xref: /openbmc/linux/net/smc/af_smc.c (revision 4cff79e9)
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 /* register a new rmb */
296 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc)
297 {
298 	/* register memory region for new rmb */
299 	if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
300 		rmb_desc->regerr = 1;
301 		return -EFAULT;
302 	}
303 	return 0;
304 }
305 
306 static int smc_clnt_conf_first_link(struct smc_sock *smc)
307 {
308 	struct smc_link_group *lgr = smc->conn.lgr;
309 	struct smc_link *link;
310 	int rest;
311 	int rc;
312 
313 	link = &lgr->lnk[SMC_SINGLE_LINK];
314 	/* receive CONFIRM LINK request from server over RoCE fabric */
315 	rest = wait_for_completion_interruptible_timeout(
316 		&link->llc_confirm,
317 		SMC_LLC_WAIT_FIRST_TIME);
318 	if (rest <= 0) {
319 		struct smc_clc_msg_decline dclc;
320 
321 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
322 				      SMC_CLC_DECLINE);
323 		return rc;
324 	}
325 
326 	if (link->llc_confirm_rc)
327 		return SMC_CLC_DECL_RMBE_EC;
328 
329 	rc = smc_ib_modify_qp_rts(link);
330 	if (rc)
331 		return SMC_CLC_DECL_INTERR;
332 
333 	smc_wr_remember_qp_attr(link);
334 
335 	if (smc_reg_rmb(link, smc->conn.rmb_desc))
336 		return SMC_CLC_DECL_INTERR;
337 
338 	/* send CONFIRM LINK response over RoCE fabric */
339 	rc = smc_llc_send_confirm_link(link,
340 				       link->smcibdev->mac[link->ibport - 1],
341 				       &link->smcibdev->gid[link->ibport - 1],
342 				       SMC_LLC_RESP);
343 	if (rc < 0)
344 		return SMC_CLC_DECL_TCL;
345 
346 	/* receive ADD LINK request from server over RoCE fabric */
347 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
348 							 SMC_LLC_WAIT_TIME);
349 	if (rest <= 0) {
350 		struct smc_clc_msg_decline dclc;
351 
352 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
353 				      SMC_CLC_DECLINE);
354 		return rc;
355 	}
356 
357 	/* send add link reject message, only one link supported for now */
358 	rc = smc_llc_send_add_link(link,
359 				   link->smcibdev->mac[link->ibport - 1],
360 				   &link->smcibdev->gid[link->ibport - 1],
361 				   SMC_LLC_RESP);
362 	if (rc < 0)
363 		return SMC_CLC_DECL_TCL;
364 
365 	link->state = SMC_LNK_ACTIVE;
366 
367 	return 0;
368 }
369 
370 static void smc_conn_save_peer_info(struct smc_sock *smc,
371 				    struct smc_clc_msg_accept_confirm *clc)
372 {
373 	smc->conn.peer_conn_idx = clc->conn_idx;
374 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
375 	smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
376 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
377 }
378 
379 static void smc_link_save_peer_info(struct smc_link *link,
380 				    struct smc_clc_msg_accept_confirm *clc)
381 {
382 	link->peer_qpn = ntoh24(clc->qpn);
383 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
384 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
385 	link->peer_psn = ntoh24(clc->psn);
386 	link->peer_mtu = clc->qp_mtu;
387 }
388 
389 /* setup for RDMA connection of client */
390 static int smc_connect_rdma(struct smc_sock *smc)
391 {
392 	struct smc_clc_msg_accept_confirm aclc;
393 	int local_contact = SMC_FIRST_CONTACT;
394 	struct smc_ib_device *smcibdev;
395 	struct smc_link *link;
396 	u8 srv_first_contact;
397 	int reason_code = 0;
398 	int rc = 0;
399 	u8 ibport;
400 
401 	sock_hold(&smc->sk); /* sock put in passive closing */
402 
403 	if (!tcp_sk(smc->clcsock->sk)->syn_smc) {
404 		/* peer has not signalled SMC-capability */
405 		smc->use_fallback = true;
406 		goto out_connected;
407 	}
408 
409 	/* IPSec connections opt out of SMC-R optimizations */
410 	if (using_ipsec(smc)) {
411 		reason_code = SMC_CLC_DECL_IPSEC;
412 		goto decline_rdma;
413 	}
414 
415 	/* PNET table look up: search active ib_device and port
416 	 * within same PNETID that also contains the ethernet device
417 	 * used for the internal TCP socket
418 	 */
419 	smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
420 	if (!smcibdev) {
421 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
422 		goto decline_rdma;
423 	}
424 
425 	/* do inband token exchange */
426 	reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
427 	if (reason_code < 0) {
428 		rc = reason_code;
429 		goto out_err;
430 	}
431 	if (reason_code > 0) /* configuration error */
432 		goto decline_rdma;
433 	/* receive SMC Accept CLC message */
434 	reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
435 				       SMC_CLC_ACCEPT);
436 	if (reason_code < 0) {
437 		rc = reason_code;
438 		goto out_err;
439 	}
440 	if (reason_code > 0)
441 		goto decline_rdma;
442 
443 	srv_first_contact = aclc.hdr.flag;
444 	mutex_lock(&smc_create_lgr_pending);
445 	local_contact = smc_conn_create(smc, smcibdev, ibport, &aclc.lcl,
446 					srv_first_contact);
447 	if (local_contact < 0) {
448 		rc = local_contact;
449 		if (rc == -ENOMEM)
450 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
451 		else if (rc == -ENOLINK)
452 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
453 		goto decline_rdma_unlock;
454 	}
455 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
456 
457 	smc_conn_save_peer_info(smc, &aclc);
458 
459 	/* create send buffer and rmb */
460 	rc = smc_buf_create(smc);
461 	if (rc) {
462 		reason_code = SMC_CLC_DECL_MEM;
463 		goto decline_rdma_unlock;
464 	}
465 
466 	if (local_contact == SMC_FIRST_CONTACT)
467 		smc_link_save_peer_info(link, &aclc);
468 
469 	rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
470 	if (rc) {
471 		reason_code = SMC_CLC_DECL_INTERR;
472 		goto decline_rdma_unlock;
473 	}
474 
475 	smc_close_init(smc);
476 	smc_rx_init(smc);
477 
478 	if (local_contact == SMC_FIRST_CONTACT) {
479 		rc = smc_ib_ready_link(link);
480 		if (rc) {
481 			reason_code = SMC_CLC_DECL_INTERR;
482 			goto decline_rdma_unlock;
483 		}
484 	} else {
485 		if (!smc->conn.rmb_desc->reused) {
486 			if (smc_reg_rmb(link, smc->conn.rmb_desc)) {
487 				reason_code = SMC_CLC_DECL_INTERR;
488 				goto decline_rdma_unlock;
489 			}
490 		}
491 	}
492 	smc_rmb_sync_sg_for_device(&smc->conn);
493 
494 	rc = smc_clc_send_confirm(smc);
495 	if (rc)
496 		goto out_err_unlock;
497 
498 	if (local_contact == SMC_FIRST_CONTACT) {
499 		/* QP confirmation over RoCE fabric */
500 		reason_code = smc_clnt_conf_first_link(smc);
501 		if (reason_code < 0) {
502 			rc = reason_code;
503 			goto out_err_unlock;
504 		}
505 		if (reason_code > 0)
506 			goto decline_rdma_unlock;
507 	}
508 
509 	mutex_unlock(&smc_create_lgr_pending);
510 	smc_tx_init(smc);
511 
512 out_connected:
513 	smc_copy_sock_settings_to_clc(smc);
514 	if (smc->sk.sk_state == SMC_INIT)
515 		smc->sk.sk_state = SMC_ACTIVE;
516 
517 	return rc ? rc : local_contact;
518 
519 decline_rdma_unlock:
520 	if (local_contact == SMC_FIRST_CONTACT)
521 		smc_lgr_forget(smc->conn.lgr);
522 	mutex_unlock(&smc_create_lgr_pending);
523 	smc_conn_free(&smc->conn);
524 decline_rdma:
525 	/* RDMA setup failed, switch back to TCP */
526 	smc->use_fallback = true;
527 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
528 		rc = smc_clc_send_decline(smc, reason_code);
529 		if (rc < 0)
530 			goto out_err;
531 	}
532 	goto out_connected;
533 
534 out_err_unlock:
535 	if (local_contact == SMC_FIRST_CONTACT)
536 		smc_lgr_forget(smc->conn.lgr);
537 	mutex_unlock(&smc_create_lgr_pending);
538 	smc_conn_free(&smc->conn);
539 out_err:
540 	if (smc->sk.sk_state == SMC_INIT)
541 		sock_put(&smc->sk); /* passive closing */
542 	return rc;
543 }
544 
545 static int smc_connect(struct socket *sock, struct sockaddr *addr,
546 		       int alen, int flags)
547 {
548 	struct sock *sk = sock->sk;
549 	struct smc_sock *smc;
550 	int rc = -EINVAL;
551 
552 	smc = smc_sk(sk);
553 
554 	/* separate smc parameter checking to be safe */
555 	if (alen < sizeof(addr->sa_family))
556 		goto out_err;
557 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
558 		goto out_err;
559 
560 	lock_sock(sk);
561 	switch (sk->sk_state) {
562 	default:
563 		goto out;
564 	case SMC_ACTIVE:
565 		rc = -EISCONN;
566 		goto out;
567 	case SMC_INIT:
568 		rc = 0;
569 		break;
570 	}
571 
572 	smc_copy_sock_settings_to_clc(smc);
573 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
574 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
575 	if (rc)
576 		goto out;
577 
578 	/* setup RDMA connection */
579 	rc = smc_connect_rdma(smc);
580 	if (rc < 0)
581 		goto out;
582 	else
583 		rc = 0; /* success cases including fallback */
584 
585 out:
586 	release_sock(sk);
587 out_err:
588 	return rc;
589 }
590 
591 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
592 {
593 	struct socket *new_clcsock = NULL;
594 	struct sock *lsk = &lsmc->sk;
595 	struct sock *new_sk;
596 	int rc;
597 
598 	release_sock(lsk);
599 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
600 	if (!new_sk) {
601 		rc = -ENOMEM;
602 		lsk->sk_err = ENOMEM;
603 		*new_smc = NULL;
604 		lock_sock(lsk);
605 		goto out;
606 	}
607 	*new_smc = smc_sk(new_sk);
608 
609 	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
610 	lock_sock(lsk);
611 	if  (rc < 0)
612 		lsk->sk_err = -rc;
613 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
614 		if (new_clcsock)
615 			sock_release(new_clcsock);
616 		new_sk->sk_state = SMC_CLOSED;
617 		sock_set_flag(new_sk, SOCK_DEAD);
618 		new_sk->sk_prot->unhash(new_sk);
619 		sock_put(new_sk); /* final */
620 		*new_smc = NULL;
621 		goto out;
622 	}
623 
624 	(*new_smc)->clcsock = new_clcsock;
625 out:
626 	return rc;
627 }
628 
629 /* add a just created sock to the accept queue of the listen sock as
630  * candidate for a following socket accept call from user space
631  */
632 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
633 {
634 	struct smc_sock *par = smc_sk(parent);
635 
636 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
637 	spin_lock(&par->accept_q_lock);
638 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
639 	spin_unlock(&par->accept_q_lock);
640 	sk_acceptq_added(parent);
641 }
642 
643 /* remove a socket from the accept queue of its parental listening socket */
644 static void smc_accept_unlink(struct sock *sk)
645 {
646 	struct smc_sock *par = smc_sk(sk)->listen_smc;
647 
648 	spin_lock(&par->accept_q_lock);
649 	list_del_init(&smc_sk(sk)->accept_q);
650 	spin_unlock(&par->accept_q_lock);
651 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
652 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
653 }
654 
655 /* remove a sock from the accept queue to bind it to a new socket created
656  * for a socket accept call from user space
657  */
658 struct sock *smc_accept_dequeue(struct sock *parent,
659 				struct socket *new_sock)
660 {
661 	struct smc_sock *isk, *n;
662 	struct sock *new_sk;
663 
664 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
665 		new_sk = (struct sock *)isk;
666 
667 		smc_accept_unlink(new_sk);
668 		if (new_sk->sk_state == SMC_CLOSED) {
669 			if (isk->clcsock) {
670 				sock_release(isk->clcsock);
671 				isk->clcsock = NULL;
672 			}
673 			new_sk->sk_prot->unhash(new_sk);
674 			sock_put(new_sk); /* final */
675 			continue;
676 		}
677 		if (new_sock)
678 			sock_graft(new_sk, new_sock);
679 		return new_sk;
680 	}
681 	return NULL;
682 }
683 
684 /* clean up for a created but never accepted sock */
685 void smc_close_non_accepted(struct sock *sk)
686 {
687 	struct smc_sock *smc = smc_sk(sk);
688 
689 	lock_sock(sk);
690 	if (!sk->sk_lingertime)
691 		/* wait for peer closing */
692 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
693 	if (!smc->use_fallback) {
694 		smc_close_active(smc);
695 		sock_set_flag(sk, SOCK_DEAD);
696 		sk->sk_shutdown |= SHUTDOWN_MASK;
697 	}
698 	if (smc->clcsock) {
699 		struct socket *tcp;
700 
701 		tcp = smc->clcsock;
702 		smc->clcsock = NULL;
703 		sock_release(tcp);
704 	}
705 	if (smc->use_fallback) {
706 		sock_put(sk); /* passive closing */
707 		sk->sk_state = SMC_CLOSED;
708 	} else {
709 		if (sk->sk_state == SMC_CLOSED)
710 			smc_conn_free(&smc->conn);
711 	}
712 	release_sock(sk);
713 	sk->sk_prot->unhash(sk);
714 	sock_put(sk); /* final sock_put */
715 }
716 
717 static int smc_serv_conf_first_link(struct smc_sock *smc)
718 {
719 	struct smc_link_group *lgr = smc->conn.lgr;
720 	struct smc_link *link;
721 	int rest;
722 	int rc;
723 
724 	link = &lgr->lnk[SMC_SINGLE_LINK];
725 
726 	if (smc_reg_rmb(link, smc->conn.rmb_desc))
727 		return SMC_CLC_DECL_INTERR;
728 
729 	/* send CONFIRM LINK request to client over the RoCE fabric */
730 	rc = smc_llc_send_confirm_link(link,
731 				       link->smcibdev->mac[link->ibport - 1],
732 				       &link->smcibdev->gid[link->ibport - 1],
733 				       SMC_LLC_REQ);
734 	if (rc < 0)
735 		return SMC_CLC_DECL_TCL;
736 
737 	/* receive CONFIRM LINK response from client over the RoCE fabric */
738 	rest = wait_for_completion_interruptible_timeout(
739 		&link->llc_confirm_resp,
740 		SMC_LLC_WAIT_FIRST_TIME);
741 	if (rest <= 0) {
742 		struct smc_clc_msg_decline dclc;
743 
744 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
745 				      SMC_CLC_DECLINE);
746 		return rc;
747 	}
748 
749 	if (link->llc_confirm_resp_rc)
750 		return SMC_CLC_DECL_RMBE_EC;
751 
752 	/* send ADD LINK request to client over the RoCE fabric */
753 	rc = smc_llc_send_add_link(link,
754 				   link->smcibdev->mac[link->ibport - 1],
755 				   &link->smcibdev->gid[link->ibport - 1],
756 				   SMC_LLC_REQ);
757 	if (rc < 0)
758 		return SMC_CLC_DECL_TCL;
759 
760 	/* receive ADD LINK response from client over the RoCE fabric */
761 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
762 							 SMC_LLC_WAIT_TIME);
763 	if (rest <= 0) {
764 		struct smc_clc_msg_decline dclc;
765 
766 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
767 				      SMC_CLC_DECLINE);
768 		return rc;
769 	}
770 
771 	link->state = SMC_LNK_ACTIVE;
772 
773 	return 0;
774 }
775 
776 /* setup for RDMA connection of server */
777 static void smc_listen_work(struct work_struct *work)
778 {
779 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
780 						smc_listen_work);
781 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
782 	struct socket *newclcsock = new_smc->clcsock;
783 	struct smc_sock *lsmc = new_smc->listen_smc;
784 	struct smc_clc_msg_accept_confirm cclc;
785 	int local_contact = SMC_REUSE_CONTACT;
786 	struct sock *newsmcsk = &new_smc->sk;
787 	struct smc_clc_msg_proposal *pclc;
788 	struct smc_ib_device *smcibdev;
789 	u8 buf[SMC_CLC_MAX_LEN];
790 	struct smc_link *link;
791 	int reason_code = 0;
792 	int rc = 0;
793 	u8 ibport;
794 
795 	/* check if peer is smc capable */
796 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
797 		new_smc->use_fallback = true;
798 		goto out_connected;
799 	}
800 
801 	/* do inband token exchange -
802 	 *wait for and receive SMC Proposal CLC message
803 	 */
804 	reason_code = smc_clc_wait_msg(new_smc, &buf, sizeof(buf),
805 				       SMC_CLC_PROPOSAL);
806 	if (reason_code < 0)
807 		goto out_err;
808 	if (reason_code > 0)
809 		goto decline_rdma;
810 
811 	/* IPSec connections opt out of SMC-R optimizations */
812 	if (using_ipsec(new_smc)) {
813 		reason_code = SMC_CLC_DECL_IPSEC;
814 		goto decline_rdma;
815 	}
816 
817 	/* PNET table look up: search active ib_device and port
818 	 * within same PNETID that also contains the ethernet device
819 	 * used for the internal TCP socket
820 	 */
821 	smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
822 	if (!smcibdev) {
823 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
824 		goto decline_rdma;
825 	}
826 
827 	pclc = (struct smc_clc_msg_proposal *)&buf;
828 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
829 
830 	rc = smc_clc_prfx_match(newclcsock, pclc_prfx);
831 	if (rc) {
832 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
833 		goto decline_rdma;
834 	}
835 
836 	/* allocate connection / link group */
837 	mutex_lock(&smc_create_lgr_pending);
838 	local_contact = smc_conn_create(new_smc, smcibdev, ibport, &pclc->lcl,
839 					0);
840 	if (local_contact < 0) {
841 		rc = local_contact;
842 		if (rc == -ENOMEM)
843 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
844 		goto decline_rdma_unlock;
845 	}
846 	link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
847 
848 	/* create send buffer and rmb */
849 	rc = smc_buf_create(new_smc);
850 	if (rc) {
851 		reason_code = SMC_CLC_DECL_MEM;
852 		goto decline_rdma_unlock;
853 	}
854 
855 	smc_close_init(new_smc);
856 	smc_rx_init(new_smc);
857 
858 	if (local_contact != SMC_FIRST_CONTACT) {
859 		if (!new_smc->conn.rmb_desc->reused) {
860 			if (smc_reg_rmb(link, new_smc->conn.rmb_desc)) {
861 				reason_code = SMC_CLC_DECL_INTERR;
862 				goto decline_rdma_unlock;
863 			}
864 		}
865 	}
866 	smc_rmb_sync_sg_for_device(&new_smc->conn);
867 
868 	rc = smc_clc_send_accept(new_smc, local_contact);
869 	if (rc)
870 		goto out_err_unlock;
871 
872 	/* receive SMC Confirm CLC message */
873 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
874 				       SMC_CLC_CONFIRM);
875 	if (reason_code < 0)
876 		goto out_err_unlock;
877 	if (reason_code > 0)
878 		goto decline_rdma_unlock;
879 	smc_conn_save_peer_info(new_smc, &cclc);
880 	if (local_contact == SMC_FIRST_CONTACT)
881 		smc_link_save_peer_info(link, &cclc);
882 
883 	rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
884 	if (rc) {
885 		reason_code = SMC_CLC_DECL_INTERR;
886 		goto decline_rdma_unlock;
887 	}
888 
889 	if (local_contact == SMC_FIRST_CONTACT) {
890 		rc = smc_ib_ready_link(link);
891 		if (rc) {
892 			reason_code = SMC_CLC_DECL_INTERR;
893 			goto decline_rdma_unlock;
894 		}
895 		/* QP confirmation over RoCE fabric */
896 		reason_code = smc_serv_conf_first_link(new_smc);
897 		if (reason_code < 0)
898 			/* peer is not aware of a problem */
899 			goto out_err_unlock;
900 		if (reason_code > 0)
901 			goto decline_rdma_unlock;
902 	}
903 
904 	smc_tx_init(new_smc);
905 	mutex_unlock(&smc_create_lgr_pending);
906 
907 out_connected:
908 	sk_refcnt_debug_inc(newsmcsk);
909 	if (newsmcsk->sk_state == SMC_INIT)
910 		newsmcsk->sk_state = SMC_ACTIVE;
911 enqueue:
912 	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
913 	if (lsmc->sk.sk_state == SMC_LISTEN) {
914 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
915 	} else { /* no longer listening */
916 		smc_close_non_accepted(newsmcsk);
917 	}
918 	release_sock(&lsmc->sk);
919 
920 	/* Wake up accept */
921 	lsmc->sk.sk_data_ready(&lsmc->sk);
922 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
923 	return;
924 
925 decline_rdma_unlock:
926 	if (local_contact == SMC_FIRST_CONTACT)
927 		smc_lgr_forget(new_smc->conn.lgr);
928 	mutex_unlock(&smc_create_lgr_pending);
929 decline_rdma:
930 	/* RDMA setup failed, switch back to TCP */
931 	smc_conn_free(&new_smc->conn);
932 	new_smc->use_fallback = true;
933 	if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
934 		if (smc_clc_send_decline(new_smc, reason_code) < 0)
935 			goto out_err;
936 	}
937 	goto out_connected;
938 
939 out_err_unlock:
940 	if (local_contact == SMC_FIRST_CONTACT)
941 		smc_lgr_forget(new_smc->conn.lgr);
942 	mutex_unlock(&smc_create_lgr_pending);
943 out_err:
944 	if (newsmcsk->sk_state == SMC_INIT)
945 		sock_put(&new_smc->sk); /* passive closing */
946 	newsmcsk->sk_state = SMC_CLOSED;
947 	smc_conn_free(&new_smc->conn);
948 	goto enqueue; /* queue new sock with sk_err set */
949 }
950 
951 static void smc_tcp_listen_work(struct work_struct *work)
952 {
953 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
954 					     tcp_listen_work);
955 	struct sock *lsk = &lsmc->sk;
956 	struct smc_sock *new_smc;
957 	int rc = 0;
958 
959 	lock_sock(lsk);
960 	while (lsk->sk_state == SMC_LISTEN) {
961 		rc = smc_clcsock_accept(lsmc, &new_smc);
962 		if (rc)
963 			goto out;
964 		if (!new_smc)
965 			continue;
966 
967 		new_smc->listen_smc = lsmc;
968 		new_smc->use_fallback = false; /* assume rdma capability first*/
969 		sock_hold(lsk); /* sock_put in smc_listen_work */
970 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
971 		smc_copy_sock_settings_to_smc(new_smc);
972 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
973 		if (!schedule_work(&new_smc->smc_listen_work))
974 			sock_put(&new_smc->sk);
975 	}
976 
977 out:
978 	release_sock(lsk);
979 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
980 }
981 
982 static int smc_listen(struct socket *sock, int backlog)
983 {
984 	struct sock *sk = sock->sk;
985 	struct smc_sock *smc;
986 	int rc;
987 
988 	smc = smc_sk(sk);
989 	lock_sock(sk);
990 
991 	rc = -EINVAL;
992 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
993 		goto out;
994 
995 	rc = 0;
996 	if (sk->sk_state == SMC_LISTEN) {
997 		sk->sk_max_ack_backlog = backlog;
998 		goto out;
999 	}
1000 	/* some socket options are handled in core, so we could not apply
1001 	 * them to the clc socket -- copy smc socket options to clc socket
1002 	 */
1003 	smc_copy_sock_settings_to_clc(smc);
1004 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1005 
1006 	rc = kernel_listen(smc->clcsock, backlog);
1007 	if (rc)
1008 		goto out;
1009 	sk->sk_max_ack_backlog = backlog;
1010 	sk->sk_ack_backlog = 0;
1011 	sk->sk_state = SMC_LISTEN;
1012 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
1013 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1014 	if (!schedule_work(&smc->tcp_listen_work))
1015 		sock_put(sk);
1016 
1017 out:
1018 	release_sock(sk);
1019 	return rc;
1020 }
1021 
1022 static int smc_accept(struct socket *sock, struct socket *new_sock,
1023 		      int flags, bool kern)
1024 {
1025 	struct sock *sk = sock->sk, *nsk;
1026 	DECLARE_WAITQUEUE(wait, current);
1027 	struct smc_sock *lsmc;
1028 	long timeo;
1029 	int rc = 0;
1030 
1031 	lsmc = smc_sk(sk);
1032 	sock_hold(sk); /* sock_put below */
1033 	lock_sock(sk);
1034 
1035 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1036 		rc = -EINVAL;
1037 		goto out;
1038 	}
1039 
1040 	/* Wait for an incoming connection */
1041 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1042 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1043 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1044 		set_current_state(TASK_INTERRUPTIBLE);
1045 		if (!timeo) {
1046 			rc = -EAGAIN;
1047 			break;
1048 		}
1049 		release_sock(sk);
1050 		timeo = schedule_timeout(timeo);
1051 		/* wakeup by sk_data_ready in smc_listen_work() */
1052 		sched_annotate_sleep();
1053 		lock_sock(sk);
1054 		if (signal_pending(current)) {
1055 			rc = sock_intr_errno(timeo);
1056 			break;
1057 		}
1058 	}
1059 	set_current_state(TASK_RUNNING);
1060 	remove_wait_queue(sk_sleep(sk), &wait);
1061 
1062 	if (!rc)
1063 		rc = sock_error(nsk);
1064 
1065 out:
1066 	release_sock(sk);
1067 	sock_put(sk); /* sock_hold above */
1068 	return rc;
1069 }
1070 
1071 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1072 		       int peer)
1073 {
1074 	struct smc_sock *smc;
1075 
1076 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1077 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1078 		return -ENOTCONN;
1079 
1080 	smc = smc_sk(sock->sk);
1081 
1082 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1083 }
1084 
1085 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1086 {
1087 	struct sock *sk = sock->sk;
1088 	struct smc_sock *smc;
1089 	int rc = -EPIPE;
1090 
1091 	smc = smc_sk(sk);
1092 	lock_sock(sk);
1093 	if ((sk->sk_state != SMC_ACTIVE) &&
1094 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1095 	    (sk->sk_state != SMC_INIT))
1096 		goto out;
1097 	if (smc->use_fallback)
1098 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1099 	else
1100 		rc = smc_tx_sendmsg(smc, msg, len);
1101 out:
1102 	release_sock(sk);
1103 	return rc;
1104 }
1105 
1106 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1107 		       int flags)
1108 {
1109 	struct sock *sk = sock->sk;
1110 	struct smc_sock *smc;
1111 	int rc = -ENOTCONN;
1112 
1113 	smc = smc_sk(sk);
1114 	lock_sock(sk);
1115 	if ((sk->sk_state == SMC_INIT) ||
1116 	    (sk->sk_state == SMC_LISTEN) ||
1117 	    (sk->sk_state == SMC_CLOSED))
1118 		goto out;
1119 
1120 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1121 		rc = 0;
1122 		goto out;
1123 	}
1124 
1125 	if (smc->use_fallback)
1126 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1127 	else
1128 		rc = smc_rx_recvmsg(smc, msg, len, flags);
1129 
1130 out:
1131 	release_sock(sk);
1132 	return rc;
1133 }
1134 
1135 static __poll_t smc_accept_poll(struct sock *parent)
1136 {
1137 	struct smc_sock *isk = smc_sk(parent);
1138 	__poll_t mask = 0;
1139 
1140 	spin_lock(&isk->accept_q_lock);
1141 	if (!list_empty(&isk->accept_q))
1142 		mask = EPOLLIN | EPOLLRDNORM;
1143 	spin_unlock(&isk->accept_q_lock);
1144 
1145 	return mask;
1146 }
1147 
1148 static __poll_t smc_poll(struct file *file, struct socket *sock,
1149 			     poll_table *wait)
1150 {
1151 	struct sock *sk = sock->sk;
1152 	__poll_t mask = 0;
1153 	struct smc_sock *smc;
1154 	int rc;
1155 
1156 	if (!sk)
1157 		return EPOLLNVAL;
1158 
1159 	smc = smc_sk(sock->sk);
1160 	sock_hold(sk);
1161 	lock_sock(sk);
1162 	if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1163 		/* delegate to CLC child sock */
1164 		release_sock(sk);
1165 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1166 		lock_sock(sk);
1167 		sk->sk_err = smc->clcsock->sk->sk_err;
1168 		if (sk->sk_err) {
1169 			mask |= EPOLLERR;
1170 		} else {
1171 			/* if non-blocking connect finished ... */
1172 			if (sk->sk_state == SMC_INIT &&
1173 			    mask & EPOLLOUT &&
1174 			    smc->clcsock->sk->sk_state != TCP_CLOSE) {
1175 				rc = smc_connect_rdma(smc);
1176 				if (rc < 0)
1177 					mask |= EPOLLERR;
1178 				/* success cases including fallback */
1179 				mask |= EPOLLOUT | EPOLLWRNORM;
1180 			}
1181 		}
1182 	} else {
1183 		if (sk->sk_state != SMC_CLOSED) {
1184 			release_sock(sk);
1185 			sock_poll_wait(file, sk_sleep(sk), wait);
1186 			lock_sock(sk);
1187 		}
1188 		if (sk->sk_err)
1189 			mask |= EPOLLERR;
1190 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1191 		    (sk->sk_state == SMC_CLOSED))
1192 			mask |= EPOLLHUP;
1193 		if (sk->sk_state == SMC_LISTEN) {
1194 			/* woken up by sk_data_ready in smc_listen_work() */
1195 			mask = smc_accept_poll(sk);
1196 		} else {
1197 			if (atomic_read(&smc->conn.sndbuf_space) ||
1198 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1199 				mask |= EPOLLOUT | EPOLLWRNORM;
1200 			} else {
1201 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1202 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1203 			}
1204 			if (atomic_read(&smc->conn.bytes_to_rcv))
1205 				mask |= EPOLLIN | EPOLLRDNORM;
1206 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1207 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1208 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1209 				mask |= EPOLLIN;
1210 		}
1211 
1212 	}
1213 	release_sock(sk);
1214 	sock_put(sk);
1215 
1216 	return mask;
1217 }
1218 
1219 static int smc_shutdown(struct socket *sock, int how)
1220 {
1221 	struct sock *sk = sock->sk;
1222 	struct smc_sock *smc;
1223 	int rc = -EINVAL;
1224 	int rc1 = 0;
1225 
1226 	smc = smc_sk(sk);
1227 
1228 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1229 		return rc;
1230 
1231 	lock_sock(sk);
1232 
1233 	rc = -ENOTCONN;
1234 	if ((sk->sk_state != SMC_LISTEN) &&
1235 	    (sk->sk_state != SMC_ACTIVE) &&
1236 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1237 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1238 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1239 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1240 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1241 		goto out;
1242 	if (smc->use_fallback) {
1243 		rc = kernel_sock_shutdown(smc->clcsock, how);
1244 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1245 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1246 			sk->sk_state = SMC_CLOSED;
1247 		goto out;
1248 	}
1249 	switch (how) {
1250 	case SHUT_RDWR:		/* shutdown in both directions */
1251 		rc = smc_close_active(smc);
1252 		break;
1253 	case SHUT_WR:
1254 		rc = smc_close_shutdown_write(smc);
1255 		break;
1256 	case SHUT_RD:
1257 		rc = 0;
1258 		/* nothing more to do because peer is not involved */
1259 		break;
1260 	}
1261 	if (smc->clcsock)
1262 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1263 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1264 	sk->sk_shutdown |= how + 1;
1265 
1266 out:
1267 	release_sock(sk);
1268 	return rc ? rc : rc1;
1269 }
1270 
1271 static int smc_setsockopt(struct socket *sock, int level, int optname,
1272 			  char __user *optval, unsigned int optlen)
1273 {
1274 	struct sock *sk = sock->sk;
1275 	struct smc_sock *smc;
1276 
1277 	smc = smc_sk(sk);
1278 
1279 	/* generic setsockopts reaching us here always apply to the
1280 	 * CLC socket
1281 	 */
1282 	return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1283 					     optval, optlen);
1284 }
1285 
1286 static int smc_getsockopt(struct socket *sock, int level, int optname,
1287 			  char __user *optval, int __user *optlen)
1288 {
1289 	struct smc_sock *smc;
1290 
1291 	smc = smc_sk(sock->sk);
1292 	/* socket options apply to the CLC socket */
1293 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1294 					     optval, optlen);
1295 }
1296 
1297 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1298 		     unsigned long arg)
1299 {
1300 	struct smc_sock *smc;
1301 
1302 	smc = smc_sk(sock->sk);
1303 	if (smc->use_fallback)
1304 		return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1305 	else
1306 		return sock_no_ioctl(sock, cmd, arg);
1307 }
1308 
1309 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1310 			    int offset, size_t size, int flags)
1311 {
1312 	struct sock *sk = sock->sk;
1313 	struct smc_sock *smc;
1314 	int rc = -EPIPE;
1315 
1316 	smc = smc_sk(sk);
1317 	lock_sock(sk);
1318 	if (sk->sk_state != SMC_ACTIVE) {
1319 		release_sock(sk);
1320 		goto out;
1321 	}
1322 	release_sock(sk);
1323 	if (smc->use_fallback)
1324 		rc = kernel_sendpage(smc->clcsock, page, offset,
1325 				     size, flags);
1326 	else
1327 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1328 
1329 out:
1330 	return rc;
1331 }
1332 
1333 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1334 			       struct pipe_inode_info *pipe, size_t len,
1335 				    unsigned int flags)
1336 {
1337 	struct sock *sk = sock->sk;
1338 	struct smc_sock *smc;
1339 	int rc = -ENOTCONN;
1340 
1341 	smc = smc_sk(sk);
1342 	lock_sock(sk);
1343 	if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1344 		goto out;
1345 	if (smc->use_fallback) {
1346 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1347 						    pipe, len, flags);
1348 	} else {
1349 		rc = -EOPNOTSUPP;
1350 	}
1351 out:
1352 	release_sock(sk);
1353 	return rc;
1354 }
1355 
1356 /* must look like tcp */
1357 static const struct proto_ops smc_sock_ops = {
1358 	.family		= PF_SMC,
1359 	.owner		= THIS_MODULE,
1360 	.release	= smc_release,
1361 	.bind		= smc_bind,
1362 	.connect	= smc_connect,
1363 	.socketpair	= sock_no_socketpair,
1364 	.accept		= smc_accept,
1365 	.getname	= smc_getname,
1366 	.poll		= smc_poll,
1367 	.ioctl		= smc_ioctl,
1368 	.listen		= smc_listen,
1369 	.shutdown	= smc_shutdown,
1370 	.setsockopt	= smc_setsockopt,
1371 	.getsockopt	= smc_getsockopt,
1372 	.sendmsg	= smc_sendmsg,
1373 	.recvmsg	= smc_recvmsg,
1374 	.mmap		= sock_no_mmap,
1375 	.sendpage	= smc_sendpage,
1376 	.splice_read	= smc_splice_read,
1377 };
1378 
1379 static int smc_create(struct net *net, struct socket *sock, int protocol,
1380 		      int kern)
1381 {
1382 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1383 	struct smc_sock *smc;
1384 	struct sock *sk;
1385 	int rc;
1386 
1387 	rc = -ESOCKTNOSUPPORT;
1388 	if (sock->type != SOCK_STREAM)
1389 		goto out;
1390 
1391 	rc = -EPROTONOSUPPORT;
1392 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1393 		goto out;
1394 
1395 	rc = -ENOBUFS;
1396 	sock->ops = &smc_sock_ops;
1397 	sk = smc_sock_alloc(net, sock, protocol);
1398 	if (!sk)
1399 		goto out;
1400 
1401 	/* create internal TCP socket for CLC handshake and fallback */
1402 	smc = smc_sk(sk);
1403 	smc->use_fallback = false; /* assume rdma capability first */
1404 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1405 			      &smc->clcsock);
1406 	if (rc) {
1407 		sk_common_release(sk);
1408 		goto out;
1409 	}
1410 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1411 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1412 
1413 out:
1414 	return rc;
1415 }
1416 
1417 static const struct net_proto_family smc_sock_family_ops = {
1418 	.family	= PF_SMC,
1419 	.owner	= THIS_MODULE,
1420 	.create	= smc_create,
1421 };
1422 
1423 static int __init smc_init(void)
1424 {
1425 	int rc;
1426 
1427 	rc = smc_pnet_init();
1428 	if (rc)
1429 		return rc;
1430 
1431 	rc = smc_llc_init();
1432 	if (rc) {
1433 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1434 		goto out_pnet;
1435 	}
1436 
1437 	rc = smc_cdc_init();
1438 	if (rc) {
1439 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1440 		goto out_pnet;
1441 	}
1442 
1443 	rc = proto_register(&smc_proto, 1);
1444 	if (rc) {
1445 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
1446 		goto out_pnet;
1447 	}
1448 
1449 	rc = proto_register(&smc_proto6, 1);
1450 	if (rc) {
1451 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
1452 		goto out_proto;
1453 	}
1454 
1455 	rc = sock_register(&smc_sock_family_ops);
1456 	if (rc) {
1457 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
1458 		goto out_proto6;
1459 	}
1460 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1461 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
1462 
1463 	rc = smc_ib_register_client();
1464 	if (rc) {
1465 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
1466 		goto out_sock;
1467 	}
1468 
1469 	static_branch_enable(&tcp_have_smc);
1470 	return 0;
1471 
1472 out_sock:
1473 	sock_unregister(PF_SMC);
1474 out_proto6:
1475 	proto_unregister(&smc_proto6);
1476 out_proto:
1477 	proto_unregister(&smc_proto);
1478 out_pnet:
1479 	smc_pnet_exit();
1480 	return rc;
1481 }
1482 
1483 static void __exit smc_exit(void)
1484 {
1485 	struct smc_link_group *lgr, *lg;
1486 	LIST_HEAD(lgr_freeing_list);
1487 
1488 	spin_lock_bh(&smc_lgr_list.lock);
1489 	if (!list_empty(&smc_lgr_list.list))
1490 		list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1491 	spin_unlock_bh(&smc_lgr_list.lock);
1492 	list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1493 		list_del_init(&lgr->list);
1494 		cancel_delayed_work_sync(&lgr->free_work);
1495 		smc_lgr_free(lgr); /* free link group */
1496 	}
1497 	static_branch_disable(&tcp_have_smc);
1498 	smc_ib_unregister_client();
1499 	sock_unregister(PF_SMC);
1500 	proto_unregister(&smc_proto6);
1501 	proto_unregister(&smc_proto);
1502 	smc_pnet_exit();
1503 }
1504 
1505 module_init(smc_init);
1506 module_exit(smc_exit);
1507 
1508 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1509 MODULE_DESCRIPTION("smc socket address family");
1510 MODULE_LICENSE("GPL");
1511 MODULE_ALIAS_NETPROTO(PF_SMC);
1512