xref: /openbmc/linux/net/smc/af_smc.c (revision 1c2dd16a)
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 	release_sock(sk);
151 
152 	sock_put(sk);
153 out:
154 	return rc;
155 }
156 
157 static void smc_destruct(struct sock *sk)
158 {
159 	if (sk->sk_state != SMC_CLOSED)
160 		return;
161 	if (!sock_flag(sk, SOCK_DEAD))
162 		return;
163 
164 	sk_refcnt_debug_dec(sk);
165 }
166 
167 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock)
168 {
169 	struct smc_sock *smc;
170 	struct sock *sk;
171 
172 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, &smc_proto, 0);
173 	if (!sk)
174 		return NULL;
175 
176 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
177 	sk->sk_state = SMC_INIT;
178 	sk->sk_destruct = smc_destruct;
179 	sk->sk_protocol = SMCPROTO_SMC;
180 	smc = smc_sk(sk);
181 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
182 	INIT_LIST_HEAD(&smc->accept_q);
183 	spin_lock_init(&smc->accept_q_lock);
184 	INIT_DELAYED_WORK(&smc->sock_put_work, smc_close_sock_put_work);
185 	sk->sk_prot->hash(sk);
186 	sk_refcnt_debug_inc(sk);
187 
188 	return sk;
189 }
190 
191 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
192 		    int addr_len)
193 {
194 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
195 	struct sock *sk = sock->sk;
196 	struct smc_sock *smc;
197 	int rc;
198 
199 	smc = smc_sk(sk);
200 
201 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
202 	rc = -EINVAL;
203 	if (addr_len < sizeof(struct sockaddr_in))
204 		goto out;
205 
206 	rc = -EAFNOSUPPORT;
207 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
208 	if ((addr->sin_family != AF_INET) &&
209 	    ((addr->sin_family != AF_UNSPEC) ||
210 	     (addr->sin_addr.s_addr != htonl(INADDR_ANY))))
211 		goto out;
212 
213 	lock_sock(sk);
214 
215 	/* Check if socket is already active */
216 	rc = -EINVAL;
217 	if (sk->sk_state != SMC_INIT)
218 		goto out_rel;
219 
220 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
221 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
222 
223 out_rel:
224 	release_sock(sk);
225 out:
226 	return rc;
227 }
228 
229 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
230 				   unsigned long mask)
231 {
232 	/* options we don't get control via setsockopt for */
233 	nsk->sk_type = osk->sk_type;
234 	nsk->sk_sndbuf = osk->sk_sndbuf;
235 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
236 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
237 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
238 	nsk->sk_mark = osk->sk_mark;
239 	nsk->sk_priority = osk->sk_priority;
240 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
241 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
242 	nsk->sk_err = osk->sk_err;
243 
244 	nsk->sk_flags &= ~mask;
245 	nsk->sk_flags |= osk->sk_flags & mask;
246 }
247 
248 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
249 			     (1UL << SOCK_KEEPOPEN) | \
250 			     (1UL << SOCK_LINGER) | \
251 			     (1UL << SOCK_BROADCAST) | \
252 			     (1UL << SOCK_TIMESTAMP) | \
253 			     (1UL << SOCK_DBG) | \
254 			     (1UL << SOCK_RCVTSTAMP) | \
255 			     (1UL << SOCK_RCVTSTAMPNS) | \
256 			     (1UL << SOCK_LOCALROUTE) | \
257 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
258 			     (1UL << SOCK_RXQ_OVFL) | \
259 			     (1UL << SOCK_WIFI_STATUS) | \
260 			     (1UL << SOCK_NOFCS) | \
261 			     (1UL << SOCK_FILTER_LOCKED))
262 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
263  * clc socket (since smc is not called for these options from net/core)
264  */
265 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
266 {
267 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
268 }
269 
270 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
271 			     (1UL << SOCK_KEEPOPEN) | \
272 			     (1UL << SOCK_LINGER) | \
273 			     (1UL << SOCK_DBG))
274 /* copy only settings and flags relevant for smc from clc to smc socket */
275 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
276 {
277 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
278 }
279 
280 /* determine subnet and mask of internal TCP socket */
281 int smc_netinfo_by_tcpsk(struct socket *clcsock,
282 			 __be32 *subnet, u8 *prefix_len)
283 {
284 	struct dst_entry *dst = sk_dst_get(clcsock->sk);
285 	struct sockaddr_in addr;
286 	int rc = -ENOENT;
287 	int len;
288 
289 	if (!dst) {
290 		rc = -ENOTCONN;
291 		goto out;
292 	}
293 	if (!dst->dev) {
294 		rc = -ENODEV;
295 		goto out_rel;
296 	}
297 
298 	/* get address to which the internal TCP socket is bound */
299 	kernel_getsockname(clcsock, (struct sockaddr *)&addr, &len);
300 	/* analyze IPv4 specific data of net_device belonging to TCP socket */
301 	for_ifa(dst->dev->ip_ptr) {
302 		if (ifa->ifa_address != addr.sin_addr.s_addr)
303 			continue;
304 		*prefix_len = inet_mask_len(ifa->ifa_mask);
305 		*subnet = ifa->ifa_address & ifa->ifa_mask;
306 		rc = 0;
307 		break;
308 	} endfor_ifa(dst->dev->ip_ptr);
309 
310 out_rel:
311 	dst_release(dst);
312 out:
313 	return rc;
314 }
315 
316 static int smc_clnt_conf_first_link(struct smc_sock *smc, union ib_gid *gid)
317 {
318 	struct smc_link_group *lgr = smc->conn.lgr;
319 	struct smc_link *link;
320 	int rest;
321 	int rc;
322 
323 	link = &lgr->lnk[SMC_SINGLE_LINK];
324 	/* receive CONFIRM LINK request from server over RoCE fabric */
325 	rest = wait_for_completion_interruptible_timeout(
326 		&link->llc_confirm,
327 		SMC_LLC_WAIT_FIRST_TIME);
328 	if (rest <= 0) {
329 		struct smc_clc_msg_decline dclc;
330 
331 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
332 				      SMC_CLC_DECLINE);
333 		return rc;
334 	}
335 
336 	rc = smc_ib_modify_qp_rts(link);
337 	if (rc)
338 		return SMC_CLC_DECL_INTERR;
339 
340 	smc_wr_remember_qp_attr(link);
341 	/* send CONFIRM LINK response over RoCE fabric */
342 	rc = smc_llc_send_confirm_link(link,
343 				       link->smcibdev->mac[link->ibport - 1],
344 				       gid, SMC_LLC_RESP);
345 	if (rc < 0)
346 		return SMC_CLC_DECL_TCL;
347 
348 	return rc;
349 }
350 
351 static void smc_conn_save_peer_info(struct smc_sock *smc,
352 				    struct smc_clc_msg_accept_confirm *clc)
353 {
354 	smc->conn.peer_conn_idx = clc->conn_idx;
355 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
356 	smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
357 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
358 }
359 
360 static void smc_link_save_peer_info(struct smc_link *link,
361 				    struct smc_clc_msg_accept_confirm *clc)
362 {
363 	link->peer_qpn = ntoh24(clc->qpn);
364 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
365 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
366 	link->peer_psn = ntoh24(clc->psn);
367 	link->peer_mtu = clc->qp_mtu;
368 }
369 
370 /* setup for RDMA connection of client */
371 static int smc_connect_rdma(struct smc_sock *smc)
372 {
373 	struct sockaddr_in *inaddr = (struct sockaddr_in *)smc->addr;
374 	struct smc_clc_msg_accept_confirm aclc;
375 	int local_contact = SMC_FIRST_CONTACT;
376 	struct smc_ib_device *smcibdev;
377 	struct smc_link *link;
378 	u8 srv_first_contact;
379 	int reason_code = 0;
380 	int rc = 0;
381 	u8 ibport;
382 
383 	/* IPSec connections opt out of SMC-R optimizations */
384 	if (using_ipsec(smc)) {
385 		reason_code = SMC_CLC_DECL_IPSEC;
386 		goto decline_rdma;
387 	}
388 
389 	/* PNET table look up: search active ib_device and port
390 	 * within same PNETID that also contains the ethernet device
391 	 * used for the internal TCP socket
392 	 */
393 	smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
394 	if (!smcibdev) {
395 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
396 		goto decline_rdma;
397 	}
398 
399 	/* do inband token exchange */
400 	reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
401 	if (reason_code < 0) {
402 		rc = reason_code;
403 		goto out_err;
404 	}
405 	if (reason_code > 0) /* configuration error */
406 		goto decline_rdma;
407 	/* receive SMC Accept CLC message */
408 	reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
409 				       SMC_CLC_ACCEPT);
410 	if (reason_code < 0) {
411 		rc = reason_code;
412 		goto out_err;
413 	}
414 	if (reason_code > 0)
415 		goto decline_rdma;
416 
417 	srv_first_contact = aclc.hdr.flag;
418 	mutex_lock(&smc_create_lgr_pending);
419 	local_contact = smc_conn_create(smc, inaddr->sin_addr.s_addr, smcibdev,
420 					ibport, &aclc.lcl, srv_first_contact);
421 	if (local_contact < 0) {
422 		rc = local_contact;
423 		if (rc == -ENOMEM)
424 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
425 		else if (rc == -ENOLINK)
426 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
427 		goto decline_rdma_unlock;
428 	}
429 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
430 
431 	smc_conn_save_peer_info(smc, &aclc);
432 
433 	rc = smc_sndbuf_create(smc);
434 	if (rc) {
435 		reason_code = SMC_CLC_DECL_MEM;
436 		goto decline_rdma_unlock;
437 	}
438 	rc = smc_rmb_create(smc);
439 	if (rc) {
440 		reason_code = SMC_CLC_DECL_MEM;
441 		goto decline_rdma_unlock;
442 	}
443 
444 	if (local_contact == SMC_FIRST_CONTACT)
445 		smc_link_save_peer_info(link, &aclc);
446 
447 	rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
448 	if (rc) {
449 		reason_code = SMC_CLC_DECL_INTERR;
450 		goto decline_rdma_unlock;
451 	}
452 
453 	smc_close_init(smc);
454 	smc_rx_init(smc);
455 
456 	if (local_contact == SMC_FIRST_CONTACT) {
457 		rc = smc_ib_ready_link(link);
458 		if (rc) {
459 			reason_code = SMC_CLC_DECL_INTERR;
460 			goto decline_rdma_unlock;
461 		}
462 	}
463 
464 	rc = smc_clc_send_confirm(smc);
465 	if (rc)
466 		goto out_err_unlock;
467 
468 	if (local_contact == SMC_FIRST_CONTACT) {
469 		/* QP confirmation over RoCE fabric */
470 		reason_code = smc_clnt_conf_first_link(
471 			smc, &smcibdev->gid[ibport - 1]);
472 		if (reason_code < 0) {
473 			rc = reason_code;
474 			goto out_err_unlock;
475 		}
476 		if (reason_code > 0)
477 			goto decline_rdma_unlock;
478 	}
479 
480 	mutex_unlock(&smc_create_lgr_pending);
481 	smc_tx_init(smc);
482 
483 out_connected:
484 	smc_copy_sock_settings_to_clc(smc);
485 	if (smc->sk.sk_state == SMC_INIT)
486 		smc->sk.sk_state = SMC_ACTIVE;
487 
488 	return rc ? rc : local_contact;
489 
490 decline_rdma_unlock:
491 	mutex_unlock(&smc_create_lgr_pending);
492 	smc_conn_free(&smc->conn);
493 decline_rdma:
494 	/* RDMA setup failed, switch back to TCP */
495 	smc->use_fallback = true;
496 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
497 		rc = smc_clc_send_decline(smc, reason_code, 0);
498 		if (rc < sizeof(struct smc_clc_msg_decline))
499 			goto out_err;
500 	}
501 	goto out_connected;
502 
503 out_err_unlock:
504 	mutex_unlock(&smc_create_lgr_pending);
505 	smc_conn_free(&smc->conn);
506 out_err:
507 	return rc;
508 }
509 
510 static int smc_connect(struct socket *sock, struct sockaddr *addr,
511 		       int alen, int flags)
512 {
513 	struct sock *sk = sock->sk;
514 	struct smc_sock *smc;
515 	int rc = -EINVAL;
516 
517 	smc = smc_sk(sk);
518 
519 	/* separate smc parameter checking to be safe */
520 	if (alen < sizeof(addr->sa_family))
521 		goto out_err;
522 	if (addr->sa_family != AF_INET)
523 		goto out_err;
524 	smc->addr = addr;	/* needed for nonblocking connect */
525 
526 	lock_sock(sk);
527 	switch (sk->sk_state) {
528 	default:
529 		goto out;
530 	case SMC_ACTIVE:
531 		rc = -EISCONN;
532 		goto out;
533 	case SMC_INIT:
534 		rc = 0;
535 		break;
536 	}
537 
538 	smc_copy_sock_settings_to_clc(smc);
539 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
540 	if (rc)
541 		goto out;
542 
543 	/* setup RDMA connection */
544 	rc = smc_connect_rdma(smc);
545 	if (rc < 0)
546 		goto out;
547 	else
548 		rc = 0; /* success cases including fallback */
549 
550 out:
551 	release_sock(sk);
552 out_err:
553 	return rc;
554 }
555 
556 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
557 {
558 	struct sock *sk = &lsmc->sk;
559 	struct socket *new_clcsock;
560 	struct sock *new_sk;
561 	int rc;
562 
563 	release_sock(&lsmc->sk);
564 	new_sk = smc_sock_alloc(sock_net(sk), NULL);
565 	if (!new_sk) {
566 		rc = -ENOMEM;
567 		lsmc->sk.sk_err = ENOMEM;
568 		*new_smc = NULL;
569 		lock_sock(&lsmc->sk);
570 		goto out;
571 	}
572 	*new_smc = smc_sk(new_sk);
573 
574 	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
575 	lock_sock(&lsmc->sk);
576 	if  (rc < 0) {
577 		lsmc->sk.sk_err = -rc;
578 		new_sk->sk_state = SMC_CLOSED;
579 		sock_set_flag(new_sk, SOCK_DEAD);
580 		sk->sk_prot->unhash(new_sk);
581 		sock_put(new_sk);
582 		*new_smc = NULL;
583 		goto out;
584 	}
585 	if (lsmc->sk.sk_state == SMC_CLOSED) {
586 		if (new_clcsock)
587 			sock_release(new_clcsock);
588 		new_sk->sk_state = SMC_CLOSED;
589 		sock_set_flag(new_sk, SOCK_DEAD);
590 		sk->sk_prot->unhash(new_sk);
591 		sock_put(new_sk);
592 		*new_smc = NULL;
593 		goto out;
594 	}
595 
596 	(*new_smc)->clcsock = new_clcsock;
597 out:
598 	return rc;
599 }
600 
601 /* add a just created sock to the accept queue of the listen sock as
602  * candidate for a following socket accept call from user space
603  */
604 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
605 {
606 	struct smc_sock *par = smc_sk(parent);
607 
608 	sock_hold(sk);
609 	spin_lock(&par->accept_q_lock);
610 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
611 	spin_unlock(&par->accept_q_lock);
612 	sk_acceptq_added(parent);
613 }
614 
615 /* remove a socket from the accept queue of its parental listening socket */
616 static void smc_accept_unlink(struct sock *sk)
617 {
618 	struct smc_sock *par = smc_sk(sk)->listen_smc;
619 
620 	spin_lock(&par->accept_q_lock);
621 	list_del_init(&smc_sk(sk)->accept_q);
622 	spin_unlock(&par->accept_q_lock);
623 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
624 	sock_put(sk);
625 }
626 
627 /* remove a sock from the accept queue to bind it to a new socket created
628  * for a socket accept call from user space
629  */
630 struct sock *smc_accept_dequeue(struct sock *parent,
631 				struct socket *new_sock)
632 {
633 	struct smc_sock *isk, *n;
634 	struct sock *new_sk;
635 
636 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
637 		new_sk = (struct sock *)isk;
638 
639 		smc_accept_unlink(new_sk);
640 		if (new_sk->sk_state == SMC_CLOSED) {
641 			new_sk->sk_prot->unhash(new_sk);
642 			sock_put(new_sk);
643 			continue;
644 		}
645 		if (new_sock)
646 			sock_graft(new_sk, new_sock);
647 		return new_sk;
648 	}
649 	return NULL;
650 }
651 
652 /* clean up for a created but never accepted sock */
653 void smc_close_non_accepted(struct sock *sk)
654 {
655 	struct smc_sock *smc = smc_sk(sk);
656 
657 	sock_hold(sk);
658 	lock_sock(sk);
659 	if (!sk->sk_lingertime)
660 		/* wait for peer closing */
661 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
662 	if (smc->use_fallback) {
663 		sk->sk_state = SMC_CLOSED;
664 	} else {
665 		smc_close_active(smc);
666 		sock_set_flag(sk, SOCK_DEAD);
667 		sk->sk_shutdown |= SHUTDOWN_MASK;
668 	}
669 	if (smc->clcsock) {
670 		struct socket *tcp;
671 
672 		tcp = smc->clcsock;
673 		smc->clcsock = NULL;
674 		sock_release(tcp);
675 	}
676 	if (smc->use_fallback) {
677 		schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
678 	} else if (sk->sk_state == SMC_CLOSED) {
679 		smc_conn_free(&smc->conn);
680 		schedule_delayed_work(&smc->sock_put_work,
681 				      SMC_CLOSE_SOCK_PUT_DELAY);
682 	}
683 	release_sock(sk);
684 	sock_put(sk);
685 }
686 
687 static int smc_serv_conf_first_link(struct smc_sock *smc)
688 {
689 	struct smc_link_group *lgr = smc->conn.lgr;
690 	struct smc_link *link;
691 	int rest;
692 	int rc;
693 
694 	link = &lgr->lnk[SMC_SINGLE_LINK];
695 	/* send CONFIRM LINK request to client over the RoCE fabric */
696 	rc = smc_llc_send_confirm_link(link,
697 				       link->smcibdev->mac[link->ibport - 1],
698 				       &link->smcibdev->gid[link->ibport - 1],
699 				       SMC_LLC_REQ);
700 	if (rc < 0)
701 		return SMC_CLC_DECL_TCL;
702 
703 	/* receive CONFIRM LINK response from client over the RoCE fabric */
704 	rest = wait_for_completion_interruptible_timeout(
705 		&link->llc_confirm_resp,
706 		SMC_LLC_WAIT_FIRST_TIME);
707 	if (rest <= 0) {
708 		struct smc_clc_msg_decline dclc;
709 
710 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
711 				      SMC_CLC_DECLINE);
712 	}
713 
714 	return rc;
715 }
716 
717 /* setup for RDMA connection of server */
718 static void smc_listen_work(struct work_struct *work)
719 {
720 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
721 						smc_listen_work);
722 	struct socket *newclcsock = new_smc->clcsock;
723 	struct smc_sock *lsmc = new_smc->listen_smc;
724 	struct smc_clc_msg_accept_confirm cclc;
725 	int local_contact = SMC_REUSE_CONTACT;
726 	struct sock *newsmcsk = &new_smc->sk;
727 	struct smc_clc_msg_proposal pclc;
728 	struct smc_ib_device *smcibdev;
729 	struct sockaddr_in peeraddr;
730 	struct smc_link *link;
731 	int reason_code = 0;
732 	int rc = 0, len;
733 	__be32 subnet;
734 	u8 prefix_len;
735 	u8 ibport;
736 
737 	/* do inband token exchange -
738 	 *wait for and receive SMC Proposal CLC message
739 	 */
740 	reason_code = smc_clc_wait_msg(new_smc, &pclc, sizeof(pclc),
741 				       SMC_CLC_PROPOSAL);
742 	if (reason_code < 0)
743 		goto out_err;
744 	if (reason_code > 0)
745 		goto decline_rdma;
746 
747 	/* IPSec connections opt out of SMC-R optimizations */
748 	if (using_ipsec(new_smc)) {
749 		reason_code = SMC_CLC_DECL_IPSEC;
750 		goto decline_rdma;
751 	}
752 
753 	/* PNET table look up: search active ib_device and port
754 	 * within same PNETID that also contains the ethernet device
755 	 * used for the internal TCP socket
756 	 */
757 	smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
758 	if (!smcibdev) {
759 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
760 		goto decline_rdma;
761 	}
762 
763 	/* determine subnet and mask from internal TCP socket */
764 	rc = smc_netinfo_by_tcpsk(newclcsock, &subnet, &prefix_len);
765 	if (rc) {
766 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
767 		goto decline_rdma;
768 	}
769 	if ((pclc.outgoing_subnet != subnet) ||
770 	    (pclc.prefix_len != prefix_len)) {
771 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
772 		goto decline_rdma;
773 	}
774 
775 	/* get address of the peer connected to the internal TCP socket */
776 	kernel_getpeername(newclcsock, (struct sockaddr *)&peeraddr, &len);
777 
778 	/* allocate connection / link group */
779 	mutex_lock(&smc_create_lgr_pending);
780 	local_contact = smc_conn_create(new_smc, peeraddr.sin_addr.s_addr,
781 					smcibdev, ibport, &pclc.lcl, 0);
782 	if (local_contact == SMC_REUSE_CONTACT)
783 		/* lock no longer needed, free it due to following
784 		 * smc_clc_wait_msg() call
785 		 */
786 		mutex_unlock(&smc_create_lgr_pending);
787 	if (local_contact < 0) {
788 		rc = local_contact;
789 		if (rc == -ENOMEM)
790 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
791 		else if (rc == -ENOLINK)
792 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
793 		goto decline_rdma;
794 	}
795 	link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
796 
797 	rc = smc_sndbuf_create(new_smc);
798 	if (rc) {
799 		reason_code = SMC_CLC_DECL_MEM;
800 		goto decline_rdma;
801 	}
802 	rc = smc_rmb_create(new_smc);
803 	if (rc) {
804 		reason_code = SMC_CLC_DECL_MEM;
805 		goto decline_rdma;
806 	}
807 
808 	smc_close_init(new_smc);
809 	smc_rx_init(new_smc);
810 
811 	rc = smc_clc_send_accept(new_smc, local_contact);
812 	if (rc)
813 		goto out_err;
814 
815 	/* receive SMC Confirm CLC message */
816 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
817 				       SMC_CLC_CONFIRM);
818 	if (reason_code < 0)
819 		goto out_err;
820 	if (reason_code > 0)
821 		goto decline_rdma;
822 	smc_conn_save_peer_info(new_smc, &cclc);
823 	if (local_contact == SMC_FIRST_CONTACT)
824 		smc_link_save_peer_info(link, &cclc);
825 
826 	rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
827 	if (rc) {
828 		reason_code = SMC_CLC_DECL_INTERR;
829 		goto decline_rdma;
830 	}
831 
832 	if (local_contact == SMC_FIRST_CONTACT) {
833 		rc = smc_ib_ready_link(link);
834 		if (rc) {
835 			reason_code = SMC_CLC_DECL_INTERR;
836 			goto decline_rdma;
837 		}
838 		/* QP confirmation over RoCE fabric */
839 		reason_code = smc_serv_conf_first_link(new_smc);
840 		if (reason_code < 0) {
841 			/* peer is not aware of a problem */
842 			rc = reason_code;
843 			goto out_err;
844 		}
845 		if (reason_code > 0)
846 			goto decline_rdma;
847 	}
848 
849 	smc_tx_init(new_smc);
850 
851 out_connected:
852 	sk_refcnt_debug_inc(newsmcsk);
853 	if (newsmcsk->sk_state == SMC_INIT)
854 		newsmcsk->sk_state = SMC_ACTIVE;
855 enqueue:
856 	if (local_contact == SMC_FIRST_CONTACT)
857 		mutex_unlock(&smc_create_lgr_pending);
858 	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
859 	if (lsmc->sk.sk_state == SMC_LISTEN) {
860 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
861 	} else { /* no longer listening */
862 		smc_close_non_accepted(newsmcsk);
863 	}
864 	release_sock(&lsmc->sk);
865 
866 	/* Wake up accept */
867 	lsmc->sk.sk_data_ready(&lsmc->sk);
868 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
869 	return;
870 
871 decline_rdma:
872 	/* RDMA setup failed, switch back to TCP */
873 	smc_conn_free(&new_smc->conn);
874 	new_smc->use_fallback = true;
875 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
876 		rc = smc_clc_send_decline(new_smc, reason_code, 0);
877 		if (rc < sizeof(struct smc_clc_msg_decline))
878 			goto out_err;
879 	}
880 	goto out_connected;
881 
882 out_err:
883 	newsmcsk->sk_state = SMC_CLOSED;
884 	smc_conn_free(&new_smc->conn);
885 	goto enqueue; /* queue new sock with sk_err set */
886 }
887 
888 static void smc_tcp_listen_work(struct work_struct *work)
889 {
890 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
891 					     tcp_listen_work);
892 	struct smc_sock *new_smc;
893 	int rc = 0;
894 
895 	lock_sock(&lsmc->sk);
896 	while (lsmc->sk.sk_state == SMC_LISTEN) {
897 		rc = smc_clcsock_accept(lsmc, &new_smc);
898 		if (rc)
899 			goto out;
900 		if (!new_smc)
901 			continue;
902 
903 		new_smc->listen_smc = lsmc;
904 		new_smc->use_fallback = false; /* assume rdma capability first*/
905 		sock_hold(&lsmc->sk); /* sock_put in smc_listen_work */
906 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
907 		smc_copy_sock_settings_to_smc(new_smc);
908 		schedule_work(&new_smc->smc_listen_work);
909 	}
910 
911 out:
912 	release_sock(&lsmc->sk);
913 	lsmc->sk.sk_data_ready(&lsmc->sk); /* no more listening, wake accept */
914 }
915 
916 static int smc_listen(struct socket *sock, int backlog)
917 {
918 	struct sock *sk = sock->sk;
919 	struct smc_sock *smc;
920 	int rc;
921 
922 	smc = smc_sk(sk);
923 	lock_sock(sk);
924 
925 	rc = -EINVAL;
926 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
927 		goto out;
928 
929 	rc = 0;
930 	if (sk->sk_state == SMC_LISTEN) {
931 		sk->sk_max_ack_backlog = backlog;
932 		goto out;
933 	}
934 	/* some socket options are handled in core, so we could not apply
935 	 * them to the clc socket -- copy smc socket options to clc socket
936 	 */
937 	smc_copy_sock_settings_to_clc(smc);
938 
939 	rc = kernel_listen(smc->clcsock, backlog);
940 	if (rc)
941 		goto out;
942 	sk->sk_max_ack_backlog = backlog;
943 	sk->sk_ack_backlog = 0;
944 	sk->sk_state = SMC_LISTEN;
945 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
946 	schedule_work(&smc->tcp_listen_work);
947 
948 out:
949 	release_sock(sk);
950 	return rc;
951 }
952 
953 static int smc_accept(struct socket *sock, struct socket *new_sock,
954 		      int flags, bool kern)
955 {
956 	struct sock *sk = sock->sk, *nsk;
957 	DECLARE_WAITQUEUE(wait, current);
958 	struct smc_sock *lsmc;
959 	long timeo;
960 	int rc = 0;
961 
962 	lsmc = smc_sk(sk);
963 	lock_sock(sk);
964 
965 	if (lsmc->sk.sk_state != SMC_LISTEN) {
966 		rc = -EINVAL;
967 		goto out;
968 	}
969 
970 	/* Wait for an incoming connection */
971 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
972 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
973 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
974 		set_current_state(TASK_INTERRUPTIBLE);
975 		if (!timeo) {
976 			rc = -EAGAIN;
977 			break;
978 		}
979 		release_sock(sk);
980 		timeo = schedule_timeout(timeo);
981 		/* wakeup by sk_data_ready in smc_listen_work() */
982 		sched_annotate_sleep();
983 		lock_sock(sk);
984 		if (signal_pending(current)) {
985 			rc = sock_intr_errno(timeo);
986 			break;
987 		}
988 	}
989 	set_current_state(TASK_RUNNING);
990 	remove_wait_queue(sk_sleep(sk), &wait);
991 
992 	if (!rc)
993 		rc = sock_error(nsk);
994 
995 out:
996 	release_sock(sk);
997 	return rc;
998 }
999 
1000 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1001 		       int *len, int peer)
1002 {
1003 	struct smc_sock *smc;
1004 
1005 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1006 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1007 		return -ENOTCONN;
1008 
1009 	smc = smc_sk(sock->sk);
1010 
1011 	return smc->clcsock->ops->getname(smc->clcsock, addr, len, peer);
1012 }
1013 
1014 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1015 {
1016 	struct sock *sk = sock->sk;
1017 	struct smc_sock *smc;
1018 	int rc = -EPIPE;
1019 
1020 	smc = smc_sk(sk);
1021 	lock_sock(sk);
1022 	if ((sk->sk_state != SMC_ACTIVE) &&
1023 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1024 	    (sk->sk_state != SMC_INIT))
1025 		goto out;
1026 	if (smc->use_fallback)
1027 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1028 	else
1029 		rc = smc_tx_sendmsg(smc, msg, len);
1030 out:
1031 	release_sock(sk);
1032 	return rc;
1033 }
1034 
1035 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1036 		       int flags)
1037 {
1038 	struct sock *sk = sock->sk;
1039 	struct smc_sock *smc;
1040 	int rc = -ENOTCONN;
1041 
1042 	smc = smc_sk(sk);
1043 	lock_sock(sk);
1044 	if ((sk->sk_state == SMC_INIT) ||
1045 	    (sk->sk_state == SMC_LISTEN) ||
1046 	    (sk->sk_state == SMC_CLOSED))
1047 		goto out;
1048 
1049 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1050 		rc = 0;
1051 		goto out;
1052 	}
1053 
1054 	if (smc->use_fallback)
1055 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1056 	else
1057 		rc = smc_rx_recvmsg(smc, msg, len, flags);
1058 
1059 out:
1060 	release_sock(sk);
1061 	return rc;
1062 }
1063 
1064 static unsigned int smc_accept_poll(struct sock *parent)
1065 {
1066 	struct smc_sock *isk;
1067 	struct sock *sk;
1068 
1069 	lock_sock(parent);
1070 	list_for_each_entry(isk, &smc_sk(parent)->accept_q, accept_q) {
1071 		sk = (struct sock *)isk;
1072 
1073 		if (sk->sk_state == SMC_ACTIVE) {
1074 			release_sock(parent);
1075 			return POLLIN | POLLRDNORM;
1076 		}
1077 	}
1078 	release_sock(parent);
1079 
1080 	return 0;
1081 }
1082 
1083 static unsigned int smc_poll(struct file *file, struct socket *sock,
1084 			     poll_table *wait)
1085 {
1086 	struct sock *sk = sock->sk;
1087 	unsigned int mask = 0;
1088 	struct smc_sock *smc;
1089 	int rc;
1090 
1091 	smc = smc_sk(sock->sk);
1092 	if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1093 		/* delegate to CLC child sock */
1094 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1095 		/* if non-blocking connect finished ... */
1096 		lock_sock(sk);
1097 		if ((sk->sk_state == SMC_INIT) && (mask & POLLOUT)) {
1098 			sk->sk_err = smc->clcsock->sk->sk_err;
1099 			if (sk->sk_err) {
1100 				mask |= POLLERR;
1101 			} else {
1102 				rc = smc_connect_rdma(smc);
1103 				if (rc < 0)
1104 					mask |= POLLERR;
1105 				else
1106 					/* success cases including fallback */
1107 					mask |= POLLOUT | POLLWRNORM;
1108 			}
1109 		}
1110 		release_sock(sk);
1111 	} else {
1112 		sock_poll_wait(file, sk_sleep(sk), wait);
1113 		if (sk->sk_state == SMC_LISTEN)
1114 			/* woken up by sk_data_ready in smc_listen_work() */
1115 			mask |= smc_accept_poll(sk);
1116 		if (sk->sk_err)
1117 			mask |= POLLERR;
1118 		if (atomic_read(&smc->conn.sndbuf_space) ||
1119 		    (sk->sk_shutdown & SEND_SHUTDOWN)) {
1120 			mask |= POLLOUT | POLLWRNORM;
1121 		} else {
1122 			sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1123 			set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1124 		}
1125 		if (atomic_read(&smc->conn.bytes_to_rcv))
1126 			mask |= POLLIN | POLLRDNORM;
1127 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1128 		    (sk->sk_state == SMC_CLOSED))
1129 			mask |= POLLHUP;
1130 		if (sk->sk_shutdown & RCV_SHUTDOWN)
1131 			mask |= POLLIN | POLLRDNORM | POLLRDHUP;
1132 		if (sk->sk_state == SMC_APPCLOSEWAIT1)
1133 			mask |= POLLIN;
1134 
1135 	}
1136 
1137 	return mask;
1138 }
1139 
1140 static int smc_shutdown(struct socket *sock, int how)
1141 {
1142 	struct sock *sk = sock->sk;
1143 	struct smc_sock *smc;
1144 	int rc = -EINVAL;
1145 	int rc1 = 0;
1146 
1147 	smc = smc_sk(sk);
1148 
1149 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1150 		return rc;
1151 
1152 	lock_sock(sk);
1153 
1154 	rc = -ENOTCONN;
1155 	if ((sk->sk_state != SMC_LISTEN) &&
1156 	    (sk->sk_state != SMC_ACTIVE) &&
1157 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1158 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1159 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1160 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1161 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1162 		goto out;
1163 	if (smc->use_fallback) {
1164 		rc = kernel_sock_shutdown(smc->clcsock, how);
1165 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1166 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1167 			sk->sk_state = SMC_CLOSED;
1168 		goto out;
1169 	}
1170 	switch (how) {
1171 	case SHUT_RDWR:		/* shutdown in both directions */
1172 		rc = smc_close_active(smc);
1173 		break;
1174 	case SHUT_WR:
1175 		rc = smc_close_shutdown_write(smc);
1176 		break;
1177 	case SHUT_RD:
1178 		if (sk->sk_state == SMC_LISTEN)
1179 			rc = smc_close_active(smc);
1180 		else
1181 			rc = 0;
1182 			/* nothing more to do because peer is not involved */
1183 		break;
1184 	}
1185 	rc1 = kernel_sock_shutdown(smc->clcsock, how);
1186 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1187 	sk->sk_shutdown |= how + 1;
1188 
1189 out:
1190 	release_sock(sk);
1191 	return rc ? rc : rc1;
1192 }
1193 
1194 static int smc_setsockopt(struct socket *sock, int level, int optname,
1195 			  char __user *optval, unsigned int optlen)
1196 {
1197 	struct sock *sk = sock->sk;
1198 	struct smc_sock *smc;
1199 
1200 	smc = smc_sk(sk);
1201 
1202 	/* generic setsockopts reaching us here always apply to the
1203 	 * CLC socket
1204 	 */
1205 	return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1206 					     optval, optlen);
1207 }
1208 
1209 static int smc_getsockopt(struct socket *sock, int level, int optname,
1210 			  char __user *optval, int __user *optlen)
1211 {
1212 	struct smc_sock *smc;
1213 
1214 	smc = smc_sk(sock->sk);
1215 	/* socket options apply to the CLC socket */
1216 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1217 					     optval, optlen);
1218 }
1219 
1220 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1221 		     unsigned long arg)
1222 {
1223 	struct smc_sock *smc;
1224 
1225 	smc = smc_sk(sock->sk);
1226 	if (smc->use_fallback)
1227 		return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1228 	else
1229 		return sock_no_ioctl(sock, cmd, arg);
1230 }
1231 
1232 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1233 			    int offset, size_t size, int flags)
1234 {
1235 	struct sock *sk = sock->sk;
1236 	struct smc_sock *smc;
1237 	int rc = -EPIPE;
1238 
1239 	smc = smc_sk(sk);
1240 	lock_sock(sk);
1241 	if (sk->sk_state != SMC_ACTIVE)
1242 		goto out;
1243 	if (smc->use_fallback)
1244 		rc = kernel_sendpage(smc->clcsock, page, offset,
1245 				     size, flags);
1246 	else
1247 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1248 
1249 out:
1250 	release_sock(sk);
1251 	return rc;
1252 }
1253 
1254 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1255 			       struct pipe_inode_info *pipe, size_t len,
1256 				    unsigned int flags)
1257 {
1258 	struct sock *sk = sock->sk;
1259 	struct smc_sock *smc;
1260 	int rc = -ENOTCONN;
1261 
1262 	smc = smc_sk(sk);
1263 	lock_sock(sk);
1264 	if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1265 		goto out;
1266 	if (smc->use_fallback) {
1267 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1268 						    pipe, len, flags);
1269 	} else {
1270 		rc = -EOPNOTSUPP;
1271 	}
1272 out:
1273 	release_sock(sk);
1274 	return rc;
1275 }
1276 
1277 /* must look like tcp */
1278 static const struct proto_ops smc_sock_ops = {
1279 	.family		= PF_SMC,
1280 	.owner		= THIS_MODULE,
1281 	.release	= smc_release,
1282 	.bind		= smc_bind,
1283 	.connect	= smc_connect,
1284 	.socketpair	= sock_no_socketpair,
1285 	.accept		= smc_accept,
1286 	.getname	= smc_getname,
1287 	.poll		= smc_poll,
1288 	.ioctl		= smc_ioctl,
1289 	.listen		= smc_listen,
1290 	.shutdown	= smc_shutdown,
1291 	.setsockopt	= smc_setsockopt,
1292 	.getsockopt	= smc_getsockopt,
1293 	.sendmsg	= smc_sendmsg,
1294 	.recvmsg	= smc_recvmsg,
1295 	.mmap		= sock_no_mmap,
1296 	.sendpage	= smc_sendpage,
1297 	.splice_read	= smc_splice_read,
1298 };
1299 
1300 static int smc_create(struct net *net, struct socket *sock, int protocol,
1301 		      int kern)
1302 {
1303 	struct smc_sock *smc;
1304 	struct sock *sk;
1305 	int rc;
1306 
1307 	rc = -ESOCKTNOSUPPORT;
1308 	if (sock->type != SOCK_STREAM)
1309 		goto out;
1310 
1311 	rc = -EPROTONOSUPPORT;
1312 	if ((protocol != IPPROTO_IP) && (protocol != IPPROTO_TCP))
1313 		goto out;
1314 
1315 	rc = -ENOBUFS;
1316 	sock->ops = &smc_sock_ops;
1317 	sk = smc_sock_alloc(net, sock);
1318 	if (!sk)
1319 		goto out;
1320 
1321 	/* create internal TCP socket for CLC handshake and fallback */
1322 	smc = smc_sk(sk);
1323 	smc->use_fallback = false; /* assume rdma capability first */
1324 	rc = sock_create_kern(net, PF_INET, SOCK_STREAM,
1325 			      IPPROTO_TCP, &smc->clcsock);
1326 	if (rc)
1327 		sk_common_release(sk);
1328 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1329 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1330 
1331 out:
1332 	return rc;
1333 }
1334 
1335 static const struct net_proto_family smc_sock_family_ops = {
1336 	.family	= PF_SMC,
1337 	.owner	= THIS_MODULE,
1338 	.create	= smc_create,
1339 };
1340 
1341 static int __init smc_init(void)
1342 {
1343 	int rc;
1344 
1345 	rc = smc_pnet_init();
1346 	if (rc)
1347 		return rc;
1348 
1349 	rc = smc_llc_init();
1350 	if (rc) {
1351 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1352 		goto out_pnet;
1353 	}
1354 
1355 	rc = smc_cdc_init();
1356 	if (rc) {
1357 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1358 		goto out_pnet;
1359 	}
1360 
1361 	rc = proto_register(&smc_proto, 1);
1362 	if (rc) {
1363 		pr_err("%s: proto_register fails with %d\n", __func__, rc);
1364 		goto out_pnet;
1365 	}
1366 
1367 	rc = sock_register(&smc_sock_family_ops);
1368 	if (rc) {
1369 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
1370 		goto out_proto;
1371 	}
1372 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1373 
1374 	rc = smc_ib_register_client();
1375 	if (rc) {
1376 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
1377 		goto out_sock;
1378 	}
1379 
1380 	return 0;
1381 
1382 out_sock:
1383 	sock_unregister(PF_SMC);
1384 out_proto:
1385 	proto_unregister(&smc_proto);
1386 out_pnet:
1387 	smc_pnet_exit();
1388 	return rc;
1389 }
1390 
1391 static void __exit smc_exit(void)
1392 {
1393 	struct smc_link_group *lgr, *lg;
1394 	LIST_HEAD(lgr_freeing_list);
1395 
1396 	spin_lock_bh(&smc_lgr_list.lock);
1397 	if (!list_empty(&smc_lgr_list.list))
1398 		list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1399 	spin_unlock_bh(&smc_lgr_list.lock);
1400 	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1401 		list_del_init(&lgr->list);
1402 		smc_lgr_free(lgr); /* free link group */
1403 	}
1404 	smc_ib_unregister_client();
1405 	sock_unregister(PF_SMC);
1406 	proto_unregister(&smc_proto);
1407 	smc_pnet_exit();
1408 }
1409 
1410 module_init(smc_init);
1411 module_exit(smc_exit);
1412 
1413 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1414 MODULE_DESCRIPTION("smc socket address family");
1415 MODULE_LICENSE("GPL");
1416 MODULE_ALIAS_NETPROTO(PF_SMC);
1417