xref: /openbmc/linux/net/smc/af_smc.c (revision aac5987a)
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  *    - non-blocking connect postponed
11  *    - IPv6 support postponed
12  *    - support for alternate links postponed
13  *    - partial support for non-blocking sockets only
14  *    - support for urgent data postponed
15  *
16  *  Copyright IBM Corp. 2016
17  *
18  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
19  *              based on prototype from Frank Blaschka
20  */
21 
22 #define KMSG_COMPONENT "smc"
23 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
24 
25 #include <linux/module.h>
26 #include <linux/socket.h>
27 #include <linux/inetdevice.h>
28 #include <linux/workqueue.h>
29 #include <linux/in.h>
30 #include <linux/sched/signal.h>
31 
32 #include <net/sock.h>
33 #include <net/tcp.h>
34 #include <net/smc.h>
35 
36 #include "smc.h"
37 #include "smc_clc.h"
38 #include "smc_llc.h"
39 #include "smc_cdc.h"
40 #include "smc_core.h"
41 #include "smc_ib.h"
42 #include "smc_pnet.h"
43 #include "smc_tx.h"
44 #include "smc_rx.h"
45 #include "smc_close.h"
46 
47 static DEFINE_MUTEX(smc_create_lgr_pending);	/* serialize link group
48 						 * creation
49 						 */
50 
51 struct smc_lgr_list smc_lgr_list = {		/* established link groups */
52 	.lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
53 	.list = LIST_HEAD_INIT(smc_lgr_list.list),
54 };
55 
56 static void smc_tcp_listen_work(struct work_struct *);
57 
58 static void smc_set_keepalive(struct sock *sk, int val)
59 {
60 	struct smc_sock *smc = smc_sk(sk);
61 
62 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
63 }
64 
65 static struct smc_hashinfo smc_v4_hashinfo = {
66 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
67 };
68 
69 int smc_hash_sk(struct sock *sk)
70 {
71 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
72 	struct hlist_head *head;
73 
74 	head = &h->ht;
75 
76 	write_lock_bh(&h->lock);
77 	sk_add_node(sk, head);
78 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
79 	write_unlock_bh(&h->lock);
80 
81 	return 0;
82 }
83 EXPORT_SYMBOL_GPL(smc_hash_sk);
84 
85 void smc_unhash_sk(struct sock *sk)
86 {
87 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
88 
89 	write_lock_bh(&h->lock);
90 	if (sk_del_node_init(sk))
91 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
92 	write_unlock_bh(&h->lock);
93 }
94 EXPORT_SYMBOL_GPL(smc_unhash_sk);
95 
96 struct proto smc_proto = {
97 	.name		= "SMC",
98 	.owner		= THIS_MODULE,
99 	.keepalive	= smc_set_keepalive,
100 	.hash		= smc_hash_sk,
101 	.unhash		= smc_unhash_sk,
102 	.obj_size	= sizeof(struct smc_sock),
103 	.h.smc_hash	= &smc_v4_hashinfo,
104 	.slab_flags	= SLAB_DESTROY_BY_RCU,
105 };
106 EXPORT_SYMBOL_GPL(smc_proto);
107 
108 static int smc_release(struct socket *sock)
109 {
110 	struct sock *sk = sock->sk;
111 	struct smc_sock *smc;
112 	int rc = 0;
113 
114 	if (!sk)
115 		goto out;
116 
117 	smc = smc_sk(sk);
118 	sock_hold(sk);
119 	if (sk->sk_state == SMC_LISTEN)
120 		/* smc_close_non_accepted() is called and acquires
121 		 * sock lock for child sockets again
122 		 */
123 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
124 	else
125 		lock_sock(sk);
126 
127 	if (smc->use_fallback) {
128 		sk->sk_state = SMC_CLOSED;
129 		sk->sk_state_change(sk);
130 	} else {
131 		rc = smc_close_active(smc);
132 		sock_set_flag(sk, SOCK_DEAD);
133 		sk->sk_shutdown |= SHUTDOWN_MASK;
134 	}
135 	if (smc->clcsock) {
136 		sock_release(smc->clcsock);
137 		smc->clcsock = NULL;
138 	}
139 
140 	/* detach socket */
141 	sock_orphan(sk);
142 	sock->sk = NULL;
143 	if (smc->use_fallback) {
144 		schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
145 	} else if (sk->sk_state == SMC_CLOSED) {
146 		smc_conn_free(&smc->conn);
147 		schedule_delayed_work(&smc->sock_put_work,
148 				      SMC_CLOSE_SOCK_PUT_DELAY);
149 	}
150 	sk->sk_prot->unhash(sk);
151 	release_sock(sk);
152 
153 	sock_put(sk);
154 out:
155 	return rc;
156 }
157 
158 static void smc_destruct(struct sock *sk)
159 {
160 	if (sk->sk_state != SMC_CLOSED)
161 		return;
162 	if (!sock_flag(sk, SOCK_DEAD))
163 		return;
164 
165 	sk_refcnt_debug_dec(sk);
166 }
167 
168 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock)
169 {
170 	struct smc_sock *smc;
171 	struct sock *sk;
172 
173 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, &smc_proto, 0);
174 	if (!sk)
175 		return NULL;
176 
177 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
178 	sk->sk_state = SMC_INIT;
179 	sk->sk_destruct = smc_destruct;
180 	sk->sk_protocol = SMCPROTO_SMC;
181 	smc = smc_sk(sk);
182 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
183 	INIT_LIST_HEAD(&smc->accept_q);
184 	spin_lock_init(&smc->accept_q_lock);
185 	INIT_DELAYED_WORK(&smc->sock_put_work, smc_close_sock_put_work);
186 	sk->sk_prot->hash(sk);
187 	sk_refcnt_debug_inc(sk);
188 
189 	return sk;
190 }
191 
192 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
193 		    int addr_len)
194 {
195 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
196 	struct sock *sk = sock->sk;
197 	struct smc_sock *smc;
198 	int rc;
199 
200 	smc = smc_sk(sk);
201 
202 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
203 	rc = -EINVAL;
204 	if (addr_len < sizeof(struct sockaddr_in))
205 		goto out;
206 
207 	rc = -EAFNOSUPPORT;
208 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
209 	if ((addr->sin_family != AF_INET) &&
210 	    ((addr->sin_family != AF_UNSPEC) ||
211 	     (addr->sin_addr.s_addr != htonl(INADDR_ANY))))
212 		goto out;
213 
214 	lock_sock(sk);
215 
216 	/* Check if socket is already active */
217 	rc = -EINVAL;
218 	if (sk->sk_state != SMC_INIT)
219 		goto out_rel;
220 
221 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
222 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
223 
224 out_rel:
225 	release_sock(sk);
226 out:
227 	return rc;
228 }
229 
230 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
231 				   unsigned long mask)
232 {
233 	/* options we don't get control via setsockopt for */
234 	nsk->sk_type = osk->sk_type;
235 	nsk->sk_sndbuf = osk->sk_sndbuf;
236 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
237 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
238 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
239 	nsk->sk_mark = osk->sk_mark;
240 	nsk->sk_priority = osk->sk_priority;
241 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
242 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
243 	nsk->sk_err = osk->sk_err;
244 
245 	nsk->sk_flags &= ~mask;
246 	nsk->sk_flags |= osk->sk_flags & mask;
247 }
248 
249 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
250 			     (1UL << SOCK_KEEPOPEN) | \
251 			     (1UL << SOCK_LINGER) | \
252 			     (1UL << SOCK_BROADCAST) | \
253 			     (1UL << SOCK_TIMESTAMP) | \
254 			     (1UL << SOCK_DBG) | \
255 			     (1UL << SOCK_RCVTSTAMP) | \
256 			     (1UL << SOCK_RCVTSTAMPNS) | \
257 			     (1UL << SOCK_LOCALROUTE) | \
258 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
259 			     (1UL << SOCK_RXQ_OVFL) | \
260 			     (1UL << SOCK_WIFI_STATUS) | \
261 			     (1UL << SOCK_NOFCS) | \
262 			     (1UL << SOCK_FILTER_LOCKED))
263 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
264  * clc socket (since smc is not called for these options from net/core)
265  */
266 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
267 {
268 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
269 }
270 
271 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
272 			     (1UL << SOCK_KEEPOPEN) | \
273 			     (1UL << SOCK_LINGER) | \
274 			     (1UL << SOCK_DBG))
275 /* copy only settings and flags relevant for smc from clc to smc socket */
276 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
277 {
278 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
279 }
280 
281 /* determine subnet and mask of internal TCP socket */
282 int smc_netinfo_by_tcpsk(struct socket *clcsock,
283 			 __be32 *subnet, u8 *prefix_len)
284 {
285 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
286 	struct sockaddr_in addr;
287 	int rc = -ENOENT;
288 	int len;
289 
290 	if (!dst) {
291 		rc = -ENOTCONN;
292 		goto out;
293 	}
294 	if (!dst->dev) {
295 		rc = -ENODEV;
296 		goto out_rel;
297 	}
298 
299 	/* get address to which the internal TCP socket is bound */
300 	kernel_getsockname(clcsock, (struct sockaddr *)&addr, &len);
301 	/* analyze IPv4 specific data of net_device belonging to TCP socket */
302 	for_ifa(dst->dev->ip_ptr) {
303 		if (ifa->ifa_address != addr.sin_addr.s_addr)
304 			continue;
305 		*prefix_len = inet_mask_len(ifa->ifa_mask);
306 		*subnet = ifa->ifa_address & ifa->ifa_mask;
307 		rc = 0;
308 		break;
309 	} endfor_ifa(dst->dev->ip_ptr);
310 
311 out_rel:
312 	dst_release(dst);
313 out:
314 	return rc;
315 }
316 
317 static int smc_clnt_conf_first_link(struct smc_sock *smc, union ib_gid *gid)
318 {
319 	struct smc_link_group *lgr = smc->conn.lgr;
320 	struct smc_link *link;
321 	int rest;
322 	int rc;
323 
324 	link = &lgr->lnk[SMC_SINGLE_LINK];
325 	/* receive CONFIRM LINK request from server over RoCE fabric */
326 	rest = wait_for_completion_interruptible_timeout(
327 		&link->llc_confirm,
328 		SMC_LLC_WAIT_FIRST_TIME);
329 	if (rest <= 0) {
330 		struct smc_clc_msg_decline dclc;
331 
332 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
333 				      SMC_CLC_DECLINE);
334 		return rc;
335 	}
336 
337 	rc = smc_ib_modify_qp_rts(link);
338 	if (rc)
339 		return SMC_CLC_DECL_INTERR;
340 
341 	smc_wr_remember_qp_attr(link);
342 	/* send CONFIRM LINK response over RoCE fabric */
343 	rc = smc_llc_send_confirm_link(link,
344 				       link->smcibdev->mac[link->ibport - 1],
345 				       gid, SMC_LLC_RESP);
346 	if (rc < 0)
347 		return SMC_CLC_DECL_TCL;
348 
349 	return rc;
350 }
351 
352 static void smc_conn_save_peer_info(struct smc_sock *smc,
353 				    struct smc_clc_msg_accept_confirm *clc)
354 {
355 	smc->conn.peer_conn_idx = clc->conn_idx;
356 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
357 	smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
358 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
359 }
360 
361 static void smc_link_save_peer_info(struct smc_link *link,
362 				    struct smc_clc_msg_accept_confirm *clc)
363 {
364 	link->peer_qpn = ntoh24(clc->qpn);
365 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
366 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
367 	link->peer_psn = ntoh24(clc->psn);
368 	link->peer_mtu = clc->qp_mtu;
369 }
370 
371 /* setup for RDMA connection of client */
372 static int smc_connect_rdma(struct smc_sock *smc)
373 {
374 	struct sockaddr_in *inaddr = (struct sockaddr_in *)smc->addr;
375 	struct smc_clc_msg_accept_confirm aclc;
376 	int local_contact = SMC_FIRST_CONTACT;
377 	struct smc_ib_device *smcibdev;
378 	struct smc_link *link;
379 	u8 srv_first_contact;
380 	int reason_code = 0;
381 	int rc = 0;
382 	u8 ibport;
383 
384 	/* IPSec connections opt out of SMC-R optimizations */
385 	if (using_ipsec(smc)) {
386 		reason_code = SMC_CLC_DECL_IPSEC;
387 		goto decline_rdma;
388 	}
389 
390 	/* PNET table look up: search active ib_device and port
391 	 * within same PNETID that also contains the ethernet device
392 	 * used for the internal TCP socket
393 	 */
394 	smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
395 	if (!smcibdev) {
396 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
397 		goto decline_rdma;
398 	}
399 
400 	/* do inband token exchange */
401 	reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
402 	if (reason_code < 0) {
403 		rc = reason_code;
404 		goto out_err;
405 	}
406 	if (reason_code > 0) /* configuration error */
407 		goto decline_rdma;
408 	/* receive SMC Accept CLC message */
409 	reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
410 				       SMC_CLC_ACCEPT);
411 	if (reason_code < 0) {
412 		rc = reason_code;
413 		goto out_err;
414 	}
415 	if (reason_code > 0)
416 		goto decline_rdma;
417 
418 	srv_first_contact = aclc.hdr.flag;
419 	mutex_lock(&smc_create_lgr_pending);
420 	local_contact = smc_conn_create(smc, inaddr->sin_addr.s_addr, smcibdev,
421 					ibport, &aclc.lcl, srv_first_contact);
422 	if (local_contact < 0) {
423 		rc = local_contact;
424 		if (rc == -ENOMEM)
425 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
426 		else if (rc == -ENOLINK)
427 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
428 		goto decline_rdma_unlock;
429 	}
430 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
431 
432 	smc_conn_save_peer_info(smc, &aclc);
433 
434 	rc = smc_sndbuf_create(smc);
435 	if (rc) {
436 		reason_code = SMC_CLC_DECL_MEM;
437 		goto decline_rdma_unlock;
438 	}
439 	rc = smc_rmb_create(smc);
440 	if (rc) {
441 		reason_code = SMC_CLC_DECL_MEM;
442 		goto decline_rdma_unlock;
443 	}
444 
445 	if (local_contact == SMC_FIRST_CONTACT)
446 		smc_link_save_peer_info(link, &aclc);
447 
448 	rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
449 	if (rc) {
450 		reason_code = SMC_CLC_DECL_INTERR;
451 		goto decline_rdma_unlock;
452 	}
453 
454 	if (local_contact == SMC_FIRST_CONTACT) {
455 		rc = smc_ib_ready_link(link);
456 		if (rc) {
457 			reason_code = SMC_CLC_DECL_INTERR;
458 			goto decline_rdma_unlock;
459 		}
460 	}
461 
462 	rc = smc_clc_send_confirm(smc);
463 	if (rc)
464 		goto out_err_unlock;
465 
466 	if (local_contact == SMC_FIRST_CONTACT) {
467 		/* QP confirmation over RoCE fabric */
468 		reason_code = smc_clnt_conf_first_link(
469 			smc, &smcibdev->gid[ibport - 1]);
470 		if (reason_code < 0) {
471 			rc = reason_code;
472 			goto out_err_unlock;
473 		}
474 		if (reason_code > 0)
475 			goto decline_rdma_unlock;
476 	}
477 
478 	mutex_unlock(&smc_create_lgr_pending);
479 	smc_tx_init(smc);
480 	smc_rx_init(smc);
481 
482 out_connected:
483 	smc_copy_sock_settings_to_clc(smc);
484 	if (smc->sk.sk_state == SMC_INIT)
485 		smc->sk.sk_state = SMC_ACTIVE;
486 
487 	return rc ? rc : local_contact;
488 
489 decline_rdma_unlock:
490 	mutex_unlock(&smc_create_lgr_pending);
491 	smc_conn_free(&smc->conn);
492 decline_rdma:
493 	/* RDMA setup failed, switch back to TCP */
494 	smc->use_fallback = true;
495 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
496 		rc = smc_clc_send_decline(smc, reason_code, 0);
497 		if (rc < sizeof(struct smc_clc_msg_decline))
498 			goto out_err;
499 	}
500 	goto out_connected;
501 
502 out_err_unlock:
503 	mutex_unlock(&smc_create_lgr_pending);
504 	smc_conn_free(&smc->conn);
505 out_err:
506 	return rc;
507 }
508 
509 static int smc_connect(struct socket *sock, struct sockaddr *addr,
510 		       int alen, int flags)
511 {
512 	struct sock *sk = sock->sk;
513 	struct smc_sock *smc;
514 	int rc = -EINVAL;
515 
516 	smc = smc_sk(sk);
517 
518 	/* separate smc parameter checking to be safe */
519 	if (alen < sizeof(addr->sa_family))
520 		goto out_err;
521 	if (addr->sa_family != AF_INET)
522 		goto out_err;
523 	smc->addr = addr;	/* needed for nonblocking connect */
524 
525 	lock_sock(sk);
526 	switch (sk->sk_state) {
527 	default:
528 		goto out;
529 	case SMC_ACTIVE:
530 		rc = -EISCONN;
531 		goto out;
532 	case SMC_INIT:
533 		rc = 0;
534 		break;
535 	}
536 
537 	smc_copy_sock_settings_to_clc(smc);
538 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
539 	if (rc)
540 		goto out;
541 
542 	/* setup RDMA connection */
543 	rc = smc_connect_rdma(smc);
544 	if (rc < 0)
545 		goto out;
546 	else
547 		rc = 0; /* success cases including fallback */
548 
549 out:
550 	release_sock(sk);
551 out_err:
552 	return rc;
553 }
554 
555 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
556 {
557 	struct sock *sk = &lsmc->sk;
558 	struct socket *new_clcsock;
559 	struct sock *new_sk;
560 	int rc;
561 
562 	release_sock(&lsmc->sk);
563 	new_sk = smc_sock_alloc(sock_net(sk), NULL);
564 	if (!new_sk) {
565 		rc = -ENOMEM;
566 		lsmc->sk.sk_err = ENOMEM;
567 		*new_smc = NULL;
568 		lock_sock(&lsmc->sk);
569 		goto out;
570 	}
571 	*new_smc = smc_sk(new_sk);
572 
573 	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
574 	lock_sock(&lsmc->sk);
575 	if  (rc < 0) {
576 		lsmc->sk.sk_err = -rc;
577 		new_sk->sk_state = SMC_CLOSED;
578 		sock_set_flag(new_sk, SOCK_DEAD);
579 		sk->sk_prot->unhash(new_sk);
580 		sock_put(new_sk);
581 		*new_smc = NULL;
582 		goto out;
583 	}
584 	if (lsmc->sk.sk_state == SMC_CLOSED) {
585 		if (new_clcsock)
586 			sock_release(new_clcsock);
587 		new_sk->sk_state = SMC_CLOSED;
588 		sock_set_flag(new_sk, SOCK_DEAD);
589 		sk->sk_prot->unhash(new_sk);
590 		sock_put(new_sk);
591 		*new_smc = NULL;
592 		goto out;
593 	}
594 
595 	(*new_smc)->clcsock = new_clcsock;
596 out:
597 	return rc;
598 }
599 
600 /* add a just created sock to the accept queue of the listen sock as
601  * candidate for a following socket accept call from user space
602  */
603 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
604 {
605 	struct smc_sock *par = smc_sk(parent);
606 
607 	sock_hold(sk);
608 	spin_lock(&par->accept_q_lock);
609 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
610 	spin_unlock(&par->accept_q_lock);
611 	sk_acceptq_added(parent);
612 }
613 
614 /* remove a socket from the accept queue of its parental listening socket */
615 static void smc_accept_unlink(struct sock *sk)
616 {
617 	struct smc_sock *par = smc_sk(sk)->listen_smc;
618 
619 	spin_lock(&par->accept_q_lock);
620 	list_del_init(&smc_sk(sk)->accept_q);
621 	spin_unlock(&par->accept_q_lock);
622 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
623 	sock_put(sk);
624 }
625 
626 /* remove a sock from the accept queue to bind it to a new socket created
627  * for a socket accept call from user space
628  */
629 struct sock *smc_accept_dequeue(struct sock *parent,
630 				struct socket *new_sock)
631 {
632 	struct smc_sock *isk, *n;
633 	struct sock *new_sk;
634 
635 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
636 		new_sk = (struct sock *)isk;
637 
638 		smc_accept_unlink(new_sk);
639 		if (new_sk->sk_state == SMC_CLOSED) {
640 			/* tbd in follow-on patch: close this sock */
641 			continue;
642 		}
643 		if (new_sock)
644 			sock_graft(new_sk, new_sock);
645 		return new_sk;
646 	}
647 	return NULL;
648 }
649 
650 /* clean up for a created but never accepted sock */
651 void smc_close_non_accepted(struct sock *sk)
652 {
653 	struct smc_sock *smc = smc_sk(sk);
654 
655 	sock_hold(sk);
656 	lock_sock(sk);
657 	if (!sk->sk_lingertime)
658 		/* wait for peer closing */
659 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
660 	if (!smc->use_fallback)
661 		smc_close_active(smc);
662 	if (smc->clcsock) {
663 		struct socket *tcp;
664 
665 		tcp = smc->clcsock;
666 		smc->clcsock = NULL;
667 		sock_release(tcp);
668 	}
669 	sock_set_flag(sk, SOCK_DEAD);
670 	sk->sk_shutdown |= SHUTDOWN_MASK;
671 	if (smc->use_fallback) {
672 		schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
673 	} else {
674 		smc_conn_free(&smc->conn);
675 		schedule_delayed_work(&smc->sock_put_work,
676 				      SMC_CLOSE_SOCK_PUT_DELAY);
677 	}
678 	release_sock(sk);
679 	sock_put(sk);
680 }
681 
682 static int smc_serv_conf_first_link(struct smc_sock *smc)
683 {
684 	struct smc_link_group *lgr = smc->conn.lgr;
685 	struct smc_link *link;
686 	int rest;
687 	int rc;
688 
689 	link = &lgr->lnk[SMC_SINGLE_LINK];
690 	/* send CONFIRM LINK request to client over the RoCE fabric */
691 	rc = smc_llc_send_confirm_link(link,
692 				       link->smcibdev->mac[link->ibport - 1],
693 				       &link->smcibdev->gid[link->ibport - 1],
694 				       SMC_LLC_REQ);
695 	if (rc < 0)
696 		return SMC_CLC_DECL_TCL;
697 
698 	/* receive CONFIRM LINK response from client over the RoCE fabric */
699 	rest = wait_for_completion_interruptible_timeout(
700 		&link->llc_confirm_resp,
701 		SMC_LLC_WAIT_FIRST_TIME);
702 	if (rest <= 0) {
703 		struct smc_clc_msg_decline dclc;
704 
705 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
706 				      SMC_CLC_DECLINE);
707 	}
708 
709 	return rc;
710 }
711 
712 /* setup for RDMA connection of server */
713 static void smc_listen_work(struct work_struct *work)
714 {
715 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
716 						smc_listen_work);
717 	struct socket *newclcsock = new_smc->clcsock;
718 	struct smc_sock *lsmc = new_smc->listen_smc;
719 	struct smc_clc_msg_accept_confirm cclc;
720 	int local_contact = SMC_REUSE_CONTACT;
721 	struct sock *newsmcsk = &new_smc->sk;
722 	struct smc_clc_msg_proposal pclc;
723 	struct smc_ib_device *smcibdev;
724 	struct sockaddr_in peeraddr;
725 	struct smc_link *link;
726 	int reason_code = 0;
727 	int rc = 0, len;
728 	__be32 subnet;
729 	u8 prefix_len;
730 	u8 ibport;
731 
732 	/* do inband token exchange -
733 	 *wait for and receive SMC Proposal CLC message
734 	 */
735 	reason_code = smc_clc_wait_msg(new_smc, &pclc, sizeof(pclc),
736 				       SMC_CLC_PROPOSAL);
737 	if (reason_code < 0)
738 		goto out_err;
739 	if (reason_code > 0)
740 		goto decline_rdma;
741 
742 	/* IPSec connections opt out of SMC-R optimizations */
743 	if (using_ipsec(new_smc)) {
744 		reason_code = SMC_CLC_DECL_IPSEC;
745 		goto decline_rdma;
746 	}
747 
748 	/* PNET table look up: search active ib_device and port
749 	 * within same PNETID that also contains the ethernet device
750 	 * used for the internal TCP socket
751 	 */
752 	smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
753 	if (!smcibdev) {
754 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
755 		goto decline_rdma;
756 	}
757 
758 	/* determine subnet and mask from internal TCP socket */
759 	rc = smc_netinfo_by_tcpsk(newclcsock, &subnet, &prefix_len);
760 	if (rc) {
761 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
762 		goto decline_rdma;
763 	}
764 	if ((pclc.outgoing_subnet != subnet) ||
765 	    (pclc.prefix_len != prefix_len)) {
766 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
767 		goto decline_rdma;
768 	}
769 
770 	/* get address of the peer connected to the internal TCP socket */
771 	kernel_getpeername(newclcsock, (struct sockaddr *)&peeraddr, &len);
772 
773 	/* allocate connection / link group */
774 	mutex_lock(&smc_create_lgr_pending);
775 	local_contact = smc_conn_create(new_smc, peeraddr.sin_addr.s_addr,
776 					smcibdev, ibport, &pclc.lcl, 0);
777 	if (local_contact == SMC_REUSE_CONTACT)
778 		/* lock no longer needed, free it due to following
779 		 * smc_clc_wait_msg() call
780 		 */
781 		mutex_unlock(&smc_create_lgr_pending);
782 	if (local_contact < 0) {
783 		rc = local_contact;
784 		if (rc == -ENOMEM)
785 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
786 		else if (rc == -ENOLINK)
787 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
788 		goto decline_rdma;
789 	}
790 	link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
791 
792 	rc = smc_sndbuf_create(new_smc);
793 	if (rc) {
794 		reason_code = SMC_CLC_DECL_MEM;
795 		goto decline_rdma;
796 	}
797 	rc = smc_rmb_create(new_smc);
798 	if (rc) {
799 		reason_code = SMC_CLC_DECL_MEM;
800 		goto decline_rdma;
801 	}
802 
803 	rc = smc_clc_send_accept(new_smc, local_contact);
804 	if (rc)
805 		goto out_err;
806 
807 	/* receive SMC Confirm CLC message */
808 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
809 				       SMC_CLC_CONFIRM);
810 	if (reason_code < 0)
811 		goto out_err;
812 	if (reason_code > 0)
813 		goto decline_rdma;
814 	smc_conn_save_peer_info(new_smc, &cclc);
815 	if (local_contact == SMC_FIRST_CONTACT)
816 		smc_link_save_peer_info(link, &cclc);
817 
818 	rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
819 	if (rc) {
820 		reason_code = SMC_CLC_DECL_INTERR;
821 		goto decline_rdma;
822 	}
823 
824 	if (local_contact == SMC_FIRST_CONTACT) {
825 		rc = smc_ib_ready_link(link);
826 		if (rc) {
827 			reason_code = SMC_CLC_DECL_INTERR;
828 			goto decline_rdma;
829 		}
830 		/* QP confirmation over RoCE fabric */
831 		reason_code = smc_serv_conf_first_link(new_smc);
832 		if (reason_code < 0) {
833 			/* peer is not aware of a problem */
834 			rc = reason_code;
835 			goto out_err;
836 		}
837 		if (reason_code > 0)
838 			goto decline_rdma;
839 	}
840 
841 	smc_tx_init(new_smc);
842 	smc_rx_init(new_smc);
843 
844 out_connected:
845 	sk_refcnt_debug_inc(newsmcsk);
846 	if (newsmcsk->sk_state == SMC_INIT)
847 		newsmcsk->sk_state = SMC_ACTIVE;
848 enqueue:
849 	if (local_contact == SMC_FIRST_CONTACT)
850 		mutex_unlock(&smc_create_lgr_pending);
851 	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
852 	if (lsmc->sk.sk_state == SMC_LISTEN) {
853 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
854 	} else { /* no longer listening */
855 		smc_close_non_accepted(newsmcsk);
856 	}
857 	release_sock(&lsmc->sk);
858 
859 	/* Wake up accept */
860 	lsmc->sk.sk_data_ready(&lsmc->sk);
861 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
862 	return;
863 
864 decline_rdma:
865 	/* RDMA setup failed, switch back to TCP */
866 	smc_conn_free(&new_smc->conn);
867 	new_smc->use_fallback = true;
868 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
869 		rc = smc_clc_send_decline(new_smc, reason_code, 0);
870 		if (rc < sizeof(struct smc_clc_msg_decline))
871 			goto out_err;
872 	}
873 	goto out_connected;
874 
875 out_err:
876 	newsmcsk->sk_state = SMC_CLOSED;
877 	smc_conn_free(&new_smc->conn);
878 	goto enqueue; /* queue new sock with sk_err set */
879 }
880 
881 static void smc_tcp_listen_work(struct work_struct *work)
882 {
883 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
884 					     tcp_listen_work);
885 	struct smc_sock *new_smc;
886 	int rc = 0;
887 
888 	lock_sock(&lsmc->sk);
889 	while (lsmc->sk.sk_state == SMC_LISTEN) {
890 		rc = smc_clcsock_accept(lsmc, &new_smc);
891 		if (rc)
892 			goto out;
893 		if (!new_smc)
894 			continue;
895 
896 		new_smc->listen_smc = lsmc;
897 		new_smc->use_fallback = false; /* assume rdma capability first*/
898 		sock_hold(&lsmc->sk); /* sock_put in smc_listen_work */
899 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
900 		smc_copy_sock_settings_to_smc(new_smc);
901 		schedule_work(&new_smc->smc_listen_work);
902 	}
903 
904 out:
905 	release_sock(&lsmc->sk);
906 	lsmc->sk.sk_data_ready(&lsmc->sk); /* no more listening, wake accept */
907 }
908 
909 static int smc_listen(struct socket *sock, int backlog)
910 {
911 	struct sock *sk = sock->sk;
912 	struct smc_sock *smc;
913 	int rc;
914 
915 	smc = smc_sk(sk);
916 	lock_sock(sk);
917 
918 	rc = -EINVAL;
919 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
920 		goto out;
921 
922 	rc = 0;
923 	if (sk->sk_state == SMC_LISTEN) {
924 		sk->sk_max_ack_backlog = backlog;
925 		goto out;
926 	}
927 	/* some socket options are handled in core, so we could not apply
928 	 * them to the clc socket -- copy smc socket options to clc socket
929 	 */
930 	smc_copy_sock_settings_to_clc(smc);
931 
932 	rc = kernel_listen(smc->clcsock, backlog);
933 	if (rc)
934 		goto out;
935 	sk->sk_max_ack_backlog = backlog;
936 	sk->sk_ack_backlog = 0;
937 	sk->sk_state = SMC_LISTEN;
938 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
939 	schedule_work(&smc->tcp_listen_work);
940 
941 out:
942 	release_sock(sk);
943 	return rc;
944 }
945 
946 static int smc_accept(struct socket *sock, struct socket *new_sock,
947 		      int flags)
948 {
949 	struct sock *sk = sock->sk, *nsk;
950 	DECLARE_WAITQUEUE(wait, current);
951 	struct smc_sock *lsmc;
952 	long timeo;
953 	int rc = 0;
954 
955 	lsmc = smc_sk(sk);
956 	lock_sock(sk);
957 
958 	if (lsmc->sk.sk_state != SMC_LISTEN) {
959 		rc = -EINVAL;
960 		goto out;
961 	}
962 
963 	/* Wait for an incoming connection */
964 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
965 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
966 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
967 		set_current_state(TASK_INTERRUPTIBLE);
968 		if (!timeo) {
969 			rc = -EAGAIN;
970 			break;
971 		}
972 		release_sock(sk);
973 		timeo = schedule_timeout(timeo);
974 		/* wakeup by sk_data_ready in smc_listen_work() */
975 		sched_annotate_sleep();
976 		lock_sock(sk);
977 		if (signal_pending(current)) {
978 			rc = sock_intr_errno(timeo);
979 			break;
980 		}
981 	}
982 	set_current_state(TASK_RUNNING);
983 	remove_wait_queue(sk_sleep(sk), &wait);
984 
985 	if (!rc)
986 		rc = sock_error(nsk);
987 
988 out:
989 	release_sock(sk);
990 	return rc;
991 }
992 
993 static int smc_getname(struct socket *sock, struct sockaddr *addr,
994 		       int *len, int peer)
995 {
996 	struct smc_sock *smc;
997 
998 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
999 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1000 		return -ENOTCONN;
1001 
1002 	smc = smc_sk(sock->sk);
1003 
1004 	return smc->clcsock->ops->getname(smc->clcsock, addr, len, peer);
1005 }
1006 
1007 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1008 {
1009 	struct sock *sk = sock->sk;
1010 	struct smc_sock *smc;
1011 	int rc = -EPIPE;
1012 
1013 	smc = smc_sk(sk);
1014 	lock_sock(sk);
1015 	if ((sk->sk_state != SMC_ACTIVE) &&
1016 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1017 	    (sk->sk_state != SMC_INIT))
1018 		goto out;
1019 	if (smc->use_fallback)
1020 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1021 	else
1022 		rc = smc_tx_sendmsg(smc, msg, len);
1023 out:
1024 	release_sock(sk);
1025 	return rc;
1026 }
1027 
1028 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1029 		       int flags)
1030 {
1031 	struct sock *sk = sock->sk;
1032 	struct smc_sock *smc;
1033 	int rc = -ENOTCONN;
1034 
1035 	smc = smc_sk(sk);
1036 	lock_sock(sk);
1037 	if ((sk->sk_state == SMC_INIT) ||
1038 	    (sk->sk_state == SMC_LISTEN) ||
1039 	    (sk->sk_state == SMC_CLOSED))
1040 		goto out;
1041 
1042 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1043 		rc = 0;
1044 		goto out;
1045 	}
1046 
1047 	if (smc->use_fallback)
1048 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1049 	else
1050 		rc = smc_rx_recvmsg(smc, msg, len, flags);
1051 
1052 out:
1053 	release_sock(sk);
1054 	return rc;
1055 }
1056 
1057 static unsigned int smc_accept_poll(struct sock *parent)
1058 {
1059 	struct smc_sock *isk;
1060 	struct sock *sk;
1061 
1062 	lock_sock(parent);
1063 	list_for_each_entry(isk, &smc_sk(parent)->accept_q, accept_q) {
1064 		sk = (struct sock *)isk;
1065 
1066 		if (sk->sk_state == SMC_ACTIVE) {
1067 			release_sock(parent);
1068 			return POLLIN | POLLRDNORM;
1069 		}
1070 	}
1071 	release_sock(parent);
1072 
1073 	return 0;
1074 }
1075 
1076 static unsigned int smc_poll(struct file *file, struct socket *sock,
1077 			     poll_table *wait)
1078 {
1079 	struct sock *sk = sock->sk;
1080 	unsigned int mask = 0;
1081 	struct smc_sock *smc;
1082 	int rc;
1083 
1084 	smc = smc_sk(sock->sk);
1085 	if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1086 		/* delegate to CLC child sock */
1087 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1088 		/* if non-blocking connect finished ... */
1089 		lock_sock(sk);
1090 		if ((sk->sk_state == SMC_INIT) && (mask & POLLOUT)) {
1091 			sk->sk_err = smc->clcsock->sk->sk_err;
1092 			if (sk->sk_err) {
1093 				mask |= POLLERR;
1094 			} else {
1095 				rc = smc_connect_rdma(smc);
1096 				if (rc < 0)
1097 					mask |= POLLERR;
1098 				else
1099 					/* success cases including fallback */
1100 					mask |= POLLOUT | POLLWRNORM;
1101 			}
1102 		}
1103 		release_sock(sk);
1104 	} else {
1105 		sock_poll_wait(file, sk_sleep(sk), wait);
1106 		if (sk->sk_state == SMC_LISTEN)
1107 			/* woken up by sk_data_ready in smc_listen_work() */
1108 			mask |= smc_accept_poll(sk);
1109 		if (sk->sk_err)
1110 			mask |= POLLERR;
1111 		if (atomic_read(&smc->conn.sndbuf_space) ||
1112 		    (sk->sk_shutdown & SEND_SHUTDOWN)) {
1113 			mask |= POLLOUT | POLLWRNORM;
1114 		} else {
1115 			sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1116 			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1117 		}
1118 		if (atomic_read(&smc->conn.bytes_to_rcv))
1119 			mask |= POLLIN | POLLRDNORM;
1120 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1121 		    (sk->sk_state == SMC_CLOSED))
1122 			mask |= POLLHUP;
1123 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1124 			mask |= POLLIN | POLLRDNORM | POLLRDHUP;
1125 		if (sk->sk_state == SMC_APPCLOSEWAIT1)
1126 			mask |= POLLIN;
1127 
1128 	}
1129 
1130 	return mask;
1131 }
1132 
1133 static int smc_shutdown(struct socket *sock, int how)
1134 {
1135 	struct sock *sk = sock->sk;
1136 	struct smc_sock *smc;
1137 	int rc = -EINVAL;
1138 	int rc1 = 0;
1139 
1140 	smc = smc_sk(sk);
1141 
1142 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1143 		return rc;
1144 
1145 	lock_sock(sk);
1146 
1147 	rc = -ENOTCONN;
1148 	if ((sk->sk_state != SMC_LISTEN) &&
1149 	    (sk->sk_state != SMC_ACTIVE) &&
1150 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1151 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1152 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1153 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1154 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1155 		goto out;
1156 	if (smc->use_fallback) {
1157 		rc = kernel_sock_shutdown(smc->clcsock, how);
1158 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1159 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1160 			sk->sk_state = SMC_CLOSED;
1161 		goto out;
1162 	}
1163 	switch (how) {
1164 	case SHUT_RDWR:		/* shutdown in both directions */
1165 		rc = smc_close_active(smc);
1166 		break;
1167 	case SHUT_WR:
1168 		rc = smc_close_shutdown_write(smc);
1169 		break;
1170 	case SHUT_RD:
1171 		if (sk->sk_state == SMC_LISTEN)
1172 			rc = smc_close_active(smc);
1173 		else
1174 			rc = 0;
1175 			/* nothing more to do because peer is not involved */
1176 		break;
1177 	}
1178 	rc1 = kernel_sock_shutdown(smc->clcsock, how);
1179 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1180 	sk->sk_shutdown |= how + 1;
1181 
1182 out:
1183 	release_sock(sk);
1184 	return rc ? rc : rc1;
1185 }
1186 
1187 static int smc_setsockopt(struct socket *sock, int level, int optname,
1188 			  char __user *optval, unsigned int optlen)
1189 {
1190 	struct sock *sk = sock->sk;
1191 	struct smc_sock *smc;
1192 
1193 	smc = smc_sk(sk);
1194 
1195 	/* generic setsockopts reaching us here always apply to the
1196 	 * CLC socket
1197 	 */
1198 	return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1199 					     optval, optlen);
1200 }
1201 
1202 static int smc_getsockopt(struct socket *sock, int level, int optname,
1203 			  char __user *optval, int __user *optlen)
1204 {
1205 	struct smc_sock *smc;
1206 
1207 	smc = smc_sk(sock->sk);
1208 	/* socket options apply to the CLC socket */
1209 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1210 					     optval, optlen);
1211 }
1212 
1213 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1214 		     unsigned long arg)
1215 {
1216 	struct smc_sock *smc;
1217 
1218 	smc = smc_sk(sock->sk);
1219 	if (smc->use_fallback)
1220 		return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1221 	else
1222 		return sock_no_ioctl(sock, cmd, arg);
1223 }
1224 
1225 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1226 			    int offset, size_t size, int flags)
1227 {
1228 	struct sock *sk = sock->sk;
1229 	struct smc_sock *smc;
1230 	int rc = -EPIPE;
1231 
1232 	smc = smc_sk(sk);
1233 	lock_sock(sk);
1234 	if (sk->sk_state != SMC_ACTIVE)
1235 		goto out;
1236 	if (smc->use_fallback)
1237 		rc = kernel_sendpage(smc->clcsock, page, offset,
1238 				     size, flags);
1239 	else
1240 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1241 
1242 out:
1243 	release_sock(sk);
1244 	return rc;
1245 }
1246 
1247 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1248 			       struct pipe_inode_info *pipe, size_t len,
1249 				    unsigned int flags)
1250 {
1251 	struct sock *sk = sock->sk;
1252 	struct smc_sock *smc;
1253 	int rc = -ENOTCONN;
1254 
1255 	smc = smc_sk(sk);
1256 	lock_sock(sk);
1257 	if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1258 		goto out;
1259 	if (smc->use_fallback) {
1260 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1261 						    pipe, len, flags);
1262 	} else {
1263 		rc = -EOPNOTSUPP;
1264 	}
1265 out:
1266 	release_sock(sk);
1267 	return rc;
1268 }
1269 
1270 /* must look like tcp */
1271 static const struct proto_ops smc_sock_ops = {
1272 	.family		= PF_SMC,
1273 	.owner		= THIS_MODULE,
1274 	.release	= smc_release,
1275 	.bind		= smc_bind,
1276 	.connect	= smc_connect,
1277 	.socketpair	= sock_no_socketpair,
1278 	.accept		= smc_accept,
1279 	.getname	= smc_getname,
1280 	.poll		= smc_poll,
1281 	.ioctl		= smc_ioctl,
1282 	.listen		= smc_listen,
1283 	.shutdown	= smc_shutdown,
1284 	.setsockopt	= smc_setsockopt,
1285 	.getsockopt	= smc_getsockopt,
1286 	.sendmsg	= smc_sendmsg,
1287 	.recvmsg	= smc_recvmsg,
1288 	.mmap		= sock_no_mmap,
1289 	.sendpage	= smc_sendpage,
1290 	.splice_read	= smc_splice_read,
1291 };
1292 
1293 static int smc_create(struct net *net, struct socket *sock, int protocol,
1294 		      int kern)
1295 {
1296 	struct smc_sock *smc;
1297 	struct sock *sk;
1298 	int rc;
1299 
1300 	rc = -ESOCKTNOSUPPORT;
1301 	if (sock->type != SOCK_STREAM)
1302 		goto out;
1303 
1304 	rc = -EPROTONOSUPPORT;
1305 	if ((protocol != IPPROTO_IP) && (protocol != IPPROTO_TCP))
1306 		goto out;
1307 
1308 	rc = -ENOBUFS;
1309 	sock->ops = &smc_sock_ops;
1310 	sk = smc_sock_alloc(net, sock);
1311 	if (!sk)
1312 		goto out;
1313 
1314 	/* create internal TCP socket for CLC handshake and fallback */
1315 	smc = smc_sk(sk);
1316 	smc->use_fallback = false; /* assume rdma capability first */
1317 	rc = sock_create_kern(net, PF_INET, SOCK_STREAM,
1318 			      IPPROTO_TCP, &smc->clcsock);
1319 	if (rc)
1320 		sk_common_release(sk);
1321 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1322 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1323 
1324 out:
1325 	return rc;
1326 }
1327 
1328 static const struct net_proto_family smc_sock_family_ops = {
1329 	.family	= PF_SMC,
1330 	.owner	= THIS_MODULE,
1331 	.create	= smc_create,
1332 };
1333 
1334 static int __init smc_init(void)
1335 {
1336 	int rc;
1337 
1338 	rc = smc_pnet_init();
1339 	if (rc)
1340 		return rc;
1341 
1342 	rc = smc_llc_init();
1343 	if (rc) {
1344 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1345 		goto out_pnet;
1346 	}
1347 
1348 	rc = smc_cdc_init();
1349 	if (rc) {
1350 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1351 		goto out_pnet;
1352 	}
1353 
1354 	rc = proto_register(&smc_proto, 1);
1355 	if (rc) {
1356 		pr_err("%s: proto_register fails with %d\n", __func__, rc);
1357 		goto out_pnet;
1358 	}
1359 
1360 	rc = sock_register(&smc_sock_family_ops);
1361 	if (rc) {
1362 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
1363 		goto out_proto;
1364 	}
1365 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1366 
1367 	rc = smc_ib_register_client();
1368 	if (rc) {
1369 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
1370 		goto out_sock;
1371 	}
1372 
1373 	return 0;
1374 
1375 out_sock:
1376 	sock_unregister(PF_SMC);
1377 out_proto:
1378 	proto_unregister(&smc_proto);
1379 out_pnet:
1380 	smc_pnet_exit();
1381 	return rc;
1382 }
1383 
1384 static void __exit smc_exit(void)
1385 {
1386 	struct smc_link_group *lgr, *lg;
1387 	LIST_HEAD(lgr_freeing_list);
1388 
1389 	spin_lock_bh(&smc_lgr_list.lock);
1390 	if (!list_empty(&smc_lgr_list.list))
1391 		list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1392 	spin_unlock_bh(&smc_lgr_list.lock);
1393 	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1394 		list_del_init(&lgr->list);
1395 		smc_lgr_free(lgr); /* free link group */
1396 	}
1397 	smc_ib_unregister_client();
1398 	sock_unregister(PF_SMC);
1399 	proto_unregister(&smc_proto);
1400 	smc_pnet_exit();
1401 }
1402 
1403 module_init(smc_init);
1404 module_exit(smc_exit);
1405 
1406 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1407 MODULE_DESCRIPTION("smc socket address family");
1408 MODULE_LICENSE("GPL");
1409 MODULE_ALIAS_NETPROTO(PF_SMC);
1410