xref: /openbmc/linux/net/smc/af_smc.c (revision 29c37341)
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 #include <linux/rcupdate_wait.h>
29 
30 #include <net/sock.h>
31 #include <net/tcp.h>
32 #include <net/smc.h>
33 #include <asm/ioctls.h>
34 
35 #include <net/net_namespace.h>
36 #include <net/netns/generic.h>
37 #include "smc_netns.h"
38 
39 #include "smc.h"
40 #include "smc_clc.h"
41 #include "smc_llc.h"
42 #include "smc_cdc.h"
43 #include "smc_core.h"
44 #include "smc_ib.h"
45 #include "smc_ism.h"
46 #include "smc_pnet.h"
47 #include "smc_tx.h"
48 #include "smc_rx.h"
49 #include "smc_close.h"
50 
51 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
52 						 * creation on server
53 						 */
54 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
55 						 * creation on client
56 						 */
57 
58 static void smc_tcp_listen_work(struct work_struct *);
59 static void smc_connect_work(struct work_struct *);
60 
61 static void smc_set_keepalive(struct sock *sk, int val)
62 {
63 	struct smc_sock *smc = smc_sk(sk);
64 
65 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
66 }
67 
68 static struct smc_hashinfo smc_v4_hashinfo = {
69 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
70 };
71 
72 static struct smc_hashinfo smc_v6_hashinfo = {
73 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
74 };
75 
76 int smc_hash_sk(struct sock *sk)
77 {
78 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
79 	struct hlist_head *head;
80 
81 	head = &h->ht;
82 
83 	write_lock_bh(&h->lock);
84 	sk_add_node(sk, head);
85 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
86 	write_unlock_bh(&h->lock);
87 
88 	return 0;
89 }
90 EXPORT_SYMBOL_GPL(smc_hash_sk);
91 
92 void smc_unhash_sk(struct sock *sk)
93 {
94 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
95 
96 	write_lock_bh(&h->lock);
97 	if (sk_del_node_init(sk))
98 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
99 	write_unlock_bh(&h->lock);
100 }
101 EXPORT_SYMBOL_GPL(smc_unhash_sk);
102 
103 struct proto smc_proto = {
104 	.name		= "SMC",
105 	.owner		= THIS_MODULE,
106 	.keepalive	= smc_set_keepalive,
107 	.hash		= smc_hash_sk,
108 	.unhash		= smc_unhash_sk,
109 	.obj_size	= sizeof(struct smc_sock),
110 	.h.smc_hash	= &smc_v4_hashinfo,
111 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
112 };
113 EXPORT_SYMBOL_GPL(smc_proto);
114 
115 struct proto smc_proto6 = {
116 	.name		= "SMC6",
117 	.owner		= THIS_MODULE,
118 	.keepalive	= smc_set_keepalive,
119 	.hash		= smc_hash_sk,
120 	.unhash		= smc_unhash_sk,
121 	.obj_size	= sizeof(struct smc_sock),
122 	.h.smc_hash	= &smc_v6_hashinfo,
123 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
124 };
125 EXPORT_SYMBOL_GPL(smc_proto6);
126 
127 static void smc_restore_fallback_changes(struct smc_sock *smc)
128 {
129 	if (smc->clcsock->file) { /* non-accepted sockets have no file yet */
130 		smc->clcsock->file->private_data = smc->sk.sk_socket;
131 		smc->clcsock->file = NULL;
132 	}
133 }
134 
135 static int __smc_release(struct smc_sock *smc)
136 {
137 	struct sock *sk = &smc->sk;
138 	int rc = 0;
139 
140 	if (!smc->use_fallback) {
141 		rc = smc_close_active(smc);
142 		sock_set_flag(sk, SOCK_DEAD);
143 		sk->sk_shutdown |= SHUTDOWN_MASK;
144 	} else {
145 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
146 			sock_put(sk); /* passive closing */
147 		if (sk->sk_state == SMC_LISTEN) {
148 			/* wake up clcsock accept */
149 			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
150 		}
151 		sk->sk_state = SMC_CLOSED;
152 		sk->sk_state_change(sk);
153 		smc_restore_fallback_changes(smc);
154 	}
155 
156 	sk->sk_prot->unhash(sk);
157 
158 	if (sk->sk_state == SMC_CLOSED) {
159 		if (smc->clcsock) {
160 			release_sock(sk);
161 			smc_clcsock_release(smc);
162 			lock_sock(sk);
163 		}
164 		if (!smc->use_fallback)
165 			smc_conn_free(&smc->conn);
166 	}
167 
168 	return rc;
169 }
170 
171 static int smc_release(struct socket *sock)
172 {
173 	struct sock *sk = sock->sk;
174 	struct smc_sock *smc;
175 	int rc = 0;
176 
177 	if (!sk)
178 		goto out;
179 
180 	sock_hold(sk); /* sock_put below */
181 	smc = smc_sk(sk);
182 
183 	/* cleanup for a dangling non-blocking connect */
184 	if (smc->connect_nonblock && sk->sk_state == SMC_INIT)
185 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
186 	flush_work(&smc->connect_work);
187 
188 	if (sk->sk_state == SMC_LISTEN)
189 		/* smc_close_non_accepted() is called and acquires
190 		 * sock lock for child sockets again
191 		 */
192 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
193 	else
194 		lock_sock(sk);
195 
196 	rc = __smc_release(smc);
197 
198 	/* detach socket */
199 	sock_orphan(sk);
200 	sock->sk = NULL;
201 	release_sock(sk);
202 
203 	sock_put(sk); /* sock_hold above */
204 	sock_put(sk); /* final sock_put */
205 out:
206 	return rc;
207 }
208 
209 static void smc_destruct(struct sock *sk)
210 {
211 	if (sk->sk_state != SMC_CLOSED)
212 		return;
213 	if (!sock_flag(sk, SOCK_DEAD))
214 		return;
215 
216 	sk_refcnt_debug_dec(sk);
217 }
218 
219 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
220 				   int protocol)
221 {
222 	struct smc_sock *smc;
223 	struct proto *prot;
224 	struct sock *sk;
225 
226 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
227 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
228 	if (!sk)
229 		return NULL;
230 
231 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
232 	sk->sk_state = SMC_INIT;
233 	sk->sk_destruct = smc_destruct;
234 	sk->sk_protocol = protocol;
235 	smc = smc_sk(sk);
236 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
237 	INIT_WORK(&smc->connect_work, smc_connect_work);
238 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
239 	INIT_LIST_HEAD(&smc->accept_q);
240 	spin_lock_init(&smc->accept_q_lock);
241 	spin_lock_init(&smc->conn.send_lock);
242 	sk->sk_prot->hash(sk);
243 	sk_refcnt_debug_inc(sk);
244 	mutex_init(&smc->clcsock_release_lock);
245 
246 	return sk;
247 }
248 
249 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
250 		    int addr_len)
251 {
252 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
253 	struct sock *sk = sock->sk;
254 	struct smc_sock *smc;
255 	int rc;
256 
257 	smc = smc_sk(sk);
258 
259 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
260 	rc = -EINVAL;
261 	if (addr_len < sizeof(struct sockaddr_in))
262 		goto out;
263 
264 	rc = -EAFNOSUPPORT;
265 	if (addr->sin_family != AF_INET &&
266 	    addr->sin_family != AF_INET6 &&
267 	    addr->sin_family != AF_UNSPEC)
268 		goto out;
269 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
270 	if (addr->sin_family == AF_UNSPEC &&
271 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
272 		goto out;
273 
274 	lock_sock(sk);
275 
276 	/* Check if socket is already active */
277 	rc = -EINVAL;
278 	if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
279 		goto out_rel;
280 
281 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
282 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
283 
284 out_rel:
285 	release_sock(sk);
286 out:
287 	return rc;
288 }
289 
290 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
291 				   unsigned long mask)
292 {
293 	/* options we don't get control via setsockopt for */
294 	nsk->sk_type = osk->sk_type;
295 	nsk->sk_sndbuf = osk->sk_sndbuf;
296 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
297 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
298 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
299 	nsk->sk_mark = osk->sk_mark;
300 	nsk->sk_priority = osk->sk_priority;
301 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
302 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
303 	nsk->sk_err = osk->sk_err;
304 
305 	nsk->sk_flags &= ~mask;
306 	nsk->sk_flags |= osk->sk_flags & mask;
307 }
308 
309 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
310 			     (1UL << SOCK_KEEPOPEN) | \
311 			     (1UL << SOCK_LINGER) | \
312 			     (1UL << SOCK_BROADCAST) | \
313 			     (1UL << SOCK_TIMESTAMP) | \
314 			     (1UL << SOCK_DBG) | \
315 			     (1UL << SOCK_RCVTSTAMP) | \
316 			     (1UL << SOCK_RCVTSTAMPNS) | \
317 			     (1UL << SOCK_LOCALROUTE) | \
318 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
319 			     (1UL << SOCK_RXQ_OVFL) | \
320 			     (1UL << SOCK_WIFI_STATUS) | \
321 			     (1UL << SOCK_NOFCS) | \
322 			     (1UL << SOCK_FILTER_LOCKED) | \
323 			     (1UL << SOCK_TSTAMP_NEW))
324 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
325  * clc socket (since smc is not called for these options from net/core)
326  */
327 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
328 {
329 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
330 }
331 
332 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
333 			     (1UL << SOCK_KEEPOPEN) | \
334 			     (1UL << SOCK_LINGER) | \
335 			     (1UL << SOCK_DBG))
336 /* copy only settings and flags relevant for smc from clc to smc socket */
337 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
338 {
339 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
340 }
341 
342 /* register the new rmb on all links */
343 static int smcr_lgr_reg_rmbs(struct smc_link *link,
344 			     struct smc_buf_desc *rmb_desc)
345 {
346 	struct smc_link_group *lgr = link->lgr;
347 	int i, rc = 0;
348 
349 	rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
350 	if (rc)
351 		return rc;
352 	/* protect against parallel smc_llc_cli_rkey_exchange() and
353 	 * parallel smcr_link_reg_rmb()
354 	 */
355 	mutex_lock(&lgr->llc_conf_mutex);
356 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
357 		if (!smc_link_active(&lgr->lnk[i]))
358 			continue;
359 		rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc);
360 		if (rc)
361 			goto out;
362 	}
363 
364 	/* exchange confirm_rkey msg with peer */
365 	rc = smc_llc_do_confirm_rkey(link, rmb_desc);
366 	if (rc) {
367 		rc = -EFAULT;
368 		goto out;
369 	}
370 	rmb_desc->is_conf_rkey = true;
371 out:
372 	mutex_unlock(&lgr->llc_conf_mutex);
373 	smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
374 	return rc;
375 }
376 
377 static int smcr_clnt_conf_first_link(struct smc_sock *smc)
378 {
379 	struct smc_link *link = smc->conn.lnk;
380 	struct smc_llc_qentry *qentry;
381 	int rc;
382 
383 	/* receive CONFIRM LINK request from server over RoCE fabric */
384 	qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
385 			      SMC_LLC_CONFIRM_LINK);
386 	if (!qentry) {
387 		struct smc_clc_msg_decline dclc;
388 
389 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
390 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
391 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
392 	}
393 	smc_llc_save_peer_uid(qentry);
394 	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
395 	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
396 	if (rc)
397 		return SMC_CLC_DECL_RMBE_EC;
398 
399 	rc = smc_ib_modify_qp_rts(link);
400 	if (rc)
401 		return SMC_CLC_DECL_ERR_RDYLNK;
402 
403 	smc_wr_remember_qp_attr(link);
404 
405 	if (smcr_link_reg_rmb(link, smc->conn.rmb_desc))
406 		return SMC_CLC_DECL_ERR_REGRMB;
407 
408 	/* confirm_rkey is implicit on 1st contact */
409 	smc->conn.rmb_desc->is_conf_rkey = true;
410 
411 	/* send CONFIRM LINK response over RoCE fabric */
412 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
413 	if (rc < 0)
414 		return SMC_CLC_DECL_TIMEOUT_CL;
415 
416 	smc_llc_link_active(link);
417 	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
418 
419 	/* optional 2nd link, receive ADD LINK request from server */
420 	qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
421 			      SMC_LLC_ADD_LINK);
422 	if (!qentry) {
423 		struct smc_clc_msg_decline dclc;
424 
425 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
426 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
427 		if (rc == -EAGAIN)
428 			rc = 0; /* no DECLINE received, go with one link */
429 		return rc;
430 	}
431 	smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
432 	smc_llc_cli_add_link(link, qentry);
433 	return 0;
434 }
435 
436 static void smcr_conn_save_peer_info(struct smc_sock *smc,
437 				     struct smc_clc_msg_accept_confirm *clc)
438 {
439 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
440 
441 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
442 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
443 	smc->conn.peer_rmbe_size = bufsize;
444 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
445 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
446 }
447 
448 static void smcd_conn_save_peer_info(struct smc_sock *smc,
449 				     struct smc_clc_msg_accept_confirm *clc)
450 {
451 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
452 
453 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
454 	smc->conn.peer_token = clc->token;
455 	/* msg header takes up space in the buffer */
456 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
457 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
458 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
459 }
460 
461 static void smc_conn_save_peer_info(struct smc_sock *smc,
462 				    struct smc_clc_msg_accept_confirm *clc)
463 {
464 	if (smc->conn.lgr->is_smcd)
465 		smcd_conn_save_peer_info(smc, clc);
466 	else
467 		smcr_conn_save_peer_info(smc, clc);
468 }
469 
470 static void smc_link_save_peer_info(struct smc_link *link,
471 				    struct smc_clc_msg_accept_confirm *clc)
472 {
473 	link->peer_qpn = ntoh24(clc->qpn);
474 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
475 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
476 	link->peer_psn = ntoh24(clc->psn);
477 	link->peer_mtu = clc->qp_mtu;
478 }
479 
480 static void smc_switch_to_fallback(struct smc_sock *smc)
481 {
482 	smc->use_fallback = true;
483 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
484 		smc->clcsock->file = smc->sk.sk_socket->file;
485 		smc->clcsock->file->private_data = smc->clcsock;
486 		smc->clcsock->wq.fasync_list =
487 			smc->sk.sk_socket->wq.fasync_list;
488 	}
489 }
490 
491 /* fall back during connect */
492 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
493 {
494 	smc_switch_to_fallback(smc);
495 	smc->fallback_rsn = reason_code;
496 	smc_copy_sock_settings_to_clc(smc);
497 	smc->connect_nonblock = 0;
498 	if (smc->sk.sk_state == SMC_INIT)
499 		smc->sk.sk_state = SMC_ACTIVE;
500 	return 0;
501 }
502 
503 /* decline and fall back during connect */
504 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
505 {
506 	int rc;
507 
508 	if (reason_code < 0) { /* error, fallback is not possible */
509 		if (smc->sk.sk_state == SMC_INIT)
510 			sock_put(&smc->sk); /* passive closing */
511 		return reason_code;
512 	}
513 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
514 		rc = smc_clc_send_decline(smc, reason_code);
515 		if (rc < 0) {
516 			if (smc->sk.sk_state == SMC_INIT)
517 				sock_put(&smc->sk); /* passive closing */
518 			return rc;
519 		}
520 	}
521 	return smc_connect_fallback(smc, reason_code);
522 }
523 
524 /* abort connecting */
525 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
526 			     int local_contact)
527 {
528 	bool is_smcd = smc->conn.lgr->is_smcd;
529 
530 	if (local_contact == SMC_FIRST_CONTACT)
531 		smc_lgr_cleanup_early(&smc->conn);
532 	else
533 		smc_conn_free(&smc->conn);
534 	if (is_smcd)
535 		/* there is only one lgr role for SMC-D; use server lock */
536 		mutex_unlock(&smc_server_lgr_pending);
537 	else
538 		mutex_unlock(&smc_client_lgr_pending);
539 
540 	smc->connect_nonblock = 0;
541 	return reason_code;
542 }
543 
544 /* check if there is a rdma device available for this connection. */
545 /* called for connect and listen */
546 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
547 {
548 	/* PNET table look up: search active ib_device and port
549 	 * within same PNETID that also contains the ethernet device
550 	 * used for the internal TCP socket
551 	 */
552 	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
553 	if (!ini->ib_dev)
554 		return SMC_CLC_DECL_NOSMCRDEV;
555 	return 0;
556 }
557 
558 /* check if there is an ISM device available for this connection. */
559 /* called for connect and listen */
560 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
561 {
562 	/* Find ISM device with same PNETID as connecting interface  */
563 	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
564 	if (!ini->ism_dev)
565 		return SMC_CLC_DECL_NOSMCDDEV;
566 	return 0;
567 }
568 
569 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
570 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
571 				      struct smc_init_info *ini)
572 {
573 	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id))
574 		return SMC_CLC_DECL_ISMVLANERR;
575 	return 0;
576 }
577 
578 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
579  * used, the VLAN ID will be registered again during the connection setup.
580  */
581 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
582 					struct smc_init_info *ini)
583 {
584 	if (!is_smcd)
585 		return 0;
586 	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id))
587 		return SMC_CLC_DECL_CNFERR;
588 	return 0;
589 }
590 
591 /* CLC handshake during connect */
592 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
593 			   struct smc_clc_msg_accept_confirm *aclc,
594 			   struct smc_init_info *ini)
595 {
596 	int rc = 0;
597 
598 	/* do inband token exchange */
599 	rc = smc_clc_send_proposal(smc, smc_type, ini);
600 	if (rc)
601 		return rc;
602 	/* receive SMC Accept CLC message */
603 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT,
604 				CLC_WAIT_TIME);
605 }
606 
607 /* setup for RDMA connection of client */
608 static int smc_connect_rdma(struct smc_sock *smc,
609 			    struct smc_clc_msg_accept_confirm *aclc,
610 			    struct smc_init_info *ini)
611 {
612 	int i, reason_code = 0;
613 	struct smc_link *link;
614 
615 	ini->is_smcd = false;
616 	ini->ib_lcl = &aclc->lcl;
617 	ini->ib_clcqpn = ntoh24(aclc->qpn);
618 	ini->srv_first_contact = aclc->hdr.flag;
619 
620 	mutex_lock(&smc_client_lgr_pending);
621 	reason_code = smc_conn_create(smc, ini);
622 	if (reason_code) {
623 		mutex_unlock(&smc_client_lgr_pending);
624 		return reason_code;
625 	}
626 
627 	smc_conn_save_peer_info(smc, aclc);
628 
629 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
630 		link = smc->conn.lnk;
631 	} else {
632 		/* set link that was assigned by server */
633 		link = NULL;
634 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
635 			struct smc_link *l = &smc->conn.lgr->lnk[i];
636 
637 			if (l->peer_qpn == ntoh24(aclc->qpn) &&
638 			    !memcmp(l->peer_gid, &aclc->lcl.gid, SMC_GID_SIZE) &&
639 			    !memcmp(l->peer_mac, &aclc->lcl.mac, sizeof(l->peer_mac))) {
640 				link = l;
641 				break;
642 			}
643 		}
644 		if (!link)
645 			return smc_connect_abort(smc, SMC_CLC_DECL_NOSRVLINK,
646 						 ini->cln_first_contact);
647 		smc->conn.lnk = link;
648 	}
649 
650 	/* create send buffer and rmb */
651 	if (smc_buf_create(smc, false))
652 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
653 					 ini->cln_first_contact);
654 
655 	if (ini->cln_first_contact == SMC_FIRST_CONTACT)
656 		smc_link_save_peer_info(link, aclc);
657 
658 	if (smc_rmb_rtoken_handling(&smc->conn, link, aclc))
659 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
660 					 ini->cln_first_contact);
661 
662 	smc_close_init(smc);
663 	smc_rx_init(smc);
664 
665 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
666 		if (smc_ib_ready_link(link))
667 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
668 						 ini->cln_first_contact);
669 	} else {
670 		if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc))
671 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
672 						 ini->cln_first_contact);
673 	}
674 	smc_rmb_sync_sg_for_device(&smc->conn);
675 
676 	reason_code = smc_clc_send_confirm(smc);
677 	if (reason_code)
678 		return smc_connect_abort(smc, reason_code,
679 					 ini->cln_first_contact);
680 
681 	smc_tx_init(smc);
682 
683 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
684 		/* QP confirmation over RoCE fabric */
685 		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
686 		reason_code = smcr_clnt_conf_first_link(smc);
687 		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
688 		if (reason_code)
689 			return smc_connect_abort(smc, reason_code,
690 						 ini->cln_first_contact);
691 	}
692 	mutex_unlock(&smc_client_lgr_pending);
693 
694 	smc_copy_sock_settings_to_clc(smc);
695 	smc->connect_nonblock = 0;
696 	if (smc->sk.sk_state == SMC_INIT)
697 		smc->sk.sk_state = SMC_ACTIVE;
698 
699 	return 0;
700 }
701 
702 /* setup for ISM connection of client */
703 static int smc_connect_ism(struct smc_sock *smc,
704 			   struct smc_clc_msg_accept_confirm *aclc,
705 			   struct smc_init_info *ini)
706 {
707 	int rc = 0;
708 
709 	ini->is_smcd = true;
710 	ini->ism_gid = aclc->gid;
711 	ini->srv_first_contact = aclc->hdr.flag;
712 
713 	/* there is only one lgr role for SMC-D; use server lock */
714 	mutex_lock(&smc_server_lgr_pending);
715 	rc = smc_conn_create(smc, ini);
716 	if (rc) {
717 		mutex_unlock(&smc_server_lgr_pending);
718 		return rc;
719 	}
720 
721 	/* Create send and receive buffers */
722 	rc = smc_buf_create(smc, true);
723 	if (rc)
724 		return smc_connect_abort(smc, (rc == -ENOSPC) ?
725 					      SMC_CLC_DECL_MAX_DMB :
726 					      SMC_CLC_DECL_MEM,
727 					 ini->cln_first_contact);
728 
729 	smc_conn_save_peer_info(smc, aclc);
730 	smc_close_init(smc);
731 	smc_rx_init(smc);
732 	smc_tx_init(smc);
733 
734 	rc = smc_clc_send_confirm(smc);
735 	if (rc)
736 		return smc_connect_abort(smc, rc, ini->cln_first_contact);
737 	mutex_unlock(&smc_server_lgr_pending);
738 
739 	smc_copy_sock_settings_to_clc(smc);
740 	smc->connect_nonblock = 0;
741 	if (smc->sk.sk_state == SMC_INIT)
742 		smc->sk.sk_state = SMC_ACTIVE;
743 
744 	return 0;
745 }
746 
747 /* perform steps before actually connecting */
748 static int __smc_connect(struct smc_sock *smc)
749 {
750 	bool ism_supported = false, rdma_supported = false;
751 	struct smc_clc_msg_accept_confirm aclc;
752 	struct smc_init_info ini = {0};
753 	int smc_type;
754 	int rc = 0;
755 
756 	if (smc->use_fallback)
757 		return smc_connect_fallback(smc, smc->fallback_rsn);
758 
759 	/* if peer has not signalled SMC-capability, fall back */
760 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
761 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
762 
763 	/* IPSec connections opt out of SMC-R optimizations */
764 	if (using_ipsec(smc))
765 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
766 
767 	/* get vlan id from IP device */
768 	if (smc_vlan_by_tcpsk(smc->clcsock, &ini))
769 		return smc_connect_decline_fallback(smc,
770 						    SMC_CLC_DECL_GETVLANERR);
771 
772 	/* check if there is an ism device available */
773 	if (!smc_find_ism_device(smc, &ini) &&
774 	    !smc_connect_ism_vlan_setup(smc, &ini)) {
775 		/* ISM is supported for this connection */
776 		ism_supported = true;
777 		smc_type = SMC_TYPE_D;
778 	}
779 
780 	/* check if there is a rdma device available */
781 	if (!smc_find_rdma_device(smc, &ini)) {
782 		/* RDMA is supported for this connection */
783 		rdma_supported = true;
784 		if (ism_supported)
785 			smc_type = SMC_TYPE_B; /* both */
786 		else
787 			smc_type = SMC_TYPE_R; /* only RDMA */
788 	}
789 
790 	/* if neither ISM nor RDMA are supported, fallback */
791 	if (!rdma_supported && !ism_supported)
792 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
793 
794 	/* perform CLC handshake */
795 	rc = smc_connect_clc(smc, smc_type, &aclc, &ini);
796 	if (rc) {
797 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
798 		return smc_connect_decline_fallback(smc, rc);
799 	}
800 
801 	/* depending on previous steps, connect using rdma or ism */
802 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
803 		rc = smc_connect_rdma(smc, &aclc, &ini);
804 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
805 		rc = smc_connect_ism(smc, &aclc, &ini);
806 	else
807 		rc = SMC_CLC_DECL_MODEUNSUPP;
808 	if (rc) {
809 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
810 		return smc_connect_decline_fallback(smc, rc);
811 	}
812 
813 	smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
814 	return 0;
815 }
816 
817 static void smc_connect_work(struct work_struct *work)
818 {
819 	struct smc_sock *smc = container_of(work, struct smc_sock,
820 					    connect_work);
821 	long timeo = smc->sk.sk_sndtimeo;
822 	int rc = 0;
823 
824 	if (!timeo)
825 		timeo = MAX_SCHEDULE_TIMEOUT;
826 	lock_sock(smc->clcsock->sk);
827 	if (smc->clcsock->sk->sk_err) {
828 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
829 	} else if ((1 << smc->clcsock->sk->sk_state) &
830 					(TCPF_SYN_SENT | TCP_SYN_RECV)) {
831 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
832 		if ((rc == -EPIPE) &&
833 		    ((1 << smc->clcsock->sk->sk_state) &
834 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
835 			rc = 0;
836 	}
837 	release_sock(smc->clcsock->sk);
838 	lock_sock(&smc->sk);
839 	if (rc != 0 || smc->sk.sk_err) {
840 		smc->sk.sk_state = SMC_CLOSED;
841 		if (rc == -EPIPE || rc == -EAGAIN)
842 			smc->sk.sk_err = EPIPE;
843 		else if (signal_pending(current))
844 			smc->sk.sk_err = -sock_intr_errno(timeo);
845 		sock_put(&smc->sk); /* passive closing */
846 		goto out;
847 	}
848 
849 	rc = __smc_connect(smc);
850 	if (rc < 0)
851 		smc->sk.sk_err = -rc;
852 
853 out:
854 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
855 		if (smc->sk.sk_err) {
856 			smc->sk.sk_state_change(&smc->sk);
857 		} else { /* allow polling before and after fallback decision */
858 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
859 			smc->sk.sk_write_space(&smc->sk);
860 		}
861 	}
862 	release_sock(&smc->sk);
863 }
864 
865 static int smc_connect(struct socket *sock, struct sockaddr *addr,
866 		       int alen, int flags)
867 {
868 	struct sock *sk = sock->sk;
869 	struct smc_sock *smc;
870 	int rc = -EINVAL;
871 
872 	smc = smc_sk(sk);
873 
874 	/* separate smc parameter checking to be safe */
875 	if (alen < sizeof(addr->sa_family))
876 		goto out_err;
877 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
878 		goto out_err;
879 
880 	lock_sock(sk);
881 	switch (sk->sk_state) {
882 	default:
883 		goto out;
884 	case SMC_ACTIVE:
885 		rc = -EISCONN;
886 		goto out;
887 	case SMC_INIT:
888 		rc = 0;
889 		break;
890 	}
891 
892 	smc_copy_sock_settings_to_clc(smc);
893 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
894 	if (smc->connect_nonblock) {
895 		rc = -EALREADY;
896 		goto out;
897 	}
898 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
899 	if (rc && rc != -EINPROGRESS)
900 		goto out;
901 
902 	sock_hold(&smc->sk); /* sock put in passive closing */
903 	if (smc->use_fallback)
904 		goto out;
905 	if (flags & O_NONBLOCK) {
906 		if (schedule_work(&smc->connect_work))
907 			smc->connect_nonblock = 1;
908 		rc = -EINPROGRESS;
909 	} else {
910 		rc = __smc_connect(smc);
911 		if (rc < 0)
912 			goto out;
913 		else
914 			rc = 0; /* success cases including fallback */
915 	}
916 
917 out:
918 	release_sock(sk);
919 out_err:
920 	return rc;
921 }
922 
923 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
924 {
925 	struct socket *new_clcsock = NULL;
926 	struct sock *lsk = &lsmc->sk;
927 	struct sock *new_sk;
928 	int rc = -EINVAL;
929 
930 	release_sock(lsk);
931 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
932 	if (!new_sk) {
933 		rc = -ENOMEM;
934 		lsk->sk_err = ENOMEM;
935 		*new_smc = NULL;
936 		lock_sock(lsk);
937 		goto out;
938 	}
939 	*new_smc = smc_sk(new_sk);
940 
941 	mutex_lock(&lsmc->clcsock_release_lock);
942 	if (lsmc->clcsock)
943 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
944 	mutex_unlock(&lsmc->clcsock_release_lock);
945 	lock_sock(lsk);
946 	if  (rc < 0)
947 		lsk->sk_err = -rc;
948 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
949 		new_sk->sk_prot->unhash(new_sk);
950 		if (new_clcsock)
951 			sock_release(new_clcsock);
952 		new_sk->sk_state = SMC_CLOSED;
953 		sock_set_flag(new_sk, SOCK_DEAD);
954 		sock_put(new_sk); /* final */
955 		*new_smc = NULL;
956 		goto out;
957 	}
958 
959 	(*new_smc)->clcsock = new_clcsock;
960 out:
961 	return rc;
962 }
963 
964 /* add a just created sock to the accept queue of the listen sock as
965  * candidate for a following socket accept call from user space
966  */
967 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
968 {
969 	struct smc_sock *par = smc_sk(parent);
970 
971 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
972 	spin_lock(&par->accept_q_lock);
973 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
974 	spin_unlock(&par->accept_q_lock);
975 	sk_acceptq_added(parent);
976 }
977 
978 /* remove a socket from the accept queue of its parental listening socket */
979 static void smc_accept_unlink(struct sock *sk)
980 {
981 	struct smc_sock *par = smc_sk(sk)->listen_smc;
982 
983 	spin_lock(&par->accept_q_lock);
984 	list_del_init(&smc_sk(sk)->accept_q);
985 	spin_unlock(&par->accept_q_lock);
986 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
987 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
988 }
989 
990 /* remove a sock from the accept queue to bind it to a new socket created
991  * for a socket accept call from user space
992  */
993 struct sock *smc_accept_dequeue(struct sock *parent,
994 				struct socket *new_sock)
995 {
996 	struct smc_sock *isk, *n;
997 	struct sock *new_sk;
998 
999 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1000 		new_sk = (struct sock *)isk;
1001 
1002 		smc_accept_unlink(new_sk);
1003 		if (new_sk->sk_state == SMC_CLOSED) {
1004 			new_sk->sk_prot->unhash(new_sk);
1005 			if (isk->clcsock) {
1006 				sock_release(isk->clcsock);
1007 				isk->clcsock = NULL;
1008 			}
1009 			sock_put(new_sk); /* final */
1010 			continue;
1011 		}
1012 		if (new_sock) {
1013 			sock_graft(new_sk, new_sock);
1014 			if (isk->use_fallback) {
1015 				smc_sk(new_sk)->clcsock->file = new_sock->file;
1016 				isk->clcsock->file->private_data = isk->clcsock;
1017 			}
1018 		}
1019 		return new_sk;
1020 	}
1021 	return NULL;
1022 }
1023 
1024 /* clean up for a created but never accepted sock */
1025 void smc_close_non_accepted(struct sock *sk)
1026 {
1027 	struct smc_sock *smc = smc_sk(sk);
1028 
1029 	sock_hold(sk); /* sock_put below */
1030 	lock_sock(sk);
1031 	if (!sk->sk_lingertime)
1032 		/* wait for peer closing */
1033 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
1034 	__smc_release(smc);
1035 	release_sock(sk);
1036 	sock_put(sk); /* sock_hold above */
1037 	sock_put(sk); /* final sock_put */
1038 }
1039 
1040 static int smcr_serv_conf_first_link(struct smc_sock *smc)
1041 {
1042 	struct smc_link *link = smc->conn.lnk;
1043 	struct smc_llc_qentry *qentry;
1044 	int rc;
1045 
1046 	if (smcr_link_reg_rmb(link, smc->conn.rmb_desc))
1047 		return SMC_CLC_DECL_ERR_REGRMB;
1048 
1049 	/* send CONFIRM LINK request to client over the RoCE fabric */
1050 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1051 	if (rc < 0)
1052 		return SMC_CLC_DECL_TIMEOUT_CL;
1053 
1054 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1055 	qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1056 			      SMC_LLC_CONFIRM_LINK);
1057 	if (!qentry) {
1058 		struct smc_clc_msg_decline dclc;
1059 
1060 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1061 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1062 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1063 	}
1064 	smc_llc_save_peer_uid(qentry);
1065 	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1066 	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1067 	if (rc)
1068 		return SMC_CLC_DECL_RMBE_EC;
1069 
1070 	/* confirm_rkey is implicit on 1st contact */
1071 	smc->conn.rmb_desc->is_conf_rkey = true;
1072 
1073 	smc_llc_link_active(link);
1074 	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1075 
1076 	/* initial contact - try to establish second link */
1077 	smc_llc_srv_add_link(link);
1078 	return 0;
1079 }
1080 
1081 /* listen worker: finish */
1082 static void smc_listen_out(struct smc_sock *new_smc)
1083 {
1084 	struct smc_sock *lsmc = new_smc->listen_smc;
1085 	struct sock *newsmcsk = &new_smc->sk;
1086 
1087 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1088 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1089 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1090 		release_sock(&lsmc->sk);
1091 	} else { /* no longer listening */
1092 		smc_close_non_accepted(newsmcsk);
1093 	}
1094 
1095 	/* Wake up accept */
1096 	lsmc->sk.sk_data_ready(&lsmc->sk);
1097 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1098 }
1099 
1100 /* listen worker: finish in state connected */
1101 static void smc_listen_out_connected(struct smc_sock *new_smc)
1102 {
1103 	struct sock *newsmcsk = &new_smc->sk;
1104 
1105 	sk_refcnt_debug_inc(newsmcsk);
1106 	if (newsmcsk->sk_state == SMC_INIT)
1107 		newsmcsk->sk_state = SMC_ACTIVE;
1108 
1109 	smc_listen_out(new_smc);
1110 }
1111 
1112 /* listen worker: finish in error state */
1113 static void smc_listen_out_err(struct smc_sock *new_smc)
1114 {
1115 	struct sock *newsmcsk = &new_smc->sk;
1116 
1117 	if (newsmcsk->sk_state == SMC_INIT)
1118 		sock_put(&new_smc->sk); /* passive closing */
1119 	newsmcsk->sk_state = SMC_CLOSED;
1120 
1121 	smc_listen_out(new_smc);
1122 }
1123 
1124 /* listen worker: decline and fall back if possible */
1125 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1126 			       int local_contact)
1127 {
1128 	/* RDMA setup failed, switch back to TCP */
1129 	if (local_contact == SMC_FIRST_CONTACT)
1130 		smc_lgr_cleanup_early(&new_smc->conn);
1131 	else
1132 		smc_conn_free(&new_smc->conn);
1133 	if (reason_code < 0) { /* error, no fallback possible */
1134 		smc_listen_out_err(new_smc);
1135 		return;
1136 	}
1137 	smc_switch_to_fallback(new_smc);
1138 	new_smc->fallback_rsn = reason_code;
1139 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1140 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1141 			smc_listen_out_err(new_smc);
1142 			return;
1143 		}
1144 	}
1145 	smc_listen_out_connected(new_smc);
1146 }
1147 
1148 /* listen worker: check prefixes */
1149 static int smc_listen_prfx_check(struct smc_sock *new_smc,
1150 				 struct smc_clc_msg_proposal *pclc)
1151 {
1152 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1153 	struct socket *newclcsock = new_smc->clcsock;
1154 
1155 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1156 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1157 		return SMC_CLC_DECL_DIFFPREFIX;
1158 
1159 	return 0;
1160 }
1161 
1162 /* listen worker: initialize connection and buffers */
1163 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1164 				struct smc_init_info *ini)
1165 {
1166 	int rc;
1167 
1168 	/* allocate connection / link group */
1169 	rc = smc_conn_create(new_smc, ini);
1170 	if (rc)
1171 		return rc;
1172 
1173 	/* create send buffer and rmb */
1174 	if (smc_buf_create(new_smc, false))
1175 		return SMC_CLC_DECL_MEM;
1176 
1177 	return 0;
1178 }
1179 
1180 /* listen worker: initialize connection and buffers for SMC-D */
1181 static int smc_listen_ism_init(struct smc_sock *new_smc,
1182 			       struct smc_clc_msg_proposal *pclc,
1183 			       struct smc_init_info *ini)
1184 {
1185 	struct smc_clc_msg_smcd *pclc_smcd;
1186 	int rc;
1187 
1188 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1189 	ini->ism_gid = pclc_smcd->gid;
1190 	rc = smc_conn_create(new_smc, ini);
1191 	if (rc)
1192 		return rc;
1193 
1194 	/* Check if peer can be reached via ISM device */
1195 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1196 			    new_smc->conn.lgr->vlan_id,
1197 			    new_smc->conn.lgr->smcd)) {
1198 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1199 			smc_lgr_cleanup_early(&new_smc->conn);
1200 		else
1201 			smc_conn_free(&new_smc->conn);
1202 		return SMC_CLC_DECL_SMCDNOTALK;
1203 	}
1204 
1205 	/* Create send and receive buffers */
1206 	rc = smc_buf_create(new_smc, true);
1207 	if (rc) {
1208 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1209 			smc_lgr_cleanup_early(&new_smc->conn);
1210 		else
1211 			smc_conn_free(&new_smc->conn);
1212 		return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
1213 					 SMC_CLC_DECL_MEM;
1214 	}
1215 
1216 	return 0;
1217 }
1218 
1219 /* listen worker: register buffers */
1220 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1221 {
1222 	struct smc_connection *conn = &new_smc->conn;
1223 
1224 	if (local_contact != SMC_FIRST_CONTACT) {
1225 		if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
1226 			return SMC_CLC_DECL_ERR_REGRMB;
1227 	}
1228 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1229 
1230 	return 0;
1231 }
1232 
1233 /* listen worker: finish RDMA setup */
1234 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1235 				  struct smc_clc_msg_accept_confirm *cclc,
1236 				  int local_contact)
1237 {
1238 	struct smc_link *link = new_smc->conn.lnk;
1239 	int reason_code = 0;
1240 
1241 	if (local_contact == SMC_FIRST_CONTACT)
1242 		smc_link_save_peer_info(link, cclc);
1243 
1244 	if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) {
1245 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1246 		goto decline;
1247 	}
1248 
1249 	if (local_contact == SMC_FIRST_CONTACT) {
1250 		if (smc_ib_ready_link(link)) {
1251 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1252 			goto decline;
1253 		}
1254 		/* QP confirmation over RoCE fabric */
1255 		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1256 		reason_code = smcr_serv_conf_first_link(new_smc);
1257 		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1258 		if (reason_code)
1259 			goto decline;
1260 	}
1261 	return 0;
1262 
1263 decline:
1264 	smc_listen_decline(new_smc, reason_code, local_contact);
1265 	return reason_code;
1266 }
1267 
1268 /* setup for RDMA connection of server */
1269 static void smc_listen_work(struct work_struct *work)
1270 {
1271 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1272 						smc_listen_work);
1273 	struct socket *newclcsock = new_smc->clcsock;
1274 	struct smc_clc_msg_accept_confirm cclc;
1275 	struct smc_clc_msg_proposal *pclc;
1276 	struct smc_init_info ini = {0};
1277 	bool ism_supported = false;
1278 	u8 buf[SMC_CLC_MAX_LEN];
1279 	int rc = 0;
1280 
1281 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1282 		return smc_listen_out_err(new_smc);
1283 
1284 	if (new_smc->use_fallback) {
1285 		smc_listen_out_connected(new_smc);
1286 		return;
1287 	}
1288 
1289 	/* check if peer is smc capable */
1290 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1291 		smc_switch_to_fallback(new_smc);
1292 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1293 		smc_listen_out_connected(new_smc);
1294 		return;
1295 	}
1296 
1297 	/* do inband token exchange -
1298 	 * wait for and receive SMC Proposal CLC message
1299 	 */
1300 	pclc = (struct smc_clc_msg_proposal *)&buf;
1301 	rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1302 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1303 	if (rc)
1304 		goto out_decl;
1305 
1306 	/* IPSec connections opt out of SMC-R optimizations */
1307 	if (using_ipsec(new_smc)) {
1308 		rc = SMC_CLC_DECL_IPSEC;
1309 		goto out_decl;
1310 	}
1311 
1312 	/* check for matching IP prefix and subnet length */
1313 	rc = smc_listen_prfx_check(new_smc, pclc);
1314 	if (rc)
1315 		goto out_decl;
1316 
1317 	/* get vlan id from IP device */
1318 	if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) {
1319 		rc = SMC_CLC_DECL_GETVLANERR;
1320 		goto out_decl;
1321 	}
1322 
1323 	mutex_lock(&smc_server_lgr_pending);
1324 	smc_close_init(new_smc);
1325 	smc_rx_init(new_smc);
1326 	smc_tx_init(new_smc);
1327 
1328 	/* check if ISM is available */
1329 	if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) {
1330 		ini.is_smcd = true; /* prepare ISM check */
1331 		rc = smc_find_ism_device(new_smc, &ini);
1332 		if (!rc)
1333 			rc = smc_listen_ism_init(new_smc, pclc, &ini);
1334 		if (!rc)
1335 			ism_supported = true;
1336 		else if (pclc->hdr.path == SMC_TYPE_D)
1337 			goto out_unlock; /* skip RDMA and decline */
1338 	}
1339 
1340 	/* check if RDMA is available */
1341 	if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */
1342 		/* prepare RDMA check */
1343 		ini.is_smcd = false;
1344 		ini.ism_dev = NULL;
1345 		ini.ib_lcl = &pclc->lcl;
1346 		rc = smc_find_rdma_device(new_smc, &ini);
1347 		if (rc) {
1348 			/* no RDMA device found */
1349 			if (pclc->hdr.path == SMC_TYPE_B)
1350 				/* neither ISM nor RDMA device found */
1351 				rc = SMC_CLC_DECL_NOSMCDEV;
1352 			goto out_unlock;
1353 		}
1354 		rc = smc_listen_rdma_init(new_smc, &ini);
1355 		if (rc)
1356 			goto out_unlock;
1357 		rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact);
1358 		if (rc)
1359 			goto out_unlock;
1360 	}
1361 
1362 	/* send SMC Accept CLC message */
1363 	rc = smc_clc_send_accept(new_smc, ini.cln_first_contact);
1364 	if (rc)
1365 		goto out_unlock;
1366 
1367 	/* SMC-D does not need this lock any more */
1368 	if (ism_supported)
1369 		mutex_unlock(&smc_server_lgr_pending);
1370 
1371 	/* receive SMC Confirm CLC message */
1372 	rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1373 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1374 	if (rc) {
1375 		if (!ism_supported)
1376 			goto out_unlock;
1377 		goto out_decl;
1378 	}
1379 
1380 	/* finish worker */
1381 	if (!ism_supported) {
1382 		rc = smc_listen_rdma_finish(new_smc, &cclc,
1383 					    ini.cln_first_contact);
1384 		mutex_unlock(&smc_server_lgr_pending);
1385 		if (rc)
1386 			return;
1387 	}
1388 	smc_conn_save_peer_info(new_smc, &cclc);
1389 	smc_listen_out_connected(new_smc);
1390 	return;
1391 
1392 out_unlock:
1393 	mutex_unlock(&smc_server_lgr_pending);
1394 out_decl:
1395 	smc_listen_decline(new_smc, rc, ini.cln_first_contact);
1396 }
1397 
1398 static void smc_tcp_listen_work(struct work_struct *work)
1399 {
1400 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1401 					     tcp_listen_work);
1402 	struct sock *lsk = &lsmc->sk;
1403 	struct smc_sock *new_smc;
1404 	int rc = 0;
1405 
1406 	lock_sock(lsk);
1407 	while (lsk->sk_state == SMC_LISTEN) {
1408 		rc = smc_clcsock_accept(lsmc, &new_smc);
1409 		if (rc)
1410 			goto out;
1411 		if (!new_smc)
1412 			continue;
1413 
1414 		new_smc->listen_smc = lsmc;
1415 		new_smc->use_fallback = lsmc->use_fallback;
1416 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1417 		sock_hold(lsk); /* sock_put in smc_listen_work */
1418 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1419 		smc_copy_sock_settings_to_smc(new_smc);
1420 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1421 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1422 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1423 		if (!schedule_work(&new_smc->smc_listen_work))
1424 			sock_put(&new_smc->sk);
1425 	}
1426 
1427 out:
1428 	release_sock(lsk);
1429 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1430 }
1431 
1432 static int smc_listen(struct socket *sock, int backlog)
1433 {
1434 	struct sock *sk = sock->sk;
1435 	struct smc_sock *smc;
1436 	int rc;
1437 
1438 	smc = smc_sk(sk);
1439 	lock_sock(sk);
1440 
1441 	rc = -EINVAL;
1442 	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
1443 	    smc->connect_nonblock)
1444 		goto out;
1445 
1446 	rc = 0;
1447 	if (sk->sk_state == SMC_LISTEN) {
1448 		sk->sk_max_ack_backlog = backlog;
1449 		goto out;
1450 	}
1451 	/* some socket options are handled in core, so we could not apply
1452 	 * them to the clc socket -- copy smc socket options to clc socket
1453 	 */
1454 	smc_copy_sock_settings_to_clc(smc);
1455 	if (!smc->use_fallback)
1456 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1457 
1458 	rc = kernel_listen(smc->clcsock, backlog);
1459 	if (rc)
1460 		goto out;
1461 	sk->sk_max_ack_backlog = backlog;
1462 	sk->sk_ack_backlog = 0;
1463 	sk->sk_state = SMC_LISTEN;
1464 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1465 	if (!schedule_work(&smc->tcp_listen_work))
1466 		sock_put(sk);
1467 
1468 out:
1469 	release_sock(sk);
1470 	return rc;
1471 }
1472 
1473 static int smc_accept(struct socket *sock, struct socket *new_sock,
1474 		      int flags, bool kern)
1475 {
1476 	struct sock *sk = sock->sk, *nsk;
1477 	DECLARE_WAITQUEUE(wait, current);
1478 	struct smc_sock *lsmc;
1479 	long timeo;
1480 	int rc = 0;
1481 
1482 	lsmc = smc_sk(sk);
1483 	sock_hold(sk); /* sock_put below */
1484 	lock_sock(sk);
1485 
1486 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1487 		rc = -EINVAL;
1488 		release_sock(sk);
1489 		goto out;
1490 	}
1491 
1492 	/* Wait for an incoming connection */
1493 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1494 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1495 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1496 		set_current_state(TASK_INTERRUPTIBLE);
1497 		if (!timeo) {
1498 			rc = -EAGAIN;
1499 			break;
1500 		}
1501 		release_sock(sk);
1502 		timeo = schedule_timeout(timeo);
1503 		/* wakeup by sk_data_ready in smc_listen_work() */
1504 		sched_annotate_sleep();
1505 		lock_sock(sk);
1506 		if (signal_pending(current)) {
1507 			rc = sock_intr_errno(timeo);
1508 			break;
1509 		}
1510 	}
1511 	set_current_state(TASK_RUNNING);
1512 	remove_wait_queue(sk_sleep(sk), &wait);
1513 
1514 	if (!rc)
1515 		rc = sock_error(nsk);
1516 	release_sock(sk);
1517 	if (rc)
1518 		goto out;
1519 
1520 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1521 		/* wait till data arrives on the socket */
1522 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1523 								MSEC_PER_SEC);
1524 		if (smc_sk(nsk)->use_fallback) {
1525 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1526 
1527 			lock_sock(clcsk);
1528 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1529 				sk_wait_data(clcsk, &timeo, NULL);
1530 			release_sock(clcsk);
1531 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1532 			lock_sock(nsk);
1533 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1534 			release_sock(nsk);
1535 		}
1536 	}
1537 
1538 out:
1539 	sock_put(sk); /* sock_hold above */
1540 	return rc;
1541 }
1542 
1543 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1544 		       int peer)
1545 {
1546 	struct smc_sock *smc;
1547 
1548 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1549 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1550 		return -ENOTCONN;
1551 
1552 	smc = smc_sk(sock->sk);
1553 
1554 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1555 }
1556 
1557 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1558 {
1559 	struct sock *sk = sock->sk;
1560 	struct smc_sock *smc;
1561 	int rc = -EPIPE;
1562 
1563 	smc = smc_sk(sk);
1564 	lock_sock(sk);
1565 	if ((sk->sk_state != SMC_ACTIVE) &&
1566 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1567 	    (sk->sk_state != SMC_INIT))
1568 		goto out;
1569 
1570 	if (msg->msg_flags & MSG_FASTOPEN) {
1571 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1572 			smc_switch_to_fallback(smc);
1573 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1574 		} else {
1575 			rc = -EINVAL;
1576 			goto out;
1577 		}
1578 	}
1579 
1580 	if (smc->use_fallback)
1581 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1582 	else
1583 		rc = smc_tx_sendmsg(smc, msg, len);
1584 out:
1585 	release_sock(sk);
1586 	return rc;
1587 }
1588 
1589 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1590 		       int flags)
1591 {
1592 	struct sock *sk = sock->sk;
1593 	struct smc_sock *smc;
1594 	int rc = -ENOTCONN;
1595 
1596 	smc = smc_sk(sk);
1597 	lock_sock(sk);
1598 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1599 		/* socket was connected before, no more data to read */
1600 		rc = 0;
1601 		goto out;
1602 	}
1603 	if ((sk->sk_state == SMC_INIT) ||
1604 	    (sk->sk_state == SMC_LISTEN) ||
1605 	    (sk->sk_state == SMC_CLOSED))
1606 		goto out;
1607 
1608 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1609 		rc = 0;
1610 		goto out;
1611 	}
1612 
1613 	if (smc->use_fallback) {
1614 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1615 	} else {
1616 		msg->msg_namelen = 0;
1617 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1618 	}
1619 
1620 out:
1621 	release_sock(sk);
1622 	return rc;
1623 }
1624 
1625 static __poll_t smc_accept_poll(struct sock *parent)
1626 {
1627 	struct smc_sock *isk = smc_sk(parent);
1628 	__poll_t mask = 0;
1629 
1630 	spin_lock(&isk->accept_q_lock);
1631 	if (!list_empty(&isk->accept_q))
1632 		mask = EPOLLIN | EPOLLRDNORM;
1633 	spin_unlock(&isk->accept_q_lock);
1634 
1635 	return mask;
1636 }
1637 
1638 static __poll_t smc_poll(struct file *file, struct socket *sock,
1639 			     poll_table *wait)
1640 {
1641 	struct sock *sk = sock->sk;
1642 	struct smc_sock *smc;
1643 	__poll_t mask = 0;
1644 
1645 	if (!sk)
1646 		return EPOLLNVAL;
1647 
1648 	smc = smc_sk(sock->sk);
1649 	if (smc->use_fallback) {
1650 		/* delegate to CLC child sock */
1651 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1652 		sk->sk_err = smc->clcsock->sk->sk_err;
1653 	} else {
1654 		if (sk->sk_state != SMC_CLOSED)
1655 			sock_poll_wait(file, sock, wait);
1656 		if (sk->sk_err)
1657 			mask |= EPOLLERR;
1658 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1659 		    (sk->sk_state == SMC_CLOSED))
1660 			mask |= EPOLLHUP;
1661 		if (sk->sk_state == SMC_LISTEN) {
1662 			/* woken up by sk_data_ready in smc_listen_work() */
1663 			mask |= smc_accept_poll(sk);
1664 		} else if (smc->use_fallback) { /* as result of connect_work()*/
1665 			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
1666 							   wait);
1667 			sk->sk_err = smc->clcsock->sk->sk_err;
1668 		} else {
1669 			if ((sk->sk_state != SMC_INIT &&
1670 			     atomic_read(&smc->conn.sndbuf_space)) ||
1671 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1672 				mask |= EPOLLOUT | EPOLLWRNORM;
1673 			} else {
1674 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1675 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1676 			}
1677 			if (atomic_read(&smc->conn.bytes_to_rcv))
1678 				mask |= EPOLLIN | EPOLLRDNORM;
1679 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1680 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1681 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1682 				mask |= EPOLLIN;
1683 			if (smc->conn.urg_state == SMC_URG_VALID)
1684 				mask |= EPOLLPRI;
1685 		}
1686 	}
1687 
1688 	return mask;
1689 }
1690 
1691 static int smc_shutdown(struct socket *sock, int how)
1692 {
1693 	struct sock *sk = sock->sk;
1694 	struct smc_sock *smc;
1695 	int rc = -EINVAL;
1696 	int rc1 = 0;
1697 
1698 	smc = smc_sk(sk);
1699 
1700 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1701 		return rc;
1702 
1703 	lock_sock(sk);
1704 
1705 	rc = -ENOTCONN;
1706 	if ((sk->sk_state != SMC_ACTIVE) &&
1707 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1708 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1709 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1710 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1711 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1712 		goto out;
1713 	if (smc->use_fallback) {
1714 		rc = kernel_sock_shutdown(smc->clcsock, how);
1715 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1716 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1717 			sk->sk_state = SMC_CLOSED;
1718 		goto out;
1719 	}
1720 	switch (how) {
1721 	case SHUT_RDWR:		/* shutdown in both directions */
1722 		rc = smc_close_active(smc);
1723 		break;
1724 	case SHUT_WR:
1725 		rc = smc_close_shutdown_write(smc);
1726 		break;
1727 	case SHUT_RD:
1728 		rc = 0;
1729 		/* nothing more to do because peer is not involved */
1730 		break;
1731 	}
1732 	if (smc->clcsock)
1733 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1734 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1735 	sk->sk_shutdown |= how + 1;
1736 
1737 out:
1738 	release_sock(sk);
1739 	return rc ? rc : rc1;
1740 }
1741 
1742 static int smc_setsockopt(struct socket *sock, int level, int optname,
1743 			  sockptr_t optval, unsigned int optlen)
1744 {
1745 	struct sock *sk = sock->sk;
1746 	struct smc_sock *smc;
1747 	int val, rc;
1748 
1749 	smc = smc_sk(sk);
1750 
1751 	/* generic setsockopts reaching us here always apply to the
1752 	 * CLC socket
1753 	 */
1754 	if (unlikely(!smc->clcsock->ops->setsockopt))
1755 		rc = -EOPNOTSUPP;
1756 	else
1757 		rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1758 						   optval, optlen);
1759 	if (smc->clcsock->sk->sk_err) {
1760 		sk->sk_err = smc->clcsock->sk->sk_err;
1761 		sk->sk_error_report(sk);
1762 	}
1763 
1764 	if (optlen < sizeof(int))
1765 		return -EINVAL;
1766 	if (copy_from_sockptr(&val, optval, sizeof(int)))
1767 		return -EFAULT;
1768 
1769 	lock_sock(sk);
1770 	if (rc || smc->use_fallback)
1771 		goto out;
1772 	switch (optname) {
1773 	case TCP_ULP:
1774 	case TCP_FASTOPEN:
1775 	case TCP_FASTOPEN_CONNECT:
1776 	case TCP_FASTOPEN_KEY:
1777 	case TCP_FASTOPEN_NO_COOKIE:
1778 		/* option not supported by SMC */
1779 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1780 			smc_switch_to_fallback(smc);
1781 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1782 		} else {
1783 			rc = -EINVAL;
1784 		}
1785 		break;
1786 	case TCP_NODELAY:
1787 		if (sk->sk_state != SMC_INIT &&
1788 		    sk->sk_state != SMC_LISTEN &&
1789 		    sk->sk_state != SMC_CLOSED) {
1790 			if (val)
1791 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1792 						 0);
1793 		}
1794 		break;
1795 	case TCP_CORK:
1796 		if (sk->sk_state != SMC_INIT &&
1797 		    sk->sk_state != SMC_LISTEN &&
1798 		    sk->sk_state != SMC_CLOSED) {
1799 			if (!val)
1800 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1801 						 0);
1802 		}
1803 		break;
1804 	case TCP_DEFER_ACCEPT:
1805 		smc->sockopt_defer_accept = val;
1806 		break;
1807 	default:
1808 		break;
1809 	}
1810 out:
1811 	release_sock(sk);
1812 
1813 	return rc;
1814 }
1815 
1816 static int smc_getsockopt(struct socket *sock, int level, int optname,
1817 			  char __user *optval, int __user *optlen)
1818 {
1819 	struct smc_sock *smc;
1820 
1821 	smc = smc_sk(sock->sk);
1822 	/* socket options apply to the CLC socket */
1823 	if (unlikely(!smc->clcsock->ops->getsockopt))
1824 		return -EOPNOTSUPP;
1825 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1826 					     optval, optlen);
1827 }
1828 
1829 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1830 		     unsigned long arg)
1831 {
1832 	union smc_host_cursor cons, urg;
1833 	struct smc_connection *conn;
1834 	struct smc_sock *smc;
1835 	int answ;
1836 
1837 	smc = smc_sk(sock->sk);
1838 	conn = &smc->conn;
1839 	lock_sock(&smc->sk);
1840 	if (smc->use_fallback) {
1841 		if (!smc->clcsock) {
1842 			release_sock(&smc->sk);
1843 			return -EBADF;
1844 		}
1845 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1846 		release_sock(&smc->sk);
1847 		return answ;
1848 	}
1849 	switch (cmd) {
1850 	case SIOCINQ: /* same as FIONREAD */
1851 		if (smc->sk.sk_state == SMC_LISTEN) {
1852 			release_sock(&smc->sk);
1853 			return -EINVAL;
1854 		}
1855 		if (smc->sk.sk_state == SMC_INIT ||
1856 		    smc->sk.sk_state == SMC_CLOSED)
1857 			answ = 0;
1858 		else
1859 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1860 		break;
1861 	case SIOCOUTQ:
1862 		/* output queue size (not send + not acked) */
1863 		if (smc->sk.sk_state == SMC_LISTEN) {
1864 			release_sock(&smc->sk);
1865 			return -EINVAL;
1866 		}
1867 		if (smc->sk.sk_state == SMC_INIT ||
1868 		    smc->sk.sk_state == SMC_CLOSED)
1869 			answ = 0;
1870 		else
1871 			answ = smc->conn.sndbuf_desc->len -
1872 					atomic_read(&smc->conn.sndbuf_space);
1873 		break;
1874 	case SIOCOUTQNSD:
1875 		/* output queue size (not send only) */
1876 		if (smc->sk.sk_state == SMC_LISTEN) {
1877 			release_sock(&smc->sk);
1878 			return -EINVAL;
1879 		}
1880 		if (smc->sk.sk_state == SMC_INIT ||
1881 		    smc->sk.sk_state == SMC_CLOSED)
1882 			answ = 0;
1883 		else
1884 			answ = smc_tx_prepared_sends(&smc->conn);
1885 		break;
1886 	case SIOCATMARK:
1887 		if (smc->sk.sk_state == SMC_LISTEN) {
1888 			release_sock(&smc->sk);
1889 			return -EINVAL;
1890 		}
1891 		if (smc->sk.sk_state == SMC_INIT ||
1892 		    smc->sk.sk_state == SMC_CLOSED) {
1893 			answ = 0;
1894 		} else {
1895 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1896 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1897 			answ = smc_curs_diff(conn->rmb_desc->len,
1898 					     &cons, &urg) == 1;
1899 		}
1900 		break;
1901 	default:
1902 		release_sock(&smc->sk);
1903 		return -ENOIOCTLCMD;
1904 	}
1905 	release_sock(&smc->sk);
1906 
1907 	return put_user(answ, (int __user *)arg);
1908 }
1909 
1910 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1911 			    int offset, size_t size, int flags)
1912 {
1913 	struct sock *sk = sock->sk;
1914 	struct smc_sock *smc;
1915 	int rc = -EPIPE;
1916 
1917 	smc = smc_sk(sk);
1918 	lock_sock(sk);
1919 	if (sk->sk_state != SMC_ACTIVE) {
1920 		release_sock(sk);
1921 		goto out;
1922 	}
1923 	release_sock(sk);
1924 	if (smc->use_fallback)
1925 		rc = kernel_sendpage(smc->clcsock, page, offset,
1926 				     size, flags);
1927 	else
1928 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1929 
1930 out:
1931 	return rc;
1932 }
1933 
1934 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1935  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1936  * updates till whenever a respective page has been fully processed.
1937  * Note that subsequent recv() calls have to wait till all splice() processing
1938  * completed.
1939  */
1940 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1941 			       struct pipe_inode_info *pipe, size_t len,
1942 			       unsigned int flags)
1943 {
1944 	struct sock *sk = sock->sk;
1945 	struct smc_sock *smc;
1946 	int rc = -ENOTCONN;
1947 
1948 	smc = smc_sk(sk);
1949 	lock_sock(sk);
1950 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1951 		/* socket was connected before, no more data to read */
1952 		rc = 0;
1953 		goto out;
1954 	}
1955 	if (sk->sk_state == SMC_INIT ||
1956 	    sk->sk_state == SMC_LISTEN ||
1957 	    sk->sk_state == SMC_CLOSED)
1958 		goto out;
1959 
1960 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1961 		rc = 0;
1962 		goto out;
1963 	}
1964 
1965 	if (smc->use_fallback) {
1966 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1967 						    pipe, len, flags);
1968 	} else {
1969 		if (*ppos) {
1970 			rc = -ESPIPE;
1971 			goto out;
1972 		}
1973 		if (flags & SPLICE_F_NONBLOCK)
1974 			flags = MSG_DONTWAIT;
1975 		else
1976 			flags = 0;
1977 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1978 	}
1979 out:
1980 	release_sock(sk);
1981 
1982 	return rc;
1983 }
1984 
1985 /* must look like tcp */
1986 static const struct proto_ops smc_sock_ops = {
1987 	.family		= PF_SMC,
1988 	.owner		= THIS_MODULE,
1989 	.release	= smc_release,
1990 	.bind		= smc_bind,
1991 	.connect	= smc_connect,
1992 	.socketpair	= sock_no_socketpair,
1993 	.accept		= smc_accept,
1994 	.getname	= smc_getname,
1995 	.poll		= smc_poll,
1996 	.ioctl		= smc_ioctl,
1997 	.listen		= smc_listen,
1998 	.shutdown	= smc_shutdown,
1999 	.setsockopt	= smc_setsockopt,
2000 	.getsockopt	= smc_getsockopt,
2001 	.sendmsg	= smc_sendmsg,
2002 	.recvmsg	= smc_recvmsg,
2003 	.mmap		= sock_no_mmap,
2004 	.sendpage	= smc_sendpage,
2005 	.splice_read	= smc_splice_read,
2006 };
2007 
2008 static int smc_create(struct net *net, struct socket *sock, int protocol,
2009 		      int kern)
2010 {
2011 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
2012 	struct smc_sock *smc;
2013 	struct sock *sk;
2014 	int rc;
2015 
2016 	rc = -ESOCKTNOSUPPORT;
2017 	if (sock->type != SOCK_STREAM)
2018 		goto out;
2019 
2020 	rc = -EPROTONOSUPPORT;
2021 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
2022 		goto out;
2023 
2024 	rc = -ENOBUFS;
2025 	sock->ops = &smc_sock_ops;
2026 	sk = smc_sock_alloc(net, sock, protocol);
2027 	if (!sk)
2028 		goto out;
2029 
2030 	/* create internal TCP socket for CLC handshake and fallback */
2031 	smc = smc_sk(sk);
2032 	smc->use_fallback = false; /* assume rdma capability first */
2033 	smc->fallback_rsn = 0;
2034 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
2035 			      &smc->clcsock);
2036 	if (rc) {
2037 		sk_common_release(sk);
2038 		goto out;
2039 	}
2040 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
2041 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
2042 
2043 out:
2044 	return rc;
2045 }
2046 
2047 static const struct net_proto_family smc_sock_family_ops = {
2048 	.family	= PF_SMC,
2049 	.owner	= THIS_MODULE,
2050 	.create	= smc_create,
2051 };
2052 
2053 unsigned int smc_net_id;
2054 
2055 static __net_init int smc_net_init(struct net *net)
2056 {
2057 	return smc_pnet_net_init(net);
2058 }
2059 
2060 static void __net_exit smc_net_exit(struct net *net)
2061 {
2062 	smc_pnet_net_exit(net);
2063 }
2064 
2065 static struct pernet_operations smc_net_ops = {
2066 	.init = smc_net_init,
2067 	.exit = smc_net_exit,
2068 	.id   = &smc_net_id,
2069 	.size = sizeof(struct smc_net),
2070 };
2071 
2072 static int __init smc_init(void)
2073 {
2074 	int rc;
2075 
2076 	rc = register_pernet_subsys(&smc_net_ops);
2077 	if (rc)
2078 		return rc;
2079 
2080 	rc = smc_pnet_init();
2081 	if (rc)
2082 		goto out_pernet_subsys;
2083 
2084 	rc = smc_core_init();
2085 	if (rc) {
2086 		pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
2087 		goto out_pnet;
2088 	}
2089 
2090 	rc = smc_llc_init();
2091 	if (rc) {
2092 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2093 		goto out_core;
2094 	}
2095 
2096 	rc = smc_cdc_init();
2097 	if (rc) {
2098 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2099 		goto out_core;
2100 	}
2101 
2102 	rc = proto_register(&smc_proto, 1);
2103 	if (rc) {
2104 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2105 		goto out_core;
2106 	}
2107 
2108 	rc = proto_register(&smc_proto6, 1);
2109 	if (rc) {
2110 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2111 		goto out_proto;
2112 	}
2113 
2114 	rc = sock_register(&smc_sock_family_ops);
2115 	if (rc) {
2116 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2117 		goto out_proto6;
2118 	}
2119 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2120 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2121 
2122 	rc = smc_ib_register_client();
2123 	if (rc) {
2124 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2125 		goto out_sock;
2126 	}
2127 
2128 	static_branch_enable(&tcp_have_smc);
2129 	return 0;
2130 
2131 out_sock:
2132 	sock_unregister(PF_SMC);
2133 out_proto6:
2134 	proto_unregister(&smc_proto6);
2135 out_proto:
2136 	proto_unregister(&smc_proto);
2137 out_core:
2138 	smc_core_exit();
2139 out_pnet:
2140 	smc_pnet_exit();
2141 out_pernet_subsys:
2142 	unregister_pernet_subsys(&smc_net_ops);
2143 
2144 	return rc;
2145 }
2146 
2147 static void __exit smc_exit(void)
2148 {
2149 	static_branch_disable(&tcp_have_smc);
2150 	sock_unregister(PF_SMC);
2151 	smc_core_exit();
2152 	smc_ib_unregister_client();
2153 	proto_unregister(&smc_proto6);
2154 	proto_unregister(&smc_proto);
2155 	smc_pnet_exit();
2156 	unregister_pernet_subsys(&smc_net_ops);
2157 	rcu_barrier();
2158 }
2159 
2160 module_init(smc_init);
2161 module_exit(smc_exit);
2162 
2163 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2164 MODULE_DESCRIPTION("smc socket address family");
2165 MODULE_LICENSE("GPL");
2166 MODULE_ALIAS_NETPROTO(PF_SMC);
2167