xref: /openbmc/linux/net/smc/af_smc.c (revision 3557b3fd)
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  *
12  *  Copyright IBM Corp. 2016, 2018
13  *
14  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
15  *              based on prototype from Frank Blaschka
16  */
17 
18 #define KMSG_COMPONENT "smc"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/module.h>
22 #include <linux/socket.h>
23 #include <linux/workqueue.h>
24 #include <linux/in.h>
25 #include <linux/sched/signal.h>
26 #include <linux/if_vlan.h>
27 
28 #include <net/sock.h>
29 #include <net/tcp.h>
30 #include <net/smc.h>
31 #include <asm/ioctls.h>
32 
33 #include <net/net_namespace.h>
34 #include <net/netns/generic.h>
35 #include "smc_netns.h"
36 
37 #include "smc.h"
38 #include "smc_clc.h"
39 #include "smc_llc.h"
40 #include "smc_cdc.h"
41 #include "smc_core.h"
42 #include "smc_ib.h"
43 #include "smc_ism.h"
44 #include "smc_pnet.h"
45 #include "smc_tx.h"
46 #include "smc_rx.h"
47 #include "smc_close.h"
48 
49 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
50 						 * creation on server
51 						 */
52 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
53 						 * creation on client
54 						 */
55 
56 static void smc_tcp_listen_work(struct work_struct *);
57 static void smc_connect_work(struct work_struct *);
58 
59 static void smc_set_keepalive(struct sock *sk, int val)
60 {
61 	struct smc_sock *smc = smc_sk(sk);
62 
63 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
64 }
65 
66 static struct smc_hashinfo smc_v4_hashinfo = {
67 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
68 };
69 
70 static struct smc_hashinfo smc_v6_hashinfo = {
71 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
72 };
73 
74 int smc_hash_sk(struct sock *sk)
75 {
76 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
77 	struct hlist_head *head;
78 
79 	head = &h->ht;
80 
81 	write_lock_bh(&h->lock);
82 	sk_add_node(sk, head);
83 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
84 	write_unlock_bh(&h->lock);
85 
86 	return 0;
87 }
88 EXPORT_SYMBOL_GPL(smc_hash_sk);
89 
90 void smc_unhash_sk(struct sock *sk)
91 {
92 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
93 
94 	write_lock_bh(&h->lock);
95 	if (sk_del_node_init(sk))
96 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
97 	write_unlock_bh(&h->lock);
98 }
99 EXPORT_SYMBOL_GPL(smc_unhash_sk);
100 
101 struct proto smc_proto = {
102 	.name		= "SMC",
103 	.owner		= THIS_MODULE,
104 	.keepalive	= smc_set_keepalive,
105 	.hash		= smc_hash_sk,
106 	.unhash		= smc_unhash_sk,
107 	.obj_size	= sizeof(struct smc_sock),
108 	.h.smc_hash	= &smc_v4_hashinfo,
109 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
110 };
111 EXPORT_SYMBOL_GPL(smc_proto);
112 
113 struct proto smc_proto6 = {
114 	.name		= "SMC6",
115 	.owner		= THIS_MODULE,
116 	.keepalive	= smc_set_keepalive,
117 	.hash		= smc_hash_sk,
118 	.unhash		= smc_unhash_sk,
119 	.obj_size	= sizeof(struct smc_sock),
120 	.h.smc_hash	= &smc_v6_hashinfo,
121 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
122 };
123 EXPORT_SYMBOL_GPL(smc_proto6);
124 
125 static int smc_release(struct socket *sock)
126 {
127 	struct sock *sk = sock->sk;
128 	struct smc_sock *smc;
129 	int rc = 0;
130 
131 	if (!sk)
132 		goto out;
133 
134 	smc = smc_sk(sk);
135 
136 	/* cleanup for a dangling non-blocking connect */
137 	if (smc->connect_nonblock && sk->sk_state == SMC_INIT)
138 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
139 	flush_work(&smc->connect_work);
140 
141 	if (sk->sk_state == SMC_LISTEN)
142 		/* smc_close_non_accepted() is called and acquires
143 		 * sock lock for child sockets again
144 		 */
145 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
146 	else
147 		lock_sock(sk);
148 
149 	if (!smc->use_fallback) {
150 		rc = smc_close_active(smc);
151 		sock_set_flag(sk, SOCK_DEAD);
152 		sk->sk_shutdown |= SHUTDOWN_MASK;
153 	} else {
154 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
155 			sock_put(sk); /* passive closing */
156 		if (sk->sk_state == SMC_LISTEN) {
157 			/* wake up clcsock accept */
158 			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
159 		}
160 		sk->sk_state = SMC_CLOSED;
161 		sk->sk_state_change(sk);
162 	}
163 
164 	sk->sk_prot->unhash(sk);
165 
166 	if (sk->sk_state == SMC_CLOSED) {
167 		if (smc->clcsock) {
168 			release_sock(sk);
169 			smc_clcsock_release(smc);
170 			lock_sock(sk);
171 		}
172 		if (!smc->use_fallback)
173 			smc_conn_free(&smc->conn);
174 	}
175 
176 	/* detach socket */
177 	sock_orphan(sk);
178 	sock->sk = NULL;
179 	release_sock(sk);
180 
181 	sock_put(sk); /* final sock_put */
182 out:
183 	return rc;
184 }
185 
186 static void smc_destruct(struct sock *sk)
187 {
188 	if (sk->sk_state != SMC_CLOSED)
189 		return;
190 	if (!sock_flag(sk, SOCK_DEAD))
191 		return;
192 
193 	sk_refcnt_debug_dec(sk);
194 }
195 
196 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
197 				   int protocol)
198 {
199 	struct smc_sock *smc;
200 	struct proto *prot;
201 	struct sock *sk;
202 
203 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
204 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
205 	if (!sk)
206 		return NULL;
207 
208 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
209 	sk->sk_state = SMC_INIT;
210 	sk->sk_destruct = smc_destruct;
211 	sk->sk_protocol = protocol;
212 	smc = smc_sk(sk);
213 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
214 	INIT_WORK(&smc->connect_work, smc_connect_work);
215 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
216 	INIT_LIST_HEAD(&smc->accept_q);
217 	spin_lock_init(&smc->accept_q_lock);
218 	spin_lock_init(&smc->conn.send_lock);
219 	sk->sk_prot->hash(sk);
220 	sk_refcnt_debug_inc(sk);
221 	mutex_init(&smc->clcsock_release_lock);
222 
223 	return sk;
224 }
225 
226 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
227 		    int addr_len)
228 {
229 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
230 	struct sock *sk = sock->sk;
231 	struct smc_sock *smc;
232 	int rc;
233 
234 	smc = smc_sk(sk);
235 
236 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
237 	rc = -EINVAL;
238 	if (addr_len < sizeof(struct sockaddr_in))
239 		goto out;
240 
241 	rc = -EAFNOSUPPORT;
242 	if (addr->sin_family != AF_INET &&
243 	    addr->sin_family != AF_INET6 &&
244 	    addr->sin_family != AF_UNSPEC)
245 		goto out;
246 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
247 	if (addr->sin_family == AF_UNSPEC &&
248 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
249 		goto out;
250 
251 	lock_sock(sk);
252 
253 	/* Check if socket is already active */
254 	rc = -EINVAL;
255 	if (sk->sk_state != SMC_INIT)
256 		goto out_rel;
257 
258 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
259 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
260 
261 out_rel:
262 	release_sock(sk);
263 out:
264 	return rc;
265 }
266 
267 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
268 				   unsigned long mask)
269 {
270 	/* options we don't get control via setsockopt for */
271 	nsk->sk_type = osk->sk_type;
272 	nsk->sk_sndbuf = osk->sk_sndbuf;
273 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
274 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
275 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
276 	nsk->sk_mark = osk->sk_mark;
277 	nsk->sk_priority = osk->sk_priority;
278 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
279 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
280 	nsk->sk_err = osk->sk_err;
281 
282 	nsk->sk_flags &= ~mask;
283 	nsk->sk_flags |= osk->sk_flags & mask;
284 }
285 
286 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
287 			     (1UL << SOCK_KEEPOPEN) | \
288 			     (1UL << SOCK_LINGER) | \
289 			     (1UL << SOCK_BROADCAST) | \
290 			     (1UL << SOCK_TIMESTAMP) | \
291 			     (1UL << SOCK_DBG) | \
292 			     (1UL << SOCK_RCVTSTAMP) | \
293 			     (1UL << SOCK_RCVTSTAMPNS) | \
294 			     (1UL << SOCK_LOCALROUTE) | \
295 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
296 			     (1UL << SOCK_RXQ_OVFL) | \
297 			     (1UL << SOCK_WIFI_STATUS) | \
298 			     (1UL << SOCK_NOFCS) | \
299 			     (1UL << SOCK_FILTER_LOCKED) | \
300 			     (1UL << SOCK_TSTAMP_NEW))
301 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
302  * clc socket (since smc is not called for these options from net/core)
303  */
304 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
305 {
306 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
307 }
308 
309 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
310 			     (1UL << SOCK_KEEPOPEN) | \
311 			     (1UL << SOCK_LINGER) | \
312 			     (1UL << SOCK_DBG))
313 /* copy only settings and flags relevant for smc from clc to smc socket */
314 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
315 {
316 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
317 }
318 
319 /* register a new rmb, send confirm_rkey msg to register with peer */
320 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc,
321 		       bool conf_rkey)
322 {
323 	if (!rmb_desc->wr_reg) {
324 		/* register memory region for new rmb */
325 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
326 			rmb_desc->regerr = 1;
327 			return -EFAULT;
328 		}
329 		rmb_desc->wr_reg = 1;
330 	}
331 	if (!conf_rkey)
332 		return 0;
333 	/* exchange confirm_rkey msg with peer */
334 	if (smc_llc_do_confirm_rkey(link, rmb_desc)) {
335 		rmb_desc->regerr = 1;
336 		return -EFAULT;
337 	}
338 	return 0;
339 }
340 
341 static int smc_clnt_conf_first_link(struct smc_sock *smc)
342 {
343 	struct net *net = sock_net(smc->clcsock->sk);
344 	struct smc_link_group *lgr = smc->conn.lgr;
345 	struct smc_link *link;
346 	int rest;
347 	int rc;
348 
349 	link = &lgr->lnk[SMC_SINGLE_LINK];
350 	/* receive CONFIRM LINK request from server over RoCE fabric */
351 	rest = wait_for_completion_interruptible_timeout(
352 		&link->llc_confirm,
353 		SMC_LLC_WAIT_FIRST_TIME);
354 	if (rest <= 0) {
355 		struct smc_clc_msg_decline dclc;
356 
357 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
358 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
359 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
360 	}
361 
362 	if (link->llc_confirm_rc)
363 		return SMC_CLC_DECL_RMBE_EC;
364 
365 	rc = smc_ib_modify_qp_rts(link);
366 	if (rc)
367 		return SMC_CLC_DECL_ERR_RDYLNK;
368 
369 	smc_wr_remember_qp_attr(link);
370 
371 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
372 		return SMC_CLC_DECL_ERR_REGRMB;
373 
374 	/* send CONFIRM LINK response over RoCE fabric */
375 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
376 	if (rc < 0)
377 		return SMC_CLC_DECL_TIMEOUT_CL;
378 
379 	/* receive ADD LINK request from server over RoCE fabric */
380 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
381 							 SMC_LLC_WAIT_TIME);
382 	if (rest <= 0) {
383 		struct smc_clc_msg_decline dclc;
384 
385 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
386 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
387 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
388 	}
389 
390 	/* send add link reject message, only one link supported for now */
391 	rc = smc_llc_send_add_link(link,
392 				   link->smcibdev->mac[link->ibport - 1],
393 				   link->gid, SMC_LLC_RESP);
394 	if (rc < 0)
395 		return SMC_CLC_DECL_TIMEOUT_AL;
396 
397 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
398 
399 	return 0;
400 }
401 
402 static void smcr_conn_save_peer_info(struct smc_sock *smc,
403 				     struct smc_clc_msg_accept_confirm *clc)
404 {
405 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
406 
407 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
408 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
409 	smc->conn.peer_rmbe_size = bufsize;
410 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
411 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
412 }
413 
414 static void smcd_conn_save_peer_info(struct smc_sock *smc,
415 				     struct smc_clc_msg_accept_confirm *clc)
416 {
417 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
418 
419 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
420 	smc->conn.peer_token = clc->token;
421 	/* msg header takes up space in the buffer */
422 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
423 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
424 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
425 }
426 
427 static void smc_conn_save_peer_info(struct smc_sock *smc,
428 				    struct smc_clc_msg_accept_confirm *clc)
429 {
430 	if (smc->conn.lgr->is_smcd)
431 		smcd_conn_save_peer_info(smc, clc);
432 	else
433 		smcr_conn_save_peer_info(smc, clc);
434 }
435 
436 static void smc_link_save_peer_info(struct smc_link *link,
437 				    struct smc_clc_msg_accept_confirm *clc)
438 {
439 	link->peer_qpn = ntoh24(clc->qpn);
440 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
441 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
442 	link->peer_psn = ntoh24(clc->psn);
443 	link->peer_mtu = clc->qp_mtu;
444 }
445 
446 static void smc_switch_to_fallback(struct smc_sock *smc)
447 {
448 	smc->use_fallback = true;
449 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
450 		smc->clcsock->file = smc->sk.sk_socket->file;
451 		smc->clcsock->file->private_data = smc->clcsock;
452 	}
453 }
454 
455 /* fall back during connect */
456 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
457 {
458 	smc_switch_to_fallback(smc);
459 	smc->fallback_rsn = reason_code;
460 	smc_copy_sock_settings_to_clc(smc);
461 	smc->connect_nonblock = 0;
462 	if (smc->sk.sk_state == SMC_INIT)
463 		smc->sk.sk_state = SMC_ACTIVE;
464 	return 0;
465 }
466 
467 /* decline and fall back during connect */
468 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
469 {
470 	int rc;
471 
472 	if (reason_code < 0) { /* error, fallback is not possible */
473 		if (smc->sk.sk_state == SMC_INIT)
474 			sock_put(&smc->sk); /* passive closing */
475 		return reason_code;
476 	}
477 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
478 		rc = smc_clc_send_decline(smc, reason_code);
479 		if (rc < 0) {
480 			if (smc->sk.sk_state == SMC_INIT)
481 				sock_put(&smc->sk); /* passive closing */
482 			return rc;
483 		}
484 	}
485 	return smc_connect_fallback(smc, reason_code);
486 }
487 
488 /* abort connecting */
489 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
490 			     int local_contact)
491 {
492 	if (local_contact == SMC_FIRST_CONTACT)
493 		smc_lgr_forget(smc->conn.lgr);
494 	if (smc->conn.lgr->is_smcd)
495 		/* there is only one lgr role for SMC-D; use server lock */
496 		mutex_unlock(&smc_server_lgr_pending);
497 	else
498 		mutex_unlock(&smc_client_lgr_pending);
499 
500 	smc_conn_free(&smc->conn);
501 	smc->connect_nonblock = 0;
502 	return reason_code;
503 }
504 
505 /* check if there is a rdma device available for this connection. */
506 /* called for connect and listen */
507 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
508 {
509 	/* PNET table look up: search active ib_device and port
510 	 * within same PNETID that also contains the ethernet device
511 	 * used for the internal TCP socket
512 	 */
513 	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
514 	if (!ini->ib_dev)
515 		return SMC_CLC_DECL_NOSMCRDEV;
516 	return 0;
517 }
518 
519 /* check if there is an ISM device available for this connection. */
520 /* called for connect and listen */
521 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
522 {
523 	/* Find ISM device with same PNETID as connecting interface  */
524 	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
525 	if (!ini->ism_dev)
526 		return SMC_CLC_DECL_NOSMCDDEV;
527 	return 0;
528 }
529 
530 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
531 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
532 				      struct smc_init_info *ini)
533 {
534 	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id))
535 		return SMC_CLC_DECL_ISMVLANERR;
536 	return 0;
537 }
538 
539 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
540  * used, the VLAN ID will be registered again during the connection setup.
541  */
542 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
543 					struct smc_init_info *ini)
544 {
545 	if (!is_smcd)
546 		return 0;
547 	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id))
548 		return SMC_CLC_DECL_CNFERR;
549 	return 0;
550 }
551 
552 /* CLC handshake during connect */
553 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
554 			   struct smc_clc_msg_accept_confirm *aclc,
555 			   struct smc_init_info *ini)
556 {
557 	int rc = 0;
558 
559 	/* do inband token exchange */
560 	rc = smc_clc_send_proposal(smc, smc_type, ini);
561 	if (rc)
562 		return rc;
563 	/* receive SMC Accept CLC message */
564 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT,
565 				CLC_WAIT_TIME);
566 }
567 
568 /* setup for RDMA connection of client */
569 static int smc_connect_rdma(struct smc_sock *smc,
570 			    struct smc_clc_msg_accept_confirm *aclc,
571 			    struct smc_init_info *ini)
572 {
573 	struct smc_link *link;
574 	int reason_code = 0;
575 
576 	ini->is_smcd = false;
577 	ini->ib_lcl = &aclc->lcl;
578 	ini->ib_clcqpn = ntoh24(aclc->qpn);
579 	ini->srv_first_contact = aclc->hdr.flag;
580 
581 	mutex_lock(&smc_client_lgr_pending);
582 	reason_code = smc_conn_create(smc, ini);
583 	if (reason_code) {
584 		mutex_unlock(&smc_client_lgr_pending);
585 		return reason_code;
586 	}
587 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
588 
589 	smc_conn_save_peer_info(smc, aclc);
590 
591 	/* create send buffer and rmb */
592 	if (smc_buf_create(smc, false))
593 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
594 					 ini->cln_first_contact);
595 
596 	if (ini->cln_first_contact == SMC_FIRST_CONTACT)
597 		smc_link_save_peer_info(link, aclc);
598 
599 	if (smc_rmb_rtoken_handling(&smc->conn, aclc))
600 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
601 					 ini->cln_first_contact);
602 
603 	smc_close_init(smc);
604 	smc_rx_init(smc);
605 
606 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
607 		if (smc_ib_ready_link(link))
608 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
609 						 ini->cln_first_contact);
610 	} else {
611 		if (smc_reg_rmb(link, smc->conn.rmb_desc, true))
612 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
613 						 ini->cln_first_contact);
614 	}
615 	smc_rmb_sync_sg_for_device(&smc->conn);
616 
617 	reason_code = smc_clc_send_confirm(smc);
618 	if (reason_code)
619 		return smc_connect_abort(smc, reason_code,
620 					 ini->cln_first_contact);
621 
622 	smc_tx_init(smc);
623 
624 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
625 		/* QP confirmation over RoCE fabric */
626 		reason_code = smc_clnt_conf_first_link(smc);
627 		if (reason_code)
628 			return smc_connect_abort(smc, reason_code,
629 						 ini->cln_first_contact);
630 	}
631 	mutex_unlock(&smc_client_lgr_pending);
632 
633 	smc_copy_sock_settings_to_clc(smc);
634 	smc->connect_nonblock = 0;
635 	if (smc->sk.sk_state == SMC_INIT)
636 		smc->sk.sk_state = SMC_ACTIVE;
637 
638 	return 0;
639 }
640 
641 /* setup for ISM connection of client */
642 static int smc_connect_ism(struct smc_sock *smc,
643 			   struct smc_clc_msg_accept_confirm *aclc,
644 			   struct smc_init_info *ini)
645 {
646 	int rc = 0;
647 
648 	ini->is_smcd = true;
649 	ini->ism_gid = aclc->gid;
650 	ini->srv_first_contact = aclc->hdr.flag;
651 
652 	/* there is only one lgr role for SMC-D; use server lock */
653 	mutex_lock(&smc_server_lgr_pending);
654 	rc = smc_conn_create(smc, ini);
655 	if (rc) {
656 		mutex_unlock(&smc_server_lgr_pending);
657 		return rc;
658 	}
659 
660 	/* Create send and receive buffers */
661 	if (smc_buf_create(smc, true))
662 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
663 					 ini->cln_first_contact);
664 
665 	smc_conn_save_peer_info(smc, aclc);
666 	smc_close_init(smc);
667 	smc_rx_init(smc);
668 	smc_tx_init(smc);
669 
670 	rc = smc_clc_send_confirm(smc);
671 	if (rc)
672 		return smc_connect_abort(smc, rc, ini->cln_first_contact);
673 	mutex_unlock(&smc_server_lgr_pending);
674 
675 	smc_copy_sock_settings_to_clc(smc);
676 	smc->connect_nonblock = 0;
677 	if (smc->sk.sk_state == SMC_INIT)
678 		smc->sk.sk_state = SMC_ACTIVE;
679 
680 	return 0;
681 }
682 
683 /* perform steps before actually connecting */
684 static int __smc_connect(struct smc_sock *smc)
685 {
686 	bool ism_supported = false, rdma_supported = false;
687 	struct smc_clc_msg_accept_confirm aclc;
688 	struct smc_init_info ini = {0};
689 	int smc_type;
690 	int rc = 0;
691 
692 	sock_hold(&smc->sk); /* sock put in passive closing */
693 
694 	if (smc->use_fallback)
695 		return smc_connect_fallback(smc, smc->fallback_rsn);
696 
697 	/* if peer has not signalled SMC-capability, fall back */
698 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
699 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
700 
701 	/* IPSec connections opt out of SMC-R optimizations */
702 	if (using_ipsec(smc))
703 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
704 
705 	/* get vlan id from IP device */
706 	if (smc_vlan_by_tcpsk(smc->clcsock, &ini))
707 		return smc_connect_decline_fallback(smc,
708 						    SMC_CLC_DECL_GETVLANERR);
709 
710 	/* check if there is an ism device available */
711 	if (!smc_find_ism_device(smc, &ini) &&
712 	    !smc_connect_ism_vlan_setup(smc, &ini)) {
713 		/* ISM is supported for this connection */
714 		ism_supported = true;
715 		smc_type = SMC_TYPE_D;
716 	}
717 
718 	/* check if there is a rdma device available */
719 	if (!smc_find_rdma_device(smc, &ini)) {
720 		/* RDMA is supported for this connection */
721 		rdma_supported = true;
722 		if (ism_supported)
723 			smc_type = SMC_TYPE_B; /* both */
724 		else
725 			smc_type = SMC_TYPE_R; /* only RDMA */
726 	}
727 
728 	/* if neither ISM nor RDMA are supported, fallback */
729 	if (!rdma_supported && !ism_supported)
730 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
731 
732 	/* perform CLC handshake */
733 	rc = smc_connect_clc(smc, smc_type, &aclc, &ini);
734 	if (rc) {
735 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
736 		return smc_connect_decline_fallback(smc, rc);
737 	}
738 
739 	/* depending on previous steps, connect using rdma or ism */
740 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
741 		rc = smc_connect_rdma(smc, &aclc, &ini);
742 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
743 		rc = smc_connect_ism(smc, &aclc, &ini);
744 	else
745 		rc = SMC_CLC_DECL_MODEUNSUPP;
746 	if (rc) {
747 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
748 		return smc_connect_decline_fallback(smc, rc);
749 	}
750 
751 	smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
752 	return 0;
753 }
754 
755 static void smc_connect_work(struct work_struct *work)
756 {
757 	struct smc_sock *smc = container_of(work, struct smc_sock,
758 					    connect_work);
759 	long timeo = smc->sk.sk_sndtimeo;
760 	int rc = 0;
761 
762 	if (!timeo)
763 		timeo = MAX_SCHEDULE_TIMEOUT;
764 	lock_sock(smc->clcsock->sk);
765 	if (smc->clcsock->sk->sk_err) {
766 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
767 	} else if ((1 << smc->clcsock->sk->sk_state) &
768 					(TCPF_SYN_SENT | TCP_SYN_RECV)) {
769 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
770 		if ((rc == -EPIPE) &&
771 		    ((1 << smc->clcsock->sk->sk_state) &
772 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
773 			rc = 0;
774 	}
775 	release_sock(smc->clcsock->sk);
776 	lock_sock(&smc->sk);
777 	if (rc != 0 || smc->sk.sk_err) {
778 		smc->sk.sk_state = SMC_CLOSED;
779 		if (rc == -EPIPE || rc == -EAGAIN)
780 			smc->sk.sk_err = EPIPE;
781 		else if (signal_pending(current))
782 			smc->sk.sk_err = -sock_intr_errno(timeo);
783 		goto out;
784 	}
785 
786 	rc = __smc_connect(smc);
787 	if (rc < 0)
788 		smc->sk.sk_err = -rc;
789 
790 out:
791 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
792 		if (smc->sk.sk_err) {
793 			smc->sk.sk_state_change(&smc->sk);
794 		} else { /* allow polling before and after fallback decision */
795 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
796 			smc->sk.sk_write_space(&smc->sk);
797 		}
798 	}
799 	release_sock(&smc->sk);
800 }
801 
802 static int smc_connect(struct socket *sock, struct sockaddr *addr,
803 		       int alen, int flags)
804 {
805 	struct sock *sk = sock->sk;
806 	struct smc_sock *smc;
807 	int rc = -EINVAL;
808 
809 	smc = smc_sk(sk);
810 
811 	/* separate smc parameter checking to be safe */
812 	if (alen < sizeof(addr->sa_family))
813 		goto out_err;
814 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
815 		goto out_err;
816 
817 	lock_sock(sk);
818 	switch (sk->sk_state) {
819 	default:
820 		goto out;
821 	case SMC_ACTIVE:
822 		rc = -EISCONN;
823 		goto out;
824 	case SMC_INIT:
825 		rc = 0;
826 		break;
827 	}
828 
829 	smc_copy_sock_settings_to_clc(smc);
830 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
831 	if (smc->connect_nonblock) {
832 		rc = -EALREADY;
833 		goto out;
834 	}
835 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
836 	if (rc && rc != -EINPROGRESS)
837 		goto out;
838 	if (flags & O_NONBLOCK) {
839 		if (schedule_work(&smc->connect_work))
840 			smc->connect_nonblock = 1;
841 		rc = -EINPROGRESS;
842 	} else {
843 		rc = __smc_connect(smc);
844 		if (rc < 0)
845 			goto out;
846 		else
847 			rc = 0; /* success cases including fallback */
848 	}
849 
850 out:
851 	release_sock(sk);
852 out_err:
853 	return rc;
854 }
855 
856 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
857 {
858 	struct socket *new_clcsock = NULL;
859 	struct sock *lsk = &lsmc->sk;
860 	struct sock *new_sk;
861 	int rc = -EINVAL;
862 
863 	release_sock(lsk);
864 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
865 	if (!new_sk) {
866 		rc = -ENOMEM;
867 		lsk->sk_err = ENOMEM;
868 		*new_smc = NULL;
869 		lock_sock(lsk);
870 		goto out;
871 	}
872 	*new_smc = smc_sk(new_sk);
873 
874 	mutex_lock(&lsmc->clcsock_release_lock);
875 	if (lsmc->clcsock)
876 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
877 	mutex_unlock(&lsmc->clcsock_release_lock);
878 	lock_sock(lsk);
879 	if  (rc < 0)
880 		lsk->sk_err = -rc;
881 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
882 		new_sk->sk_prot->unhash(new_sk);
883 		if (new_clcsock)
884 			sock_release(new_clcsock);
885 		new_sk->sk_state = SMC_CLOSED;
886 		sock_set_flag(new_sk, SOCK_DEAD);
887 		sock_put(new_sk); /* final */
888 		*new_smc = NULL;
889 		goto out;
890 	}
891 
892 	(*new_smc)->clcsock = new_clcsock;
893 out:
894 	return rc;
895 }
896 
897 /* add a just created sock to the accept queue of the listen sock as
898  * candidate for a following socket accept call from user space
899  */
900 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
901 {
902 	struct smc_sock *par = smc_sk(parent);
903 
904 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
905 	spin_lock(&par->accept_q_lock);
906 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
907 	spin_unlock(&par->accept_q_lock);
908 	sk_acceptq_added(parent);
909 }
910 
911 /* remove a socket from the accept queue of its parental listening socket */
912 static void smc_accept_unlink(struct sock *sk)
913 {
914 	struct smc_sock *par = smc_sk(sk)->listen_smc;
915 
916 	spin_lock(&par->accept_q_lock);
917 	list_del_init(&smc_sk(sk)->accept_q);
918 	spin_unlock(&par->accept_q_lock);
919 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
920 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
921 }
922 
923 /* remove a sock from the accept queue to bind it to a new socket created
924  * for a socket accept call from user space
925  */
926 struct sock *smc_accept_dequeue(struct sock *parent,
927 				struct socket *new_sock)
928 {
929 	struct smc_sock *isk, *n;
930 	struct sock *new_sk;
931 
932 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
933 		new_sk = (struct sock *)isk;
934 
935 		smc_accept_unlink(new_sk);
936 		if (new_sk->sk_state == SMC_CLOSED) {
937 			new_sk->sk_prot->unhash(new_sk);
938 			if (isk->clcsock) {
939 				sock_release(isk->clcsock);
940 				isk->clcsock = NULL;
941 			}
942 			sock_put(new_sk); /* final */
943 			continue;
944 		}
945 		if (new_sock) {
946 			sock_graft(new_sk, new_sock);
947 			if (isk->use_fallback) {
948 				smc_sk(new_sk)->clcsock->file = new_sock->file;
949 				isk->clcsock->file->private_data = isk->clcsock;
950 			}
951 		}
952 		return new_sk;
953 	}
954 	return NULL;
955 }
956 
957 /* clean up for a created but never accepted sock */
958 void smc_close_non_accepted(struct sock *sk)
959 {
960 	struct smc_sock *smc = smc_sk(sk);
961 
962 	lock_sock(sk);
963 	if (!sk->sk_lingertime)
964 		/* wait for peer closing */
965 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
966 	if (!smc->use_fallback) {
967 		smc_close_active(smc);
968 		sock_set_flag(sk, SOCK_DEAD);
969 		sk->sk_shutdown |= SHUTDOWN_MASK;
970 	}
971 	sk->sk_prot->unhash(sk);
972 	if (smc->clcsock) {
973 		struct socket *tcp;
974 
975 		tcp = smc->clcsock;
976 		smc->clcsock = NULL;
977 		sock_release(tcp);
978 	}
979 	if (smc->use_fallback) {
980 		sock_put(sk); /* passive closing */
981 		sk->sk_state = SMC_CLOSED;
982 	} else {
983 		if (sk->sk_state == SMC_CLOSED)
984 			smc_conn_free(&smc->conn);
985 	}
986 	release_sock(sk);
987 	sock_put(sk); /* final sock_put */
988 }
989 
990 static int smc_serv_conf_first_link(struct smc_sock *smc)
991 {
992 	struct net *net = sock_net(smc->clcsock->sk);
993 	struct smc_link_group *lgr = smc->conn.lgr;
994 	struct smc_link *link;
995 	int rest;
996 	int rc;
997 
998 	link = &lgr->lnk[SMC_SINGLE_LINK];
999 
1000 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
1001 		return SMC_CLC_DECL_ERR_REGRMB;
1002 
1003 	/* send CONFIRM LINK request to client over the RoCE fabric */
1004 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1005 	if (rc < 0)
1006 		return SMC_CLC_DECL_TIMEOUT_CL;
1007 
1008 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1009 	rest = wait_for_completion_interruptible_timeout(
1010 		&link->llc_confirm_resp,
1011 		SMC_LLC_WAIT_FIRST_TIME);
1012 	if (rest <= 0) {
1013 		struct smc_clc_msg_decline dclc;
1014 
1015 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1016 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1017 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1018 	}
1019 
1020 	if (link->llc_confirm_resp_rc)
1021 		return SMC_CLC_DECL_RMBE_EC;
1022 
1023 	/* send ADD LINK request to client over the RoCE fabric */
1024 	rc = smc_llc_send_add_link(link,
1025 				   link->smcibdev->mac[link->ibport - 1],
1026 				   link->gid, SMC_LLC_REQ);
1027 	if (rc < 0)
1028 		return SMC_CLC_DECL_TIMEOUT_AL;
1029 
1030 	/* receive ADD LINK response from client over the RoCE fabric */
1031 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
1032 							 SMC_LLC_WAIT_TIME);
1033 	if (rest <= 0) {
1034 		struct smc_clc_msg_decline dclc;
1035 
1036 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1037 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1038 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
1039 	}
1040 
1041 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
1042 
1043 	return 0;
1044 }
1045 
1046 /* listen worker: finish */
1047 static void smc_listen_out(struct smc_sock *new_smc)
1048 {
1049 	struct smc_sock *lsmc = new_smc->listen_smc;
1050 	struct sock *newsmcsk = &new_smc->sk;
1051 
1052 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1053 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1054 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1055 		release_sock(&lsmc->sk);
1056 	} else { /* no longer listening */
1057 		smc_close_non_accepted(newsmcsk);
1058 	}
1059 
1060 	/* Wake up accept */
1061 	lsmc->sk.sk_data_ready(&lsmc->sk);
1062 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1063 }
1064 
1065 /* listen worker: finish in state connected */
1066 static void smc_listen_out_connected(struct smc_sock *new_smc)
1067 {
1068 	struct sock *newsmcsk = &new_smc->sk;
1069 
1070 	sk_refcnt_debug_inc(newsmcsk);
1071 	if (newsmcsk->sk_state == SMC_INIT)
1072 		newsmcsk->sk_state = SMC_ACTIVE;
1073 
1074 	smc_listen_out(new_smc);
1075 }
1076 
1077 /* listen worker: finish in error state */
1078 static void smc_listen_out_err(struct smc_sock *new_smc)
1079 {
1080 	struct sock *newsmcsk = &new_smc->sk;
1081 
1082 	if (newsmcsk->sk_state == SMC_INIT)
1083 		sock_put(&new_smc->sk); /* passive closing */
1084 	newsmcsk->sk_state = SMC_CLOSED;
1085 	smc_conn_free(&new_smc->conn);
1086 
1087 	smc_listen_out(new_smc);
1088 }
1089 
1090 /* listen worker: decline and fall back if possible */
1091 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1092 			       int local_contact)
1093 {
1094 	/* RDMA setup failed, switch back to TCP */
1095 	if (local_contact == SMC_FIRST_CONTACT)
1096 		smc_lgr_forget(new_smc->conn.lgr);
1097 	if (reason_code < 0) { /* error, no fallback possible */
1098 		smc_listen_out_err(new_smc);
1099 		return;
1100 	}
1101 	smc_conn_free(&new_smc->conn);
1102 	smc_switch_to_fallback(new_smc);
1103 	new_smc->fallback_rsn = reason_code;
1104 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1105 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1106 			smc_listen_out_err(new_smc);
1107 			return;
1108 		}
1109 	}
1110 	smc_listen_out_connected(new_smc);
1111 }
1112 
1113 /* listen worker: check prefixes */
1114 static int smc_listen_prfx_check(struct smc_sock *new_smc,
1115 				 struct smc_clc_msg_proposal *pclc)
1116 {
1117 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1118 	struct socket *newclcsock = new_smc->clcsock;
1119 
1120 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1121 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1122 		return SMC_CLC_DECL_DIFFPREFIX;
1123 
1124 	return 0;
1125 }
1126 
1127 /* listen worker: initialize connection and buffers */
1128 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1129 				struct smc_init_info *ini)
1130 {
1131 	int rc;
1132 
1133 	/* allocate connection / link group */
1134 	rc = smc_conn_create(new_smc, ini);
1135 	if (rc)
1136 		return rc;
1137 
1138 	/* create send buffer and rmb */
1139 	if (smc_buf_create(new_smc, false))
1140 		return SMC_CLC_DECL_MEM;
1141 
1142 	return 0;
1143 }
1144 
1145 /* listen worker: initialize connection and buffers for SMC-D */
1146 static int smc_listen_ism_init(struct smc_sock *new_smc,
1147 			       struct smc_clc_msg_proposal *pclc,
1148 			       struct smc_init_info *ini)
1149 {
1150 	struct smc_clc_msg_smcd *pclc_smcd;
1151 	int rc;
1152 
1153 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1154 	ini->ism_gid = pclc_smcd->gid;
1155 	rc = smc_conn_create(new_smc, ini);
1156 	if (rc)
1157 		return rc;
1158 
1159 	/* Check if peer can be reached via ISM device */
1160 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1161 			    new_smc->conn.lgr->vlan_id,
1162 			    new_smc->conn.lgr->smcd)) {
1163 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1164 			smc_lgr_forget(new_smc->conn.lgr);
1165 		smc_conn_free(&new_smc->conn);
1166 		return SMC_CLC_DECL_SMCDNOTALK;
1167 	}
1168 
1169 	/* Create send and receive buffers */
1170 	if (smc_buf_create(new_smc, true)) {
1171 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1172 			smc_lgr_forget(new_smc->conn.lgr);
1173 		smc_conn_free(&new_smc->conn);
1174 		return SMC_CLC_DECL_MEM;
1175 	}
1176 
1177 	return 0;
1178 }
1179 
1180 /* listen worker: register buffers */
1181 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1182 {
1183 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1184 
1185 	if (local_contact != SMC_FIRST_CONTACT) {
1186 		if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true))
1187 			return SMC_CLC_DECL_ERR_REGRMB;
1188 	}
1189 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1190 
1191 	return 0;
1192 }
1193 
1194 /* listen worker: finish RDMA setup */
1195 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1196 				  struct smc_clc_msg_accept_confirm *cclc,
1197 				  int local_contact)
1198 {
1199 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1200 	int reason_code = 0;
1201 
1202 	if (local_contact == SMC_FIRST_CONTACT)
1203 		smc_link_save_peer_info(link, cclc);
1204 
1205 	if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) {
1206 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1207 		goto decline;
1208 	}
1209 
1210 	if (local_contact == SMC_FIRST_CONTACT) {
1211 		if (smc_ib_ready_link(link)) {
1212 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1213 			goto decline;
1214 		}
1215 		/* QP confirmation over RoCE fabric */
1216 		reason_code = smc_serv_conf_first_link(new_smc);
1217 		if (reason_code)
1218 			goto decline;
1219 	}
1220 	return 0;
1221 
1222 decline:
1223 	smc_listen_decline(new_smc, reason_code, local_contact);
1224 	return reason_code;
1225 }
1226 
1227 /* setup for RDMA connection of server */
1228 static void smc_listen_work(struct work_struct *work)
1229 {
1230 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1231 						smc_listen_work);
1232 	struct socket *newclcsock = new_smc->clcsock;
1233 	struct smc_clc_msg_accept_confirm cclc;
1234 	struct smc_clc_msg_proposal *pclc;
1235 	struct smc_init_info ini = {0};
1236 	bool ism_supported = false;
1237 	u8 buf[SMC_CLC_MAX_LEN];
1238 	int rc = 0;
1239 
1240 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1241 		return smc_listen_out_err(new_smc);
1242 
1243 	if (new_smc->use_fallback) {
1244 		smc_listen_out_connected(new_smc);
1245 		return;
1246 	}
1247 
1248 	/* check if peer is smc capable */
1249 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1250 		smc_switch_to_fallback(new_smc);
1251 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1252 		smc_listen_out_connected(new_smc);
1253 		return;
1254 	}
1255 
1256 	/* do inband token exchange -
1257 	 * wait for and receive SMC Proposal CLC message
1258 	 */
1259 	pclc = (struct smc_clc_msg_proposal *)&buf;
1260 	rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1261 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1262 	if (rc)
1263 		goto out_decl;
1264 
1265 	/* IPSec connections opt out of SMC-R optimizations */
1266 	if (using_ipsec(new_smc)) {
1267 		rc = SMC_CLC_DECL_IPSEC;
1268 		goto out_decl;
1269 	}
1270 
1271 	/* check for matching IP prefix and subnet length */
1272 	rc = smc_listen_prfx_check(new_smc, pclc);
1273 	if (rc)
1274 		goto out_decl;
1275 
1276 	/* get vlan id from IP device */
1277 	if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) {
1278 		rc = SMC_CLC_DECL_GETVLANERR;
1279 		goto out_decl;
1280 	}
1281 
1282 	mutex_lock(&smc_server_lgr_pending);
1283 	smc_close_init(new_smc);
1284 	smc_rx_init(new_smc);
1285 	smc_tx_init(new_smc);
1286 
1287 	/* check if ISM is available */
1288 	if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) {
1289 		ini.is_smcd = true; /* prepare ISM check */
1290 		rc = smc_find_ism_device(new_smc, &ini);
1291 		if (!rc)
1292 			rc = smc_listen_ism_init(new_smc, pclc, &ini);
1293 		if (!rc)
1294 			ism_supported = true;
1295 		else if (pclc->hdr.path == SMC_TYPE_D)
1296 			goto out_unlock; /* skip RDMA and decline */
1297 	}
1298 
1299 	/* check if RDMA is available */
1300 	if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */
1301 		/* prepare RDMA check */
1302 		memset(&ini, 0, sizeof(ini));
1303 		ini.is_smcd = false;
1304 		ini.ib_lcl = &pclc->lcl;
1305 		rc = smc_find_rdma_device(new_smc, &ini);
1306 		if (rc) {
1307 			/* no RDMA device found */
1308 			if (pclc->hdr.path == SMC_TYPE_B)
1309 				/* neither ISM nor RDMA device found */
1310 				rc = SMC_CLC_DECL_NOSMCDEV;
1311 			goto out_unlock;
1312 		}
1313 		rc = smc_listen_rdma_init(new_smc, &ini);
1314 		if (rc)
1315 			goto out_unlock;
1316 		rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact);
1317 		if (rc)
1318 			goto out_unlock;
1319 	}
1320 
1321 	/* send SMC Accept CLC message */
1322 	rc = smc_clc_send_accept(new_smc, ini.cln_first_contact);
1323 	if (rc)
1324 		goto out_unlock;
1325 
1326 	/* SMC-D does not need this lock any more */
1327 	if (ism_supported)
1328 		mutex_unlock(&smc_server_lgr_pending);
1329 
1330 	/* receive SMC Confirm CLC message */
1331 	rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1332 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1333 	if (rc) {
1334 		if (!ism_supported)
1335 			goto out_unlock;
1336 		goto out_decl;
1337 	}
1338 
1339 	/* finish worker */
1340 	if (!ism_supported) {
1341 		rc = smc_listen_rdma_finish(new_smc, &cclc,
1342 					    ini.cln_first_contact);
1343 		mutex_unlock(&smc_server_lgr_pending);
1344 		if (rc)
1345 			return;
1346 	}
1347 	smc_conn_save_peer_info(new_smc, &cclc);
1348 	smc_listen_out_connected(new_smc);
1349 	return;
1350 
1351 out_unlock:
1352 	mutex_unlock(&smc_server_lgr_pending);
1353 out_decl:
1354 	smc_listen_decline(new_smc, rc, ini.cln_first_contact);
1355 }
1356 
1357 static void smc_tcp_listen_work(struct work_struct *work)
1358 {
1359 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1360 					     tcp_listen_work);
1361 	struct sock *lsk = &lsmc->sk;
1362 	struct smc_sock *new_smc;
1363 	int rc = 0;
1364 
1365 	lock_sock(lsk);
1366 	while (lsk->sk_state == SMC_LISTEN) {
1367 		rc = smc_clcsock_accept(lsmc, &new_smc);
1368 		if (rc)
1369 			goto out;
1370 		if (!new_smc)
1371 			continue;
1372 
1373 		new_smc->listen_smc = lsmc;
1374 		new_smc->use_fallback = lsmc->use_fallback;
1375 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1376 		sock_hold(lsk); /* sock_put in smc_listen_work */
1377 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1378 		smc_copy_sock_settings_to_smc(new_smc);
1379 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1380 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1381 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1382 		if (!schedule_work(&new_smc->smc_listen_work))
1383 			sock_put(&new_smc->sk);
1384 	}
1385 
1386 out:
1387 	release_sock(lsk);
1388 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1389 }
1390 
1391 static int smc_listen(struct socket *sock, int backlog)
1392 {
1393 	struct sock *sk = sock->sk;
1394 	struct smc_sock *smc;
1395 	int rc;
1396 
1397 	smc = smc_sk(sk);
1398 	lock_sock(sk);
1399 
1400 	rc = -EINVAL;
1401 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
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) {
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 && sk->sk_state != SMC_LISTEN) {
1744 			if (val && !smc->use_fallback)
1745 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1746 						 0);
1747 		}
1748 		break;
1749 	case TCP_CORK:
1750 		if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
1751 			if (!val && !smc->use_fallback)
1752 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1753 						 0);
1754 		}
1755 		break;
1756 	case TCP_DEFER_ACCEPT:
1757 		smc->sockopt_defer_accept = val;
1758 		break;
1759 	default:
1760 		break;
1761 	}
1762 	release_sock(sk);
1763 
1764 	return rc;
1765 }
1766 
1767 static int smc_getsockopt(struct socket *sock, int level, int optname,
1768 			  char __user *optval, int __user *optlen)
1769 {
1770 	struct smc_sock *smc;
1771 
1772 	smc = smc_sk(sock->sk);
1773 	/* socket options apply to the CLC socket */
1774 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1775 					     optval, optlen);
1776 }
1777 
1778 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1779 		     unsigned long arg)
1780 {
1781 	union smc_host_cursor cons, urg;
1782 	struct smc_connection *conn;
1783 	struct smc_sock *smc;
1784 	int answ;
1785 
1786 	smc = smc_sk(sock->sk);
1787 	conn = &smc->conn;
1788 	lock_sock(&smc->sk);
1789 	if (smc->use_fallback) {
1790 		if (!smc->clcsock) {
1791 			release_sock(&smc->sk);
1792 			return -EBADF;
1793 		}
1794 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1795 		release_sock(&smc->sk);
1796 		return answ;
1797 	}
1798 	switch (cmd) {
1799 	case SIOCINQ: /* same as FIONREAD */
1800 		if (smc->sk.sk_state == SMC_LISTEN) {
1801 			release_sock(&smc->sk);
1802 			return -EINVAL;
1803 		}
1804 		if (smc->sk.sk_state == SMC_INIT ||
1805 		    smc->sk.sk_state == SMC_CLOSED)
1806 			answ = 0;
1807 		else
1808 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1809 		break;
1810 	case SIOCOUTQ:
1811 		/* output queue size (not send + not acked) */
1812 		if (smc->sk.sk_state == SMC_LISTEN) {
1813 			release_sock(&smc->sk);
1814 			return -EINVAL;
1815 		}
1816 		if (smc->sk.sk_state == SMC_INIT ||
1817 		    smc->sk.sk_state == SMC_CLOSED)
1818 			answ = 0;
1819 		else
1820 			answ = smc->conn.sndbuf_desc->len -
1821 					atomic_read(&smc->conn.sndbuf_space);
1822 		break;
1823 	case SIOCOUTQNSD:
1824 		/* output queue size (not send only) */
1825 		if (smc->sk.sk_state == SMC_LISTEN) {
1826 			release_sock(&smc->sk);
1827 			return -EINVAL;
1828 		}
1829 		if (smc->sk.sk_state == SMC_INIT ||
1830 		    smc->sk.sk_state == SMC_CLOSED)
1831 			answ = 0;
1832 		else
1833 			answ = smc_tx_prepared_sends(&smc->conn);
1834 		break;
1835 	case SIOCATMARK:
1836 		if (smc->sk.sk_state == SMC_LISTEN) {
1837 			release_sock(&smc->sk);
1838 			return -EINVAL;
1839 		}
1840 		if (smc->sk.sk_state == SMC_INIT ||
1841 		    smc->sk.sk_state == SMC_CLOSED) {
1842 			answ = 0;
1843 		} else {
1844 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1845 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1846 			answ = smc_curs_diff(conn->rmb_desc->len,
1847 					     &cons, &urg) == 1;
1848 		}
1849 		break;
1850 	default:
1851 		release_sock(&smc->sk);
1852 		return -ENOIOCTLCMD;
1853 	}
1854 	release_sock(&smc->sk);
1855 
1856 	return put_user(answ, (int __user *)arg);
1857 }
1858 
1859 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1860 			    int offset, size_t size, int flags)
1861 {
1862 	struct sock *sk = sock->sk;
1863 	struct smc_sock *smc;
1864 	int rc = -EPIPE;
1865 
1866 	smc = smc_sk(sk);
1867 	lock_sock(sk);
1868 	if (sk->sk_state != SMC_ACTIVE) {
1869 		release_sock(sk);
1870 		goto out;
1871 	}
1872 	release_sock(sk);
1873 	if (smc->use_fallback)
1874 		rc = kernel_sendpage(smc->clcsock, page, offset,
1875 				     size, flags);
1876 	else
1877 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1878 
1879 out:
1880 	return rc;
1881 }
1882 
1883 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1884  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1885  * updates till whenever a respective page has been fully processed.
1886  * Note that subsequent recv() calls have to wait till all splice() processing
1887  * completed.
1888  */
1889 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1890 			       struct pipe_inode_info *pipe, size_t len,
1891 			       unsigned int flags)
1892 {
1893 	struct sock *sk = sock->sk;
1894 	struct smc_sock *smc;
1895 	int rc = -ENOTCONN;
1896 
1897 	smc = smc_sk(sk);
1898 	lock_sock(sk);
1899 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1900 		/* socket was connected before, no more data to read */
1901 		rc = 0;
1902 		goto out;
1903 	}
1904 	if (sk->sk_state == SMC_INIT ||
1905 	    sk->sk_state == SMC_LISTEN ||
1906 	    sk->sk_state == SMC_CLOSED)
1907 		goto out;
1908 
1909 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1910 		rc = 0;
1911 		goto out;
1912 	}
1913 
1914 	if (smc->use_fallback) {
1915 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1916 						    pipe, len, flags);
1917 	} else {
1918 		if (*ppos) {
1919 			rc = -ESPIPE;
1920 			goto out;
1921 		}
1922 		if (flags & SPLICE_F_NONBLOCK)
1923 			flags = MSG_DONTWAIT;
1924 		else
1925 			flags = 0;
1926 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1927 	}
1928 out:
1929 	release_sock(sk);
1930 
1931 	return rc;
1932 }
1933 
1934 /* must look like tcp */
1935 static const struct proto_ops smc_sock_ops = {
1936 	.family		= PF_SMC,
1937 	.owner		= THIS_MODULE,
1938 	.release	= smc_release,
1939 	.bind		= smc_bind,
1940 	.connect	= smc_connect,
1941 	.socketpair	= sock_no_socketpair,
1942 	.accept		= smc_accept,
1943 	.getname	= smc_getname,
1944 	.poll		= smc_poll,
1945 	.ioctl		= smc_ioctl,
1946 	.listen		= smc_listen,
1947 	.shutdown	= smc_shutdown,
1948 	.setsockopt	= smc_setsockopt,
1949 	.getsockopt	= smc_getsockopt,
1950 	.sendmsg	= smc_sendmsg,
1951 	.recvmsg	= smc_recvmsg,
1952 	.mmap		= sock_no_mmap,
1953 	.sendpage	= smc_sendpage,
1954 	.splice_read	= smc_splice_read,
1955 };
1956 
1957 static int smc_create(struct net *net, struct socket *sock, int protocol,
1958 		      int kern)
1959 {
1960 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1961 	struct smc_sock *smc;
1962 	struct sock *sk;
1963 	int rc;
1964 
1965 	rc = -ESOCKTNOSUPPORT;
1966 	if (sock->type != SOCK_STREAM)
1967 		goto out;
1968 
1969 	rc = -EPROTONOSUPPORT;
1970 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1971 		goto out;
1972 
1973 	rc = -ENOBUFS;
1974 	sock->ops = &smc_sock_ops;
1975 	sk = smc_sock_alloc(net, sock, protocol);
1976 	if (!sk)
1977 		goto out;
1978 
1979 	/* create internal TCP socket for CLC handshake and fallback */
1980 	smc = smc_sk(sk);
1981 	smc->use_fallback = false; /* assume rdma capability first */
1982 	smc->fallback_rsn = 0;
1983 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1984 			      &smc->clcsock);
1985 	if (rc) {
1986 		sk_common_release(sk);
1987 		goto out;
1988 	}
1989 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1990 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1991 
1992 out:
1993 	return rc;
1994 }
1995 
1996 static const struct net_proto_family smc_sock_family_ops = {
1997 	.family	= PF_SMC,
1998 	.owner	= THIS_MODULE,
1999 	.create	= smc_create,
2000 };
2001 
2002 unsigned int smc_net_id;
2003 
2004 static __net_init int smc_net_init(struct net *net)
2005 {
2006 	return smc_pnet_net_init(net);
2007 }
2008 
2009 static void __net_exit smc_net_exit(struct net *net)
2010 {
2011 	smc_pnet_net_exit(net);
2012 }
2013 
2014 static struct pernet_operations smc_net_ops = {
2015 	.init = smc_net_init,
2016 	.exit = smc_net_exit,
2017 	.id   = &smc_net_id,
2018 	.size = sizeof(struct smc_net),
2019 };
2020 
2021 static int __init smc_init(void)
2022 {
2023 	int rc;
2024 
2025 	rc = register_pernet_subsys(&smc_net_ops);
2026 	if (rc)
2027 		return rc;
2028 
2029 	rc = smc_pnet_init();
2030 	if (rc)
2031 		return rc;
2032 
2033 	rc = smc_llc_init();
2034 	if (rc) {
2035 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2036 		goto out_pnet;
2037 	}
2038 
2039 	rc = smc_cdc_init();
2040 	if (rc) {
2041 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2042 		goto out_pnet;
2043 	}
2044 
2045 	rc = proto_register(&smc_proto, 1);
2046 	if (rc) {
2047 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2048 		goto out_pnet;
2049 	}
2050 
2051 	rc = proto_register(&smc_proto6, 1);
2052 	if (rc) {
2053 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2054 		goto out_proto;
2055 	}
2056 
2057 	rc = sock_register(&smc_sock_family_ops);
2058 	if (rc) {
2059 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2060 		goto out_proto6;
2061 	}
2062 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2063 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2064 
2065 	rc = smc_ib_register_client();
2066 	if (rc) {
2067 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2068 		goto out_sock;
2069 	}
2070 
2071 	static_branch_enable(&tcp_have_smc);
2072 	return 0;
2073 
2074 out_sock:
2075 	sock_unregister(PF_SMC);
2076 out_proto6:
2077 	proto_unregister(&smc_proto6);
2078 out_proto:
2079 	proto_unregister(&smc_proto);
2080 out_pnet:
2081 	smc_pnet_exit();
2082 	return rc;
2083 }
2084 
2085 static void __exit smc_exit(void)
2086 {
2087 	smc_core_exit();
2088 	static_branch_disable(&tcp_have_smc);
2089 	smc_ib_unregister_client();
2090 	sock_unregister(PF_SMC);
2091 	proto_unregister(&smc_proto6);
2092 	proto_unregister(&smc_proto);
2093 	smc_pnet_exit();
2094 	unregister_pernet_subsys(&smc_net_ops);
2095 }
2096 
2097 module_init(smc_init);
2098 module_exit(smc_exit);
2099 
2100 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2101 MODULE_DESCRIPTION("smc socket address family");
2102 MODULE_LICENSE("GPL");
2103 MODULE_ALIAS_NETPROTO(PF_SMC);
2104