xref: /openbmc/linux/net/smc/af_smc.c (revision 2359ccdd)
1 /*
2  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
3  *
4  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
5  *  applies to SOCK_STREAM sockets only
6  *  offers an alternative communication option for TCP-protocol sockets
7  *  applicable with RoCE-cards only
8  *
9  *  Initial restrictions:
10  *    - support for alternate links postponed
11  *    - partial support for non-blocking sockets only
12  *    - support for urgent data postponed
13  *
14  *  Copyright IBM Corp. 2016, 2018
15  *
16  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
17  *              based on prototype from Frank Blaschka
18  */
19 
20 #define KMSG_COMPONENT "smc"
21 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22 
23 #include <linux/module.h>
24 #include <linux/socket.h>
25 #include <linux/workqueue.h>
26 #include <linux/in.h>
27 #include <linux/sched/signal.h>
28 
29 #include <net/sock.h>
30 #include <net/tcp.h>
31 #include <net/smc.h>
32 
33 #include "smc.h"
34 #include "smc_clc.h"
35 #include "smc_llc.h"
36 #include "smc_cdc.h"
37 #include "smc_core.h"
38 #include "smc_ib.h"
39 #include "smc_pnet.h"
40 #include "smc_tx.h"
41 #include "smc_rx.h"
42 #include "smc_close.h"
43 
44 static DEFINE_MUTEX(smc_create_lgr_pending);	/* serialize link group
45 						 * creation
46 						 */
47 
48 struct smc_lgr_list smc_lgr_list = {		/* established link groups */
49 	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
50 	.list = LIST_HEAD_INIT(smc_lgr_list.list),
51 };
52 
53 static void smc_tcp_listen_work(struct work_struct *);
54 
55 static void smc_set_keepalive(struct sock *sk, int val)
56 {
57 	struct smc_sock *smc = smc_sk(sk);
58 
59 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
60 }
61 
62 static struct smc_hashinfo smc_v4_hashinfo = {
63 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
64 };
65 
66 static struct smc_hashinfo smc_v6_hashinfo = {
67 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
68 };
69 
70 int smc_hash_sk(struct sock *sk)
71 {
72 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
73 	struct hlist_head *head;
74 
75 	head = &h->ht;
76 
77 	write_lock_bh(&h->lock);
78 	sk_add_node(sk, head);
79 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
80 	write_unlock_bh(&h->lock);
81 
82 	return 0;
83 }
84 EXPORT_SYMBOL_GPL(smc_hash_sk);
85 
86 void smc_unhash_sk(struct sock *sk)
87 {
88 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
89 
90 	write_lock_bh(&h->lock);
91 	if (sk_del_node_init(sk))
92 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
93 	write_unlock_bh(&h->lock);
94 }
95 EXPORT_SYMBOL_GPL(smc_unhash_sk);
96 
97 struct proto smc_proto = {
98 	.name		= "SMC",
99 	.owner		= THIS_MODULE,
100 	.keepalive	= smc_set_keepalive,
101 	.hash		= smc_hash_sk,
102 	.unhash		= smc_unhash_sk,
103 	.obj_size	= sizeof(struct smc_sock),
104 	.h.smc_hash	= &smc_v4_hashinfo,
105 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
106 };
107 EXPORT_SYMBOL_GPL(smc_proto);
108 
109 struct proto smc_proto6 = {
110 	.name		= "SMC6",
111 	.owner		= THIS_MODULE,
112 	.keepalive	= smc_set_keepalive,
113 	.hash		= smc_hash_sk,
114 	.unhash		= smc_unhash_sk,
115 	.obj_size	= sizeof(struct smc_sock),
116 	.h.smc_hash	= &smc_v6_hashinfo,
117 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
118 };
119 EXPORT_SYMBOL_GPL(smc_proto6);
120 
121 static int smc_release(struct socket *sock)
122 {
123 	struct sock *sk = sock->sk;
124 	struct smc_sock *smc;
125 	int rc = 0;
126 
127 	if (!sk)
128 		goto out;
129 
130 	smc = smc_sk(sk);
131 	if (sk->sk_state == SMC_LISTEN)
132 		/* smc_close_non_accepted() is called and acquires
133 		 * sock lock for child sockets again
134 		 */
135 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
136 	else
137 		lock_sock(sk);
138 
139 	if (!smc->use_fallback) {
140 		rc = smc_close_active(smc);
141 		sock_set_flag(sk, SOCK_DEAD);
142 		sk->sk_shutdown |= SHUTDOWN_MASK;
143 	}
144 	if (smc->clcsock) {
145 		sock_release(smc->clcsock);
146 		smc->clcsock = NULL;
147 	}
148 	if (smc->use_fallback) {
149 		sock_put(sk); /* passive closing */
150 		sk->sk_state = SMC_CLOSED;
151 		sk->sk_state_change(sk);
152 	}
153 
154 	/* detach socket */
155 	sock_orphan(sk);
156 	sock->sk = NULL;
157 	if (!smc->use_fallback && sk->sk_state == SMC_CLOSED)
158 		smc_conn_free(&smc->conn);
159 	release_sock(sk);
160 
161 	sk->sk_prot->unhash(sk);
162 	sock_put(sk); /* final sock_put */
163 out:
164 	return rc;
165 }
166 
167 static void smc_destruct(struct sock *sk)
168 {
169 	if (sk->sk_state != SMC_CLOSED)
170 		return;
171 	if (!sock_flag(sk, SOCK_DEAD))
172 		return;
173 
174 	sk_refcnt_debug_dec(sk);
175 }
176 
177 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
178 				   int protocol)
179 {
180 	struct smc_sock *smc;
181 	struct proto *prot;
182 	struct sock *sk;
183 
184 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
185 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
186 	if (!sk)
187 		return NULL;
188 
189 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
190 	sk->sk_state = SMC_INIT;
191 	sk->sk_destruct = smc_destruct;
192 	sk->sk_protocol = protocol;
193 	smc = smc_sk(sk);
194 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
195 	INIT_LIST_HEAD(&smc->accept_q);
196 	spin_lock_init(&smc->accept_q_lock);
197 	sk->sk_prot->hash(sk);
198 	sk_refcnt_debug_inc(sk);
199 
200 	return sk;
201 }
202 
203 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
204 		    int addr_len)
205 {
206 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
207 	struct sock *sk = sock->sk;
208 	struct smc_sock *smc;
209 	int rc;
210 
211 	smc = smc_sk(sk);
212 
213 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
214 	rc = -EINVAL;
215 	if (addr_len < sizeof(struct sockaddr_in))
216 		goto out;
217 
218 	rc = -EAFNOSUPPORT;
219 	if (addr->sin_family != AF_INET &&
220 	    addr->sin_family != AF_INET6 &&
221 	    addr->sin_family != AF_UNSPEC)
222 		goto out;
223 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
224 	if (addr->sin_family == AF_UNSPEC &&
225 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
226 		goto out;
227 
228 	lock_sock(sk);
229 
230 	/* Check if socket is already active */
231 	rc = -EINVAL;
232 	if (sk->sk_state != SMC_INIT)
233 		goto out_rel;
234 
235 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
236 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
237 
238 out_rel:
239 	release_sock(sk);
240 out:
241 	return rc;
242 }
243 
244 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
245 				   unsigned long mask)
246 {
247 	/* options we don't get control via setsockopt for */
248 	nsk->sk_type = osk->sk_type;
249 	nsk->sk_sndbuf = osk->sk_sndbuf;
250 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
251 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
252 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
253 	nsk->sk_mark = osk->sk_mark;
254 	nsk->sk_priority = osk->sk_priority;
255 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
256 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
257 	nsk->sk_err = osk->sk_err;
258 
259 	nsk->sk_flags &= ~mask;
260 	nsk->sk_flags |= osk->sk_flags & mask;
261 }
262 
263 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
264 			     (1UL << SOCK_KEEPOPEN) | \
265 			     (1UL << SOCK_LINGER) | \
266 			     (1UL << SOCK_BROADCAST) | \
267 			     (1UL << SOCK_TIMESTAMP) | \
268 			     (1UL << SOCK_DBG) | \
269 			     (1UL << SOCK_RCVTSTAMP) | \
270 			     (1UL << SOCK_RCVTSTAMPNS) | \
271 			     (1UL << SOCK_LOCALROUTE) | \
272 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
273 			     (1UL << SOCK_RXQ_OVFL) | \
274 			     (1UL << SOCK_WIFI_STATUS) | \
275 			     (1UL << SOCK_NOFCS) | \
276 			     (1UL << SOCK_FILTER_LOCKED))
277 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
278  * clc socket (since smc is not called for these options from net/core)
279  */
280 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
281 {
282 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
283 }
284 
285 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
286 			     (1UL << SOCK_KEEPOPEN) | \
287 			     (1UL << SOCK_LINGER) | \
288 			     (1UL << SOCK_DBG))
289 /* copy only settings and flags relevant for smc from clc to smc socket */
290 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
291 {
292 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
293 }
294 
295 static int smc_clnt_conf_first_link(struct smc_sock *smc)
296 {
297 	struct smc_link_group *lgr = smc->conn.lgr;
298 	struct smc_link *link;
299 	int rest;
300 	int rc;
301 
302 	link = &lgr->lnk[SMC_SINGLE_LINK];
303 	/* receive CONFIRM LINK request from server over RoCE fabric */
304 	rest = wait_for_completion_interruptible_timeout(
305 		&link->llc_confirm,
306 		SMC_LLC_WAIT_FIRST_TIME);
307 	if (rest <= 0) {
308 		struct smc_clc_msg_decline dclc;
309 
310 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
311 				      SMC_CLC_DECLINE);
312 		return rc;
313 	}
314 
315 	if (link->llc_confirm_rc)
316 		return SMC_CLC_DECL_RMBE_EC;
317 
318 	rc = smc_ib_modify_qp_rts(link);
319 	if (rc)
320 		return SMC_CLC_DECL_INTERR;
321 
322 	smc_wr_remember_qp_attr(link);
323 
324 	rc = smc_wr_reg_send(link,
325 			     smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
326 	if (rc)
327 		return SMC_CLC_DECL_INTERR;
328 
329 	/* send CONFIRM LINK response over RoCE fabric */
330 	rc = smc_llc_send_confirm_link(link,
331 				       link->smcibdev->mac[link->ibport - 1],
332 				       &link->smcibdev->gid[link->ibport - 1],
333 				       SMC_LLC_RESP);
334 	if (rc < 0)
335 		return SMC_CLC_DECL_TCL;
336 
337 	/* receive ADD LINK request from server over RoCE fabric */
338 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
339 							 SMC_LLC_WAIT_TIME);
340 	if (rest <= 0) {
341 		struct smc_clc_msg_decline dclc;
342 
343 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
344 				      SMC_CLC_DECLINE);
345 		return rc;
346 	}
347 
348 	/* send add link reject message, only one link supported for now */
349 	rc = smc_llc_send_add_link(link,
350 				   link->smcibdev->mac[link->ibport - 1],
351 				   &link->smcibdev->gid[link->ibport - 1],
352 				   SMC_LLC_RESP);
353 	if (rc < 0)
354 		return SMC_CLC_DECL_TCL;
355 
356 	link->state = SMC_LNK_ACTIVE;
357 
358 	return 0;
359 }
360 
361 static void smc_conn_save_peer_info(struct smc_sock *smc,
362 				    struct smc_clc_msg_accept_confirm *clc)
363 {
364 	smc->conn.peer_conn_idx = clc->conn_idx;
365 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
366 	smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
367 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
368 }
369 
370 static void smc_link_save_peer_info(struct smc_link *link,
371 				    struct smc_clc_msg_accept_confirm *clc)
372 {
373 	link->peer_qpn = ntoh24(clc->qpn);
374 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
375 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
376 	link->peer_psn = ntoh24(clc->psn);
377 	link->peer_mtu = clc->qp_mtu;
378 }
379 
380 /* setup for RDMA connection of client */
381 static int smc_connect_rdma(struct smc_sock *smc)
382 {
383 	struct smc_clc_msg_accept_confirm aclc;
384 	int local_contact = SMC_FIRST_CONTACT;
385 	struct smc_ib_device *smcibdev;
386 	struct smc_link *link;
387 	u8 srv_first_contact;
388 	int reason_code = 0;
389 	int rc = 0;
390 	u8 ibport;
391 
392 	sock_hold(&smc->sk); /* sock put in passive closing */
393 
394 	if (!tcp_sk(smc->clcsock->sk)->syn_smc) {
395 		/* peer has not signalled SMC-capability */
396 		smc->use_fallback = true;
397 		goto out_connected;
398 	}
399 
400 	/* IPSec connections opt out of SMC-R optimizations */
401 	if (using_ipsec(smc)) {
402 		reason_code = SMC_CLC_DECL_IPSEC;
403 		goto decline_rdma;
404 	}
405 
406 	/* PNET table look up: search active ib_device and port
407 	 * within same PNETID that also contains the ethernet device
408 	 * used for the internal TCP socket
409 	 */
410 	smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
411 	if (!smcibdev) {
412 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
413 		goto decline_rdma;
414 	}
415 
416 	/* do inband token exchange */
417 	reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
418 	if (reason_code < 0) {
419 		rc = reason_code;
420 		goto out_err;
421 	}
422 	if (reason_code > 0) /* configuration error */
423 		goto decline_rdma;
424 	/* receive SMC Accept CLC message */
425 	reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
426 				       SMC_CLC_ACCEPT);
427 	if (reason_code < 0) {
428 		rc = reason_code;
429 		goto out_err;
430 	}
431 	if (reason_code > 0)
432 		goto decline_rdma;
433 
434 	srv_first_contact = aclc.hdr.flag;
435 	mutex_lock(&smc_create_lgr_pending);
436 	local_contact = smc_conn_create(smc, smcibdev, ibport, &aclc.lcl,
437 					srv_first_contact);
438 	if (local_contact < 0) {
439 		rc = local_contact;
440 		if (rc == -ENOMEM)
441 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
442 		else if (rc == -ENOLINK)
443 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
444 		goto decline_rdma_unlock;
445 	}
446 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
447 
448 	smc_conn_save_peer_info(smc, &aclc);
449 
450 	/* create send buffer and rmb */
451 	rc = smc_buf_create(smc);
452 	if (rc) {
453 		reason_code = SMC_CLC_DECL_MEM;
454 		goto decline_rdma_unlock;
455 	}
456 
457 	if (local_contact == SMC_FIRST_CONTACT)
458 		smc_link_save_peer_info(link, &aclc);
459 
460 	rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
461 	if (rc) {
462 		reason_code = SMC_CLC_DECL_INTERR;
463 		goto decline_rdma_unlock;
464 	}
465 
466 	smc_close_init(smc);
467 	smc_rx_init(smc);
468 
469 	if (local_contact == SMC_FIRST_CONTACT) {
470 		rc = smc_ib_ready_link(link);
471 		if (rc) {
472 			reason_code = SMC_CLC_DECL_INTERR;
473 			goto decline_rdma_unlock;
474 		}
475 	} else {
476 		struct smc_buf_desc *buf_desc = smc->conn.rmb_desc;
477 
478 		if (!buf_desc->reused) {
479 			/* register memory region for new rmb */
480 			rc = smc_wr_reg_send(link,
481 					     buf_desc->mr_rx[SMC_SINGLE_LINK]);
482 			if (rc) {
483 				reason_code = SMC_CLC_DECL_INTERR;
484 				goto decline_rdma_unlock;
485 			}
486 		}
487 	}
488 	smc_rmb_sync_sg_for_device(&smc->conn);
489 
490 	rc = smc_clc_send_confirm(smc);
491 	if (rc)
492 		goto out_err_unlock;
493 
494 	if (local_contact == SMC_FIRST_CONTACT) {
495 		/* QP confirmation over RoCE fabric */
496 		reason_code = smc_clnt_conf_first_link(smc);
497 		if (reason_code < 0) {
498 			rc = reason_code;
499 			goto out_err_unlock;
500 		}
501 		if (reason_code > 0)
502 			goto decline_rdma_unlock;
503 	}
504 
505 	mutex_unlock(&smc_create_lgr_pending);
506 	smc_tx_init(smc);
507 
508 out_connected:
509 	smc_copy_sock_settings_to_clc(smc);
510 	if (smc->sk.sk_state == SMC_INIT)
511 		smc->sk.sk_state = SMC_ACTIVE;
512 
513 	return rc ? rc : local_contact;
514 
515 decline_rdma_unlock:
516 	if (local_contact == SMC_FIRST_CONTACT)
517 		smc_lgr_forget(smc->conn.lgr);
518 	mutex_unlock(&smc_create_lgr_pending);
519 	smc_conn_free(&smc->conn);
520 decline_rdma:
521 	/* RDMA setup failed, switch back to TCP */
522 	smc->use_fallback = true;
523 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
524 		rc = smc_clc_send_decline(smc, reason_code);
525 		if (rc < 0)
526 			goto out_err;
527 	}
528 	goto out_connected;
529 
530 out_err_unlock:
531 	if (local_contact == SMC_FIRST_CONTACT)
532 		smc_lgr_forget(smc->conn.lgr);
533 	mutex_unlock(&smc_create_lgr_pending);
534 	smc_conn_free(&smc->conn);
535 out_err:
536 	if (smc->sk.sk_state == SMC_INIT)
537 		sock_put(&smc->sk); /* passive closing */
538 	return rc;
539 }
540 
541 static int smc_connect(struct socket *sock, struct sockaddr *addr,
542 		       int alen, int flags)
543 {
544 	struct sock *sk = sock->sk;
545 	struct smc_sock *smc;
546 	int rc = -EINVAL;
547 
548 	smc = smc_sk(sk);
549 
550 	/* separate smc parameter checking to be safe */
551 	if (alen < sizeof(addr->sa_family))
552 		goto out_err;
553 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
554 		goto out_err;
555 
556 	lock_sock(sk);
557 	switch (sk->sk_state) {
558 	default:
559 		goto out;
560 	case SMC_ACTIVE:
561 		rc = -EISCONN;
562 		goto out;
563 	case SMC_INIT:
564 		rc = 0;
565 		break;
566 	}
567 
568 	smc_copy_sock_settings_to_clc(smc);
569 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
570 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
571 	if (rc)
572 		goto out;
573 
574 	/* setup RDMA connection */
575 	rc = smc_connect_rdma(smc);
576 	if (rc < 0)
577 		goto out;
578 	else
579 		rc = 0; /* success cases including fallback */
580 
581 out:
582 	release_sock(sk);
583 out_err:
584 	return rc;
585 }
586 
587 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
588 {
589 	struct socket *new_clcsock = NULL;
590 	struct sock *lsk = &lsmc->sk;
591 	struct sock *new_sk;
592 	int rc;
593 
594 	release_sock(lsk);
595 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
596 	if (!new_sk) {
597 		rc = -ENOMEM;
598 		lsk->sk_err = ENOMEM;
599 		*new_smc = NULL;
600 		lock_sock(lsk);
601 		goto out;
602 	}
603 	*new_smc = smc_sk(new_sk);
604 
605 	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
606 	lock_sock(lsk);
607 	if  (rc < 0)
608 		lsk->sk_err = -rc;
609 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
610 		if (new_clcsock)
611 			sock_release(new_clcsock);
612 		new_sk->sk_state = SMC_CLOSED;
613 		sock_set_flag(new_sk, SOCK_DEAD);
614 		new_sk->sk_prot->unhash(new_sk);
615 		sock_put(new_sk); /* final */
616 		*new_smc = NULL;
617 		goto out;
618 	}
619 
620 	(*new_smc)->clcsock = new_clcsock;
621 out:
622 	return rc;
623 }
624 
625 /* add a just created sock to the accept queue of the listen sock as
626  * candidate for a following socket accept call from user space
627  */
628 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
629 {
630 	struct smc_sock *par = smc_sk(parent);
631 
632 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
633 	spin_lock(&par->accept_q_lock);
634 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
635 	spin_unlock(&par->accept_q_lock);
636 	sk_acceptq_added(parent);
637 }
638 
639 /* remove a socket from the accept queue of its parental listening socket */
640 static void smc_accept_unlink(struct sock *sk)
641 {
642 	struct smc_sock *par = smc_sk(sk)->listen_smc;
643 
644 	spin_lock(&par->accept_q_lock);
645 	list_del_init(&smc_sk(sk)->accept_q);
646 	spin_unlock(&par->accept_q_lock);
647 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
648 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
649 }
650 
651 /* remove a sock from the accept queue to bind it to a new socket created
652  * for a socket accept call from user space
653  */
654 struct sock *smc_accept_dequeue(struct sock *parent,
655 				struct socket *new_sock)
656 {
657 	struct smc_sock *isk, *n;
658 	struct sock *new_sk;
659 
660 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
661 		new_sk = (struct sock *)isk;
662 
663 		smc_accept_unlink(new_sk);
664 		if (new_sk->sk_state == SMC_CLOSED) {
665 			if (isk->clcsock) {
666 				sock_release(isk->clcsock);
667 				isk->clcsock = NULL;
668 			}
669 			new_sk->sk_prot->unhash(new_sk);
670 			sock_put(new_sk); /* final */
671 			continue;
672 		}
673 		if (new_sock)
674 			sock_graft(new_sk, new_sock);
675 		return new_sk;
676 	}
677 	return NULL;
678 }
679 
680 /* clean up for a created but never accepted sock */
681 void smc_close_non_accepted(struct sock *sk)
682 {
683 	struct smc_sock *smc = smc_sk(sk);
684 
685 	lock_sock(sk);
686 	if (!sk->sk_lingertime)
687 		/* wait for peer closing */
688 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
689 	if (!smc->use_fallback) {
690 		smc_close_active(smc);
691 		sock_set_flag(sk, SOCK_DEAD);
692 		sk->sk_shutdown |= SHUTDOWN_MASK;
693 	}
694 	if (smc->clcsock) {
695 		struct socket *tcp;
696 
697 		tcp = smc->clcsock;
698 		smc->clcsock = NULL;
699 		sock_release(tcp);
700 	}
701 	if (smc->use_fallback) {
702 		sock_put(sk); /* passive closing */
703 		sk->sk_state = SMC_CLOSED;
704 	} else {
705 		if (sk->sk_state == SMC_CLOSED)
706 			smc_conn_free(&smc->conn);
707 	}
708 	release_sock(sk);
709 	sk->sk_prot->unhash(sk);
710 	sock_put(sk); /* final sock_put */
711 }
712 
713 static int smc_serv_conf_first_link(struct smc_sock *smc)
714 {
715 	struct smc_link_group *lgr = smc->conn.lgr;
716 	struct smc_link *link;
717 	int rest;
718 	int rc;
719 
720 	link = &lgr->lnk[SMC_SINGLE_LINK];
721 
722 	rc = smc_wr_reg_send(link,
723 			     smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
724 	if (rc)
725 		return SMC_CLC_DECL_INTERR;
726 
727 	/* send CONFIRM LINK request to client over the RoCE fabric */
728 	rc = smc_llc_send_confirm_link(link,
729 				       link->smcibdev->mac[link->ibport - 1],
730 				       &link->smcibdev->gid[link->ibport - 1],
731 				       SMC_LLC_REQ);
732 	if (rc < 0)
733 		return SMC_CLC_DECL_TCL;
734 
735 	/* receive CONFIRM LINK response from client over the RoCE fabric */
736 	rest = wait_for_completion_interruptible_timeout(
737 		&link->llc_confirm_resp,
738 		SMC_LLC_WAIT_FIRST_TIME);
739 	if (rest <= 0) {
740 		struct smc_clc_msg_decline dclc;
741 
742 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
743 				      SMC_CLC_DECLINE);
744 		return rc;
745 	}
746 
747 	if (link->llc_confirm_resp_rc)
748 		return SMC_CLC_DECL_RMBE_EC;
749 
750 	/* send ADD LINK request to client over the RoCE fabric */
751 	rc = smc_llc_send_add_link(link,
752 				   link->smcibdev->mac[link->ibport - 1],
753 				   &link->smcibdev->gid[link->ibport - 1],
754 				   SMC_LLC_REQ);
755 	if (rc < 0)
756 		return SMC_CLC_DECL_TCL;
757 
758 	/* receive ADD LINK response from client over the RoCE fabric */
759 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
760 							 SMC_LLC_WAIT_TIME);
761 	if (rest <= 0) {
762 		struct smc_clc_msg_decline dclc;
763 
764 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
765 				      SMC_CLC_DECLINE);
766 		return rc;
767 	}
768 
769 	link->state = SMC_LNK_ACTIVE;
770 
771 	return 0;
772 }
773 
774 /* setup for RDMA connection of server */
775 static void smc_listen_work(struct work_struct *work)
776 {
777 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
778 						smc_listen_work);
779 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
780 	struct socket *newclcsock = new_smc->clcsock;
781 	struct smc_sock *lsmc = new_smc->listen_smc;
782 	struct smc_clc_msg_accept_confirm cclc;
783 	int local_contact = SMC_REUSE_CONTACT;
784 	struct sock *newsmcsk = &new_smc->sk;
785 	struct smc_clc_msg_proposal *pclc;
786 	struct smc_ib_device *smcibdev;
787 	u8 buf[SMC_CLC_MAX_LEN];
788 	struct smc_link *link;
789 	int reason_code = 0;
790 	int rc = 0;
791 	u8 ibport;
792 
793 	/* check if peer is smc capable */
794 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
795 		new_smc->use_fallback = true;
796 		goto out_connected;
797 	}
798 
799 	/* do inband token exchange -
800 	 *wait for and receive SMC Proposal CLC message
801 	 */
802 	reason_code = smc_clc_wait_msg(new_smc, &buf, sizeof(buf),
803 				       SMC_CLC_PROPOSAL);
804 	if (reason_code < 0)
805 		goto out_err;
806 	if (reason_code > 0)
807 		goto decline_rdma;
808 
809 	/* IPSec connections opt out of SMC-R optimizations */
810 	if (using_ipsec(new_smc)) {
811 		reason_code = SMC_CLC_DECL_IPSEC;
812 		goto decline_rdma;
813 	}
814 
815 	/* PNET table look up: search active ib_device and port
816 	 * within same PNETID that also contains the ethernet device
817 	 * used for the internal TCP socket
818 	 */
819 	smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
820 	if (!smcibdev) {
821 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
822 		goto decline_rdma;
823 	}
824 
825 	pclc = (struct smc_clc_msg_proposal *)&buf;
826 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
827 
828 	rc = smc_clc_prfx_match(newclcsock, pclc_prfx);
829 	if (rc) {
830 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
831 		goto decline_rdma;
832 	}
833 
834 	/* allocate connection / link group */
835 	mutex_lock(&smc_create_lgr_pending);
836 	local_contact = smc_conn_create(new_smc, smcibdev, ibport, &pclc->lcl,
837 					0);
838 	if (local_contact < 0) {
839 		rc = local_contact;
840 		if (rc == -ENOMEM)
841 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
842 		goto decline_rdma_unlock;
843 	}
844 	link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
845 
846 	/* create send buffer and rmb */
847 	rc = smc_buf_create(new_smc);
848 	if (rc) {
849 		reason_code = SMC_CLC_DECL_MEM;
850 		goto decline_rdma_unlock;
851 	}
852 
853 	smc_close_init(new_smc);
854 	smc_rx_init(new_smc);
855 
856 	if (local_contact != SMC_FIRST_CONTACT) {
857 		struct smc_buf_desc *buf_desc = new_smc->conn.rmb_desc;
858 
859 		if (!buf_desc->reused) {
860 			/* register memory region for new rmb */
861 			rc = smc_wr_reg_send(link,
862 					     buf_desc->mr_rx[SMC_SINGLE_LINK]);
863 			if (rc) {
864 				reason_code = SMC_CLC_DECL_INTERR;
865 				goto decline_rdma_unlock;
866 			}
867 		}
868 	}
869 	smc_rmb_sync_sg_for_device(&new_smc->conn);
870 
871 	rc = smc_clc_send_accept(new_smc, local_contact);
872 	if (rc)
873 		goto out_err_unlock;
874 
875 	/* receive SMC Confirm CLC message */
876 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
877 				       SMC_CLC_CONFIRM);
878 	if (reason_code < 0)
879 		goto out_err_unlock;
880 	if (reason_code > 0)
881 		goto decline_rdma_unlock;
882 	smc_conn_save_peer_info(new_smc, &cclc);
883 	if (local_contact == SMC_FIRST_CONTACT)
884 		smc_link_save_peer_info(link, &cclc);
885 
886 	rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
887 	if (rc) {
888 		reason_code = SMC_CLC_DECL_INTERR;
889 		goto decline_rdma_unlock;
890 	}
891 
892 	if (local_contact == SMC_FIRST_CONTACT) {
893 		rc = smc_ib_ready_link(link);
894 		if (rc) {
895 			reason_code = SMC_CLC_DECL_INTERR;
896 			goto decline_rdma_unlock;
897 		}
898 		/* QP confirmation over RoCE fabric */
899 		reason_code = smc_serv_conf_first_link(new_smc);
900 		if (reason_code < 0)
901 			/* peer is not aware of a problem */
902 			goto out_err_unlock;
903 		if (reason_code > 0)
904 			goto decline_rdma_unlock;
905 	}
906 
907 	smc_tx_init(new_smc);
908 	mutex_unlock(&smc_create_lgr_pending);
909 
910 out_connected:
911 	sk_refcnt_debug_inc(newsmcsk);
912 	if (newsmcsk->sk_state == SMC_INIT)
913 		newsmcsk->sk_state = SMC_ACTIVE;
914 enqueue:
915 	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
916 	if (lsmc->sk.sk_state == SMC_LISTEN) {
917 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
918 	} else { /* no longer listening */
919 		smc_close_non_accepted(newsmcsk);
920 	}
921 	release_sock(&lsmc->sk);
922 
923 	/* Wake up accept */
924 	lsmc->sk.sk_data_ready(&lsmc->sk);
925 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
926 	return;
927 
928 decline_rdma_unlock:
929 	if (local_contact == SMC_FIRST_CONTACT)
930 		smc_lgr_forget(new_smc->conn.lgr);
931 	mutex_unlock(&smc_create_lgr_pending);
932 decline_rdma:
933 	/* RDMA setup failed, switch back to TCP */
934 	smc_conn_free(&new_smc->conn);
935 	new_smc->use_fallback = true;
936 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
937 		if (smc_clc_send_decline(new_smc, reason_code) < 0)
938 			goto out_err;
939 	}
940 	goto out_connected;
941 
942 out_err_unlock:
943 	if (local_contact == SMC_FIRST_CONTACT)
944 		smc_lgr_forget(new_smc->conn.lgr);
945 	mutex_unlock(&smc_create_lgr_pending);
946 out_err:
947 	if (newsmcsk->sk_state == SMC_INIT)
948 		sock_put(&new_smc->sk); /* passive closing */
949 	newsmcsk->sk_state = SMC_CLOSED;
950 	smc_conn_free(&new_smc->conn);
951 	goto enqueue; /* queue new sock with sk_err set */
952 }
953 
954 static void smc_tcp_listen_work(struct work_struct *work)
955 {
956 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
957 					     tcp_listen_work);
958 	struct sock *lsk = &lsmc->sk;
959 	struct smc_sock *new_smc;
960 	int rc = 0;
961 
962 	lock_sock(lsk);
963 	while (lsk->sk_state == SMC_LISTEN) {
964 		rc = smc_clcsock_accept(lsmc, &new_smc);
965 		if (rc)
966 			goto out;
967 		if (!new_smc)
968 			continue;
969 
970 		new_smc->listen_smc = lsmc;
971 		new_smc->use_fallback = false; /* assume rdma capability first*/
972 		sock_hold(lsk); /* sock_put in smc_listen_work */
973 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
974 		smc_copy_sock_settings_to_smc(new_smc);
975 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
976 		if (!schedule_work(&new_smc->smc_listen_work))
977 			sock_put(&new_smc->sk);
978 	}
979 
980 out:
981 	if (lsmc->clcsock) {
982 		sock_release(lsmc->clcsock);
983 		lsmc->clcsock = NULL;
984 	}
985 	release_sock(lsk);
986 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
987 }
988 
989 static int smc_listen(struct socket *sock, int backlog)
990 {
991 	struct sock *sk = sock->sk;
992 	struct smc_sock *smc;
993 	int rc;
994 
995 	smc = smc_sk(sk);
996 	lock_sock(sk);
997 
998 	rc = -EINVAL;
999 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
1000 		goto out;
1001 
1002 	rc = 0;
1003 	if (sk->sk_state == SMC_LISTEN) {
1004 		sk->sk_max_ack_backlog = backlog;
1005 		goto out;
1006 	}
1007 	/* some socket options are handled in core, so we could not apply
1008 	 * them to the clc socket -- copy smc socket options to clc socket
1009 	 */
1010 	smc_copy_sock_settings_to_clc(smc);
1011 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1012 
1013 	rc = kernel_listen(smc->clcsock, backlog);
1014 	if (rc)
1015 		goto out;
1016 	sk->sk_max_ack_backlog = backlog;
1017 	sk->sk_ack_backlog = 0;
1018 	sk->sk_state = SMC_LISTEN;
1019 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
1020 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1021 	if (!schedule_work(&smc->tcp_listen_work))
1022 		sock_put(sk);
1023 
1024 out:
1025 	release_sock(sk);
1026 	return rc;
1027 }
1028 
1029 static int smc_accept(struct socket *sock, struct socket *new_sock,
1030 		      int flags, bool kern)
1031 {
1032 	struct sock *sk = sock->sk, *nsk;
1033 	DECLARE_WAITQUEUE(wait, current);
1034 	struct smc_sock *lsmc;
1035 	long timeo;
1036 	int rc = 0;
1037 
1038 	lsmc = smc_sk(sk);
1039 	sock_hold(sk); /* sock_put below */
1040 	lock_sock(sk);
1041 
1042 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1043 		rc = -EINVAL;
1044 		goto out;
1045 	}
1046 
1047 	/* Wait for an incoming connection */
1048 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1049 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1050 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1051 		set_current_state(TASK_INTERRUPTIBLE);
1052 		if (!timeo) {
1053 			rc = -EAGAIN;
1054 			break;
1055 		}
1056 		release_sock(sk);
1057 		timeo = schedule_timeout(timeo);
1058 		/* wakeup by sk_data_ready in smc_listen_work() */
1059 		sched_annotate_sleep();
1060 		lock_sock(sk);
1061 		if (signal_pending(current)) {
1062 			rc = sock_intr_errno(timeo);
1063 			break;
1064 		}
1065 	}
1066 	set_current_state(TASK_RUNNING);
1067 	remove_wait_queue(sk_sleep(sk), &wait);
1068 
1069 	if (!rc)
1070 		rc = sock_error(nsk);
1071 
1072 out:
1073 	release_sock(sk);
1074 	sock_put(sk); /* sock_hold above */
1075 	return rc;
1076 }
1077 
1078 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1079 		       int peer)
1080 {
1081 	struct smc_sock *smc;
1082 
1083 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1084 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1085 		return -ENOTCONN;
1086 
1087 	smc = smc_sk(sock->sk);
1088 
1089 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1090 }
1091 
1092 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1093 {
1094 	struct sock *sk = sock->sk;
1095 	struct smc_sock *smc;
1096 	int rc = -EPIPE;
1097 
1098 	smc = smc_sk(sk);
1099 	lock_sock(sk);
1100 	if ((sk->sk_state != SMC_ACTIVE) &&
1101 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1102 	    (sk->sk_state != SMC_INIT))
1103 		goto out;
1104 	if (smc->use_fallback)
1105 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1106 	else
1107 		rc = smc_tx_sendmsg(smc, msg, len);
1108 out:
1109 	release_sock(sk);
1110 	return rc;
1111 }
1112 
1113 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1114 		       int flags)
1115 {
1116 	struct sock *sk = sock->sk;
1117 	struct smc_sock *smc;
1118 	int rc = -ENOTCONN;
1119 
1120 	smc = smc_sk(sk);
1121 	lock_sock(sk);
1122 	if ((sk->sk_state == SMC_INIT) ||
1123 	    (sk->sk_state == SMC_LISTEN) ||
1124 	    (sk->sk_state == SMC_CLOSED))
1125 		goto out;
1126 
1127 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1128 		rc = 0;
1129 		goto out;
1130 	}
1131 
1132 	if (smc->use_fallback)
1133 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1134 	else
1135 		rc = smc_rx_recvmsg(smc, msg, len, flags);
1136 
1137 out:
1138 	release_sock(sk);
1139 	return rc;
1140 }
1141 
1142 static __poll_t smc_accept_poll(struct sock *parent)
1143 {
1144 	struct smc_sock *isk = smc_sk(parent);
1145 	__poll_t mask = 0;
1146 
1147 	spin_lock(&isk->accept_q_lock);
1148 	if (!list_empty(&isk->accept_q))
1149 		mask = EPOLLIN | EPOLLRDNORM;
1150 	spin_unlock(&isk->accept_q_lock);
1151 
1152 	return mask;
1153 }
1154 
1155 static __poll_t smc_poll(struct file *file, struct socket *sock,
1156 			     poll_table *wait)
1157 {
1158 	struct sock *sk = sock->sk;
1159 	__poll_t mask = 0;
1160 	struct smc_sock *smc;
1161 	int rc;
1162 
1163 	if (!sk)
1164 		return EPOLLNVAL;
1165 
1166 	smc = smc_sk(sock->sk);
1167 	sock_hold(sk);
1168 	lock_sock(sk);
1169 	if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1170 		/* delegate to CLC child sock */
1171 		release_sock(sk);
1172 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1173 		/* if non-blocking connect finished ... */
1174 		lock_sock(sk);
1175 		if ((sk->sk_state == SMC_INIT) && (mask & EPOLLOUT)) {
1176 			sk->sk_err = smc->clcsock->sk->sk_err;
1177 			if (sk->sk_err) {
1178 				mask |= EPOLLERR;
1179 			} else {
1180 				rc = smc_connect_rdma(smc);
1181 				if (rc < 0)
1182 					mask |= EPOLLERR;
1183 				/* success cases including fallback */
1184 				mask |= EPOLLOUT | EPOLLWRNORM;
1185 			}
1186 		}
1187 	} else {
1188 		if (sk->sk_state != SMC_CLOSED) {
1189 			release_sock(sk);
1190 			sock_poll_wait(file, sk_sleep(sk), wait);
1191 			lock_sock(sk);
1192 		}
1193 		if (sk->sk_err)
1194 			mask |= EPOLLERR;
1195 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1196 		    (sk->sk_state == SMC_CLOSED))
1197 			mask |= EPOLLHUP;
1198 		if (sk->sk_state == SMC_LISTEN) {
1199 			/* woken up by sk_data_ready in smc_listen_work() */
1200 			mask = smc_accept_poll(sk);
1201 		} else {
1202 			if (atomic_read(&smc->conn.sndbuf_space) ||
1203 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1204 				mask |= EPOLLOUT | EPOLLWRNORM;
1205 			} else {
1206 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1207 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1208 			}
1209 			if (atomic_read(&smc->conn.bytes_to_rcv))
1210 				mask |= EPOLLIN | EPOLLRDNORM;
1211 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1212 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1213 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1214 				mask |= EPOLLIN;
1215 		}
1216 
1217 	}
1218 	release_sock(sk);
1219 	sock_put(sk);
1220 
1221 	return mask;
1222 }
1223 
1224 static int smc_shutdown(struct socket *sock, int how)
1225 {
1226 	struct sock *sk = sock->sk;
1227 	struct smc_sock *smc;
1228 	int rc = -EINVAL;
1229 	int rc1 = 0;
1230 
1231 	smc = smc_sk(sk);
1232 
1233 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1234 		return rc;
1235 
1236 	lock_sock(sk);
1237 
1238 	rc = -ENOTCONN;
1239 	if ((sk->sk_state != SMC_LISTEN) &&
1240 	    (sk->sk_state != SMC_ACTIVE) &&
1241 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1242 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1243 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1244 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1245 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1246 		goto out;
1247 	if (smc->use_fallback) {
1248 		rc = kernel_sock_shutdown(smc->clcsock, how);
1249 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1250 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1251 			sk->sk_state = SMC_CLOSED;
1252 		goto out;
1253 	}
1254 	switch (how) {
1255 	case SHUT_RDWR:		/* shutdown in both directions */
1256 		rc = smc_close_active(smc);
1257 		break;
1258 	case SHUT_WR:
1259 		rc = smc_close_shutdown_write(smc);
1260 		break;
1261 	case SHUT_RD:
1262 		if (sk->sk_state == SMC_LISTEN)
1263 			rc = smc_close_active(smc);
1264 		else
1265 			rc = 0;
1266 			/* nothing more to do because peer is not involved */
1267 		break;
1268 	}
1269 	rc1 = kernel_sock_shutdown(smc->clcsock, how);
1270 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1271 	sk->sk_shutdown |= how + 1;
1272 
1273 out:
1274 	release_sock(sk);
1275 	return rc ? rc : rc1;
1276 }
1277 
1278 static int smc_setsockopt(struct socket *sock, int level, int optname,
1279 			  char __user *optval, unsigned int optlen)
1280 {
1281 	struct sock *sk = sock->sk;
1282 	struct smc_sock *smc;
1283 
1284 	smc = smc_sk(sk);
1285 
1286 	/* generic setsockopts reaching us here always apply to the
1287 	 * CLC socket
1288 	 */
1289 	return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1290 					     optval, optlen);
1291 }
1292 
1293 static int smc_getsockopt(struct socket *sock, int level, int optname,
1294 			  char __user *optval, int __user *optlen)
1295 {
1296 	struct smc_sock *smc;
1297 
1298 	smc = smc_sk(sock->sk);
1299 	/* socket options apply to the CLC socket */
1300 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1301 					     optval, optlen);
1302 }
1303 
1304 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1305 		     unsigned long arg)
1306 {
1307 	struct smc_sock *smc;
1308 
1309 	smc = smc_sk(sock->sk);
1310 	if (smc->use_fallback)
1311 		return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1312 	else
1313 		return sock_no_ioctl(sock, cmd, arg);
1314 }
1315 
1316 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1317 			    int offset, size_t size, int flags)
1318 {
1319 	struct sock *sk = sock->sk;
1320 	struct smc_sock *smc;
1321 	int rc = -EPIPE;
1322 
1323 	smc = smc_sk(sk);
1324 	lock_sock(sk);
1325 	if (sk->sk_state != SMC_ACTIVE)
1326 		goto out;
1327 	if (smc->use_fallback)
1328 		rc = kernel_sendpage(smc->clcsock, page, offset,
1329 				     size, flags);
1330 	else
1331 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1332 
1333 out:
1334 	release_sock(sk);
1335 	return rc;
1336 }
1337 
1338 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1339 			       struct pipe_inode_info *pipe, size_t len,
1340 				    unsigned int flags)
1341 {
1342 	struct sock *sk = sock->sk;
1343 	struct smc_sock *smc;
1344 	int rc = -ENOTCONN;
1345 
1346 	smc = smc_sk(sk);
1347 	lock_sock(sk);
1348 	if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1349 		goto out;
1350 	if (smc->use_fallback) {
1351 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1352 						    pipe, len, flags);
1353 	} else {
1354 		rc = -EOPNOTSUPP;
1355 	}
1356 out:
1357 	release_sock(sk);
1358 	return rc;
1359 }
1360 
1361 /* must look like tcp */
1362 static const struct proto_ops smc_sock_ops = {
1363 	.family		= PF_SMC,
1364 	.owner		= THIS_MODULE,
1365 	.release	= smc_release,
1366 	.bind		= smc_bind,
1367 	.connect	= smc_connect,
1368 	.socketpair	= sock_no_socketpair,
1369 	.accept		= smc_accept,
1370 	.getname	= smc_getname,
1371 	.poll		= smc_poll,
1372 	.ioctl		= smc_ioctl,
1373 	.listen		= smc_listen,
1374 	.shutdown	= smc_shutdown,
1375 	.setsockopt	= smc_setsockopt,
1376 	.getsockopt	= smc_getsockopt,
1377 	.sendmsg	= smc_sendmsg,
1378 	.recvmsg	= smc_recvmsg,
1379 	.mmap		= sock_no_mmap,
1380 	.sendpage	= smc_sendpage,
1381 	.splice_read	= smc_splice_read,
1382 };
1383 
1384 static int smc_create(struct net *net, struct socket *sock, int protocol,
1385 		      int kern)
1386 {
1387 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1388 	struct smc_sock *smc;
1389 	struct sock *sk;
1390 	int rc;
1391 
1392 	rc = -ESOCKTNOSUPPORT;
1393 	if (sock->type != SOCK_STREAM)
1394 		goto out;
1395 
1396 	rc = -EPROTONOSUPPORT;
1397 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1398 		goto out;
1399 
1400 	rc = -ENOBUFS;
1401 	sock->ops = &smc_sock_ops;
1402 	sk = smc_sock_alloc(net, sock, protocol);
1403 	if (!sk)
1404 		goto out;
1405 
1406 	/* create internal TCP socket for CLC handshake and fallback */
1407 	smc = smc_sk(sk);
1408 	smc->use_fallback = false; /* assume rdma capability first */
1409 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1410 			      &smc->clcsock);
1411 	if (rc) {
1412 		sk_common_release(sk);
1413 		goto out;
1414 	}
1415 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1416 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1417 
1418 out:
1419 	return rc;
1420 }
1421 
1422 static const struct net_proto_family smc_sock_family_ops = {
1423 	.family	= PF_SMC,
1424 	.owner	= THIS_MODULE,
1425 	.create	= smc_create,
1426 };
1427 
1428 static int __init smc_init(void)
1429 {
1430 	int rc;
1431 
1432 	rc = smc_pnet_init();
1433 	if (rc)
1434 		return rc;
1435 
1436 	rc = smc_llc_init();
1437 	if (rc) {
1438 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1439 		goto out_pnet;
1440 	}
1441 
1442 	rc = smc_cdc_init();
1443 	if (rc) {
1444 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1445 		goto out_pnet;
1446 	}
1447 
1448 	rc = proto_register(&smc_proto, 1);
1449 	if (rc) {
1450 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
1451 		goto out_pnet;
1452 	}
1453 
1454 	rc = proto_register(&smc_proto6, 1);
1455 	if (rc) {
1456 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
1457 		goto out_proto;
1458 	}
1459 
1460 	rc = sock_register(&smc_sock_family_ops);
1461 	if (rc) {
1462 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
1463 		goto out_proto6;
1464 	}
1465 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1466 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
1467 
1468 	rc = smc_ib_register_client();
1469 	if (rc) {
1470 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
1471 		goto out_sock;
1472 	}
1473 
1474 	static_branch_enable(&tcp_have_smc);
1475 	return 0;
1476 
1477 out_sock:
1478 	sock_unregister(PF_SMC);
1479 out_proto6:
1480 	proto_unregister(&smc_proto6);
1481 out_proto:
1482 	proto_unregister(&smc_proto);
1483 out_pnet:
1484 	smc_pnet_exit();
1485 	return rc;
1486 }
1487 
1488 static void __exit smc_exit(void)
1489 {
1490 	struct smc_link_group *lgr, *lg;
1491 	LIST_HEAD(lgr_freeing_list);
1492 
1493 	spin_lock_bh(&smc_lgr_list.lock);
1494 	if (!list_empty(&smc_lgr_list.list))
1495 		list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1496 	spin_unlock_bh(&smc_lgr_list.lock);
1497 	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1498 		list_del_init(&lgr->list);
1499 		cancel_delayed_work_sync(&lgr->free_work);
1500 		smc_lgr_free(lgr); /* free link group */
1501 	}
1502 	static_branch_disable(&tcp_have_smc);
1503 	smc_ib_unregister_client();
1504 	sock_unregister(PF_SMC);
1505 	proto_unregister(&smc_proto6);
1506 	proto_unregister(&smc_proto);
1507 	smc_pnet_exit();
1508 }
1509 
1510 module_init(smc_init);
1511 module_exit(smc_exit);
1512 
1513 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1514 MODULE_DESCRIPTION("smc socket address family");
1515 MODULE_LICENSE("GPL");
1516 MODULE_ALIAS_NETPROTO(PF_SMC);
1517