xref: /openbmc/linux/net/smc/af_smc.c (revision 05cf4fe738242183f1237f1b3a28b4479348c0a1)
1 /*
2  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
3  *
4  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
5  *  applies to SOCK_STREAM sockets only
6  *  offers an alternative communication option for TCP-protocol sockets
7  *  applicable with RoCE-cards only
8  *
9  *  Initial restrictions:
10  *    - support for alternate links postponed
11  *
12  *  Copyright IBM Corp. 2016, 2018
13  *
14  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
15  *              based on prototype from Frank Blaschka
16  */
17 
18 #define KMSG_COMPONENT "smc"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/module.h>
22 #include <linux/socket.h>
23 #include <linux/workqueue.h>
24 #include <linux/in.h>
25 #include <linux/sched/signal.h>
26 #include <linux/if_vlan.h>
27 
28 #include <net/sock.h>
29 #include <net/tcp.h>
30 #include <net/smc.h>
31 #include <asm/ioctls.h>
32 
33 #include "smc.h"
34 #include "smc_clc.h"
35 #include "smc_llc.h"
36 #include "smc_cdc.h"
37 #include "smc_core.h"
38 #include "smc_ib.h"
39 #include "smc_ism.h"
40 #include "smc_pnet.h"
41 #include "smc_tx.h"
42 #include "smc_rx.h"
43 #include "smc_close.h"
44 
45 static DEFINE_MUTEX(smc_create_lgr_pending);	/* serialize link group
46 						 * creation
47 						 */
48 
49 static void smc_tcp_listen_work(struct work_struct *);
50 static void smc_connect_work(struct work_struct *);
51 
52 static void smc_set_keepalive(struct sock *sk, int val)
53 {
54 	struct smc_sock *smc = smc_sk(sk);
55 
56 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
57 }
58 
59 static struct smc_hashinfo smc_v4_hashinfo = {
60 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
61 };
62 
63 static struct smc_hashinfo smc_v6_hashinfo = {
64 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
65 };
66 
67 int smc_hash_sk(struct sock *sk)
68 {
69 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
70 	struct hlist_head *head;
71 
72 	head = &h->ht;
73 
74 	write_lock_bh(&h->lock);
75 	sk_add_node(sk, head);
76 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
77 	write_unlock_bh(&h->lock);
78 
79 	return 0;
80 }
81 EXPORT_SYMBOL_GPL(smc_hash_sk);
82 
83 void smc_unhash_sk(struct sock *sk)
84 {
85 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
86 
87 	write_lock_bh(&h->lock);
88 	if (sk_del_node_init(sk))
89 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
90 	write_unlock_bh(&h->lock);
91 }
92 EXPORT_SYMBOL_GPL(smc_unhash_sk);
93 
94 struct proto smc_proto = {
95 	.name		= "SMC",
96 	.owner		= THIS_MODULE,
97 	.keepalive	= smc_set_keepalive,
98 	.hash		= smc_hash_sk,
99 	.unhash		= smc_unhash_sk,
100 	.obj_size	= sizeof(struct smc_sock),
101 	.h.smc_hash	= &smc_v4_hashinfo,
102 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
103 };
104 EXPORT_SYMBOL_GPL(smc_proto);
105 
106 struct proto smc_proto6 = {
107 	.name		= "SMC6",
108 	.owner		= THIS_MODULE,
109 	.keepalive	= smc_set_keepalive,
110 	.hash		= smc_hash_sk,
111 	.unhash		= smc_unhash_sk,
112 	.obj_size	= sizeof(struct smc_sock),
113 	.h.smc_hash	= &smc_v6_hashinfo,
114 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
115 };
116 EXPORT_SYMBOL_GPL(smc_proto6);
117 
118 static int smc_release(struct socket *sock)
119 {
120 	struct sock *sk = sock->sk;
121 	struct smc_sock *smc;
122 	int rc = 0;
123 
124 	if (!sk)
125 		goto out;
126 
127 	smc = smc_sk(sk);
128 
129 	/* cleanup for a dangling non-blocking connect */
130 	if (smc->connect_info && sk->sk_state == SMC_INIT)
131 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
132 	flush_work(&smc->connect_work);
133 	kfree(smc->connect_info);
134 	smc->connect_info = NULL;
135 
136 	if (sk->sk_state == SMC_LISTEN)
137 		/* smc_close_non_accepted() is called and acquires
138 		 * sock lock for child sockets again
139 		 */
140 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
141 	else
142 		lock_sock(sk);
143 
144 	if (!smc->use_fallback) {
145 		rc = smc_close_active(smc);
146 		sock_set_flag(sk, SOCK_DEAD);
147 		sk->sk_shutdown |= SHUTDOWN_MASK;
148 	}
149 	if (smc->clcsock) {
150 		sock_release(smc->clcsock);
151 		smc->clcsock = NULL;
152 	}
153 	if (smc->use_fallback) {
154 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
155 			sock_put(sk); /* passive closing */
156 		sk->sk_state = SMC_CLOSED;
157 		sk->sk_state_change(sk);
158 	}
159 
160 	/* detach socket */
161 	sock_orphan(sk);
162 	sock->sk = NULL;
163 	if (!smc->use_fallback && sk->sk_state == SMC_CLOSED)
164 		smc_conn_free(&smc->conn);
165 	release_sock(sk);
166 
167 	sk->sk_prot->unhash(sk);
168 	sock_put(sk); /* final sock_put */
169 out:
170 	return rc;
171 }
172 
173 static void smc_destruct(struct sock *sk)
174 {
175 	if (sk->sk_state != SMC_CLOSED)
176 		return;
177 	if (!sock_flag(sk, SOCK_DEAD))
178 		return;
179 
180 	sk_refcnt_debug_dec(sk);
181 }
182 
183 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
184 				   int protocol)
185 {
186 	struct smc_sock *smc;
187 	struct proto *prot;
188 	struct sock *sk;
189 
190 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
191 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
192 	if (!sk)
193 		return NULL;
194 
195 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
196 	sk->sk_state = SMC_INIT;
197 	sk->sk_destruct = smc_destruct;
198 	sk->sk_protocol = protocol;
199 	smc = smc_sk(sk);
200 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
201 	INIT_WORK(&smc->connect_work, smc_connect_work);
202 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
203 	INIT_LIST_HEAD(&smc->accept_q);
204 	spin_lock_init(&smc->accept_q_lock);
205 	spin_lock_init(&smc->conn.send_lock);
206 	sk->sk_prot->hash(sk);
207 	sk_refcnt_debug_inc(sk);
208 
209 	return sk;
210 }
211 
212 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
213 		    int addr_len)
214 {
215 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
216 	struct sock *sk = sock->sk;
217 	struct smc_sock *smc;
218 	int rc;
219 
220 	smc = smc_sk(sk);
221 
222 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
223 	rc = -EINVAL;
224 	if (addr_len < sizeof(struct sockaddr_in))
225 		goto out;
226 
227 	rc = -EAFNOSUPPORT;
228 	if (addr->sin_family != AF_INET &&
229 	    addr->sin_family != AF_INET6 &&
230 	    addr->sin_family != AF_UNSPEC)
231 		goto out;
232 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
233 	if (addr->sin_family == AF_UNSPEC &&
234 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
235 		goto out;
236 
237 	lock_sock(sk);
238 
239 	/* Check if socket is already active */
240 	rc = -EINVAL;
241 	if (sk->sk_state != SMC_INIT)
242 		goto out_rel;
243 
244 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
245 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
246 
247 out_rel:
248 	release_sock(sk);
249 out:
250 	return rc;
251 }
252 
253 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
254 				   unsigned long mask)
255 {
256 	/* options we don't get control via setsockopt for */
257 	nsk->sk_type = osk->sk_type;
258 	nsk->sk_sndbuf = osk->sk_sndbuf;
259 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
260 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
261 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
262 	nsk->sk_mark = osk->sk_mark;
263 	nsk->sk_priority = osk->sk_priority;
264 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
265 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
266 	nsk->sk_err = osk->sk_err;
267 
268 	nsk->sk_flags &= ~mask;
269 	nsk->sk_flags |= osk->sk_flags & mask;
270 }
271 
272 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
273 			     (1UL << SOCK_KEEPOPEN) | \
274 			     (1UL << SOCK_LINGER) | \
275 			     (1UL << SOCK_BROADCAST) | \
276 			     (1UL << SOCK_TIMESTAMP) | \
277 			     (1UL << SOCK_DBG) | \
278 			     (1UL << SOCK_RCVTSTAMP) | \
279 			     (1UL << SOCK_RCVTSTAMPNS) | \
280 			     (1UL << SOCK_LOCALROUTE) | \
281 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
282 			     (1UL << SOCK_RXQ_OVFL) | \
283 			     (1UL << SOCK_WIFI_STATUS) | \
284 			     (1UL << SOCK_NOFCS) | \
285 			     (1UL << SOCK_FILTER_LOCKED))
286 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
287  * clc socket (since smc is not called for these options from net/core)
288  */
289 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
290 {
291 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
292 }
293 
294 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
295 			     (1UL << SOCK_KEEPOPEN) | \
296 			     (1UL << SOCK_LINGER) | \
297 			     (1UL << SOCK_DBG))
298 /* copy only settings and flags relevant for smc from clc to smc socket */
299 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
300 {
301 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
302 }
303 
304 /* register a new rmb, optionally send confirm_rkey msg to register with peer */
305 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc,
306 		       bool conf_rkey)
307 {
308 	/* register memory region for new rmb */
309 	if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
310 		rmb_desc->regerr = 1;
311 		return -EFAULT;
312 	}
313 	if (!conf_rkey)
314 		return 0;
315 	/* exchange confirm_rkey msg with peer */
316 	if (smc_llc_do_confirm_rkey(link, rmb_desc)) {
317 		rmb_desc->regerr = 1;
318 		return -EFAULT;
319 	}
320 	return 0;
321 }
322 
323 static int smc_clnt_conf_first_link(struct smc_sock *smc)
324 {
325 	struct net *net = sock_net(smc->clcsock->sk);
326 	struct smc_link_group *lgr = smc->conn.lgr;
327 	struct smc_link *link;
328 	int rest;
329 	int rc;
330 
331 	link = &lgr->lnk[SMC_SINGLE_LINK];
332 	/* receive CONFIRM LINK request from server over RoCE fabric */
333 	rest = wait_for_completion_interruptible_timeout(
334 		&link->llc_confirm,
335 		SMC_LLC_WAIT_FIRST_TIME);
336 	if (rest <= 0) {
337 		struct smc_clc_msg_decline dclc;
338 
339 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
340 				      SMC_CLC_DECLINE);
341 		return rc;
342 	}
343 
344 	if (link->llc_confirm_rc)
345 		return SMC_CLC_DECL_RMBE_EC;
346 
347 	rc = smc_ib_modify_qp_rts(link);
348 	if (rc)
349 		return SMC_CLC_DECL_ERR_RDYLNK;
350 
351 	smc_wr_remember_qp_attr(link);
352 
353 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
354 		return SMC_CLC_DECL_ERR_REGRMB;
355 
356 	/* send CONFIRM LINK response over RoCE fabric */
357 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
358 	if (rc < 0)
359 		return SMC_CLC_DECL_TIMEOUT_CL;
360 
361 	/* receive ADD LINK request from server over RoCE fabric */
362 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
363 							 SMC_LLC_WAIT_TIME);
364 	if (rest <= 0) {
365 		struct smc_clc_msg_decline dclc;
366 
367 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
368 				      SMC_CLC_DECLINE);
369 		return rc;
370 	}
371 
372 	/* send add link reject message, only one link supported for now */
373 	rc = smc_llc_send_add_link(link,
374 				   link->smcibdev->mac[link->ibport - 1],
375 				   link->gid, SMC_LLC_RESP);
376 	if (rc < 0)
377 		return SMC_CLC_DECL_TIMEOUT_AL;
378 
379 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
380 
381 	return 0;
382 }
383 
384 static void smcr_conn_save_peer_info(struct smc_sock *smc,
385 				     struct smc_clc_msg_accept_confirm *clc)
386 {
387 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
388 
389 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
390 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
391 	smc->conn.peer_rmbe_size = bufsize;
392 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
393 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
394 }
395 
396 static void smcd_conn_save_peer_info(struct smc_sock *smc,
397 				     struct smc_clc_msg_accept_confirm *clc)
398 {
399 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
400 
401 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
402 	smc->conn.peer_token = clc->token;
403 	/* msg header takes up space in the buffer */
404 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
405 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
406 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
407 }
408 
409 static void smc_conn_save_peer_info(struct smc_sock *smc,
410 				    struct smc_clc_msg_accept_confirm *clc)
411 {
412 	if (smc->conn.lgr->is_smcd)
413 		smcd_conn_save_peer_info(smc, clc);
414 	else
415 		smcr_conn_save_peer_info(smc, clc);
416 }
417 
418 static void smc_link_save_peer_info(struct smc_link *link,
419 				    struct smc_clc_msg_accept_confirm *clc)
420 {
421 	link->peer_qpn = ntoh24(clc->qpn);
422 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
423 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
424 	link->peer_psn = ntoh24(clc->psn);
425 	link->peer_mtu = clc->qp_mtu;
426 }
427 
428 /* fall back during connect */
429 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
430 {
431 	smc->use_fallback = true;
432 	smc->fallback_rsn = reason_code;
433 	smc_copy_sock_settings_to_clc(smc);
434 	if (smc->sk.sk_state == SMC_INIT)
435 		smc->sk.sk_state = SMC_ACTIVE;
436 	return 0;
437 }
438 
439 /* decline and fall back during connect */
440 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
441 {
442 	int rc;
443 
444 	if (reason_code < 0) { /* error, fallback is not possible */
445 		if (smc->sk.sk_state == SMC_INIT)
446 			sock_put(&smc->sk); /* passive closing */
447 		return reason_code;
448 	}
449 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
450 		rc = smc_clc_send_decline(smc, reason_code);
451 		if (rc < 0) {
452 			if (smc->sk.sk_state == SMC_INIT)
453 				sock_put(&smc->sk); /* passive closing */
454 			return rc;
455 		}
456 	}
457 	return smc_connect_fallback(smc, reason_code);
458 }
459 
460 /* abort connecting */
461 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
462 			     int local_contact)
463 {
464 	if (local_contact == SMC_FIRST_CONTACT)
465 		smc_lgr_forget(smc->conn.lgr);
466 	mutex_unlock(&smc_create_lgr_pending);
467 	smc_conn_free(&smc->conn);
468 	return reason_code;
469 }
470 
471 /* check if there is a rdma device available for this connection. */
472 /* called for connect and listen */
473 static int smc_check_rdma(struct smc_sock *smc, struct smc_ib_device **ibdev,
474 			  u8 *ibport, unsigned short vlan_id, u8 gid[])
475 {
476 	int reason_code = 0;
477 
478 	/* PNET table look up: search active ib_device and port
479 	 * within same PNETID that also contains the ethernet device
480 	 * used for the internal TCP socket
481 	 */
482 	smc_pnet_find_roce_resource(smc->clcsock->sk, ibdev, ibport, vlan_id,
483 				    gid);
484 	if (!(*ibdev))
485 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
486 
487 	return reason_code;
488 }
489 
490 /* check if there is an ISM device available for this connection. */
491 /* called for connect and listen */
492 static int smc_check_ism(struct smc_sock *smc, struct smcd_dev **ismdev)
493 {
494 	/* Find ISM device with same PNETID as connecting interface  */
495 	smc_pnet_find_ism_resource(smc->clcsock->sk, ismdev);
496 	if (!(*ismdev))
497 		return SMC_CLC_DECL_CNFERR; /* configuration error */
498 	return 0;
499 }
500 
501 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
502 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
503 				      struct smcd_dev *ismdev,
504 				      unsigned short vlan_id)
505 {
506 	if (vlan_id && smc_ism_get_vlan(ismdev, vlan_id))
507 		return SMC_CLC_DECL_CNFERR;
508 	return 0;
509 }
510 
511 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
512  * used, the VLAN ID will be registered again during the connection setup.
513  */
514 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
515 					struct smcd_dev *ismdev,
516 					unsigned short vlan_id)
517 {
518 	if (!is_smcd)
519 		return 0;
520 	if (vlan_id && smc_ism_put_vlan(ismdev, vlan_id))
521 		return SMC_CLC_DECL_CNFERR;
522 	return 0;
523 }
524 
525 /* CLC handshake during connect */
526 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
527 			   struct smc_clc_msg_accept_confirm *aclc,
528 			   struct smc_ib_device *ibdev, u8 ibport,
529 			   u8 gid[], struct smcd_dev *ismdev)
530 {
531 	int rc = 0;
532 
533 	/* do inband token exchange */
534 	rc = smc_clc_send_proposal(smc, smc_type, ibdev, ibport, gid, ismdev);
535 	if (rc)
536 		return rc;
537 	/* receive SMC Accept CLC message */
538 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT);
539 }
540 
541 /* setup for RDMA connection of client */
542 static int smc_connect_rdma(struct smc_sock *smc,
543 			    struct smc_clc_msg_accept_confirm *aclc,
544 			    struct smc_ib_device *ibdev, u8 ibport)
545 {
546 	int local_contact = SMC_FIRST_CONTACT;
547 	struct smc_link *link;
548 	int reason_code = 0;
549 
550 	mutex_lock(&smc_create_lgr_pending);
551 	local_contact = smc_conn_create(smc, false, aclc->hdr.flag, ibdev,
552 					ibport, ntoh24(aclc->qpn), &aclc->lcl,
553 					NULL, 0);
554 	if (local_contact < 0) {
555 		if (local_contact == -ENOMEM)
556 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
557 		else if (local_contact == -ENOLINK)
558 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
559 		else
560 			reason_code = SMC_CLC_DECL_INTERR; /* other error */
561 		return smc_connect_abort(smc, reason_code, 0);
562 	}
563 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
564 
565 	smc_conn_save_peer_info(smc, aclc);
566 
567 	/* create send buffer and rmb */
568 	if (smc_buf_create(smc, false))
569 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM, local_contact);
570 
571 	if (local_contact == SMC_FIRST_CONTACT)
572 		smc_link_save_peer_info(link, aclc);
573 
574 	if (smc_rmb_rtoken_handling(&smc->conn, aclc))
575 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
576 					 local_contact);
577 
578 	smc_close_init(smc);
579 	smc_rx_init(smc);
580 
581 	if (local_contact == SMC_FIRST_CONTACT) {
582 		if (smc_ib_ready_link(link))
583 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
584 						 local_contact);
585 	} else {
586 		if (!smc->conn.rmb_desc->reused &&
587 		    smc_reg_rmb(link, smc->conn.rmb_desc, true))
588 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
589 						 local_contact);
590 	}
591 	smc_rmb_sync_sg_for_device(&smc->conn);
592 
593 	reason_code = smc_clc_send_confirm(smc);
594 	if (reason_code)
595 		return smc_connect_abort(smc, reason_code, local_contact);
596 
597 	smc_tx_init(smc);
598 
599 	if (local_contact == SMC_FIRST_CONTACT) {
600 		/* QP confirmation over RoCE fabric */
601 		reason_code = smc_clnt_conf_first_link(smc);
602 		if (reason_code)
603 			return smc_connect_abort(smc, reason_code,
604 						 local_contact);
605 	}
606 	mutex_unlock(&smc_create_lgr_pending);
607 
608 	smc_copy_sock_settings_to_clc(smc);
609 	if (smc->sk.sk_state == SMC_INIT)
610 		smc->sk.sk_state = SMC_ACTIVE;
611 
612 	return 0;
613 }
614 
615 /* setup for ISM connection of client */
616 static int smc_connect_ism(struct smc_sock *smc,
617 			   struct smc_clc_msg_accept_confirm *aclc,
618 			   struct smcd_dev *ismdev)
619 {
620 	int local_contact = SMC_FIRST_CONTACT;
621 	int rc = 0;
622 
623 	mutex_lock(&smc_create_lgr_pending);
624 	local_contact = smc_conn_create(smc, true, aclc->hdr.flag, NULL, 0, 0,
625 					NULL, ismdev, aclc->gid);
626 	if (local_contact < 0)
627 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 0);
628 
629 	/* Create send and receive buffers */
630 	if (smc_buf_create(smc, true))
631 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM, local_contact);
632 
633 	smc_conn_save_peer_info(smc, aclc);
634 	smc_close_init(smc);
635 	smc_rx_init(smc);
636 	smc_tx_init(smc);
637 
638 	rc = smc_clc_send_confirm(smc);
639 	if (rc)
640 		return smc_connect_abort(smc, rc, local_contact);
641 	mutex_unlock(&smc_create_lgr_pending);
642 
643 	smc_copy_sock_settings_to_clc(smc);
644 	if (smc->sk.sk_state == SMC_INIT)
645 		smc->sk.sk_state = SMC_ACTIVE;
646 
647 	return 0;
648 }
649 
650 /* perform steps before actually connecting */
651 static int __smc_connect(struct smc_sock *smc)
652 {
653 	bool ism_supported = false, rdma_supported = false;
654 	struct smc_clc_msg_accept_confirm aclc;
655 	struct smc_ib_device *ibdev;
656 	struct smcd_dev *ismdev;
657 	u8 gid[SMC_GID_SIZE];
658 	unsigned short vlan;
659 	int smc_type;
660 	int rc = 0;
661 	u8 ibport;
662 
663 	sock_hold(&smc->sk); /* sock put in passive closing */
664 
665 	if (smc->use_fallback)
666 		return smc_connect_fallback(smc, smc->fallback_rsn);
667 
668 	/* if peer has not signalled SMC-capability, fall back */
669 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
670 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
671 
672 	/* IPSec connections opt out of SMC-R optimizations */
673 	if (using_ipsec(smc))
674 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
675 
676 	/* check for VLAN ID */
677 	if (smc_vlan_by_tcpsk(smc->clcsock, &vlan))
678 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_CNFERR);
679 
680 	/* check if there is an ism device available */
681 	if (!smc_check_ism(smc, &ismdev) &&
682 	    !smc_connect_ism_vlan_setup(smc, ismdev, vlan)) {
683 		/* ISM is supported for this connection */
684 		ism_supported = true;
685 		smc_type = SMC_TYPE_D;
686 	}
687 
688 	/* check if there is a rdma device available */
689 	if (!smc_check_rdma(smc, &ibdev, &ibport, vlan, gid)) {
690 		/* RDMA is supported for this connection */
691 		rdma_supported = true;
692 		if (ism_supported)
693 			smc_type = SMC_TYPE_B; /* both */
694 		else
695 			smc_type = SMC_TYPE_R; /* only RDMA */
696 	}
697 
698 	/* if neither ISM nor RDMA are supported, fallback */
699 	if (!rdma_supported && !ism_supported)
700 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
701 
702 	/* perform CLC handshake */
703 	rc = smc_connect_clc(smc, smc_type, &aclc, ibdev, ibport, gid, ismdev);
704 	if (rc) {
705 		smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
706 		return smc_connect_decline_fallback(smc, rc);
707 	}
708 
709 	/* depending on previous steps, connect using rdma or ism */
710 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
711 		rc = smc_connect_rdma(smc, &aclc, ibdev, ibport);
712 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
713 		rc = smc_connect_ism(smc, &aclc, ismdev);
714 	else
715 		rc = SMC_CLC_DECL_MODEUNSUPP;
716 	if (rc) {
717 		smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
718 		return smc_connect_decline_fallback(smc, rc);
719 	}
720 
721 	smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
722 	return 0;
723 }
724 
725 static void smc_connect_work(struct work_struct *work)
726 {
727 	struct smc_sock *smc = container_of(work, struct smc_sock,
728 					    connect_work);
729 	int rc;
730 
731 	lock_sock(&smc->sk);
732 	rc = kernel_connect(smc->clcsock, &smc->connect_info->addr,
733 			    smc->connect_info->alen, smc->connect_info->flags);
734 	if (smc->clcsock->sk->sk_err) {
735 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
736 		goto out;
737 	}
738 	if (rc < 0) {
739 		smc->sk.sk_err = -rc;
740 		goto out;
741 	}
742 
743 	rc = __smc_connect(smc);
744 	if (rc < 0)
745 		smc->sk.sk_err = -rc;
746 
747 out:
748 	if (smc->sk.sk_err)
749 		smc->sk.sk_state_change(&smc->sk);
750 	else
751 		smc->sk.sk_write_space(&smc->sk);
752 	kfree(smc->connect_info);
753 	smc->connect_info = NULL;
754 	release_sock(&smc->sk);
755 }
756 
757 static int smc_connect(struct socket *sock, struct sockaddr *addr,
758 		       int alen, int flags)
759 {
760 	struct sock *sk = sock->sk;
761 	struct smc_sock *smc;
762 	int rc = -EINVAL;
763 
764 	smc = smc_sk(sk);
765 
766 	/* separate smc parameter checking to be safe */
767 	if (alen < sizeof(addr->sa_family))
768 		goto out_err;
769 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
770 		goto out_err;
771 
772 	lock_sock(sk);
773 	switch (sk->sk_state) {
774 	default:
775 		goto out;
776 	case SMC_ACTIVE:
777 		rc = -EISCONN;
778 		goto out;
779 	case SMC_INIT:
780 		rc = 0;
781 		break;
782 	}
783 
784 	smc_copy_sock_settings_to_clc(smc);
785 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
786 	if (flags & O_NONBLOCK) {
787 		if (smc->connect_info) {
788 			rc = -EALREADY;
789 			goto out;
790 		}
791 		smc->connect_info = kzalloc(alen + 2 * sizeof(int), GFP_KERNEL);
792 		if (!smc->connect_info) {
793 			rc = -ENOMEM;
794 			goto out;
795 		}
796 		smc->connect_info->alen = alen;
797 		smc->connect_info->flags = flags ^ O_NONBLOCK;
798 		memcpy(&smc->connect_info->addr, addr, alen);
799 		schedule_work(&smc->connect_work);
800 		rc = -EINPROGRESS;
801 	} else {
802 		rc = kernel_connect(smc->clcsock, addr, alen, flags);
803 		if (rc)
804 			goto out;
805 
806 		rc = __smc_connect(smc);
807 		if (rc < 0)
808 			goto out;
809 		else
810 			rc = 0; /* success cases including fallback */
811 	}
812 
813 out:
814 	release_sock(sk);
815 out_err:
816 	return rc;
817 }
818 
819 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
820 {
821 	struct socket *new_clcsock = NULL;
822 	struct sock *lsk = &lsmc->sk;
823 	struct sock *new_sk;
824 	int rc;
825 
826 	release_sock(lsk);
827 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
828 	if (!new_sk) {
829 		rc = -ENOMEM;
830 		lsk->sk_err = ENOMEM;
831 		*new_smc = NULL;
832 		lock_sock(lsk);
833 		goto out;
834 	}
835 	*new_smc = smc_sk(new_sk);
836 
837 	rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
838 	lock_sock(lsk);
839 	if  (rc < 0)
840 		lsk->sk_err = -rc;
841 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
842 		if (new_clcsock)
843 			sock_release(new_clcsock);
844 		new_sk->sk_state = SMC_CLOSED;
845 		sock_set_flag(new_sk, SOCK_DEAD);
846 		new_sk->sk_prot->unhash(new_sk);
847 		sock_put(new_sk); /* final */
848 		*new_smc = NULL;
849 		goto out;
850 	}
851 
852 	(*new_smc)->clcsock = new_clcsock;
853 out:
854 	return rc;
855 }
856 
857 /* add a just created sock to the accept queue of the listen sock as
858  * candidate for a following socket accept call from user space
859  */
860 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
861 {
862 	struct smc_sock *par = smc_sk(parent);
863 
864 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
865 	spin_lock(&par->accept_q_lock);
866 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
867 	spin_unlock(&par->accept_q_lock);
868 	sk_acceptq_added(parent);
869 }
870 
871 /* remove a socket from the accept queue of its parental listening socket */
872 static void smc_accept_unlink(struct sock *sk)
873 {
874 	struct smc_sock *par = smc_sk(sk)->listen_smc;
875 
876 	spin_lock(&par->accept_q_lock);
877 	list_del_init(&smc_sk(sk)->accept_q);
878 	spin_unlock(&par->accept_q_lock);
879 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
880 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
881 }
882 
883 /* remove a sock from the accept queue to bind it to a new socket created
884  * for a socket accept call from user space
885  */
886 struct sock *smc_accept_dequeue(struct sock *parent,
887 				struct socket *new_sock)
888 {
889 	struct smc_sock *isk, *n;
890 	struct sock *new_sk;
891 
892 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
893 		new_sk = (struct sock *)isk;
894 
895 		smc_accept_unlink(new_sk);
896 		if (new_sk->sk_state == SMC_CLOSED) {
897 			if (isk->clcsock) {
898 				sock_release(isk->clcsock);
899 				isk->clcsock = NULL;
900 			}
901 			new_sk->sk_prot->unhash(new_sk);
902 			sock_put(new_sk); /* final */
903 			continue;
904 		}
905 		if (new_sock)
906 			sock_graft(new_sk, new_sock);
907 		return new_sk;
908 	}
909 	return NULL;
910 }
911 
912 /* clean up for a created but never accepted sock */
913 void smc_close_non_accepted(struct sock *sk)
914 {
915 	struct smc_sock *smc = smc_sk(sk);
916 
917 	lock_sock(sk);
918 	if (!sk->sk_lingertime)
919 		/* wait for peer closing */
920 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
921 	if (!smc->use_fallback) {
922 		smc_close_active(smc);
923 		sock_set_flag(sk, SOCK_DEAD);
924 		sk->sk_shutdown |= SHUTDOWN_MASK;
925 	}
926 	if (smc->clcsock) {
927 		struct socket *tcp;
928 
929 		tcp = smc->clcsock;
930 		smc->clcsock = NULL;
931 		sock_release(tcp);
932 	}
933 	if (smc->use_fallback) {
934 		sock_put(sk); /* passive closing */
935 		sk->sk_state = SMC_CLOSED;
936 	} else {
937 		if (sk->sk_state == SMC_CLOSED)
938 			smc_conn_free(&smc->conn);
939 	}
940 	release_sock(sk);
941 	sk->sk_prot->unhash(sk);
942 	sock_put(sk); /* final sock_put */
943 }
944 
945 static int smc_serv_conf_first_link(struct smc_sock *smc)
946 {
947 	struct net *net = sock_net(smc->clcsock->sk);
948 	struct smc_link_group *lgr = smc->conn.lgr;
949 	struct smc_link *link;
950 	int rest;
951 	int rc;
952 
953 	link = &lgr->lnk[SMC_SINGLE_LINK];
954 
955 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
956 		return SMC_CLC_DECL_ERR_REGRMB;
957 
958 	/* send CONFIRM LINK request to client over the RoCE fabric */
959 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
960 	if (rc < 0)
961 		return SMC_CLC_DECL_TIMEOUT_CL;
962 
963 	/* receive CONFIRM LINK response from client over the RoCE fabric */
964 	rest = wait_for_completion_interruptible_timeout(
965 		&link->llc_confirm_resp,
966 		SMC_LLC_WAIT_FIRST_TIME);
967 	if (rest <= 0) {
968 		struct smc_clc_msg_decline dclc;
969 
970 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
971 				      SMC_CLC_DECLINE);
972 		return rc;
973 	}
974 
975 	if (link->llc_confirm_resp_rc)
976 		return SMC_CLC_DECL_RMBE_EC;
977 
978 	/* send ADD LINK request to client over the RoCE fabric */
979 	rc = smc_llc_send_add_link(link,
980 				   link->smcibdev->mac[link->ibport - 1],
981 				   link->gid, SMC_LLC_REQ);
982 	if (rc < 0)
983 		return SMC_CLC_DECL_TIMEOUT_AL;
984 
985 	/* receive ADD LINK response from client over the RoCE fabric */
986 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
987 							 SMC_LLC_WAIT_TIME);
988 	if (rest <= 0) {
989 		struct smc_clc_msg_decline dclc;
990 
991 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
992 				      SMC_CLC_DECLINE);
993 		return rc;
994 	}
995 
996 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
997 
998 	return 0;
999 }
1000 
1001 /* listen worker: finish */
1002 static void smc_listen_out(struct smc_sock *new_smc)
1003 {
1004 	struct smc_sock *lsmc = new_smc->listen_smc;
1005 	struct sock *newsmcsk = &new_smc->sk;
1006 
1007 	lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1008 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1009 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1010 	} else { /* no longer listening */
1011 		smc_close_non_accepted(newsmcsk);
1012 	}
1013 	release_sock(&lsmc->sk);
1014 
1015 	/* Wake up accept */
1016 	lsmc->sk.sk_data_ready(&lsmc->sk);
1017 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1018 }
1019 
1020 /* listen worker: finish in state connected */
1021 static void smc_listen_out_connected(struct smc_sock *new_smc)
1022 {
1023 	struct sock *newsmcsk = &new_smc->sk;
1024 
1025 	sk_refcnt_debug_inc(newsmcsk);
1026 	if (newsmcsk->sk_state == SMC_INIT)
1027 		newsmcsk->sk_state = SMC_ACTIVE;
1028 
1029 	smc_listen_out(new_smc);
1030 }
1031 
1032 /* listen worker: finish in error state */
1033 static void smc_listen_out_err(struct smc_sock *new_smc)
1034 {
1035 	struct sock *newsmcsk = &new_smc->sk;
1036 
1037 	if (newsmcsk->sk_state == SMC_INIT)
1038 		sock_put(&new_smc->sk); /* passive closing */
1039 	newsmcsk->sk_state = SMC_CLOSED;
1040 	smc_conn_free(&new_smc->conn);
1041 
1042 	smc_listen_out(new_smc);
1043 }
1044 
1045 /* listen worker: decline and fall back if possible */
1046 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1047 			       int local_contact)
1048 {
1049 	/* RDMA setup failed, switch back to TCP */
1050 	if (local_contact == SMC_FIRST_CONTACT)
1051 		smc_lgr_forget(new_smc->conn.lgr);
1052 	if (reason_code < 0) { /* error, no fallback possible */
1053 		smc_listen_out_err(new_smc);
1054 		return;
1055 	}
1056 	smc_conn_free(&new_smc->conn);
1057 	new_smc->use_fallback = true;
1058 	new_smc->fallback_rsn = reason_code;
1059 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1060 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1061 			smc_listen_out_err(new_smc);
1062 			return;
1063 		}
1064 	}
1065 	smc_listen_out_connected(new_smc);
1066 }
1067 
1068 /* listen worker: check prefixes */
1069 static int smc_listen_rdma_check(struct smc_sock *new_smc,
1070 				 struct smc_clc_msg_proposal *pclc)
1071 {
1072 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1073 	struct socket *newclcsock = new_smc->clcsock;
1074 
1075 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1076 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1077 		return SMC_CLC_DECL_CNFERR;
1078 
1079 	return 0;
1080 }
1081 
1082 /* listen worker: initialize connection and buffers */
1083 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1084 				struct smc_clc_msg_proposal *pclc,
1085 				struct smc_ib_device *ibdev, u8 ibport,
1086 				int *local_contact)
1087 {
1088 	/* allocate connection / link group */
1089 	*local_contact = smc_conn_create(new_smc, false, 0, ibdev, ibport, 0,
1090 					 &pclc->lcl, NULL, 0);
1091 	if (*local_contact < 0) {
1092 		if (*local_contact == -ENOMEM)
1093 			return SMC_CLC_DECL_MEM;/* insufficient memory*/
1094 		return SMC_CLC_DECL_INTERR; /* other error */
1095 	}
1096 
1097 	/* create send buffer and rmb */
1098 	if (smc_buf_create(new_smc, false))
1099 		return SMC_CLC_DECL_MEM;
1100 
1101 	return 0;
1102 }
1103 
1104 /* listen worker: initialize connection and buffers for SMC-D */
1105 static int smc_listen_ism_init(struct smc_sock *new_smc,
1106 			       struct smc_clc_msg_proposal *pclc,
1107 			       struct smcd_dev *ismdev,
1108 			       int *local_contact)
1109 {
1110 	struct smc_clc_msg_smcd *pclc_smcd;
1111 
1112 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1113 	*local_contact = smc_conn_create(new_smc, true, 0, NULL, 0, 0, NULL,
1114 					 ismdev, pclc_smcd->gid);
1115 	if (*local_contact < 0) {
1116 		if (*local_contact == -ENOMEM)
1117 			return SMC_CLC_DECL_MEM;/* insufficient memory*/
1118 		return SMC_CLC_DECL_INTERR; /* other error */
1119 	}
1120 
1121 	/* Check if peer can be reached via ISM device */
1122 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1123 			    new_smc->conn.lgr->vlan_id,
1124 			    new_smc->conn.lgr->smcd)) {
1125 		if (*local_contact == SMC_FIRST_CONTACT)
1126 			smc_lgr_forget(new_smc->conn.lgr);
1127 		smc_conn_free(&new_smc->conn);
1128 		return SMC_CLC_DECL_CNFERR;
1129 	}
1130 
1131 	/* Create send and receive buffers */
1132 	if (smc_buf_create(new_smc, true)) {
1133 		if (*local_contact == SMC_FIRST_CONTACT)
1134 			smc_lgr_forget(new_smc->conn.lgr);
1135 		smc_conn_free(&new_smc->conn);
1136 		return SMC_CLC_DECL_MEM;
1137 	}
1138 
1139 	return 0;
1140 }
1141 
1142 /* listen worker: register buffers */
1143 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1144 {
1145 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1146 
1147 	if (local_contact != SMC_FIRST_CONTACT) {
1148 		if (!new_smc->conn.rmb_desc->reused) {
1149 			if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true))
1150 				return SMC_CLC_DECL_ERR_REGRMB;
1151 		}
1152 	}
1153 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1154 
1155 	return 0;
1156 }
1157 
1158 /* listen worker: finish RDMA setup */
1159 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1160 				  struct smc_clc_msg_accept_confirm *cclc,
1161 				  int local_contact)
1162 {
1163 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1164 	int reason_code = 0;
1165 
1166 	if (local_contact == SMC_FIRST_CONTACT)
1167 		smc_link_save_peer_info(link, cclc);
1168 
1169 	if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) {
1170 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1171 		goto decline;
1172 	}
1173 
1174 	if (local_contact == SMC_FIRST_CONTACT) {
1175 		if (smc_ib_ready_link(link)) {
1176 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1177 			goto decline;
1178 		}
1179 		/* QP confirmation over RoCE fabric */
1180 		reason_code = smc_serv_conf_first_link(new_smc);
1181 		if (reason_code)
1182 			goto decline;
1183 	}
1184 	return 0;
1185 
1186 decline:
1187 	mutex_unlock(&smc_create_lgr_pending);
1188 	smc_listen_decline(new_smc, reason_code, local_contact);
1189 	return reason_code;
1190 }
1191 
1192 /* setup for RDMA connection of server */
1193 static void smc_listen_work(struct work_struct *work)
1194 {
1195 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1196 						smc_listen_work);
1197 	struct socket *newclcsock = new_smc->clcsock;
1198 	struct smc_clc_msg_accept_confirm cclc;
1199 	struct smc_clc_msg_proposal *pclc;
1200 	struct smc_ib_device *ibdev;
1201 	bool ism_supported = false;
1202 	struct smcd_dev *ismdev;
1203 	u8 buf[SMC_CLC_MAX_LEN];
1204 	int local_contact = 0;
1205 	unsigned short vlan;
1206 	int reason_code = 0;
1207 	int rc = 0;
1208 	u8 ibport;
1209 
1210 	if (new_smc->use_fallback) {
1211 		smc_listen_out_connected(new_smc);
1212 		return;
1213 	}
1214 
1215 	/* check if peer is smc capable */
1216 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1217 		new_smc->use_fallback = true;
1218 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1219 		smc_listen_out_connected(new_smc);
1220 		return;
1221 	}
1222 
1223 	/* do inband token exchange -
1224 	 * wait for and receive SMC Proposal CLC message
1225 	 */
1226 	pclc = (struct smc_clc_msg_proposal *)&buf;
1227 	reason_code = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1228 				       SMC_CLC_PROPOSAL);
1229 	if (reason_code) {
1230 		smc_listen_decline(new_smc, reason_code, 0);
1231 		return;
1232 	}
1233 
1234 	/* IPSec connections opt out of SMC-R optimizations */
1235 	if (using_ipsec(new_smc)) {
1236 		smc_listen_decline(new_smc, SMC_CLC_DECL_IPSEC, 0);
1237 		return;
1238 	}
1239 
1240 	mutex_lock(&smc_create_lgr_pending);
1241 	smc_close_init(new_smc);
1242 	smc_rx_init(new_smc);
1243 	smc_tx_init(new_smc);
1244 
1245 	/* check if ISM is available */
1246 	if ((pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) &&
1247 	    !smc_check_ism(new_smc, &ismdev) &&
1248 	    !smc_listen_ism_init(new_smc, pclc, ismdev, &local_contact)) {
1249 		ism_supported = true;
1250 	}
1251 
1252 	/* check if RDMA is available */
1253 	if (!ism_supported &&
1254 	    ((pclc->hdr.path != SMC_TYPE_R && pclc->hdr.path != SMC_TYPE_B) ||
1255 	     smc_vlan_by_tcpsk(new_smc->clcsock, &vlan) ||
1256 	     smc_check_rdma(new_smc, &ibdev, &ibport, vlan, NULL) ||
1257 	     smc_listen_rdma_check(new_smc, pclc) ||
1258 	     smc_listen_rdma_init(new_smc, pclc, ibdev, ibport,
1259 				  &local_contact) ||
1260 	     smc_listen_rdma_reg(new_smc, local_contact))) {
1261 		/* SMC not supported, decline */
1262 		mutex_unlock(&smc_create_lgr_pending);
1263 		smc_listen_decline(new_smc, SMC_CLC_DECL_MODEUNSUPP,
1264 				   local_contact);
1265 		return;
1266 	}
1267 
1268 	/* send SMC Accept CLC message */
1269 	rc = smc_clc_send_accept(new_smc, local_contact);
1270 	if (rc) {
1271 		mutex_unlock(&smc_create_lgr_pending);
1272 		smc_listen_decline(new_smc, rc, local_contact);
1273 		return;
1274 	}
1275 
1276 	/* receive SMC Confirm CLC message */
1277 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1278 				       SMC_CLC_CONFIRM);
1279 	if (reason_code) {
1280 		mutex_unlock(&smc_create_lgr_pending);
1281 		smc_listen_decline(new_smc, reason_code, local_contact);
1282 		return;
1283 	}
1284 
1285 	/* finish worker */
1286 	if (!ism_supported) {
1287 		if (smc_listen_rdma_finish(new_smc, &cclc, local_contact))
1288 			return;
1289 	}
1290 	smc_conn_save_peer_info(new_smc, &cclc);
1291 	mutex_unlock(&smc_create_lgr_pending);
1292 	smc_listen_out_connected(new_smc);
1293 }
1294 
1295 static void smc_tcp_listen_work(struct work_struct *work)
1296 {
1297 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1298 					     tcp_listen_work);
1299 	struct sock *lsk = &lsmc->sk;
1300 	struct smc_sock *new_smc;
1301 	int rc = 0;
1302 
1303 	lock_sock(lsk);
1304 	while (lsk->sk_state == SMC_LISTEN) {
1305 		rc = smc_clcsock_accept(lsmc, &new_smc);
1306 		if (rc)
1307 			goto out;
1308 		if (!new_smc)
1309 			continue;
1310 
1311 		new_smc->listen_smc = lsmc;
1312 		new_smc->use_fallback = lsmc->use_fallback;
1313 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1314 		sock_hold(lsk); /* sock_put in smc_listen_work */
1315 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1316 		smc_copy_sock_settings_to_smc(new_smc);
1317 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1318 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1319 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1320 		if (!schedule_work(&new_smc->smc_listen_work))
1321 			sock_put(&new_smc->sk);
1322 	}
1323 
1324 out:
1325 	release_sock(lsk);
1326 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1327 }
1328 
1329 static int smc_listen(struct socket *sock, int backlog)
1330 {
1331 	struct sock *sk = sock->sk;
1332 	struct smc_sock *smc;
1333 	int rc;
1334 
1335 	smc = smc_sk(sk);
1336 	lock_sock(sk);
1337 
1338 	rc = -EINVAL;
1339 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
1340 		goto out;
1341 
1342 	rc = 0;
1343 	if (sk->sk_state == SMC_LISTEN) {
1344 		sk->sk_max_ack_backlog = backlog;
1345 		goto out;
1346 	}
1347 	/* some socket options are handled in core, so we could not apply
1348 	 * them to the clc socket -- copy smc socket options to clc socket
1349 	 */
1350 	smc_copy_sock_settings_to_clc(smc);
1351 	if (!smc->use_fallback)
1352 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1353 
1354 	rc = kernel_listen(smc->clcsock, backlog);
1355 	if (rc)
1356 		goto out;
1357 	sk->sk_max_ack_backlog = backlog;
1358 	sk->sk_ack_backlog = 0;
1359 	sk->sk_state = SMC_LISTEN;
1360 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
1361 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1362 	if (!schedule_work(&smc->tcp_listen_work))
1363 		sock_put(sk);
1364 
1365 out:
1366 	release_sock(sk);
1367 	return rc;
1368 }
1369 
1370 static int smc_accept(struct socket *sock, struct socket *new_sock,
1371 		      int flags, bool kern)
1372 {
1373 	struct sock *sk = sock->sk, *nsk;
1374 	DECLARE_WAITQUEUE(wait, current);
1375 	struct smc_sock *lsmc;
1376 	long timeo;
1377 	int rc = 0;
1378 
1379 	lsmc = smc_sk(sk);
1380 	sock_hold(sk); /* sock_put below */
1381 	lock_sock(sk);
1382 
1383 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1384 		rc = -EINVAL;
1385 		release_sock(sk);
1386 		goto out;
1387 	}
1388 
1389 	/* Wait for an incoming connection */
1390 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1391 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1392 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1393 		set_current_state(TASK_INTERRUPTIBLE);
1394 		if (!timeo) {
1395 			rc = -EAGAIN;
1396 			break;
1397 		}
1398 		release_sock(sk);
1399 		timeo = schedule_timeout(timeo);
1400 		/* wakeup by sk_data_ready in smc_listen_work() */
1401 		sched_annotate_sleep();
1402 		lock_sock(sk);
1403 		if (signal_pending(current)) {
1404 			rc = sock_intr_errno(timeo);
1405 			break;
1406 		}
1407 	}
1408 	set_current_state(TASK_RUNNING);
1409 	remove_wait_queue(sk_sleep(sk), &wait);
1410 
1411 	if (!rc)
1412 		rc = sock_error(nsk);
1413 	release_sock(sk);
1414 	if (rc)
1415 		goto out;
1416 
1417 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1418 		/* wait till data arrives on the socket */
1419 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1420 								MSEC_PER_SEC);
1421 		if (smc_sk(nsk)->use_fallback) {
1422 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1423 
1424 			lock_sock(clcsk);
1425 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1426 				sk_wait_data(clcsk, &timeo, NULL);
1427 			release_sock(clcsk);
1428 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1429 			lock_sock(nsk);
1430 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1431 			release_sock(nsk);
1432 		}
1433 	}
1434 
1435 out:
1436 	sock_put(sk); /* sock_hold above */
1437 	return rc;
1438 }
1439 
1440 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1441 		       int peer)
1442 {
1443 	struct smc_sock *smc;
1444 
1445 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1446 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1447 		return -ENOTCONN;
1448 
1449 	smc = smc_sk(sock->sk);
1450 
1451 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1452 }
1453 
1454 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1455 {
1456 	struct sock *sk = sock->sk;
1457 	struct smc_sock *smc;
1458 	int rc = -EPIPE;
1459 
1460 	smc = smc_sk(sk);
1461 	lock_sock(sk);
1462 	if ((sk->sk_state != SMC_ACTIVE) &&
1463 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1464 	    (sk->sk_state != SMC_INIT))
1465 		goto out;
1466 
1467 	if (msg->msg_flags & MSG_FASTOPEN) {
1468 		if (sk->sk_state == SMC_INIT) {
1469 			smc->use_fallback = true;
1470 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1471 		} else {
1472 			rc = -EINVAL;
1473 			goto out;
1474 		}
1475 	}
1476 
1477 	if (smc->use_fallback)
1478 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1479 	else
1480 		rc = smc_tx_sendmsg(smc, msg, len);
1481 out:
1482 	release_sock(sk);
1483 	return rc;
1484 }
1485 
1486 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1487 		       int flags)
1488 {
1489 	struct sock *sk = sock->sk;
1490 	struct smc_sock *smc;
1491 	int rc = -ENOTCONN;
1492 
1493 	smc = smc_sk(sk);
1494 	lock_sock(sk);
1495 	if ((sk->sk_state == SMC_INIT) ||
1496 	    (sk->sk_state == SMC_LISTEN) ||
1497 	    (sk->sk_state == SMC_CLOSED))
1498 		goto out;
1499 
1500 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1501 		rc = 0;
1502 		goto out;
1503 	}
1504 
1505 	if (smc->use_fallback) {
1506 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1507 	} else {
1508 		msg->msg_namelen = 0;
1509 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1510 	}
1511 
1512 out:
1513 	release_sock(sk);
1514 	return rc;
1515 }
1516 
1517 static __poll_t smc_accept_poll(struct sock *parent)
1518 {
1519 	struct smc_sock *isk = smc_sk(parent);
1520 	__poll_t mask = 0;
1521 
1522 	spin_lock(&isk->accept_q_lock);
1523 	if (!list_empty(&isk->accept_q))
1524 		mask = EPOLLIN | EPOLLRDNORM;
1525 	spin_unlock(&isk->accept_q_lock);
1526 
1527 	return mask;
1528 }
1529 
1530 static __poll_t smc_poll(struct file *file, struct socket *sock,
1531 			     poll_table *wait)
1532 {
1533 	struct sock *sk = sock->sk;
1534 	__poll_t mask = 0;
1535 	struct smc_sock *smc;
1536 
1537 	if (!sk)
1538 		return EPOLLNVAL;
1539 
1540 	smc = smc_sk(sock->sk);
1541 	if (smc->use_fallback) {
1542 		/* delegate to CLC child sock */
1543 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1544 		sk->sk_err = smc->clcsock->sk->sk_err;
1545 		if (sk->sk_err)
1546 			mask |= EPOLLERR;
1547 	} else {
1548 		if (sk->sk_state != SMC_CLOSED)
1549 			sock_poll_wait(file, sock, wait);
1550 		if (sk->sk_err)
1551 			mask |= EPOLLERR;
1552 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1553 		    (sk->sk_state == SMC_CLOSED))
1554 			mask |= EPOLLHUP;
1555 		if (sk->sk_state == SMC_LISTEN) {
1556 			/* woken up by sk_data_ready in smc_listen_work() */
1557 			mask = smc_accept_poll(sk);
1558 		} else {
1559 			if (atomic_read(&smc->conn.sndbuf_space) ||
1560 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1561 				mask |= EPOLLOUT | EPOLLWRNORM;
1562 			} else {
1563 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1564 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1565 			}
1566 			if (atomic_read(&smc->conn.bytes_to_rcv))
1567 				mask |= EPOLLIN | EPOLLRDNORM;
1568 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1569 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1570 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1571 				mask |= EPOLLIN;
1572 			if (smc->conn.urg_state == SMC_URG_VALID)
1573 				mask |= EPOLLPRI;
1574 		}
1575 	}
1576 
1577 	return mask;
1578 }
1579 
1580 static int smc_shutdown(struct socket *sock, int how)
1581 {
1582 	struct sock *sk = sock->sk;
1583 	struct smc_sock *smc;
1584 	int rc = -EINVAL;
1585 	int rc1 = 0;
1586 
1587 	smc = smc_sk(sk);
1588 
1589 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1590 		return rc;
1591 
1592 	lock_sock(sk);
1593 
1594 	rc = -ENOTCONN;
1595 	if ((sk->sk_state != SMC_ACTIVE) &&
1596 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1597 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1598 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1599 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1600 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1601 		goto out;
1602 	if (smc->use_fallback) {
1603 		rc = kernel_sock_shutdown(smc->clcsock, how);
1604 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1605 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1606 			sk->sk_state = SMC_CLOSED;
1607 		goto out;
1608 	}
1609 	switch (how) {
1610 	case SHUT_RDWR:		/* shutdown in both directions */
1611 		rc = smc_close_active(smc);
1612 		break;
1613 	case SHUT_WR:
1614 		rc = smc_close_shutdown_write(smc);
1615 		break;
1616 	case SHUT_RD:
1617 		rc = 0;
1618 		/* nothing more to do because peer is not involved */
1619 		break;
1620 	}
1621 	if (smc->clcsock)
1622 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1623 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1624 	sk->sk_shutdown |= how + 1;
1625 
1626 out:
1627 	release_sock(sk);
1628 	return rc ? rc : rc1;
1629 }
1630 
1631 static int smc_setsockopt(struct socket *sock, int level, int optname,
1632 			  char __user *optval, unsigned int optlen)
1633 {
1634 	struct sock *sk = sock->sk;
1635 	struct smc_sock *smc;
1636 	int val, rc;
1637 
1638 	smc = smc_sk(sk);
1639 
1640 	/* generic setsockopts reaching us here always apply to the
1641 	 * CLC socket
1642 	 */
1643 	rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1644 					   optval, optlen);
1645 	if (smc->clcsock->sk->sk_err) {
1646 		sk->sk_err = smc->clcsock->sk->sk_err;
1647 		sk->sk_error_report(sk);
1648 	}
1649 	if (rc)
1650 		return rc;
1651 
1652 	if (optlen < sizeof(int))
1653 		return -EINVAL;
1654 	if (get_user(val, (int __user *)optval))
1655 		return -EFAULT;
1656 
1657 	lock_sock(sk);
1658 	switch (optname) {
1659 	case TCP_ULP:
1660 	case TCP_FASTOPEN:
1661 	case TCP_FASTOPEN_CONNECT:
1662 	case TCP_FASTOPEN_KEY:
1663 	case TCP_FASTOPEN_NO_COOKIE:
1664 		/* option not supported by SMC */
1665 		if (sk->sk_state == SMC_INIT) {
1666 			smc->use_fallback = true;
1667 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1668 		} else {
1669 			if (!smc->use_fallback)
1670 				rc = -EINVAL;
1671 		}
1672 		break;
1673 	case TCP_NODELAY:
1674 		if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
1675 			if (val && !smc->use_fallback)
1676 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1677 						 0);
1678 		}
1679 		break;
1680 	case TCP_CORK:
1681 		if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
1682 			if (!val && !smc->use_fallback)
1683 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1684 						 0);
1685 		}
1686 		break;
1687 	case TCP_DEFER_ACCEPT:
1688 		smc->sockopt_defer_accept = val;
1689 		break;
1690 	default:
1691 		break;
1692 	}
1693 	release_sock(sk);
1694 
1695 	return rc;
1696 }
1697 
1698 static int smc_getsockopt(struct socket *sock, int level, int optname,
1699 			  char __user *optval, int __user *optlen)
1700 {
1701 	struct smc_sock *smc;
1702 
1703 	smc = smc_sk(sock->sk);
1704 	/* socket options apply to the CLC socket */
1705 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1706 					     optval, optlen);
1707 }
1708 
1709 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1710 		     unsigned long arg)
1711 {
1712 	union smc_host_cursor cons, urg;
1713 	struct smc_connection *conn;
1714 	struct smc_sock *smc;
1715 	int answ;
1716 
1717 	smc = smc_sk(sock->sk);
1718 	conn = &smc->conn;
1719 	lock_sock(&smc->sk);
1720 	if (smc->use_fallback) {
1721 		if (!smc->clcsock) {
1722 			release_sock(&smc->sk);
1723 			return -EBADF;
1724 		}
1725 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1726 		release_sock(&smc->sk);
1727 		return answ;
1728 	}
1729 	switch (cmd) {
1730 	case SIOCINQ: /* same as FIONREAD */
1731 		if (smc->sk.sk_state == SMC_LISTEN) {
1732 			release_sock(&smc->sk);
1733 			return -EINVAL;
1734 		}
1735 		if (smc->sk.sk_state == SMC_INIT ||
1736 		    smc->sk.sk_state == SMC_CLOSED)
1737 			answ = 0;
1738 		else
1739 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1740 		break;
1741 	case SIOCOUTQ:
1742 		/* output queue size (not send + not acked) */
1743 		if (smc->sk.sk_state == SMC_LISTEN) {
1744 			release_sock(&smc->sk);
1745 			return -EINVAL;
1746 		}
1747 		if (smc->sk.sk_state == SMC_INIT ||
1748 		    smc->sk.sk_state == SMC_CLOSED)
1749 			answ = 0;
1750 		else
1751 			answ = smc->conn.sndbuf_desc->len -
1752 					atomic_read(&smc->conn.sndbuf_space);
1753 		break;
1754 	case SIOCOUTQNSD:
1755 		/* output queue size (not send only) */
1756 		if (smc->sk.sk_state == SMC_LISTEN) {
1757 			release_sock(&smc->sk);
1758 			return -EINVAL;
1759 		}
1760 		if (smc->sk.sk_state == SMC_INIT ||
1761 		    smc->sk.sk_state == SMC_CLOSED)
1762 			answ = 0;
1763 		else
1764 			answ = smc_tx_prepared_sends(&smc->conn);
1765 		break;
1766 	case SIOCATMARK:
1767 		if (smc->sk.sk_state == SMC_LISTEN) {
1768 			release_sock(&smc->sk);
1769 			return -EINVAL;
1770 		}
1771 		if (smc->sk.sk_state == SMC_INIT ||
1772 		    smc->sk.sk_state == SMC_CLOSED) {
1773 			answ = 0;
1774 		} else {
1775 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1776 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1777 			answ = smc_curs_diff(conn->rmb_desc->len,
1778 					     &cons, &urg) == 1;
1779 		}
1780 		break;
1781 	default:
1782 		release_sock(&smc->sk);
1783 		return -ENOIOCTLCMD;
1784 	}
1785 	release_sock(&smc->sk);
1786 
1787 	return put_user(answ, (int __user *)arg);
1788 }
1789 
1790 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1791 			    int offset, size_t size, int flags)
1792 {
1793 	struct sock *sk = sock->sk;
1794 	struct smc_sock *smc;
1795 	int rc = -EPIPE;
1796 
1797 	smc = smc_sk(sk);
1798 	lock_sock(sk);
1799 	if (sk->sk_state != SMC_ACTIVE) {
1800 		release_sock(sk);
1801 		goto out;
1802 	}
1803 	release_sock(sk);
1804 	if (smc->use_fallback)
1805 		rc = kernel_sendpage(smc->clcsock, page, offset,
1806 				     size, flags);
1807 	else
1808 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1809 
1810 out:
1811 	return rc;
1812 }
1813 
1814 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1815  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1816  * updates till whenever a respective page has been fully processed.
1817  * Note that subsequent recv() calls have to wait till all splice() processing
1818  * completed.
1819  */
1820 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1821 			       struct pipe_inode_info *pipe, size_t len,
1822 			       unsigned int flags)
1823 {
1824 	struct sock *sk = sock->sk;
1825 	struct smc_sock *smc;
1826 	int rc = -ENOTCONN;
1827 
1828 	smc = smc_sk(sk);
1829 	lock_sock(sk);
1830 
1831 	if (sk->sk_state == SMC_INIT ||
1832 	    sk->sk_state == SMC_LISTEN ||
1833 	    sk->sk_state == SMC_CLOSED)
1834 		goto out;
1835 
1836 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1837 		rc = 0;
1838 		goto out;
1839 	}
1840 
1841 	if (smc->use_fallback) {
1842 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1843 						    pipe, len, flags);
1844 	} else {
1845 		if (*ppos) {
1846 			rc = -ESPIPE;
1847 			goto out;
1848 		}
1849 		if (flags & SPLICE_F_NONBLOCK)
1850 			flags = MSG_DONTWAIT;
1851 		else
1852 			flags = 0;
1853 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1854 	}
1855 out:
1856 	release_sock(sk);
1857 
1858 	return rc;
1859 }
1860 
1861 /* must look like tcp */
1862 static const struct proto_ops smc_sock_ops = {
1863 	.family		= PF_SMC,
1864 	.owner		= THIS_MODULE,
1865 	.release	= smc_release,
1866 	.bind		= smc_bind,
1867 	.connect	= smc_connect,
1868 	.socketpair	= sock_no_socketpair,
1869 	.accept		= smc_accept,
1870 	.getname	= smc_getname,
1871 	.poll		= smc_poll,
1872 	.ioctl		= smc_ioctl,
1873 	.listen		= smc_listen,
1874 	.shutdown	= smc_shutdown,
1875 	.setsockopt	= smc_setsockopt,
1876 	.getsockopt	= smc_getsockopt,
1877 	.sendmsg	= smc_sendmsg,
1878 	.recvmsg	= smc_recvmsg,
1879 	.mmap		= sock_no_mmap,
1880 	.sendpage	= smc_sendpage,
1881 	.splice_read	= smc_splice_read,
1882 };
1883 
1884 static int smc_create(struct net *net, struct socket *sock, int protocol,
1885 		      int kern)
1886 {
1887 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1888 	struct smc_sock *smc;
1889 	struct sock *sk;
1890 	int rc;
1891 
1892 	rc = -ESOCKTNOSUPPORT;
1893 	if (sock->type != SOCK_STREAM)
1894 		goto out;
1895 
1896 	rc = -EPROTONOSUPPORT;
1897 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1898 		goto out;
1899 
1900 	rc = -ENOBUFS;
1901 	sock->ops = &smc_sock_ops;
1902 	sk = smc_sock_alloc(net, sock, protocol);
1903 	if (!sk)
1904 		goto out;
1905 
1906 	/* create internal TCP socket for CLC handshake and fallback */
1907 	smc = smc_sk(sk);
1908 	smc->use_fallback = false; /* assume rdma capability first */
1909 	smc->fallback_rsn = 0;
1910 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1911 			      &smc->clcsock);
1912 	if (rc) {
1913 		sk_common_release(sk);
1914 		goto out;
1915 	}
1916 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1917 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1918 
1919 out:
1920 	return rc;
1921 }
1922 
1923 static const struct net_proto_family smc_sock_family_ops = {
1924 	.family	= PF_SMC,
1925 	.owner	= THIS_MODULE,
1926 	.create	= smc_create,
1927 };
1928 
1929 static int __init smc_init(void)
1930 {
1931 	int rc;
1932 
1933 	rc = smc_pnet_init();
1934 	if (rc)
1935 		return rc;
1936 
1937 	rc = smc_llc_init();
1938 	if (rc) {
1939 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1940 		goto out_pnet;
1941 	}
1942 
1943 	rc = smc_cdc_init();
1944 	if (rc) {
1945 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1946 		goto out_pnet;
1947 	}
1948 
1949 	rc = proto_register(&smc_proto, 1);
1950 	if (rc) {
1951 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
1952 		goto out_pnet;
1953 	}
1954 
1955 	rc = proto_register(&smc_proto6, 1);
1956 	if (rc) {
1957 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
1958 		goto out_proto;
1959 	}
1960 
1961 	rc = sock_register(&smc_sock_family_ops);
1962 	if (rc) {
1963 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
1964 		goto out_proto6;
1965 	}
1966 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1967 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
1968 
1969 	rc = smc_ib_register_client();
1970 	if (rc) {
1971 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
1972 		goto out_sock;
1973 	}
1974 
1975 	static_branch_enable(&tcp_have_smc);
1976 	return 0;
1977 
1978 out_sock:
1979 	sock_unregister(PF_SMC);
1980 out_proto6:
1981 	proto_unregister(&smc_proto6);
1982 out_proto:
1983 	proto_unregister(&smc_proto);
1984 out_pnet:
1985 	smc_pnet_exit();
1986 	return rc;
1987 }
1988 
1989 static void __exit smc_exit(void)
1990 {
1991 	smc_core_exit();
1992 	static_branch_disable(&tcp_have_smc);
1993 	smc_ib_unregister_client();
1994 	sock_unregister(PF_SMC);
1995 	proto_unregister(&smc_proto6);
1996 	proto_unregister(&smc_proto);
1997 	smc_pnet_exit();
1998 }
1999 
2000 module_init(smc_init);
2001 module_exit(smc_exit);
2002 
2003 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2004 MODULE_DESCRIPTION("smc socket address family");
2005 MODULE_LICENSE("GPL");
2006 MODULE_ALIAS_NETPROTO(PF_SMC);
2007