	default:
3271		release_sock(&smc->sk);
3272		return -ENOIOCTLCMD;
3273	}
3274	release_sock(&smc->sk);
3275
3276	return put_user(answ, (int __user *)arg);
3277}
3278
3279/* Map the affected portions of the rmbe into an spd, note the number of bytes
3280 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3281 * updates till whenever a respective page has been fully processed.
3282 * Note that subsequent recv() calls have to wait till all splice() processing
3283 * completed.
3284 */
3285ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3286			struct pipe_inode_info *pipe, size_t len,
3287			unsigned int flags)
3288{
3289	struct sock *sk = sock->sk;
3290	struct smc_sock *smc;
3291	int rc = -ENOTCONN;
3292
3293	smc = smc_sk(sk);
3294	lock_sock(sk);
3295	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3296		/* socket was connected before, no more data to read */
3297		rc = 0;
3298		goto out;
3299	}
3300	if (sk->sk_state == SMC_INIT ||
3301	    sk->sk_state == SMC_LISTEN ||
3302	    sk->sk_state == SMC_CLOSED)
3303		goto out;
3304
3305	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3306		rc = 0;
3307		goto out;
3308	}
3309
3310	if (smc->use_fallback) {
3311		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3312						    pipe, len, flags);
3313	} else {
3314		if (*ppos) {
3315			rc = -ESPIPE;
3316			goto out;
3317		}
3318		if (flags & SPLICE_F_NONBLOCK)
3319			flags = MSG_DONTWAIT;
3320		else
3321			flags = 0;
3322		SMC_STAT_INC(smc, splice_cnt);
3323		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3324	}
3325out:
3326	release_sock(sk);
3327
3328	return rc;
3329}
3330
3331/* must look like tcp */
3332static const struct proto_ops smc_sock_ops = {
3333	.family		= PF_SMC,
3334	.owner		= THIS_MODULE,
3335	.release	= smc_release,
3336	.bind		= smc_bind,
3337	.connect	= smc_connect,
3338	.socketpair	= sock_no_socketpair,
3339	.accept		= smc_accept,
3340	.getname	= smc_getname,
3341	.poll		= smc_poll,
3342	.ioctl		= smc_ioctl,
3343	.listen		= smc_listen,
3344	.shutdown	= smc_shutdown,
3345	.setsockopt	= smc_setsockopt,
3346	.getsockopt	= smc_getsockopt,
3347	.sendmsg	= smc_sendmsg,
3348	.recvmsg	= smc_recvmsg,
3349	.mmap		= sock_no_mmap,
3350	.splice_read	= smc_splice_read,
3351};
3352
3353int smc_create_clcsk(struct net *net, struct sock *sk, int family)
3354{
3355	struct smc_sock *smc = smc_sk(sk);
3356	int rc;
3357
3358	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3359			      &smc->clcsock);
3360	if (rc)
3361		return rc;
3362
3363	/* smc_clcsock_release() does not wait smc->clcsock->sk's
3364	 * destruction;  its sk_state might not be TCP_CLOSE after
3365	 * smc->sk is close()d, and TCP timers can be fired later,
3366	 * which need net ref.
3367	 */
3368	sk = smc->clcsock->sk;
3369	sk_net_refcnt_upgrade(sk);
3370	return 0;
3371}
3372
3373static int smc_create(struct net *net, struct socket *sock, int protocol,
3374		      int kern)
3375{
3376	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3377	struct sock *sk;
3378	int rc;
3379
3380	rc = -ESOCKTNOSUPPORT;
3381	if (sock->type != SOCK_STREAM)
3382		goto out;
3383
3384	rc = -EPROTONOSUPPORT;
3385	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3386		goto out;
3387
3388	rc = -ENOBUFS;
3389	sock->ops = &smc_sock_ops;
3390	sock->state = SS_UNCONNECTED;
3391	sk = smc_sock_alloc(net, sock, protocol);
3392	if (!sk)
3393		goto out;
3394
3395	rc = smc_create_clcsk(net, sk, family);
3396	if (rc) {
3397		sk_common_release(sk);
3398		sock->sk = NULL;
3399	}
3400out:
3401	return rc;
3402}
3403
3404static const struct net_proto_family smc_sock_family_ops = {
3405	.family	= PF_SMC,
3406	.owner	= THIS_MODULE,
3407	.create	= smc_create,
3408};
3409
3410unsigned int smc_net_id;
3411
3412static __net_init int smc_net_init(struct net *net)
3413{
3414	int rc;
3415
3416	rc = smc_sysctl_net_init(net);
3417	if (rc)
3418		return rc;
3419	return smc_pnet_net_init(net);
3420}
3421
3422static void __net_exit smc_net_exit(struct net *net)
3423{
3424	smc_sysctl_net_exit(net);
3425	smc_pnet_net_exit(net);
3426}
3427
3428static __net_init int smc_net_stat_init(struct net *net)
3429{
3430	return smc_stats_init(net);
3431}
3432
3433static void __net_exit smc_net_stat_exit(struct net *net)
3434{
3435	smc_stats_exit(net);
3436}
3437
3438static struct pernet_operations smc_net_ops = {
3439	.init = smc_net_init,
3440	.exit = smc_net_exit,
3441	.id   = &smc_net_id,
3442	.size = sizeof(struct smc_net),
3443};
3444
3445static struct pernet_operations smc_net_stat_ops = {
3446	.init = smc_net_stat_init,
3447	.exit = smc_net_stat_exit,
3448};
3449
3450static int __init smc_init(void)
3451{
3452	int rc;
3453
3454	rc = register_pernet_subsys(&smc_net_ops);
3455	if (rc)
3456		return rc;
3457
3458	rc = register_pernet_subsys(&smc_net_stat_ops);
3459	if (rc)
3460		goto out_pernet_subsys;
3461
3462	rc = smc_ism_init();
3463	if (rc)
3464		goto out_pernet_subsys_stat;
3465	smc_clc_init();
3466
3467	rc = smc_nl_init();
3468	if (rc)
3469		goto out_ism;
3470
3471	rc = smc_pnet_init();
3472	if (rc)
3473		goto out_nl;
3474
3475	rc = -ENOMEM;
3476
3477	smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", WQ_PERCPU, 0);
3478	if (!smc_tcp_ls_wq)
3479		goto out_pnet;
3480
3481	smc_hs_wq = alloc_workqueue("smc_hs_wq", WQ_PERCPU, 0);
3482	if (!smc_hs_wq)
3483		goto out_alloc_tcp_ls_wq;
3484
3485	smc_close_wq = alloc_workqueue("smc_close_wq", WQ_PERCPU, 0);
3486	if (!smc_close_wq)
3487		goto out_alloc_hs_wq;
3488
3489	rc = smc_core_init();
3490	if (rc) {
3491		pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3492		goto out_alloc_wqs;
3493	}
3494
3495	rc = smc_llc_init();
3496	if (rc) {
3497		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3498		goto out_core;
3499	}
3500
3501	rc = smc_cdc_init();
3502	if (rc) {
3503		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3504		goto out_core;
3505	}
3506
3507	rc = proto_register(&smc_proto, 1);
3508	if (rc) {
3509		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3510		goto out_core;
3511	}
3512
3513	rc = proto_register(&smc_proto6, 1);
3514	if (rc) {
3515		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3516		goto out_proto;
3517	}
3518
3519	rc = sock_register(&smc_sock_family_ops);
3520	if (rc) {
3521		pr_err("%s: sock_register fails with %d\n", __func__, rc);
3522		goto out_proto6;
3523	}
3524	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3525	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3526
3527	rc = smc_ib_register_client();
3528	if (rc) {
3529		pr_err("%s: ib_register fails with %d\n", __func__, rc);
3530		goto out_sock;
3531	}
3532	rc = smc_inet_init();
3533	if (rc) {
3534		pr_err("%s: smc_inet_init fails with %d\n", __func__, rc);
3535		goto out_ib;
3536	}
3537	rc = bpf_smc_hs_ctrl_init();
3538	if (rc) {
3539		pr_err("%s: bpf_smc_hs_ctrl_init fails with %d\n", __func__,
3540		       rc);
3541		goto out_inet;
3542	}
3543	static_branch_enable(&tcp_have_smc);
3544	return 0;
3545out_inet:
3546	smc_inet_exit();
3547out_ib:
3548	smc_ib_unregister_client();
3549out_sock:
3550	sock_unregister(PF_SMC);
3551out_proto6:
3552	proto_unregister(&smc_proto6);
3553out_proto:
3554	proto_unregister(&smc_proto);
3555out_core:
3556	smc_core_exit();
3557out_alloc_wqs:
3558	destroy_workqueue(smc_close_wq);
3559out_alloc_hs_wq:
3560	destroy_workqueue(smc_hs_wq);
3561out_alloc_tcp_ls_wq:
3562	destroy_workqueue(smc_tcp_ls_wq);
3563out_pnet:
3564	smc_pnet_exit();
3565out_nl:
3566	smc_nl_exit();
3567out_ism:
3568	smc_clc_exit();
3569	smc_ism_exit();
3570out_pernet_subsys_stat:
3571	unregister_pernet_subsys(&smc_net_stat_ops);
3572out_pernet_subsys:
3573	unregister_pernet_subsys(&smc_net_ops);
3574
3575	return rc;
3576}
3577
3578static void __exit smc_exit(void)
3579{
3580	static_branch_disable(&tcp_have_smc);
3581	smc_inet_exit();
3582	sock_unregister(PF_SMC);
3583	smc_core_exit();
3584	smc_ib_unregister_client();
3585	smc_ism_exit();
3586	destroy_workqueue(smc_close_wq);
3587	destroy_workqueue(smc_tcp_ls_wq);
3588	destroy_workqueue(smc_hs_wq);
3589	proto_unregister(&smc_proto6);
3590	proto_unregister(&smc_proto);
3591	smc_pnet_exit();
3592	smc_nl_exit();
3593	smc_clc_exit();
3594	unregister_pernet_subsys(&smc_net_stat_ops);
3595	unregister_pernet_subsys(&smc_net_ops);
3596	rcu_barrier();
3597}
3598
3599module_init(smc_init);
3600module_exit(smc_exit);
3601
3602MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3603MODULE_DESCRIPTION("smc socket address family");
3604MODULE_LICENSE("GPL");
3605MODULE_ALIAS_NETPROTO(PF_SMC);
3606/* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */
3607MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1);
3608#if IS_ENABLED(CONFIG_IPV6)
3609MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1);
3610#endif /* CONFIG_IPV6 */
3611MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);