xref: /openbmc/linux/net/smc/af_smc.c (revision 4c5a116a)
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 	if (smc_buf_create(smc, true))
723 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
724 					 ini->cln_first_contact);
725 
726 	smc_conn_save_peer_info(smc, aclc);
727 	smc_close_init(smc);
728 	smc_rx_init(smc);
729 	smc_tx_init(smc);
730 
731 	rc = smc_clc_send_confirm(smc);
732 	if (rc)
733 		return smc_connect_abort(smc, rc, ini->cln_first_contact);
734 	mutex_unlock(&smc_server_lgr_pending);
735 
736 	smc_copy_sock_settings_to_clc(smc);
737 	smc->connect_nonblock = 0;
738 	if (smc->sk.sk_state == SMC_INIT)
739 		smc->sk.sk_state = SMC_ACTIVE;
740 
741 	return 0;
742 }
743 
744 /* perform steps before actually connecting */
745 static int __smc_connect(struct smc_sock *smc)
746 {
747 	bool ism_supported = false, rdma_supported = false;
748 	struct smc_clc_msg_accept_confirm aclc;
749 	struct smc_init_info ini = {0};
750 	int smc_type;
751 	int rc = 0;
752 
753 	if (smc->use_fallback)
754 		return smc_connect_fallback(smc, smc->fallback_rsn);
755 
756 	/* if peer has not signalled SMC-capability, fall back */
757 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
758 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
759 
760 	/* IPSec connections opt out of SMC-R optimizations */
761 	if (using_ipsec(smc))
762 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
763 
764 	/* get vlan id from IP device */
765 	if (smc_vlan_by_tcpsk(smc->clcsock, &ini))
766 		return smc_connect_decline_fallback(smc,
767 						    SMC_CLC_DECL_GETVLANERR);
768 
769 	/* check if there is an ism device available */
770 	if (!smc_find_ism_device(smc, &ini) &&
771 	    !smc_connect_ism_vlan_setup(smc, &ini)) {
772 		/* ISM is supported for this connection */
773 		ism_supported = true;
774 		smc_type = SMC_TYPE_D;
775 	}
776 
777 	/* check if there is a rdma device available */
778 	if (!smc_find_rdma_device(smc, &ini)) {
779 		/* RDMA is supported for this connection */
780 		rdma_supported = true;
781 		if (ism_supported)
782 			smc_type = SMC_TYPE_B; /* both */
783 		else
784 			smc_type = SMC_TYPE_R; /* only RDMA */
785 	}
786 
787 	/* if neither ISM nor RDMA are supported, fallback */
788 	if (!rdma_supported && !ism_supported)
789 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
790 
791 	/* perform CLC handshake */
792 	rc = smc_connect_clc(smc, smc_type, &aclc, &ini);
793 	if (rc) {
794 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
795 		return smc_connect_decline_fallback(smc, rc);
796 	}
797 
798 	/* depending on previous steps, connect using rdma or ism */
799 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
800 		rc = smc_connect_rdma(smc, &aclc, &ini);
801 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
802 		rc = smc_connect_ism(smc, &aclc, &ini);
803 	else
804 		rc = SMC_CLC_DECL_MODEUNSUPP;
805 	if (rc) {
806 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
807 		return smc_connect_decline_fallback(smc, rc);
808 	}
809 
810 	smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
811 	return 0;
812 }
813 
814 static void smc_connect_work(struct work_struct *work)
815 {
816 	struct smc_sock *smc = container_of(work, struct smc_sock,
817 					    connect_work);
818 	long timeo = smc->sk.sk_sndtimeo;
819 	int rc = 0;
820 
821 	if (!timeo)
822 		timeo = MAX_SCHEDULE_TIMEOUT;
823 	lock_sock(smc->clcsock->sk);
824 	if (smc->clcsock->sk->sk_err) {
825 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
826 	} else if ((1 << smc->clcsock->sk->sk_state) &
827 					(TCPF_SYN_SENT | TCP_SYN_RECV)) {
828 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
829 		if ((rc == -EPIPE) &&
830 		    ((1 << smc->clcsock->sk->sk_state) &
831 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
832 			rc = 0;
833 	}
834 	release_sock(smc->clcsock->sk);
835 	lock_sock(&smc->sk);
836 	if (rc != 0 || smc->sk.sk_err) {
837 		smc->sk.sk_state = SMC_CLOSED;
838 		if (rc == -EPIPE || rc == -EAGAIN)
839 			smc->sk.sk_err = EPIPE;
840 		else if (signal_pending(current))
841 			smc->sk.sk_err = -sock_intr_errno(timeo);
842 		sock_put(&smc->sk); /* passive closing */
843 		goto out;
844 	}
845 
846 	rc = __smc_connect(smc);
847 	if (rc < 0)
848 		smc->sk.sk_err = -rc;
849 
850 out:
851 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
852 		if (smc->sk.sk_err) {
853 			smc->sk.sk_state_change(&smc->sk);
854 		} else { /* allow polling before and after fallback decision */
855 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
856 			smc->sk.sk_write_space(&smc->sk);
857 		}
858 	}
859 	release_sock(&smc->sk);
860 }
861 
862 static int smc_connect(struct socket *sock, struct sockaddr *addr,
863 		       int alen, int flags)
864 {
865 	struct sock *sk = sock->sk;
866 	struct smc_sock *smc;
867 	int rc = -EINVAL;
868 
869 	smc = smc_sk(sk);
870 
871 	/* separate smc parameter checking to be safe */
872 	if (alen < sizeof(addr->sa_family))
873 		goto out_err;
874 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
875 		goto out_err;
876 
877 	lock_sock(sk);
878 	switch (sk->sk_state) {
879 	default:
880 		goto out;
881 	case SMC_ACTIVE:
882 		rc = -EISCONN;
883 		goto out;
884 	case SMC_INIT:
885 		rc = 0;
886 		break;
887 	}
888 
889 	smc_copy_sock_settings_to_clc(smc);
890 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
891 	if (smc->connect_nonblock) {
892 		rc = -EALREADY;
893 		goto out;
894 	}
895 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
896 	if (rc && rc != -EINPROGRESS)
897 		goto out;
898 
899 	sock_hold(&smc->sk); /* sock put in passive closing */
900 	if (smc->use_fallback)
901 		goto out;
902 	if (flags & O_NONBLOCK) {
903 		if (schedule_work(&smc->connect_work))
904 			smc->connect_nonblock = 1;
905 		rc = -EINPROGRESS;
906 	} else {
907 		rc = __smc_connect(smc);
908 		if (rc < 0)
909 			goto out;
910 		else
911 			rc = 0; /* success cases including fallback */
912 	}
913 
914 out:
915 	release_sock(sk);
916 out_err:
917 	return rc;
918 }
919 
920 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
921 {
922 	struct socket *new_clcsock = NULL;
923 	struct sock *lsk = &lsmc->sk;
924 	struct sock *new_sk;
925 	int rc = -EINVAL;
926 
927 	release_sock(lsk);
928 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
929 	if (!new_sk) {
930 		rc = -ENOMEM;
931 		lsk->sk_err = ENOMEM;
932 		*new_smc = NULL;
933 		lock_sock(lsk);
934 		goto out;
935 	}
936 	*new_smc = smc_sk(new_sk);
937 
938 	mutex_lock(&lsmc->clcsock_release_lock);
939 	if (lsmc->clcsock)
940 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
941 	mutex_unlock(&lsmc->clcsock_release_lock);
942 	lock_sock(lsk);
943 	if  (rc < 0)
944 		lsk->sk_err = -rc;
945 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
946 		new_sk->sk_prot->unhash(new_sk);
947 		if (new_clcsock)
948 			sock_release(new_clcsock);
949 		new_sk->sk_state = SMC_CLOSED;
950 		sock_set_flag(new_sk, SOCK_DEAD);
951 		sock_put(new_sk); /* final */
952 		*new_smc = NULL;
953 		goto out;
954 	}
955 
956 	(*new_smc)->clcsock = new_clcsock;
957 out:
958 	return rc;
959 }
960 
961 /* add a just created sock to the accept queue of the listen sock as
962  * candidate for a following socket accept call from user space
963  */
964 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
965 {
966 	struct smc_sock *par = smc_sk(parent);
967 
968 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
969 	spin_lock(&par->accept_q_lock);
970 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
971 	spin_unlock(&par->accept_q_lock);
972 	sk_acceptq_added(parent);
973 }
974 
975 /* remove a socket from the accept queue of its parental listening socket */
976 static void smc_accept_unlink(struct sock *sk)
977 {
978 	struct smc_sock *par = smc_sk(sk)->listen_smc;
979 
980 	spin_lock(&par->accept_q_lock);
981 	list_del_init(&smc_sk(sk)->accept_q);
982 	spin_unlock(&par->accept_q_lock);
983 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
984 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
985 }
986 
987 /* remove a sock from the accept queue to bind it to a new socket created
988  * for a socket accept call from user space
989  */
990 struct sock *smc_accept_dequeue(struct sock *parent,
991 				struct socket *new_sock)
992 {
993 	struct smc_sock *isk, *n;
994 	struct sock *new_sk;
995 
996 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
997 		new_sk = (struct sock *)isk;
998 
999 		smc_accept_unlink(new_sk);
1000 		if (new_sk->sk_state == SMC_CLOSED) {
1001 			new_sk->sk_prot->unhash(new_sk);
1002 			if (isk->clcsock) {
1003 				sock_release(isk->clcsock);
1004 				isk->clcsock = NULL;
1005 			}
1006 			sock_put(new_sk); /* final */
1007 			continue;
1008 		}
1009 		if (new_sock) {
1010 			sock_graft(new_sk, new_sock);
1011 			if (isk->use_fallback) {
1012 				smc_sk(new_sk)->clcsock->file = new_sock->file;
1013 				isk->clcsock->file->private_data = isk->clcsock;
1014 			}
1015 		}
1016 		return new_sk;
1017 	}
1018 	return NULL;
1019 }
1020 
1021 /* clean up for a created but never accepted sock */
1022 void smc_close_non_accepted(struct sock *sk)
1023 {
1024 	struct smc_sock *smc = smc_sk(sk);
1025 
1026 	sock_hold(sk); /* sock_put below */
1027 	lock_sock(sk);
1028 	if (!sk->sk_lingertime)
1029 		/* wait for peer closing */
1030 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
1031 	__smc_release(smc);
1032 	release_sock(sk);
1033 	sock_put(sk); /* sock_hold above */
1034 	sock_put(sk); /* final sock_put */
1035 }
1036 
1037 static int smcr_serv_conf_first_link(struct smc_sock *smc)
1038 {
1039 	struct smc_link *link = smc->conn.lnk;
1040 	struct smc_llc_qentry *qentry;
1041 	int rc;
1042 
1043 	if (smcr_link_reg_rmb(link, smc->conn.rmb_desc))
1044 		return SMC_CLC_DECL_ERR_REGRMB;
1045 
1046 	/* send CONFIRM LINK request to client over the RoCE fabric */
1047 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1048 	if (rc < 0)
1049 		return SMC_CLC_DECL_TIMEOUT_CL;
1050 
1051 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1052 	qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1053 			      SMC_LLC_CONFIRM_LINK);
1054 	if (!qentry) {
1055 		struct smc_clc_msg_decline dclc;
1056 
1057 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1058 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1059 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1060 	}
1061 	smc_llc_save_peer_uid(qentry);
1062 	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1063 	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1064 	if (rc)
1065 		return SMC_CLC_DECL_RMBE_EC;
1066 
1067 	/* confirm_rkey is implicit on 1st contact */
1068 	smc->conn.rmb_desc->is_conf_rkey = true;
1069 
1070 	smc_llc_link_active(link);
1071 	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1072 
1073 	/* initial contact - try to establish second link */
1074 	smc_llc_srv_add_link(link);
1075 	return 0;
1076 }
1077 
1078 /* listen worker: finish */
1079 static void smc_listen_out(struct smc_sock *new_smc)
1080 {
1081 	struct smc_sock *lsmc = new_smc->listen_smc;
1082 	struct sock *newsmcsk = &new_smc->sk;
1083 
1084 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1085 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1086 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1087 		release_sock(&lsmc->sk);
1088 	} else { /* no longer listening */
1089 		smc_close_non_accepted(newsmcsk);
1090 	}
1091 
1092 	/* Wake up accept */
1093 	lsmc->sk.sk_data_ready(&lsmc->sk);
1094 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1095 }
1096 
1097 /* listen worker: finish in state connected */
1098 static void smc_listen_out_connected(struct smc_sock *new_smc)
1099 {
1100 	struct sock *newsmcsk = &new_smc->sk;
1101 
1102 	sk_refcnt_debug_inc(newsmcsk);
1103 	if (newsmcsk->sk_state == SMC_INIT)
1104 		newsmcsk->sk_state = SMC_ACTIVE;
1105 
1106 	smc_listen_out(new_smc);
1107 }
1108 
1109 /* listen worker: finish in error state */
1110 static void smc_listen_out_err(struct smc_sock *new_smc)
1111 {
1112 	struct sock *newsmcsk = &new_smc->sk;
1113 
1114 	if (newsmcsk->sk_state == SMC_INIT)
1115 		sock_put(&new_smc->sk); /* passive closing */
1116 	newsmcsk->sk_state = SMC_CLOSED;
1117 
1118 	smc_listen_out(new_smc);
1119 }
1120 
1121 /* listen worker: decline and fall back if possible */
1122 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1123 			       int local_contact)
1124 {
1125 	/* RDMA setup failed, switch back to TCP */
1126 	if (local_contact == SMC_FIRST_CONTACT)
1127 		smc_lgr_cleanup_early(&new_smc->conn);
1128 	else
1129 		smc_conn_free(&new_smc->conn);
1130 	if (reason_code < 0) { /* error, no fallback possible */
1131 		smc_listen_out_err(new_smc);
1132 		return;
1133 	}
1134 	smc_switch_to_fallback(new_smc);
1135 	new_smc->fallback_rsn = reason_code;
1136 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1137 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1138 			smc_listen_out_err(new_smc);
1139 			return;
1140 		}
1141 	}
1142 	smc_listen_out_connected(new_smc);
1143 }
1144 
1145 /* listen worker: check prefixes */
1146 static int smc_listen_prfx_check(struct smc_sock *new_smc,
1147 				 struct smc_clc_msg_proposal *pclc)
1148 {
1149 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1150 	struct socket *newclcsock = new_smc->clcsock;
1151 
1152 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1153 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1154 		return SMC_CLC_DECL_DIFFPREFIX;
1155 
1156 	return 0;
1157 }
1158 
1159 /* listen worker: initialize connection and buffers */
1160 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1161 				struct smc_init_info *ini)
1162 {
1163 	int rc;
1164 
1165 	/* allocate connection / link group */
1166 	rc = smc_conn_create(new_smc, ini);
1167 	if (rc)
1168 		return rc;
1169 
1170 	/* create send buffer and rmb */
1171 	if (smc_buf_create(new_smc, false))
1172 		return SMC_CLC_DECL_MEM;
1173 
1174 	return 0;
1175 }
1176 
1177 /* listen worker: initialize connection and buffers for SMC-D */
1178 static int smc_listen_ism_init(struct smc_sock *new_smc,
1179 			       struct smc_clc_msg_proposal *pclc,
1180 			       struct smc_init_info *ini)
1181 {
1182 	struct smc_clc_msg_smcd *pclc_smcd;
1183 	int rc;
1184 
1185 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1186 	ini->ism_gid = pclc_smcd->gid;
1187 	rc = smc_conn_create(new_smc, ini);
1188 	if (rc)
1189 		return rc;
1190 
1191 	/* Check if peer can be reached via ISM device */
1192 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1193 			    new_smc->conn.lgr->vlan_id,
1194 			    new_smc->conn.lgr->smcd)) {
1195 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1196 			smc_lgr_cleanup_early(&new_smc->conn);
1197 		else
1198 			smc_conn_free(&new_smc->conn);
1199 		return SMC_CLC_DECL_SMCDNOTALK;
1200 	}
1201 
1202 	/* Create send and receive buffers */
1203 	if (smc_buf_create(new_smc, true)) {
1204 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1205 			smc_lgr_cleanup_early(&new_smc->conn);
1206 		else
1207 			smc_conn_free(&new_smc->conn);
1208 		return SMC_CLC_DECL_MEM;
1209 	}
1210 
1211 	return 0;
1212 }
1213 
1214 /* listen worker: register buffers */
1215 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1216 {
1217 	struct smc_connection *conn = &new_smc->conn;
1218 
1219 	if (local_contact != SMC_FIRST_CONTACT) {
1220 		if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
1221 			return SMC_CLC_DECL_ERR_REGRMB;
1222 	}
1223 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1224 
1225 	return 0;
1226 }
1227 
1228 /* listen worker: finish RDMA setup */
1229 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1230 				  struct smc_clc_msg_accept_confirm *cclc,
1231 				  int local_contact)
1232 {
1233 	struct smc_link *link = new_smc->conn.lnk;
1234 	int reason_code = 0;
1235 
1236 	if (local_contact == SMC_FIRST_CONTACT)
1237 		smc_link_save_peer_info(link, cclc);
1238 
1239 	if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) {
1240 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1241 		goto decline;
1242 	}
1243 
1244 	if (local_contact == SMC_FIRST_CONTACT) {
1245 		if (smc_ib_ready_link(link)) {
1246 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1247 			goto decline;
1248 		}
1249 		/* QP confirmation over RoCE fabric */
1250 		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1251 		reason_code = smcr_serv_conf_first_link(new_smc);
1252 		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1253 		if (reason_code)
1254 			goto decline;
1255 	}
1256 	return 0;
1257 
1258 decline:
1259 	smc_listen_decline(new_smc, reason_code, local_contact);
1260 	return reason_code;
1261 }
1262 
1263 /* setup for RDMA connection of server */
1264 static void smc_listen_work(struct work_struct *work)
1265 {
1266 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1267 						smc_listen_work);
1268 	struct socket *newclcsock = new_smc->clcsock;
1269 	struct smc_clc_msg_accept_confirm cclc;
1270 	struct smc_clc_msg_proposal *pclc;
1271 	struct smc_init_info ini = {0};
1272 	bool ism_supported = false;
1273 	u8 buf[SMC_CLC_MAX_LEN];
1274 	int rc = 0;
1275 
1276 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1277 		return smc_listen_out_err(new_smc);
1278 
1279 	if (new_smc->use_fallback) {
1280 		smc_listen_out_connected(new_smc);
1281 		return;
1282 	}
1283 
1284 	/* check if peer is smc capable */
1285 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1286 		smc_switch_to_fallback(new_smc);
1287 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1288 		smc_listen_out_connected(new_smc);
1289 		return;
1290 	}
1291 
1292 	/* do inband token exchange -
1293 	 * wait for and receive SMC Proposal CLC message
1294 	 */
1295 	pclc = (struct smc_clc_msg_proposal *)&buf;
1296 	rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1297 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1298 	if (rc)
1299 		goto out_decl;
1300 
1301 	/* IPSec connections opt out of SMC-R optimizations */
1302 	if (using_ipsec(new_smc)) {
1303 		rc = SMC_CLC_DECL_IPSEC;
1304 		goto out_decl;
1305 	}
1306 
1307 	/* check for matching IP prefix and subnet length */
1308 	rc = smc_listen_prfx_check(new_smc, pclc);
1309 	if (rc)
1310 		goto out_decl;
1311 
1312 	/* get vlan id from IP device */
1313 	if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) {
1314 		rc = SMC_CLC_DECL_GETVLANERR;
1315 		goto out_decl;
1316 	}
1317 
1318 	mutex_lock(&smc_server_lgr_pending);
1319 	smc_close_init(new_smc);
1320 	smc_rx_init(new_smc);
1321 	smc_tx_init(new_smc);
1322 
1323 	/* check if ISM is available */
1324 	if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) {
1325 		ini.is_smcd = true; /* prepare ISM check */
1326 		rc = smc_find_ism_device(new_smc, &ini);
1327 		if (!rc)
1328 			rc = smc_listen_ism_init(new_smc, pclc, &ini);
1329 		if (!rc)
1330 			ism_supported = true;
1331 		else if (pclc->hdr.path == SMC_TYPE_D)
1332 			goto out_unlock; /* skip RDMA and decline */
1333 	}
1334 
1335 	/* check if RDMA is available */
1336 	if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */
1337 		/* prepare RDMA check */
1338 		ini.is_smcd = false;
1339 		ini.ism_dev = NULL;
1340 		ini.ib_lcl = &pclc->lcl;
1341 		rc = smc_find_rdma_device(new_smc, &ini);
1342 		if (rc) {
1343 			/* no RDMA device found */
1344 			if (pclc->hdr.path == SMC_TYPE_B)
1345 				/* neither ISM nor RDMA device found */
1346 				rc = SMC_CLC_DECL_NOSMCDEV;
1347 			goto out_unlock;
1348 		}
1349 		rc = smc_listen_rdma_init(new_smc, &ini);
1350 		if (rc)
1351 			goto out_unlock;
1352 		rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact);
1353 		if (rc)
1354 			goto out_unlock;
1355 	}
1356 
1357 	/* send SMC Accept CLC message */
1358 	rc = smc_clc_send_accept(new_smc, ini.cln_first_contact);
1359 	if (rc)
1360 		goto out_unlock;
1361 
1362 	/* SMC-D does not need this lock any more */
1363 	if (ism_supported)
1364 		mutex_unlock(&smc_server_lgr_pending);
1365 
1366 	/* receive SMC Confirm CLC message */
1367 	rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1368 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1369 	if (rc) {
1370 		if (!ism_supported)
1371 			goto out_unlock;
1372 		goto out_decl;
1373 	}
1374 
1375 	/* finish worker */
1376 	if (!ism_supported) {
1377 		rc = smc_listen_rdma_finish(new_smc, &cclc,
1378 					    ini.cln_first_contact);
1379 		mutex_unlock(&smc_server_lgr_pending);
1380 		if (rc)
1381 			return;
1382 	}
1383 	smc_conn_save_peer_info(new_smc, &cclc);
1384 	smc_listen_out_connected(new_smc);
1385 	return;
1386 
1387 out_unlock:
1388 	mutex_unlock(&smc_server_lgr_pending);
1389 out_decl:
1390 	smc_listen_decline(new_smc, rc, ini.cln_first_contact);
1391 }
1392 
1393 static void smc_tcp_listen_work(struct work_struct *work)
1394 {
1395 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1396 					     tcp_listen_work);
1397 	struct sock *lsk = &lsmc->sk;
1398 	struct smc_sock *new_smc;
1399 	int rc = 0;
1400 
1401 	lock_sock(lsk);
1402 	while (lsk->sk_state == SMC_LISTEN) {
1403 		rc = smc_clcsock_accept(lsmc, &new_smc);
1404 		if (rc)
1405 			goto out;
1406 		if (!new_smc)
1407 			continue;
1408 
1409 		new_smc->listen_smc = lsmc;
1410 		new_smc->use_fallback = lsmc->use_fallback;
1411 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1412 		sock_hold(lsk); /* sock_put in smc_listen_work */
1413 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1414 		smc_copy_sock_settings_to_smc(new_smc);
1415 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1416 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1417 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1418 		if (!schedule_work(&new_smc->smc_listen_work))
1419 			sock_put(&new_smc->sk);
1420 	}
1421 
1422 out:
1423 	release_sock(lsk);
1424 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1425 }
1426 
1427 static int smc_listen(struct socket *sock, int backlog)
1428 {
1429 	struct sock *sk = sock->sk;
1430 	struct smc_sock *smc;
1431 	int rc;
1432 
1433 	smc = smc_sk(sk);
1434 	lock_sock(sk);
1435 
1436 	rc = -EINVAL;
1437 	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
1438 	    smc->connect_nonblock)
1439 		goto out;
1440 
1441 	rc = 0;
1442 	if (sk->sk_state == SMC_LISTEN) {
1443 		sk->sk_max_ack_backlog = backlog;
1444 		goto out;
1445 	}
1446 	/* some socket options are handled in core, so we could not apply
1447 	 * them to the clc socket -- copy smc socket options to clc socket
1448 	 */
1449 	smc_copy_sock_settings_to_clc(smc);
1450 	if (!smc->use_fallback)
1451 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1452 
1453 	rc = kernel_listen(smc->clcsock, backlog);
1454 	if (rc)
1455 		goto out;
1456 	sk->sk_max_ack_backlog = backlog;
1457 	sk->sk_ack_backlog = 0;
1458 	sk->sk_state = SMC_LISTEN;
1459 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1460 	if (!schedule_work(&smc->tcp_listen_work))
1461 		sock_put(sk);
1462 
1463 out:
1464 	release_sock(sk);
1465 	return rc;
1466 }
1467 
1468 static int smc_accept(struct socket *sock, struct socket *new_sock,
1469 		      int flags, bool kern)
1470 {
1471 	struct sock *sk = sock->sk, *nsk;
1472 	DECLARE_WAITQUEUE(wait, current);
1473 	struct smc_sock *lsmc;
1474 	long timeo;
1475 	int rc = 0;
1476 
1477 	lsmc = smc_sk(sk);
1478 	sock_hold(sk); /* sock_put below */
1479 	lock_sock(sk);
1480 
1481 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1482 		rc = -EINVAL;
1483 		release_sock(sk);
1484 		goto out;
1485 	}
1486 
1487 	/* Wait for an incoming connection */
1488 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1489 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1490 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1491 		set_current_state(TASK_INTERRUPTIBLE);
1492 		if (!timeo) {
1493 			rc = -EAGAIN;
1494 			break;
1495 		}
1496 		release_sock(sk);
1497 		timeo = schedule_timeout(timeo);
1498 		/* wakeup by sk_data_ready in smc_listen_work() */
1499 		sched_annotate_sleep();
1500 		lock_sock(sk);
1501 		if (signal_pending(current)) {
1502 			rc = sock_intr_errno(timeo);
1503 			break;
1504 		}
1505 	}
1506 	set_current_state(TASK_RUNNING);
1507 	remove_wait_queue(sk_sleep(sk), &wait);
1508 
1509 	if (!rc)
1510 		rc = sock_error(nsk);
1511 	release_sock(sk);
1512 	if (rc)
1513 		goto out;
1514 
1515 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1516 		/* wait till data arrives on the socket */
1517 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1518 								MSEC_PER_SEC);
1519 		if (smc_sk(nsk)->use_fallback) {
1520 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1521 
1522 			lock_sock(clcsk);
1523 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1524 				sk_wait_data(clcsk, &timeo, NULL);
1525 			release_sock(clcsk);
1526 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1527 			lock_sock(nsk);
1528 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1529 			release_sock(nsk);
1530 		}
1531 	}
1532 
1533 out:
1534 	sock_put(sk); /* sock_hold above */
1535 	return rc;
1536 }
1537 
1538 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1539 		       int peer)
1540 {
1541 	struct smc_sock *smc;
1542 
1543 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1544 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1545 		return -ENOTCONN;
1546 
1547 	smc = smc_sk(sock->sk);
1548 
1549 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1550 }
1551 
1552 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1553 {
1554 	struct sock *sk = sock->sk;
1555 	struct smc_sock *smc;
1556 	int rc = -EPIPE;
1557 
1558 	smc = smc_sk(sk);
1559 	lock_sock(sk);
1560 	if ((sk->sk_state != SMC_ACTIVE) &&
1561 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1562 	    (sk->sk_state != SMC_INIT))
1563 		goto out;
1564 
1565 	if (msg->msg_flags & MSG_FASTOPEN) {
1566 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1567 			smc_switch_to_fallback(smc);
1568 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1569 		} else {
1570 			rc = -EINVAL;
1571 			goto out;
1572 		}
1573 	}
1574 
1575 	if (smc->use_fallback)
1576 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1577 	else
1578 		rc = smc_tx_sendmsg(smc, msg, len);
1579 out:
1580 	release_sock(sk);
1581 	return rc;
1582 }
1583 
1584 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1585 		       int flags)
1586 {
1587 	struct sock *sk = sock->sk;
1588 	struct smc_sock *smc;
1589 	int rc = -ENOTCONN;
1590 
1591 	smc = smc_sk(sk);
1592 	lock_sock(sk);
1593 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1594 		/* socket was connected before, no more data to read */
1595 		rc = 0;
1596 		goto out;
1597 	}
1598 	if ((sk->sk_state == SMC_INIT) ||
1599 	    (sk->sk_state == SMC_LISTEN) ||
1600 	    (sk->sk_state == SMC_CLOSED))
1601 		goto out;
1602 
1603 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1604 		rc = 0;
1605 		goto out;
1606 	}
1607 
1608 	if (smc->use_fallback) {
1609 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1610 	} else {
1611 		msg->msg_namelen = 0;
1612 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1613 	}
1614 
1615 out:
1616 	release_sock(sk);
1617 	return rc;
1618 }
1619 
1620 static __poll_t smc_accept_poll(struct sock *parent)
1621 {
1622 	struct smc_sock *isk = smc_sk(parent);
1623 	__poll_t mask = 0;
1624 
1625 	spin_lock(&isk->accept_q_lock);
1626 	if (!list_empty(&isk->accept_q))
1627 		mask = EPOLLIN | EPOLLRDNORM;
1628 	spin_unlock(&isk->accept_q_lock);
1629 
1630 	return mask;
1631 }
1632 
1633 static __poll_t smc_poll(struct file *file, struct socket *sock,
1634 			     poll_table *wait)
1635 {
1636 	struct sock *sk = sock->sk;
1637 	struct smc_sock *smc;
1638 	__poll_t mask = 0;
1639 
1640 	if (!sk)
1641 		return EPOLLNVAL;
1642 
1643 	smc = smc_sk(sock->sk);
1644 	if (smc->use_fallback) {
1645 		/* delegate to CLC child sock */
1646 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1647 		sk->sk_err = smc->clcsock->sk->sk_err;
1648 	} else {
1649 		if (sk->sk_state != SMC_CLOSED)
1650 			sock_poll_wait(file, sock, wait);
1651 		if (sk->sk_err)
1652 			mask |= EPOLLERR;
1653 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1654 		    (sk->sk_state == SMC_CLOSED))
1655 			mask |= EPOLLHUP;
1656 		if (sk->sk_state == SMC_LISTEN) {
1657 			/* woken up by sk_data_ready in smc_listen_work() */
1658 			mask |= smc_accept_poll(sk);
1659 		} else if (smc->use_fallback) { /* as result of connect_work()*/
1660 			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
1661 							   wait);
1662 			sk->sk_err = smc->clcsock->sk->sk_err;
1663 		} else {
1664 			if ((sk->sk_state != SMC_INIT &&
1665 			     atomic_read(&smc->conn.sndbuf_space)) ||
1666 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1667 				mask |= EPOLLOUT | EPOLLWRNORM;
1668 			} else {
1669 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1670 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1671 			}
1672 			if (atomic_read(&smc->conn.bytes_to_rcv))
1673 				mask |= EPOLLIN | EPOLLRDNORM;
1674 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1675 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1676 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1677 				mask |= EPOLLIN;
1678 			if (smc->conn.urg_state == SMC_URG_VALID)
1679 				mask |= EPOLLPRI;
1680 		}
1681 	}
1682 
1683 	return mask;
1684 }
1685 
1686 static int smc_shutdown(struct socket *sock, int how)
1687 {
1688 	struct sock *sk = sock->sk;
1689 	struct smc_sock *smc;
1690 	int rc = -EINVAL;
1691 	int rc1 = 0;
1692 
1693 	smc = smc_sk(sk);
1694 
1695 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1696 		return rc;
1697 
1698 	lock_sock(sk);
1699 
1700 	rc = -ENOTCONN;
1701 	if ((sk->sk_state != SMC_ACTIVE) &&
1702 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1703 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1704 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1705 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1706 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1707 		goto out;
1708 	if (smc->use_fallback) {
1709 		rc = kernel_sock_shutdown(smc->clcsock, how);
1710 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1711 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1712 			sk->sk_state = SMC_CLOSED;
1713 		goto out;
1714 	}
1715 	switch (how) {
1716 	case SHUT_RDWR:		/* shutdown in both directions */
1717 		rc = smc_close_active(smc);
1718 		break;
1719 	case SHUT_WR:
1720 		rc = smc_close_shutdown_write(smc);
1721 		break;
1722 	case SHUT_RD:
1723 		rc = 0;
1724 		/* nothing more to do because peer is not involved */
1725 		break;
1726 	}
1727 	if (smc->clcsock)
1728 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1729 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1730 	sk->sk_shutdown |= how + 1;
1731 
1732 out:
1733 	release_sock(sk);
1734 	return rc ? rc : rc1;
1735 }
1736 
1737 static int smc_setsockopt(struct socket *sock, int level, int optname,
1738 			  char __user *optval, unsigned int optlen)
1739 {
1740 	struct sock *sk = sock->sk;
1741 	struct smc_sock *smc;
1742 	int val, rc;
1743 
1744 	smc = smc_sk(sk);
1745 
1746 	/* generic setsockopts reaching us here always apply to the
1747 	 * CLC socket
1748 	 */
1749 	rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1750 					   optval, optlen);
1751 	if (smc->clcsock->sk->sk_err) {
1752 		sk->sk_err = smc->clcsock->sk->sk_err;
1753 		sk->sk_error_report(sk);
1754 	}
1755 
1756 	if (optlen < sizeof(int))
1757 		return -EINVAL;
1758 	if (get_user(val, (int __user *)optval))
1759 		return -EFAULT;
1760 
1761 	lock_sock(sk);
1762 	if (rc || smc->use_fallback)
1763 		goto out;
1764 	switch (optname) {
1765 	case TCP_ULP:
1766 	case TCP_FASTOPEN:
1767 	case TCP_FASTOPEN_CONNECT:
1768 	case TCP_FASTOPEN_KEY:
1769 	case TCP_FASTOPEN_NO_COOKIE:
1770 		/* option not supported by SMC */
1771 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1772 			smc_switch_to_fallback(smc);
1773 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1774 		} else {
1775 			rc = -EINVAL;
1776 		}
1777 		break;
1778 	case TCP_NODELAY:
1779 		if (sk->sk_state != SMC_INIT &&
1780 		    sk->sk_state != SMC_LISTEN &&
1781 		    sk->sk_state != SMC_CLOSED) {
1782 			if (val)
1783 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1784 						 0);
1785 		}
1786 		break;
1787 	case TCP_CORK:
1788 		if (sk->sk_state != SMC_INIT &&
1789 		    sk->sk_state != SMC_LISTEN &&
1790 		    sk->sk_state != SMC_CLOSED) {
1791 			if (!val)
1792 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1793 						 0);
1794 		}
1795 		break;
1796 	case TCP_DEFER_ACCEPT:
1797 		smc->sockopt_defer_accept = val;
1798 		break;
1799 	default:
1800 		break;
1801 	}
1802 out:
1803 	release_sock(sk);
1804 
1805 	return rc;
1806 }
1807 
1808 static int smc_getsockopt(struct socket *sock, int level, int optname,
1809 			  char __user *optval, int __user *optlen)
1810 {
1811 	struct smc_sock *smc;
1812 
1813 	smc = smc_sk(sock->sk);
1814 	/* socket options apply to the CLC socket */
1815 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1816 					     optval, optlen);
1817 }
1818 
1819 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1820 		     unsigned long arg)
1821 {
1822 	union smc_host_cursor cons, urg;
1823 	struct smc_connection *conn;
1824 	struct smc_sock *smc;
1825 	int answ;
1826 
1827 	smc = smc_sk(sock->sk);
1828 	conn = &smc->conn;
1829 	lock_sock(&smc->sk);
1830 	if (smc->use_fallback) {
1831 		if (!smc->clcsock) {
1832 			release_sock(&smc->sk);
1833 			return -EBADF;
1834 		}
1835 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1836 		release_sock(&smc->sk);
1837 		return answ;
1838 	}
1839 	switch (cmd) {
1840 	case SIOCINQ: /* same as FIONREAD */
1841 		if (smc->sk.sk_state == SMC_LISTEN) {
1842 			release_sock(&smc->sk);
1843 			return -EINVAL;
1844 		}
1845 		if (smc->sk.sk_state == SMC_INIT ||
1846 		    smc->sk.sk_state == SMC_CLOSED)
1847 			answ = 0;
1848 		else
1849 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1850 		break;
1851 	case SIOCOUTQ:
1852 		/* output queue size (not send + not acked) */
1853 		if (smc->sk.sk_state == SMC_LISTEN) {
1854 			release_sock(&smc->sk);
1855 			return -EINVAL;
1856 		}
1857 		if (smc->sk.sk_state == SMC_INIT ||
1858 		    smc->sk.sk_state == SMC_CLOSED)
1859 			answ = 0;
1860 		else
1861 			answ = smc->conn.sndbuf_desc->len -
1862 					atomic_read(&smc->conn.sndbuf_space);
1863 		break;
1864 	case SIOCOUTQNSD:
1865 		/* output queue size (not send only) */
1866 		if (smc->sk.sk_state == SMC_LISTEN) {
1867 			release_sock(&smc->sk);
1868 			return -EINVAL;
1869 		}
1870 		if (smc->sk.sk_state == SMC_INIT ||
1871 		    smc->sk.sk_state == SMC_CLOSED)
1872 			answ = 0;
1873 		else
1874 			answ = smc_tx_prepared_sends(&smc->conn);
1875 		break;
1876 	case SIOCATMARK:
1877 		if (smc->sk.sk_state == SMC_LISTEN) {
1878 			release_sock(&smc->sk);
1879 			return -EINVAL;
1880 		}
1881 		if (smc->sk.sk_state == SMC_INIT ||
1882 		    smc->sk.sk_state == SMC_CLOSED) {
1883 			answ = 0;
1884 		} else {
1885 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1886 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1887 			answ = smc_curs_diff(conn->rmb_desc->len,
1888 					     &cons, &urg) == 1;
1889 		}
1890 		break;
1891 	default:
1892 		release_sock(&smc->sk);
1893 		return -ENOIOCTLCMD;
1894 	}
1895 	release_sock(&smc->sk);
1896 
1897 	return put_user(answ, (int __user *)arg);
1898 }
1899 
1900 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1901 			    int offset, size_t size, int flags)
1902 {
1903 	struct sock *sk = sock->sk;
1904 	struct smc_sock *smc;
1905 	int rc = -EPIPE;
1906 
1907 	smc = smc_sk(sk);
1908 	lock_sock(sk);
1909 	if (sk->sk_state != SMC_ACTIVE) {
1910 		release_sock(sk);
1911 		goto out;
1912 	}
1913 	release_sock(sk);
1914 	if (smc->use_fallback)
1915 		rc = kernel_sendpage(smc->clcsock, page, offset,
1916 				     size, flags);
1917 	else
1918 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1919 
1920 out:
1921 	return rc;
1922 }
1923 
1924 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1925  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1926  * updates till whenever a respective page has been fully processed.
1927  * Note that subsequent recv() calls have to wait till all splice() processing
1928  * completed.
1929  */
1930 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1931 			       struct pipe_inode_info *pipe, size_t len,
1932 			       unsigned int flags)
1933 {
1934 	struct sock *sk = sock->sk;
1935 	struct smc_sock *smc;
1936 	int rc = -ENOTCONN;
1937 
1938 	smc = smc_sk(sk);
1939 	lock_sock(sk);
1940 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1941 		/* socket was connected before, no more data to read */
1942 		rc = 0;
1943 		goto out;
1944 	}
1945 	if (sk->sk_state == SMC_INIT ||
1946 	    sk->sk_state == SMC_LISTEN ||
1947 	    sk->sk_state == SMC_CLOSED)
1948 		goto out;
1949 
1950 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1951 		rc = 0;
1952 		goto out;
1953 	}
1954 
1955 	if (smc->use_fallback) {
1956 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1957 						    pipe, len, flags);
1958 	} else {
1959 		if (*ppos) {
1960 			rc = -ESPIPE;
1961 			goto out;
1962 		}
1963 		if (flags & SPLICE_F_NONBLOCK)
1964 			flags = MSG_DONTWAIT;
1965 		else
1966 			flags = 0;
1967 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1968 	}
1969 out:
1970 	release_sock(sk);
1971 
1972 	return rc;
1973 }
1974 
1975 /* must look like tcp */
1976 static const struct proto_ops smc_sock_ops = {
1977 	.family		= PF_SMC,
1978 	.owner		= THIS_MODULE,
1979 	.release	= smc_release,
1980 	.bind		= smc_bind,
1981 	.connect	= smc_connect,
1982 	.socketpair	= sock_no_socketpair,
1983 	.accept		= smc_accept,
1984 	.getname	= smc_getname,
1985 	.poll		= smc_poll,
1986 	.ioctl		= smc_ioctl,
1987 	.listen		= smc_listen,
1988 	.shutdown	= smc_shutdown,
1989 	.setsockopt	= smc_setsockopt,
1990 	.getsockopt	= smc_getsockopt,
1991 	.sendmsg	= smc_sendmsg,
1992 	.recvmsg	= smc_recvmsg,
1993 	.mmap		= sock_no_mmap,
1994 	.sendpage	= smc_sendpage,
1995 	.splice_read	= smc_splice_read,
1996 };
1997 
1998 static int smc_create(struct net *net, struct socket *sock, int protocol,
1999 		      int kern)
2000 {
2001 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
2002 	struct smc_sock *smc;
2003 	struct sock *sk;
2004 	int rc;
2005 
2006 	rc = -ESOCKTNOSUPPORT;
2007 	if (sock->type != SOCK_STREAM)
2008 		goto out;
2009 
2010 	rc = -EPROTONOSUPPORT;
2011 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
2012 		goto out;
2013 
2014 	rc = -ENOBUFS;
2015 	sock->ops = &smc_sock_ops;
2016 	sk = smc_sock_alloc(net, sock, protocol);
2017 	if (!sk)
2018 		goto out;
2019 
2020 	/* create internal TCP socket for CLC handshake and fallback */
2021 	smc = smc_sk(sk);
2022 	smc->use_fallback = false; /* assume rdma capability first */
2023 	smc->fallback_rsn = 0;
2024 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
2025 			      &smc->clcsock);
2026 	if (rc) {
2027 		sk_common_release(sk);
2028 		goto out;
2029 	}
2030 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
2031 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
2032 
2033 out:
2034 	return rc;
2035 }
2036 
2037 static const struct net_proto_family smc_sock_family_ops = {
2038 	.family	= PF_SMC,
2039 	.owner	= THIS_MODULE,
2040 	.create	= smc_create,
2041 };
2042 
2043 unsigned int smc_net_id;
2044 
2045 static __net_init int smc_net_init(struct net *net)
2046 {
2047 	return smc_pnet_net_init(net);
2048 }
2049 
2050 static void __net_exit smc_net_exit(struct net *net)
2051 {
2052 	smc_pnet_net_exit(net);
2053 }
2054 
2055 static struct pernet_operations smc_net_ops = {
2056 	.init = smc_net_init,
2057 	.exit = smc_net_exit,
2058 	.id   = &smc_net_id,
2059 	.size = sizeof(struct smc_net),
2060 };
2061 
2062 static int __init smc_init(void)
2063 {
2064 	int rc;
2065 
2066 	rc = register_pernet_subsys(&smc_net_ops);
2067 	if (rc)
2068 		return rc;
2069 
2070 	rc = smc_pnet_init();
2071 	if (rc)
2072 		goto out_pernet_subsys;
2073 
2074 	rc = smc_core_init();
2075 	if (rc) {
2076 		pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
2077 		goto out_pnet;
2078 	}
2079 
2080 	rc = smc_llc_init();
2081 	if (rc) {
2082 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2083 		goto out_core;
2084 	}
2085 
2086 	rc = smc_cdc_init();
2087 	if (rc) {
2088 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2089 		goto out_core;
2090 	}
2091 
2092 	rc = proto_register(&smc_proto, 1);
2093 	if (rc) {
2094 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2095 		goto out_core;
2096 	}
2097 
2098 	rc = proto_register(&smc_proto6, 1);
2099 	if (rc) {
2100 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2101 		goto out_proto;
2102 	}
2103 
2104 	rc = sock_register(&smc_sock_family_ops);
2105 	if (rc) {
2106 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2107 		goto out_proto6;
2108 	}
2109 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2110 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2111 
2112 	rc = smc_ib_register_client();
2113 	if (rc) {
2114 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2115 		goto out_sock;
2116 	}
2117 
2118 	static_branch_enable(&tcp_have_smc);
2119 	return 0;
2120 
2121 out_sock:
2122 	sock_unregister(PF_SMC);
2123 out_proto6:
2124 	proto_unregister(&smc_proto6);
2125 out_proto:
2126 	proto_unregister(&smc_proto);
2127 out_core:
2128 	smc_core_exit();
2129 out_pnet:
2130 	smc_pnet_exit();
2131 out_pernet_subsys:
2132 	unregister_pernet_subsys(&smc_net_ops);
2133 
2134 	return rc;
2135 }
2136 
2137 static void __exit smc_exit(void)
2138 {
2139 	static_branch_disable(&tcp_have_smc);
2140 	sock_unregister(PF_SMC);
2141 	smc_core_exit();
2142 	smc_ib_unregister_client();
2143 	proto_unregister(&smc_proto6);
2144 	proto_unregister(&smc_proto);
2145 	smc_pnet_exit();
2146 	unregister_pernet_subsys(&smc_net_ops);
2147 	rcu_barrier();
2148 }
2149 
2150 module_init(smc_init);
2151 module_exit(smc_exit);
2152 
2153 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2154 MODULE_DESCRIPTION("smc socket address family");
2155 MODULE_LICENSE("GPL");
2156 MODULE_ALIAS_NETPROTO(PF_SMC);
2157