xref: /openbmc/linux/net/smc/af_smc.c (revision 8e20ba2e)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
6  *  applies to SOCK_STREAM sockets only
7  *  offers an alternative communication option for TCP-protocol sockets
8  *  applicable with RoCE-cards only
9  *
10  *  Initial restrictions:
11  *    - support for alternate links postponed
12  *
13  *  Copyright IBM Corp. 2016, 2018
14  *
15  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
16  *              based on prototype from Frank Blaschka
17  */
18 
19 #define KMSG_COMPONENT "smc"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/module.h>
23 #include <linux/socket.h>
24 #include <linux/workqueue.h>
25 #include <linux/in.h>
26 #include <linux/sched/signal.h>
27 #include <linux/if_vlan.h>
28 
29 #include <net/sock.h>
30 #include <net/tcp.h>
31 #include <net/smc.h>
32 #include <asm/ioctls.h>
33 
34 #include <net/net_namespace.h>
35 #include <net/netns/generic.h>
36 #include "smc_netns.h"
37 
38 #include "smc.h"
39 #include "smc_clc.h"
40 #include "smc_llc.h"
41 #include "smc_cdc.h"
42 #include "smc_core.h"
43 #include "smc_ib.h"
44 #include "smc_ism.h"
45 #include "smc_pnet.h"
46 #include "smc_tx.h"
47 #include "smc_rx.h"
48 #include "smc_close.h"
49 
50 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
51 						 * creation on server
52 						 */
53 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
54 						 * creation on client
55 						 */
56 
57 static void smc_tcp_listen_work(struct work_struct *);
58 static void smc_connect_work(struct work_struct *);
59 
60 static void smc_set_keepalive(struct sock *sk, int val)
61 {
62 	struct smc_sock *smc = smc_sk(sk);
63 
64 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
65 }
66 
67 static struct smc_hashinfo smc_v4_hashinfo = {
68 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
69 };
70 
71 static struct smc_hashinfo smc_v6_hashinfo = {
72 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
73 };
74 
75 int smc_hash_sk(struct sock *sk)
76 {
77 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
78 	struct hlist_head *head;
79 
80 	head = &h->ht;
81 
82 	write_lock_bh(&h->lock);
83 	sk_add_node(sk, head);
84 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
85 	write_unlock_bh(&h->lock);
86 
87 	return 0;
88 }
89 EXPORT_SYMBOL_GPL(smc_hash_sk);
90 
91 void smc_unhash_sk(struct sock *sk)
92 {
93 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
94 
95 	write_lock_bh(&h->lock);
96 	if (sk_del_node_init(sk))
97 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
98 	write_unlock_bh(&h->lock);
99 }
100 EXPORT_SYMBOL_GPL(smc_unhash_sk);
101 
102 struct proto smc_proto = {
103 	.name		= "SMC",
104 	.owner		= THIS_MODULE,
105 	.keepalive	= smc_set_keepalive,
106 	.hash		= smc_hash_sk,
107 	.unhash		= smc_unhash_sk,
108 	.obj_size	= sizeof(struct smc_sock),
109 	.h.smc_hash	= &smc_v4_hashinfo,
110 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
111 };
112 EXPORT_SYMBOL_GPL(smc_proto);
113 
114 struct proto smc_proto6 = {
115 	.name		= "SMC6",
116 	.owner		= THIS_MODULE,
117 	.keepalive	= smc_set_keepalive,
118 	.hash		= smc_hash_sk,
119 	.unhash		= smc_unhash_sk,
120 	.obj_size	= sizeof(struct smc_sock),
121 	.h.smc_hash	= &smc_v6_hashinfo,
122 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
123 };
124 EXPORT_SYMBOL_GPL(smc_proto6);
125 
126 static void smc_restore_fallback_changes(struct smc_sock *smc)
127 {
128 	smc->clcsock->file->private_data = smc->sk.sk_socket;
129 	smc->clcsock->file = NULL;
130 }
131 
132 static int __smc_release(struct smc_sock *smc)
133 {
134 	struct sock *sk = &smc->sk;
135 	int rc = 0;
136 
137 	if (!smc->use_fallback) {
138 		rc = smc_close_active(smc);
139 		sock_set_flag(sk, SOCK_DEAD);
140 		sk->sk_shutdown |= SHUTDOWN_MASK;
141 	} else {
142 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
143 			sock_put(sk); /* passive closing */
144 		if (sk->sk_state == SMC_LISTEN) {
145 			/* wake up clcsock accept */
146 			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
147 		}
148 		sk->sk_state = SMC_CLOSED;
149 		sk->sk_state_change(sk);
150 		smc_restore_fallback_changes(smc);
151 	}
152 
153 	sk->sk_prot->unhash(sk);
154 
155 	if (sk->sk_state == SMC_CLOSED) {
156 		if (smc->clcsock) {
157 			release_sock(sk);
158 			smc_clcsock_release(smc);
159 			lock_sock(sk);
160 		}
161 		if (!smc->use_fallback)
162 			smc_conn_free(&smc->conn);
163 	}
164 
165 	return rc;
166 }
167 
168 static int smc_release(struct socket *sock)
169 {
170 	struct sock *sk = sock->sk;
171 	struct smc_sock *smc;
172 	int rc = 0;
173 
174 	if (!sk)
175 		goto out;
176 
177 	smc = smc_sk(sk);
178 
179 	/* cleanup for a dangling non-blocking connect */
180 	if (smc->connect_nonblock && sk->sk_state == SMC_INIT)
181 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
182 	flush_work(&smc->connect_work);
183 
184 	if (sk->sk_state == SMC_LISTEN)
185 		/* smc_close_non_accepted() is called and acquires
186 		 * sock lock for child sockets again
187 		 */
188 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
189 	else
190 		lock_sock(sk);
191 
192 	rc = __smc_release(smc);
193 
194 	/* detach socket */
195 	sock_orphan(sk);
196 	sock->sk = NULL;
197 	release_sock(sk);
198 
199 	sock_put(sk); /* final sock_put */
200 out:
201 	return rc;
202 }
203 
204 static void smc_destruct(struct sock *sk)
205 {
206 	if (sk->sk_state != SMC_CLOSED)
207 		return;
208 	if (!sock_flag(sk, SOCK_DEAD))
209 		return;
210 
211 	sk_refcnt_debug_dec(sk);
212 }
213 
214 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
215 				   int protocol)
216 {
217 	struct smc_sock *smc;
218 	struct proto *prot;
219 	struct sock *sk;
220 
221 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
222 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
223 	if (!sk)
224 		return NULL;
225 
226 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
227 	sk->sk_state = SMC_INIT;
228 	sk->sk_destruct = smc_destruct;
229 	sk->sk_protocol = protocol;
230 	smc = smc_sk(sk);
231 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
232 	INIT_WORK(&smc->connect_work, smc_connect_work);
233 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
234 	INIT_LIST_HEAD(&smc->accept_q);
235 	spin_lock_init(&smc->accept_q_lock);
236 	spin_lock_init(&smc->conn.send_lock);
237 	sk->sk_prot->hash(sk);
238 	sk_refcnt_debug_inc(sk);
239 	mutex_init(&smc->clcsock_release_lock);
240 
241 	return sk;
242 }
243 
244 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
245 		    int addr_len)
246 {
247 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
248 	struct sock *sk = sock->sk;
249 	struct smc_sock *smc;
250 	int rc;
251 
252 	smc = smc_sk(sk);
253 
254 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
255 	rc = -EINVAL;
256 	if (addr_len < sizeof(struct sockaddr_in))
257 		goto out;
258 
259 	rc = -EAFNOSUPPORT;
260 	if (addr->sin_family != AF_INET &&
261 	    addr->sin_family != AF_INET6 &&
262 	    addr->sin_family != AF_UNSPEC)
263 		goto out;
264 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
265 	if (addr->sin_family == AF_UNSPEC &&
266 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
267 		goto out;
268 
269 	lock_sock(sk);
270 
271 	/* Check if socket is already active */
272 	rc = -EINVAL;
273 	if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
274 		goto out_rel;
275 
276 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
277 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
278 
279 out_rel:
280 	release_sock(sk);
281 out:
282 	return rc;
283 }
284 
285 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
286 				   unsigned long mask)
287 {
288 	/* options we don't get control via setsockopt for */
289 	nsk->sk_type = osk->sk_type;
290 	nsk->sk_sndbuf = osk->sk_sndbuf;
291 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
292 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
293 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
294 	nsk->sk_mark = osk->sk_mark;
295 	nsk->sk_priority = osk->sk_priority;
296 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
297 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
298 	nsk->sk_err = osk->sk_err;
299 
300 	nsk->sk_flags &= ~mask;
301 	nsk->sk_flags |= osk->sk_flags & mask;
302 }
303 
304 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
305 			     (1UL << SOCK_KEEPOPEN) | \
306 			     (1UL << SOCK_LINGER) | \
307 			     (1UL << SOCK_BROADCAST) | \
308 			     (1UL << SOCK_TIMESTAMP) | \
309 			     (1UL << SOCK_DBG) | \
310 			     (1UL << SOCK_RCVTSTAMP) | \
311 			     (1UL << SOCK_RCVTSTAMPNS) | \
312 			     (1UL << SOCK_LOCALROUTE) | \
313 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
314 			     (1UL << SOCK_RXQ_OVFL) | \
315 			     (1UL << SOCK_WIFI_STATUS) | \
316 			     (1UL << SOCK_NOFCS) | \
317 			     (1UL << SOCK_FILTER_LOCKED) | \
318 			     (1UL << SOCK_TSTAMP_NEW))
319 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
320  * clc socket (since smc is not called for these options from net/core)
321  */
322 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
323 {
324 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
325 }
326 
327 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
328 			     (1UL << SOCK_KEEPOPEN) | \
329 			     (1UL << SOCK_LINGER) | \
330 			     (1UL << SOCK_DBG))
331 /* copy only settings and flags relevant for smc from clc to smc socket */
332 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
333 {
334 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
335 }
336 
337 /* register a new rmb, send confirm_rkey msg to register with peer */
338 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc,
339 		       bool conf_rkey)
340 {
341 	if (!rmb_desc->wr_reg) {
342 		/* register memory region for new rmb */
343 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
344 			rmb_desc->regerr = 1;
345 			return -EFAULT;
346 		}
347 		rmb_desc->wr_reg = 1;
348 	}
349 	if (!conf_rkey)
350 		return 0;
351 	/* exchange confirm_rkey msg with peer */
352 	if (smc_llc_do_confirm_rkey(link, rmb_desc)) {
353 		rmb_desc->regerr = 1;
354 		return -EFAULT;
355 	}
356 	return 0;
357 }
358 
359 static int smc_clnt_conf_first_link(struct smc_sock *smc)
360 {
361 	struct net *net = sock_net(smc->clcsock->sk);
362 	struct smc_link_group *lgr = smc->conn.lgr;
363 	struct smc_link *link;
364 	int rest;
365 	int rc;
366 
367 	link = &lgr->lnk[SMC_SINGLE_LINK];
368 	/* receive CONFIRM LINK request from server over RoCE fabric */
369 	rest = wait_for_completion_interruptible_timeout(
370 		&link->llc_confirm,
371 		SMC_LLC_WAIT_FIRST_TIME);
372 	if (rest <= 0) {
373 		struct smc_clc_msg_decline dclc;
374 
375 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
376 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
377 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
378 	}
379 
380 	if (link->llc_confirm_rc)
381 		return SMC_CLC_DECL_RMBE_EC;
382 
383 	rc = smc_ib_modify_qp_rts(link);
384 	if (rc)
385 		return SMC_CLC_DECL_ERR_RDYLNK;
386 
387 	smc_wr_remember_qp_attr(link);
388 
389 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
390 		return SMC_CLC_DECL_ERR_REGRMB;
391 
392 	/* send CONFIRM LINK response over RoCE fabric */
393 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
394 	if (rc < 0)
395 		return SMC_CLC_DECL_TIMEOUT_CL;
396 
397 	/* receive ADD LINK request from server over RoCE fabric */
398 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
399 							 SMC_LLC_WAIT_TIME);
400 	if (rest <= 0) {
401 		struct smc_clc_msg_decline dclc;
402 
403 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
404 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
405 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
406 	}
407 
408 	/* send add link reject message, only one link supported for now */
409 	rc = smc_llc_send_add_link(link,
410 				   link->smcibdev->mac[link->ibport - 1],
411 				   link->gid, SMC_LLC_RESP);
412 	if (rc < 0)
413 		return SMC_CLC_DECL_TIMEOUT_AL;
414 
415 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
416 
417 	return 0;
418 }
419 
420 static void smcr_conn_save_peer_info(struct smc_sock *smc,
421 				     struct smc_clc_msg_accept_confirm *clc)
422 {
423 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
424 
425 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
426 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
427 	smc->conn.peer_rmbe_size = bufsize;
428 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
429 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
430 }
431 
432 static void smcd_conn_save_peer_info(struct smc_sock *smc,
433 				     struct smc_clc_msg_accept_confirm *clc)
434 {
435 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
436 
437 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
438 	smc->conn.peer_token = clc->token;
439 	/* msg header takes up space in the buffer */
440 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
441 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
442 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
443 }
444 
445 static void smc_conn_save_peer_info(struct smc_sock *smc,
446 				    struct smc_clc_msg_accept_confirm *clc)
447 {
448 	if (smc->conn.lgr->is_smcd)
449 		smcd_conn_save_peer_info(smc, clc);
450 	else
451 		smcr_conn_save_peer_info(smc, clc);
452 }
453 
454 static void smc_link_save_peer_info(struct smc_link *link,
455 				    struct smc_clc_msg_accept_confirm *clc)
456 {
457 	link->peer_qpn = ntoh24(clc->qpn);
458 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
459 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
460 	link->peer_psn = ntoh24(clc->psn);
461 	link->peer_mtu = clc->qp_mtu;
462 }
463 
464 static void smc_switch_to_fallback(struct smc_sock *smc)
465 {
466 	smc->use_fallback = true;
467 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
468 		smc->clcsock->file = smc->sk.sk_socket->file;
469 		smc->clcsock->file->private_data = smc->clcsock;
470 	}
471 }
472 
473 /* fall back during connect */
474 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
475 {
476 	smc_switch_to_fallback(smc);
477 	smc->fallback_rsn = reason_code;
478 	smc_copy_sock_settings_to_clc(smc);
479 	smc->connect_nonblock = 0;
480 	if (smc->sk.sk_state == SMC_INIT)
481 		smc->sk.sk_state = SMC_ACTIVE;
482 	return 0;
483 }
484 
485 /* decline and fall back during connect */
486 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
487 {
488 	int rc;
489 
490 	if (reason_code < 0) { /* error, fallback is not possible */
491 		if (smc->sk.sk_state == SMC_INIT)
492 			sock_put(&smc->sk); /* passive closing */
493 		return reason_code;
494 	}
495 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
496 		rc = smc_clc_send_decline(smc, reason_code);
497 		if (rc < 0) {
498 			if (smc->sk.sk_state == SMC_INIT)
499 				sock_put(&smc->sk); /* passive closing */
500 			return rc;
501 		}
502 	}
503 	return smc_connect_fallback(smc, reason_code);
504 }
505 
506 /* abort connecting */
507 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
508 			     int local_contact)
509 {
510 	if (local_contact == SMC_FIRST_CONTACT)
511 		smc_lgr_forget(smc->conn.lgr);
512 	if (smc->conn.lgr->is_smcd)
513 		/* there is only one lgr role for SMC-D; use server lock */
514 		mutex_unlock(&smc_server_lgr_pending);
515 	else
516 		mutex_unlock(&smc_client_lgr_pending);
517 
518 	smc_conn_free(&smc->conn);
519 	smc->connect_nonblock = 0;
520 	return reason_code;
521 }
522 
523 /* check if there is a rdma device available for this connection. */
524 /* called for connect and listen */
525 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
526 {
527 	/* PNET table look up: search active ib_device and port
528 	 * within same PNETID that also contains the ethernet device
529 	 * used for the internal TCP socket
530 	 */
531 	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
532 	if (!ini->ib_dev)
533 		return SMC_CLC_DECL_NOSMCRDEV;
534 	return 0;
535 }
536 
537 /* check if there is an ISM device available for this connection. */
538 /* called for connect and listen */
539 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
540 {
541 	/* Find ISM device with same PNETID as connecting interface  */
542 	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
543 	if (!ini->ism_dev)
544 		return SMC_CLC_DECL_NOSMCDDEV;
545 	return 0;
546 }
547 
548 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
549 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
550 				      struct smc_init_info *ini)
551 {
552 	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id))
553 		return SMC_CLC_DECL_ISMVLANERR;
554 	return 0;
555 }
556 
557 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
558  * used, the VLAN ID will be registered again during the connection setup.
559  */
560 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
561 					struct smc_init_info *ini)
562 {
563 	if (!is_smcd)
564 		return 0;
565 	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id))
566 		return SMC_CLC_DECL_CNFERR;
567 	return 0;
568 }
569 
570 /* CLC handshake during connect */
571 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
572 			   struct smc_clc_msg_accept_confirm *aclc,
573 			   struct smc_init_info *ini)
574 {
575 	int rc = 0;
576 
577 	/* do inband token exchange */
578 	rc = smc_clc_send_proposal(smc, smc_type, ini);
579 	if (rc)
580 		return rc;
581 	/* receive SMC Accept CLC message */
582 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT,
583 				CLC_WAIT_TIME);
584 }
585 
586 /* setup for RDMA connection of client */
587 static int smc_connect_rdma(struct smc_sock *smc,
588 			    struct smc_clc_msg_accept_confirm *aclc,
589 			    struct smc_init_info *ini)
590 {
591 	struct smc_link *link;
592 	int reason_code = 0;
593 
594 	ini->is_smcd = false;
595 	ini->ib_lcl = &aclc->lcl;
596 	ini->ib_clcqpn = ntoh24(aclc->qpn);
597 	ini->srv_first_contact = aclc->hdr.flag;
598 
599 	mutex_lock(&smc_client_lgr_pending);
600 	reason_code = smc_conn_create(smc, ini);
601 	if (reason_code) {
602 		mutex_unlock(&smc_client_lgr_pending);
603 		return reason_code;
604 	}
605 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
606 
607 	smc_conn_save_peer_info(smc, aclc);
608 
609 	/* create send buffer and rmb */
610 	if (smc_buf_create(smc, false))
611 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
612 					 ini->cln_first_contact);
613 
614 	if (ini->cln_first_contact == SMC_FIRST_CONTACT)
615 		smc_link_save_peer_info(link, aclc);
616 
617 	if (smc_rmb_rtoken_handling(&smc->conn, aclc))
618 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
619 					 ini->cln_first_contact);
620 
621 	smc_close_init(smc);
622 	smc_rx_init(smc);
623 
624 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
625 		if (smc_ib_ready_link(link))
626 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
627 						 ini->cln_first_contact);
628 	} else {
629 		if (smc_reg_rmb(link, smc->conn.rmb_desc, true))
630 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
631 						 ini->cln_first_contact);
632 	}
633 	smc_rmb_sync_sg_for_device(&smc->conn);
634 
635 	reason_code = smc_clc_send_confirm(smc);
636 	if (reason_code)
637 		return smc_connect_abort(smc, reason_code,
638 					 ini->cln_first_contact);
639 
640 	smc_tx_init(smc);
641 
642 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
643 		/* QP confirmation over RoCE fabric */
644 		reason_code = smc_clnt_conf_first_link(smc);
645 		if (reason_code)
646 			return smc_connect_abort(smc, reason_code,
647 						 ini->cln_first_contact);
648 	}
649 	mutex_unlock(&smc_client_lgr_pending);
650 
651 	smc_copy_sock_settings_to_clc(smc);
652 	smc->connect_nonblock = 0;
653 	if (smc->sk.sk_state == SMC_INIT)
654 		smc->sk.sk_state = SMC_ACTIVE;
655 
656 	return 0;
657 }
658 
659 /* setup for ISM connection of client */
660 static int smc_connect_ism(struct smc_sock *smc,
661 			   struct smc_clc_msg_accept_confirm *aclc,
662 			   struct smc_init_info *ini)
663 {
664 	int rc = 0;
665 
666 	ini->is_smcd = true;
667 	ini->ism_gid = aclc->gid;
668 	ini->srv_first_contact = aclc->hdr.flag;
669 
670 	/* there is only one lgr role for SMC-D; use server lock */
671 	mutex_lock(&smc_server_lgr_pending);
672 	rc = smc_conn_create(smc, ini);
673 	if (rc) {
674 		mutex_unlock(&smc_server_lgr_pending);
675 		return rc;
676 	}
677 
678 	/* Create send and receive buffers */
679 	if (smc_buf_create(smc, true))
680 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
681 					 ini->cln_first_contact);
682 
683 	smc_conn_save_peer_info(smc, aclc);
684 	smc_close_init(smc);
685 	smc_rx_init(smc);
686 	smc_tx_init(smc);
687 
688 	rc = smc_clc_send_confirm(smc);
689 	if (rc)
690 		return smc_connect_abort(smc, rc, ini->cln_first_contact);
691 	mutex_unlock(&smc_server_lgr_pending);
692 
693 	smc_copy_sock_settings_to_clc(smc);
694 	smc->connect_nonblock = 0;
695 	if (smc->sk.sk_state == SMC_INIT)
696 		smc->sk.sk_state = SMC_ACTIVE;
697 
698 	return 0;
699 }
700 
701 /* perform steps before actually connecting */
702 static int __smc_connect(struct smc_sock *smc)
703 {
704 	bool ism_supported = false, rdma_supported = false;
705 	struct smc_clc_msg_accept_confirm aclc;
706 	struct smc_init_info ini = {0};
707 	int smc_type;
708 	int rc = 0;
709 
710 	if (smc->use_fallback)
711 		return smc_connect_fallback(smc, smc->fallback_rsn);
712 
713 	/* if peer has not signalled SMC-capability, fall back */
714 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
715 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
716 
717 	/* IPSec connections opt out of SMC-R optimizations */
718 	if (using_ipsec(smc))
719 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
720 
721 	/* get vlan id from IP device */
722 	if (smc_vlan_by_tcpsk(smc->clcsock, &ini))
723 		return smc_connect_decline_fallback(smc,
724 						    SMC_CLC_DECL_GETVLANERR);
725 
726 	/* check if there is an ism device available */
727 	if (!smc_find_ism_device(smc, &ini) &&
728 	    !smc_connect_ism_vlan_setup(smc, &ini)) {
729 		/* ISM is supported for this connection */
730 		ism_supported = true;
731 		smc_type = SMC_TYPE_D;
732 	}
733 
734 	/* check if there is a rdma device available */
735 	if (!smc_find_rdma_device(smc, &ini)) {
736 		/* RDMA is supported for this connection */
737 		rdma_supported = true;
738 		if (ism_supported)
739 			smc_type = SMC_TYPE_B; /* both */
740 		else
741 			smc_type = SMC_TYPE_R; /* only RDMA */
742 	}
743 
744 	/* if neither ISM nor RDMA are supported, fallback */
745 	if (!rdma_supported && !ism_supported)
746 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
747 
748 	/* perform CLC handshake */
749 	rc = smc_connect_clc(smc, smc_type, &aclc, &ini);
750 	if (rc) {
751 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
752 		return smc_connect_decline_fallback(smc, rc);
753 	}
754 
755 	/* depending on previous steps, connect using rdma or ism */
756 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
757 		rc = smc_connect_rdma(smc, &aclc, &ini);
758 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
759 		rc = smc_connect_ism(smc, &aclc, &ini);
760 	else
761 		rc = SMC_CLC_DECL_MODEUNSUPP;
762 	if (rc) {
763 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
764 		return smc_connect_decline_fallback(smc, rc);
765 	}
766 
767 	smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
768 	return 0;
769 }
770 
771 static void smc_connect_work(struct work_struct *work)
772 {
773 	struct smc_sock *smc = container_of(work, struct smc_sock,
774 					    connect_work);
775 	long timeo = smc->sk.sk_sndtimeo;
776 	int rc = 0;
777 
778 	if (!timeo)
779 		timeo = MAX_SCHEDULE_TIMEOUT;
780 	lock_sock(smc->clcsock->sk);
781 	if (smc->clcsock->sk->sk_err) {
782 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
783 	} else if ((1 << smc->clcsock->sk->sk_state) &
784 					(TCPF_SYN_SENT | TCP_SYN_RECV)) {
785 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
786 		if ((rc == -EPIPE) &&
787 		    ((1 << smc->clcsock->sk->sk_state) &
788 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
789 			rc = 0;
790 	}
791 	release_sock(smc->clcsock->sk);
792 	lock_sock(&smc->sk);
793 	if (rc != 0 || smc->sk.sk_err) {
794 		smc->sk.sk_state = SMC_CLOSED;
795 		if (rc == -EPIPE || rc == -EAGAIN)
796 			smc->sk.sk_err = EPIPE;
797 		else if (signal_pending(current))
798 			smc->sk.sk_err = -sock_intr_errno(timeo);
799 		goto out;
800 	}
801 
802 	rc = __smc_connect(smc);
803 	if (rc < 0)
804 		smc->sk.sk_err = -rc;
805 
806 out:
807 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
808 		if (smc->sk.sk_err) {
809 			smc->sk.sk_state_change(&smc->sk);
810 		} else { /* allow polling before and after fallback decision */
811 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
812 			smc->sk.sk_write_space(&smc->sk);
813 		}
814 	}
815 	release_sock(&smc->sk);
816 }
817 
818 static int smc_connect(struct socket *sock, struct sockaddr *addr,
819 		       int alen, int flags)
820 {
821 	struct sock *sk = sock->sk;
822 	struct smc_sock *smc;
823 	int rc = -EINVAL;
824 
825 	smc = smc_sk(sk);
826 
827 	/* separate smc parameter checking to be safe */
828 	if (alen < sizeof(addr->sa_family))
829 		goto out_err;
830 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
831 		goto out_err;
832 
833 	lock_sock(sk);
834 	switch (sk->sk_state) {
835 	default:
836 		goto out;
837 	case SMC_ACTIVE:
838 		rc = -EISCONN;
839 		goto out;
840 	case SMC_INIT:
841 		rc = 0;
842 		break;
843 	}
844 
845 	smc_copy_sock_settings_to_clc(smc);
846 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
847 	if (smc->connect_nonblock) {
848 		rc = -EALREADY;
849 		goto out;
850 	}
851 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
852 	if (rc && rc != -EINPROGRESS)
853 		goto out;
854 
855 	sock_hold(&smc->sk); /* sock put in passive closing */
856 	if (flags & O_NONBLOCK) {
857 		if (schedule_work(&smc->connect_work))
858 			smc->connect_nonblock = 1;
859 		rc = -EINPROGRESS;
860 	} else {
861 		rc = __smc_connect(smc);
862 		if (rc < 0)
863 			goto out;
864 		else
865 			rc = 0; /* success cases including fallback */
866 	}
867 
868 out:
869 	release_sock(sk);
870 out_err:
871 	return rc;
872 }
873 
874 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
875 {
876 	struct socket *new_clcsock = NULL;
877 	struct sock *lsk = &lsmc->sk;
878 	struct sock *new_sk;
879 	int rc = -EINVAL;
880 
881 	release_sock(lsk);
882 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
883 	if (!new_sk) {
884 		rc = -ENOMEM;
885 		lsk->sk_err = ENOMEM;
886 		*new_smc = NULL;
887 		lock_sock(lsk);
888 		goto out;
889 	}
890 	*new_smc = smc_sk(new_sk);
891 
892 	mutex_lock(&lsmc->clcsock_release_lock);
893 	if (lsmc->clcsock)
894 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
895 	mutex_unlock(&lsmc->clcsock_release_lock);
896 	lock_sock(lsk);
897 	if  (rc < 0)
898 		lsk->sk_err = -rc;
899 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
900 		new_sk->sk_prot->unhash(new_sk);
901 		if (new_clcsock)
902 			sock_release(new_clcsock);
903 		new_sk->sk_state = SMC_CLOSED;
904 		sock_set_flag(new_sk, SOCK_DEAD);
905 		sock_put(new_sk); /* final */
906 		*new_smc = NULL;
907 		goto out;
908 	}
909 
910 	(*new_smc)->clcsock = new_clcsock;
911 out:
912 	return rc;
913 }
914 
915 /* add a just created sock to the accept queue of the listen sock as
916  * candidate for a following socket accept call from user space
917  */
918 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
919 {
920 	struct smc_sock *par = smc_sk(parent);
921 
922 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
923 	spin_lock(&par->accept_q_lock);
924 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
925 	spin_unlock(&par->accept_q_lock);
926 	sk_acceptq_added(parent);
927 }
928 
929 /* remove a socket from the accept queue of its parental listening socket */
930 static void smc_accept_unlink(struct sock *sk)
931 {
932 	struct smc_sock *par = smc_sk(sk)->listen_smc;
933 
934 	spin_lock(&par->accept_q_lock);
935 	list_del_init(&smc_sk(sk)->accept_q);
936 	spin_unlock(&par->accept_q_lock);
937 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
938 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
939 }
940 
941 /* remove a sock from the accept queue to bind it to a new socket created
942  * for a socket accept call from user space
943  */
944 struct sock *smc_accept_dequeue(struct sock *parent,
945 				struct socket *new_sock)
946 {
947 	struct smc_sock *isk, *n;
948 	struct sock *new_sk;
949 
950 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
951 		new_sk = (struct sock *)isk;
952 
953 		smc_accept_unlink(new_sk);
954 		if (new_sk->sk_state == SMC_CLOSED) {
955 			new_sk->sk_prot->unhash(new_sk);
956 			if (isk->clcsock) {
957 				sock_release(isk->clcsock);
958 				isk->clcsock = NULL;
959 			}
960 			sock_put(new_sk); /* final */
961 			continue;
962 		}
963 		if (new_sock) {
964 			sock_graft(new_sk, new_sock);
965 			if (isk->use_fallback) {
966 				smc_sk(new_sk)->clcsock->file = new_sock->file;
967 				isk->clcsock->file->private_data = isk->clcsock;
968 			}
969 		}
970 		return new_sk;
971 	}
972 	return NULL;
973 }
974 
975 /* clean up for a created but never accepted sock */
976 void smc_close_non_accepted(struct sock *sk)
977 {
978 	struct smc_sock *smc = smc_sk(sk);
979 
980 	lock_sock(sk);
981 	if (!sk->sk_lingertime)
982 		/* wait for peer closing */
983 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
984 	__smc_release(smc);
985 	release_sock(sk);
986 	sock_put(sk); /* final sock_put */
987 }
988 
989 static int smc_serv_conf_first_link(struct smc_sock *smc)
990 {
991 	struct net *net = sock_net(smc->clcsock->sk);
992 	struct smc_link_group *lgr = smc->conn.lgr;
993 	struct smc_link *link;
994 	int rest;
995 	int rc;
996 
997 	link = &lgr->lnk[SMC_SINGLE_LINK];
998 
999 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
1000 		return SMC_CLC_DECL_ERR_REGRMB;
1001 
1002 	/* send CONFIRM LINK request to client over the RoCE fabric */
1003 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1004 	if (rc < 0)
1005 		return SMC_CLC_DECL_TIMEOUT_CL;
1006 
1007 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1008 	rest = wait_for_completion_interruptible_timeout(
1009 		&link->llc_confirm_resp,
1010 		SMC_LLC_WAIT_FIRST_TIME);
1011 	if (rest <= 0) {
1012 		struct smc_clc_msg_decline dclc;
1013 
1014 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1015 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1016 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1017 	}
1018 
1019 	if (link->llc_confirm_resp_rc)
1020 		return SMC_CLC_DECL_RMBE_EC;
1021 
1022 	/* send ADD LINK request to client over the RoCE fabric */
1023 	rc = smc_llc_send_add_link(link,
1024 				   link->smcibdev->mac[link->ibport - 1],
1025 				   link->gid, SMC_LLC_REQ);
1026 	if (rc < 0)
1027 		return SMC_CLC_DECL_TIMEOUT_AL;
1028 
1029 	/* receive ADD LINK response from client over the RoCE fabric */
1030 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
1031 							 SMC_LLC_WAIT_TIME);
1032 	if (rest <= 0) {
1033 		struct smc_clc_msg_decline dclc;
1034 
1035 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1036 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1037 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
1038 	}
1039 
1040 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
1041 
1042 	return 0;
1043 }
1044 
1045 /* listen worker: finish */
1046 static void smc_listen_out(struct smc_sock *new_smc)
1047 {
1048 	struct smc_sock *lsmc = new_smc->listen_smc;
1049 	struct sock *newsmcsk = &new_smc->sk;
1050 
1051 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1052 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1053 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1054 		release_sock(&lsmc->sk);
1055 	} else { /* no longer listening */
1056 		smc_close_non_accepted(newsmcsk);
1057 	}
1058 
1059 	/* Wake up accept */
1060 	lsmc->sk.sk_data_ready(&lsmc->sk);
1061 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1062 }
1063 
1064 /* listen worker: finish in state connected */
1065 static void smc_listen_out_connected(struct smc_sock *new_smc)
1066 {
1067 	struct sock *newsmcsk = &new_smc->sk;
1068 
1069 	sk_refcnt_debug_inc(newsmcsk);
1070 	if (newsmcsk->sk_state == SMC_INIT)
1071 		newsmcsk->sk_state = SMC_ACTIVE;
1072 
1073 	smc_listen_out(new_smc);
1074 }
1075 
1076 /* listen worker: finish in error state */
1077 static void smc_listen_out_err(struct smc_sock *new_smc)
1078 {
1079 	struct sock *newsmcsk = &new_smc->sk;
1080 
1081 	if (newsmcsk->sk_state == SMC_INIT)
1082 		sock_put(&new_smc->sk); /* passive closing */
1083 	newsmcsk->sk_state = SMC_CLOSED;
1084 	smc_conn_free(&new_smc->conn);
1085 
1086 	smc_listen_out(new_smc);
1087 }
1088 
1089 /* listen worker: decline and fall back if possible */
1090 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1091 			       int local_contact)
1092 {
1093 	/* RDMA setup failed, switch back to TCP */
1094 	if (local_contact == SMC_FIRST_CONTACT)
1095 		smc_lgr_forget(new_smc->conn.lgr);
1096 	if (reason_code < 0) { /* error, no fallback possible */
1097 		smc_listen_out_err(new_smc);
1098 		return;
1099 	}
1100 	smc_conn_free(&new_smc->conn);
1101 	smc_switch_to_fallback(new_smc);
1102 	new_smc->fallback_rsn = reason_code;
1103 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1104 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1105 			smc_listen_out_err(new_smc);
1106 			return;
1107 		}
1108 	}
1109 	smc_listen_out_connected(new_smc);
1110 }
1111 
1112 /* listen worker: check prefixes */
1113 static int smc_listen_prfx_check(struct smc_sock *new_smc,
1114 				 struct smc_clc_msg_proposal *pclc)
1115 {
1116 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1117 	struct socket *newclcsock = new_smc->clcsock;
1118 
1119 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1120 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1121 		return SMC_CLC_DECL_DIFFPREFIX;
1122 
1123 	return 0;
1124 }
1125 
1126 /* listen worker: initialize connection and buffers */
1127 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1128 				struct smc_init_info *ini)
1129 {
1130 	int rc;
1131 
1132 	/* allocate connection / link group */
1133 	rc = smc_conn_create(new_smc, ini);
1134 	if (rc)
1135 		return rc;
1136 
1137 	/* create send buffer and rmb */
1138 	if (smc_buf_create(new_smc, false))
1139 		return SMC_CLC_DECL_MEM;
1140 
1141 	return 0;
1142 }
1143 
1144 /* listen worker: initialize connection and buffers for SMC-D */
1145 static int smc_listen_ism_init(struct smc_sock *new_smc,
1146 			       struct smc_clc_msg_proposal *pclc,
1147 			       struct smc_init_info *ini)
1148 {
1149 	struct smc_clc_msg_smcd *pclc_smcd;
1150 	int rc;
1151 
1152 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1153 	ini->ism_gid = pclc_smcd->gid;
1154 	rc = smc_conn_create(new_smc, ini);
1155 	if (rc)
1156 		return rc;
1157 
1158 	/* Check if peer can be reached via ISM device */
1159 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1160 			    new_smc->conn.lgr->vlan_id,
1161 			    new_smc->conn.lgr->smcd)) {
1162 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1163 			smc_lgr_forget(new_smc->conn.lgr);
1164 		smc_conn_free(&new_smc->conn);
1165 		return SMC_CLC_DECL_SMCDNOTALK;
1166 	}
1167 
1168 	/* Create send and receive buffers */
1169 	if (smc_buf_create(new_smc, true)) {
1170 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1171 			smc_lgr_forget(new_smc->conn.lgr);
1172 		smc_conn_free(&new_smc->conn);
1173 		return SMC_CLC_DECL_MEM;
1174 	}
1175 
1176 	return 0;
1177 }
1178 
1179 /* listen worker: register buffers */
1180 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1181 {
1182 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1183 
1184 	if (local_contact != SMC_FIRST_CONTACT) {
1185 		if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true))
1186 			return SMC_CLC_DECL_ERR_REGRMB;
1187 	}
1188 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1189 
1190 	return 0;
1191 }
1192 
1193 /* listen worker: finish RDMA setup */
1194 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1195 				  struct smc_clc_msg_accept_confirm *cclc,
1196 				  int local_contact)
1197 {
1198 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1199 	int reason_code = 0;
1200 
1201 	if (local_contact == SMC_FIRST_CONTACT)
1202 		smc_link_save_peer_info(link, cclc);
1203 
1204 	if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) {
1205 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1206 		goto decline;
1207 	}
1208 
1209 	if (local_contact == SMC_FIRST_CONTACT) {
1210 		if (smc_ib_ready_link(link)) {
1211 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1212 			goto decline;
1213 		}
1214 		/* QP confirmation over RoCE fabric */
1215 		reason_code = smc_serv_conf_first_link(new_smc);
1216 		if (reason_code)
1217 			goto decline;
1218 	}
1219 	return 0;
1220 
1221 decline:
1222 	smc_listen_decline(new_smc, reason_code, local_contact);
1223 	return reason_code;
1224 }
1225 
1226 /* setup for RDMA connection of server */
1227 static void smc_listen_work(struct work_struct *work)
1228 {
1229 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1230 						smc_listen_work);
1231 	struct socket *newclcsock = new_smc->clcsock;
1232 	struct smc_clc_msg_accept_confirm cclc;
1233 	struct smc_clc_msg_proposal *pclc;
1234 	struct smc_init_info ini = {0};
1235 	bool ism_supported = false;
1236 	u8 buf[SMC_CLC_MAX_LEN];
1237 	int rc = 0;
1238 
1239 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1240 		return smc_listen_out_err(new_smc);
1241 
1242 	if (new_smc->use_fallback) {
1243 		smc_listen_out_connected(new_smc);
1244 		return;
1245 	}
1246 
1247 	/* check if peer is smc capable */
1248 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1249 		smc_switch_to_fallback(new_smc);
1250 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1251 		smc_listen_out_connected(new_smc);
1252 		return;
1253 	}
1254 
1255 	/* do inband token exchange -
1256 	 * wait for and receive SMC Proposal CLC message
1257 	 */
1258 	pclc = (struct smc_clc_msg_proposal *)&buf;
1259 	rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1260 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1261 	if (rc)
1262 		goto out_decl;
1263 
1264 	/* IPSec connections opt out of SMC-R optimizations */
1265 	if (using_ipsec(new_smc)) {
1266 		rc = SMC_CLC_DECL_IPSEC;
1267 		goto out_decl;
1268 	}
1269 
1270 	/* check for matching IP prefix and subnet length */
1271 	rc = smc_listen_prfx_check(new_smc, pclc);
1272 	if (rc)
1273 		goto out_decl;
1274 
1275 	/* get vlan id from IP device */
1276 	if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) {
1277 		rc = SMC_CLC_DECL_GETVLANERR;
1278 		goto out_decl;
1279 	}
1280 
1281 	mutex_lock(&smc_server_lgr_pending);
1282 	smc_close_init(new_smc);
1283 	smc_rx_init(new_smc);
1284 	smc_tx_init(new_smc);
1285 
1286 	/* check if ISM is available */
1287 	if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) {
1288 		ini.is_smcd = true; /* prepare ISM check */
1289 		rc = smc_find_ism_device(new_smc, &ini);
1290 		if (!rc)
1291 			rc = smc_listen_ism_init(new_smc, pclc, &ini);
1292 		if (!rc)
1293 			ism_supported = true;
1294 		else if (pclc->hdr.path == SMC_TYPE_D)
1295 			goto out_unlock; /* skip RDMA and decline */
1296 	}
1297 
1298 	/* check if RDMA is available */
1299 	if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */
1300 		/* prepare RDMA check */
1301 		ini.is_smcd = false;
1302 		ini.ism_dev = NULL;
1303 		ini.ib_lcl = &pclc->lcl;
1304 		rc = smc_find_rdma_device(new_smc, &ini);
1305 		if (rc) {
1306 			/* no RDMA device found */
1307 			if (pclc->hdr.path == SMC_TYPE_B)
1308 				/* neither ISM nor RDMA device found */
1309 				rc = SMC_CLC_DECL_NOSMCDEV;
1310 			goto out_unlock;
1311 		}
1312 		rc = smc_listen_rdma_init(new_smc, &ini);
1313 		if (rc)
1314 			goto out_unlock;
1315 		rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact);
1316 		if (rc)
1317 			goto out_unlock;
1318 	}
1319 
1320 	/* send SMC Accept CLC message */
1321 	rc = smc_clc_send_accept(new_smc, ini.cln_first_contact);
1322 	if (rc)
1323 		goto out_unlock;
1324 
1325 	/* SMC-D does not need this lock any more */
1326 	if (ism_supported)
1327 		mutex_unlock(&smc_server_lgr_pending);
1328 
1329 	/* receive SMC Confirm CLC message */
1330 	rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1331 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1332 	if (rc) {
1333 		if (!ism_supported)
1334 			goto out_unlock;
1335 		goto out_decl;
1336 	}
1337 
1338 	/* finish worker */
1339 	if (!ism_supported) {
1340 		rc = smc_listen_rdma_finish(new_smc, &cclc,
1341 					    ini.cln_first_contact);
1342 		mutex_unlock(&smc_server_lgr_pending);
1343 		if (rc)
1344 			return;
1345 	}
1346 	smc_conn_save_peer_info(new_smc, &cclc);
1347 	smc_listen_out_connected(new_smc);
1348 	return;
1349 
1350 out_unlock:
1351 	mutex_unlock(&smc_server_lgr_pending);
1352 out_decl:
1353 	smc_listen_decline(new_smc, rc, ini.cln_first_contact);
1354 }
1355 
1356 static void smc_tcp_listen_work(struct work_struct *work)
1357 {
1358 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1359 					     tcp_listen_work);
1360 	struct sock *lsk = &lsmc->sk;
1361 	struct smc_sock *new_smc;
1362 	int rc = 0;
1363 
1364 	lock_sock(lsk);
1365 	while (lsk->sk_state == SMC_LISTEN) {
1366 		rc = smc_clcsock_accept(lsmc, &new_smc);
1367 		if (rc)
1368 			goto out;
1369 		if (!new_smc)
1370 			continue;
1371 
1372 		new_smc->listen_smc = lsmc;
1373 		new_smc->use_fallback = lsmc->use_fallback;
1374 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1375 		sock_hold(lsk); /* sock_put in smc_listen_work */
1376 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1377 		smc_copy_sock_settings_to_smc(new_smc);
1378 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1379 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1380 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1381 		if (!schedule_work(&new_smc->smc_listen_work))
1382 			sock_put(&new_smc->sk);
1383 	}
1384 
1385 out:
1386 	release_sock(lsk);
1387 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1388 }
1389 
1390 static int smc_listen(struct socket *sock, int backlog)
1391 {
1392 	struct sock *sk = sock->sk;
1393 	struct smc_sock *smc;
1394 	int rc;
1395 
1396 	smc = smc_sk(sk);
1397 	lock_sock(sk);
1398 
1399 	rc = -EINVAL;
1400 	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
1401 	    smc->connect_nonblock)
1402 		goto out;
1403 
1404 	rc = 0;
1405 	if (sk->sk_state == SMC_LISTEN) {
1406 		sk->sk_max_ack_backlog = backlog;
1407 		goto out;
1408 	}
1409 	/* some socket options are handled in core, so we could not apply
1410 	 * them to the clc socket -- copy smc socket options to clc socket
1411 	 */
1412 	smc_copy_sock_settings_to_clc(smc);
1413 	if (!smc->use_fallback)
1414 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1415 
1416 	rc = kernel_listen(smc->clcsock, backlog);
1417 	if (rc)
1418 		goto out;
1419 	sk->sk_max_ack_backlog = backlog;
1420 	sk->sk_ack_backlog = 0;
1421 	sk->sk_state = SMC_LISTEN;
1422 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1423 	if (!schedule_work(&smc->tcp_listen_work))
1424 		sock_put(sk);
1425 
1426 out:
1427 	release_sock(sk);
1428 	return rc;
1429 }
1430 
1431 static int smc_accept(struct socket *sock, struct socket *new_sock,
1432 		      int flags, bool kern)
1433 {
1434 	struct sock *sk = sock->sk, *nsk;
1435 	DECLARE_WAITQUEUE(wait, current);
1436 	struct smc_sock *lsmc;
1437 	long timeo;
1438 	int rc = 0;
1439 
1440 	lsmc = smc_sk(sk);
1441 	sock_hold(sk); /* sock_put below */
1442 	lock_sock(sk);
1443 
1444 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1445 		rc = -EINVAL;
1446 		release_sock(sk);
1447 		goto out;
1448 	}
1449 
1450 	/* Wait for an incoming connection */
1451 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1452 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1453 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1454 		set_current_state(TASK_INTERRUPTIBLE);
1455 		if (!timeo) {
1456 			rc = -EAGAIN;
1457 			break;
1458 		}
1459 		release_sock(sk);
1460 		timeo = schedule_timeout(timeo);
1461 		/* wakeup by sk_data_ready in smc_listen_work() */
1462 		sched_annotate_sleep();
1463 		lock_sock(sk);
1464 		if (signal_pending(current)) {
1465 			rc = sock_intr_errno(timeo);
1466 			break;
1467 		}
1468 	}
1469 	set_current_state(TASK_RUNNING);
1470 	remove_wait_queue(sk_sleep(sk), &wait);
1471 
1472 	if (!rc)
1473 		rc = sock_error(nsk);
1474 	release_sock(sk);
1475 	if (rc)
1476 		goto out;
1477 
1478 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1479 		/* wait till data arrives on the socket */
1480 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1481 								MSEC_PER_SEC);
1482 		if (smc_sk(nsk)->use_fallback) {
1483 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1484 
1485 			lock_sock(clcsk);
1486 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1487 				sk_wait_data(clcsk, &timeo, NULL);
1488 			release_sock(clcsk);
1489 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1490 			lock_sock(nsk);
1491 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1492 			release_sock(nsk);
1493 		}
1494 	}
1495 
1496 out:
1497 	sock_put(sk); /* sock_hold above */
1498 	return rc;
1499 }
1500 
1501 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1502 		       int peer)
1503 {
1504 	struct smc_sock *smc;
1505 
1506 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1507 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1508 		return -ENOTCONN;
1509 
1510 	smc = smc_sk(sock->sk);
1511 
1512 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1513 }
1514 
1515 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1516 {
1517 	struct sock *sk = sock->sk;
1518 	struct smc_sock *smc;
1519 	int rc = -EPIPE;
1520 
1521 	smc = smc_sk(sk);
1522 	lock_sock(sk);
1523 	if ((sk->sk_state != SMC_ACTIVE) &&
1524 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1525 	    (sk->sk_state != SMC_INIT))
1526 		goto out;
1527 
1528 	if (msg->msg_flags & MSG_FASTOPEN) {
1529 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1530 			smc_switch_to_fallback(smc);
1531 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1532 		} else {
1533 			rc = -EINVAL;
1534 			goto out;
1535 		}
1536 	}
1537 
1538 	if (smc->use_fallback)
1539 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1540 	else
1541 		rc = smc_tx_sendmsg(smc, msg, len);
1542 out:
1543 	release_sock(sk);
1544 	return rc;
1545 }
1546 
1547 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1548 		       int flags)
1549 {
1550 	struct sock *sk = sock->sk;
1551 	struct smc_sock *smc;
1552 	int rc = -ENOTCONN;
1553 
1554 	smc = smc_sk(sk);
1555 	lock_sock(sk);
1556 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1557 		/* socket was connected before, no more data to read */
1558 		rc = 0;
1559 		goto out;
1560 	}
1561 	if ((sk->sk_state == SMC_INIT) ||
1562 	    (sk->sk_state == SMC_LISTEN) ||
1563 	    (sk->sk_state == SMC_CLOSED))
1564 		goto out;
1565 
1566 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1567 		rc = 0;
1568 		goto out;
1569 	}
1570 
1571 	if (smc->use_fallback) {
1572 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1573 	} else {
1574 		msg->msg_namelen = 0;
1575 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1576 	}
1577 
1578 out:
1579 	release_sock(sk);
1580 	return rc;
1581 }
1582 
1583 static __poll_t smc_accept_poll(struct sock *parent)
1584 {
1585 	struct smc_sock *isk = smc_sk(parent);
1586 	__poll_t mask = 0;
1587 
1588 	spin_lock(&isk->accept_q_lock);
1589 	if (!list_empty(&isk->accept_q))
1590 		mask = EPOLLIN | EPOLLRDNORM;
1591 	spin_unlock(&isk->accept_q_lock);
1592 
1593 	return mask;
1594 }
1595 
1596 static __poll_t smc_poll(struct file *file, struct socket *sock,
1597 			     poll_table *wait)
1598 {
1599 	struct sock *sk = sock->sk;
1600 	struct smc_sock *smc;
1601 	__poll_t mask = 0;
1602 
1603 	if (!sk)
1604 		return EPOLLNVAL;
1605 
1606 	smc = smc_sk(sock->sk);
1607 	if (smc->use_fallback) {
1608 		/* delegate to CLC child sock */
1609 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1610 		sk->sk_err = smc->clcsock->sk->sk_err;
1611 	} else {
1612 		if (sk->sk_state != SMC_CLOSED)
1613 			sock_poll_wait(file, sock, wait);
1614 		if (sk->sk_err)
1615 			mask |= EPOLLERR;
1616 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1617 		    (sk->sk_state == SMC_CLOSED))
1618 			mask |= EPOLLHUP;
1619 		if (sk->sk_state == SMC_LISTEN) {
1620 			/* woken up by sk_data_ready in smc_listen_work() */
1621 			mask |= smc_accept_poll(sk);
1622 		} else if (smc->use_fallback) { /* as result of connect_work()*/
1623 			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
1624 							   wait);
1625 			sk->sk_err = smc->clcsock->sk->sk_err;
1626 		} else {
1627 			if ((sk->sk_state != SMC_INIT &&
1628 			     atomic_read(&smc->conn.sndbuf_space)) ||
1629 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1630 				mask |= EPOLLOUT | EPOLLWRNORM;
1631 			} else {
1632 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1633 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1634 			}
1635 			if (atomic_read(&smc->conn.bytes_to_rcv))
1636 				mask |= EPOLLIN | EPOLLRDNORM;
1637 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1638 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1639 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1640 				mask |= EPOLLIN;
1641 			if (smc->conn.urg_state == SMC_URG_VALID)
1642 				mask |= EPOLLPRI;
1643 		}
1644 	}
1645 
1646 	return mask;
1647 }
1648 
1649 static int smc_shutdown(struct socket *sock, int how)
1650 {
1651 	struct sock *sk = sock->sk;
1652 	struct smc_sock *smc;
1653 	int rc = -EINVAL;
1654 	int rc1 = 0;
1655 
1656 	smc = smc_sk(sk);
1657 
1658 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1659 		return rc;
1660 
1661 	lock_sock(sk);
1662 
1663 	rc = -ENOTCONN;
1664 	if ((sk->sk_state != SMC_ACTIVE) &&
1665 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1666 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1667 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1668 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1669 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1670 		goto out;
1671 	if (smc->use_fallback) {
1672 		rc = kernel_sock_shutdown(smc->clcsock, how);
1673 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1674 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1675 			sk->sk_state = SMC_CLOSED;
1676 		goto out;
1677 	}
1678 	switch (how) {
1679 	case SHUT_RDWR:		/* shutdown in both directions */
1680 		rc = smc_close_active(smc);
1681 		break;
1682 	case SHUT_WR:
1683 		rc = smc_close_shutdown_write(smc);
1684 		break;
1685 	case SHUT_RD:
1686 		rc = 0;
1687 		/* nothing more to do because peer is not involved */
1688 		break;
1689 	}
1690 	if (smc->clcsock)
1691 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1692 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1693 	sk->sk_shutdown |= how + 1;
1694 
1695 out:
1696 	release_sock(sk);
1697 	return rc ? rc : rc1;
1698 }
1699 
1700 static int smc_setsockopt(struct socket *sock, int level, int optname,
1701 			  char __user *optval, unsigned int optlen)
1702 {
1703 	struct sock *sk = sock->sk;
1704 	struct smc_sock *smc;
1705 	int val, rc;
1706 
1707 	smc = smc_sk(sk);
1708 
1709 	/* generic setsockopts reaching us here always apply to the
1710 	 * CLC socket
1711 	 */
1712 	rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1713 					   optval, optlen);
1714 	if (smc->clcsock->sk->sk_err) {
1715 		sk->sk_err = smc->clcsock->sk->sk_err;
1716 		sk->sk_error_report(sk);
1717 	}
1718 	if (rc)
1719 		return rc;
1720 
1721 	if (optlen < sizeof(int))
1722 		return -EINVAL;
1723 	if (get_user(val, (int __user *)optval))
1724 		return -EFAULT;
1725 
1726 	lock_sock(sk);
1727 	switch (optname) {
1728 	case TCP_ULP:
1729 	case TCP_FASTOPEN:
1730 	case TCP_FASTOPEN_CONNECT:
1731 	case TCP_FASTOPEN_KEY:
1732 	case TCP_FASTOPEN_NO_COOKIE:
1733 		/* option not supported by SMC */
1734 		if (sk->sk_state == SMC_INIT) {
1735 			smc_switch_to_fallback(smc);
1736 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1737 		} else {
1738 			if (!smc->use_fallback)
1739 				rc = -EINVAL;
1740 		}
1741 		break;
1742 	case TCP_NODELAY:
1743 		if (sk->sk_state != SMC_INIT &&
1744 		    sk->sk_state != SMC_LISTEN &&
1745 		    sk->sk_state != SMC_CLOSED) {
1746 			if (val && !smc->use_fallback)
1747 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1748 						 0);
1749 		}
1750 		break;
1751 	case TCP_CORK:
1752 		if (sk->sk_state != SMC_INIT &&
1753 		    sk->sk_state != SMC_LISTEN &&
1754 		    sk->sk_state != SMC_CLOSED) {
1755 			if (!val && !smc->use_fallback)
1756 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1757 						 0);
1758 		}
1759 		break;
1760 	case TCP_DEFER_ACCEPT:
1761 		smc->sockopt_defer_accept = val;
1762 		break;
1763 	default:
1764 		break;
1765 	}
1766 	release_sock(sk);
1767 
1768 	return rc;
1769 }
1770 
1771 static int smc_getsockopt(struct socket *sock, int level, int optname,
1772 			  char __user *optval, int __user *optlen)
1773 {
1774 	struct smc_sock *smc;
1775 
1776 	smc = smc_sk(sock->sk);
1777 	/* socket options apply to the CLC socket */
1778 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1779 					     optval, optlen);
1780 }
1781 
1782 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1783 		     unsigned long arg)
1784 {
1785 	union smc_host_cursor cons, urg;
1786 	struct smc_connection *conn;
1787 	struct smc_sock *smc;
1788 	int answ;
1789 
1790 	smc = smc_sk(sock->sk);
1791 	conn = &smc->conn;
1792 	lock_sock(&smc->sk);
1793 	if (smc->use_fallback) {
1794 		if (!smc->clcsock) {
1795 			release_sock(&smc->sk);
1796 			return -EBADF;
1797 		}
1798 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1799 		release_sock(&smc->sk);
1800 		return answ;
1801 	}
1802 	switch (cmd) {
1803 	case SIOCINQ: /* same as FIONREAD */
1804 		if (smc->sk.sk_state == SMC_LISTEN) {
1805 			release_sock(&smc->sk);
1806 			return -EINVAL;
1807 		}
1808 		if (smc->sk.sk_state == SMC_INIT ||
1809 		    smc->sk.sk_state == SMC_CLOSED)
1810 			answ = 0;
1811 		else
1812 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1813 		break;
1814 	case SIOCOUTQ:
1815 		/* output queue size (not send + not acked) */
1816 		if (smc->sk.sk_state == SMC_LISTEN) {
1817 			release_sock(&smc->sk);
1818 			return -EINVAL;
1819 		}
1820 		if (smc->sk.sk_state == SMC_INIT ||
1821 		    smc->sk.sk_state == SMC_CLOSED)
1822 			answ = 0;
1823 		else
1824 			answ = smc->conn.sndbuf_desc->len -
1825 					atomic_read(&smc->conn.sndbuf_space);
1826 		break;
1827 	case SIOCOUTQNSD:
1828 		/* output queue size (not send only) */
1829 		if (smc->sk.sk_state == SMC_LISTEN) {
1830 			release_sock(&smc->sk);
1831 			return -EINVAL;
1832 		}
1833 		if (smc->sk.sk_state == SMC_INIT ||
1834 		    smc->sk.sk_state == SMC_CLOSED)
1835 			answ = 0;
1836 		else
1837 			answ = smc_tx_prepared_sends(&smc->conn);
1838 		break;
1839 	case SIOCATMARK:
1840 		if (smc->sk.sk_state == SMC_LISTEN) {
1841 			release_sock(&smc->sk);
1842 			return -EINVAL;
1843 		}
1844 		if (smc->sk.sk_state == SMC_INIT ||
1845 		    smc->sk.sk_state == SMC_CLOSED) {
1846 			answ = 0;
1847 		} else {
1848 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1849 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1850 			answ = smc_curs_diff(conn->rmb_desc->len,
1851 					     &cons, &urg) == 1;
1852 		}
1853 		break;
1854 	default:
1855 		release_sock(&smc->sk);
1856 		return -ENOIOCTLCMD;
1857 	}
1858 	release_sock(&smc->sk);
1859 
1860 	return put_user(answ, (int __user *)arg);
1861 }
1862 
1863 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1864 			    int offset, size_t size, int flags)
1865 {
1866 	struct sock *sk = sock->sk;
1867 	struct smc_sock *smc;
1868 	int rc = -EPIPE;
1869 
1870 	smc = smc_sk(sk);
1871 	lock_sock(sk);
1872 	if (sk->sk_state != SMC_ACTIVE) {
1873 		release_sock(sk);
1874 		goto out;
1875 	}
1876 	release_sock(sk);
1877 	if (smc->use_fallback)
1878 		rc = kernel_sendpage(smc->clcsock, page, offset,
1879 				     size, flags);
1880 	else
1881 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1882 
1883 out:
1884 	return rc;
1885 }
1886 
1887 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1888  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1889  * updates till whenever a respective page has been fully processed.
1890  * Note that subsequent recv() calls have to wait till all splice() processing
1891  * completed.
1892  */
1893 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1894 			       struct pipe_inode_info *pipe, size_t len,
1895 			       unsigned int flags)
1896 {
1897 	struct sock *sk = sock->sk;
1898 	struct smc_sock *smc;
1899 	int rc = -ENOTCONN;
1900 
1901 	smc = smc_sk(sk);
1902 	lock_sock(sk);
1903 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1904 		/* socket was connected before, no more data to read */
1905 		rc = 0;
1906 		goto out;
1907 	}
1908 	if (sk->sk_state == SMC_INIT ||
1909 	    sk->sk_state == SMC_LISTEN ||
1910 	    sk->sk_state == SMC_CLOSED)
1911 		goto out;
1912 
1913 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1914 		rc = 0;
1915 		goto out;
1916 	}
1917 
1918 	if (smc->use_fallback) {
1919 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1920 						    pipe, len, flags);
1921 	} else {
1922 		if (*ppos) {
1923 			rc = -ESPIPE;
1924 			goto out;
1925 		}
1926 		if (flags & SPLICE_F_NONBLOCK)
1927 			flags = MSG_DONTWAIT;
1928 		else
1929 			flags = 0;
1930 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1931 	}
1932 out:
1933 	release_sock(sk);
1934 
1935 	return rc;
1936 }
1937 
1938 /* must look like tcp */
1939 static const struct proto_ops smc_sock_ops = {
1940 	.family		= PF_SMC,
1941 	.owner		= THIS_MODULE,
1942 	.release	= smc_release,
1943 	.bind		= smc_bind,
1944 	.connect	= smc_connect,
1945 	.socketpair	= sock_no_socketpair,
1946 	.accept		= smc_accept,
1947 	.getname	= smc_getname,
1948 	.poll		= smc_poll,
1949 	.ioctl		= smc_ioctl,
1950 	.listen		= smc_listen,
1951 	.shutdown	= smc_shutdown,
1952 	.setsockopt	= smc_setsockopt,
1953 	.getsockopt	= smc_getsockopt,
1954 	.sendmsg	= smc_sendmsg,
1955 	.recvmsg	= smc_recvmsg,
1956 	.mmap		= sock_no_mmap,
1957 	.sendpage	= smc_sendpage,
1958 	.splice_read	= smc_splice_read,
1959 };
1960 
1961 static int smc_create(struct net *net, struct socket *sock, int protocol,
1962 		      int kern)
1963 {
1964 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1965 	struct smc_sock *smc;
1966 	struct sock *sk;
1967 	int rc;
1968 
1969 	rc = -ESOCKTNOSUPPORT;
1970 	if (sock->type != SOCK_STREAM)
1971 		goto out;
1972 
1973 	rc = -EPROTONOSUPPORT;
1974 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1975 		goto out;
1976 
1977 	rc = -ENOBUFS;
1978 	sock->ops = &smc_sock_ops;
1979 	sk = smc_sock_alloc(net, sock, protocol);
1980 	if (!sk)
1981 		goto out;
1982 
1983 	/* create internal TCP socket for CLC handshake and fallback */
1984 	smc = smc_sk(sk);
1985 	smc->use_fallback = false; /* assume rdma capability first */
1986 	smc->fallback_rsn = 0;
1987 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1988 			      &smc->clcsock);
1989 	if (rc) {
1990 		sk_common_release(sk);
1991 		goto out;
1992 	}
1993 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1994 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1995 
1996 out:
1997 	return rc;
1998 }
1999 
2000 static const struct net_proto_family smc_sock_family_ops = {
2001 	.family	= PF_SMC,
2002 	.owner	= THIS_MODULE,
2003 	.create	= smc_create,
2004 };
2005 
2006 unsigned int smc_net_id;
2007 
2008 static __net_init int smc_net_init(struct net *net)
2009 {
2010 	return smc_pnet_net_init(net);
2011 }
2012 
2013 static void __net_exit smc_net_exit(struct net *net)
2014 {
2015 	smc_pnet_net_exit(net);
2016 }
2017 
2018 static struct pernet_operations smc_net_ops = {
2019 	.init = smc_net_init,
2020 	.exit = smc_net_exit,
2021 	.id   = &smc_net_id,
2022 	.size = sizeof(struct smc_net),
2023 };
2024 
2025 static int __init smc_init(void)
2026 {
2027 	int rc;
2028 
2029 	rc = register_pernet_subsys(&smc_net_ops);
2030 	if (rc)
2031 		return rc;
2032 
2033 	rc = smc_pnet_init();
2034 	if (rc)
2035 		goto out_pernet_subsys;
2036 
2037 	rc = smc_llc_init();
2038 	if (rc) {
2039 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2040 		goto out_pnet;
2041 	}
2042 
2043 	rc = smc_cdc_init();
2044 	if (rc) {
2045 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2046 		goto out_pnet;
2047 	}
2048 
2049 	rc = proto_register(&smc_proto, 1);
2050 	if (rc) {
2051 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2052 		goto out_pnet;
2053 	}
2054 
2055 	rc = proto_register(&smc_proto6, 1);
2056 	if (rc) {
2057 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2058 		goto out_proto;
2059 	}
2060 
2061 	rc = sock_register(&smc_sock_family_ops);
2062 	if (rc) {
2063 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2064 		goto out_proto6;
2065 	}
2066 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2067 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2068 
2069 	rc = smc_ib_register_client();
2070 	if (rc) {
2071 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2072 		goto out_sock;
2073 	}
2074 
2075 	static_branch_enable(&tcp_have_smc);
2076 	return 0;
2077 
2078 out_sock:
2079 	sock_unregister(PF_SMC);
2080 out_proto6:
2081 	proto_unregister(&smc_proto6);
2082 out_proto:
2083 	proto_unregister(&smc_proto);
2084 out_pnet:
2085 	smc_pnet_exit();
2086 out_pernet_subsys:
2087 	unregister_pernet_subsys(&smc_net_ops);
2088 
2089 	return rc;
2090 }
2091 
2092 static void __exit smc_exit(void)
2093 {
2094 	smc_core_exit();
2095 	static_branch_disable(&tcp_have_smc);
2096 	smc_ib_unregister_client();
2097 	sock_unregister(PF_SMC);
2098 	proto_unregister(&smc_proto6);
2099 	proto_unregister(&smc_proto);
2100 	smc_pnet_exit();
2101 	unregister_pernet_subsys(&smc_net_ops);
2102 }
2103 
2104 module_init(smc_init);
2105 module_exit(smc_exit);
2106 
2107 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2108 MODULE_DESCRIPTION("smc socket address family");
2109 MODULE_LICENSE("GPL");
2110 MODULE_ALIAS_NETPROTO(PF_SMC);
2111