xref: /openbmc/linux/net/smc/af_smc.c (revision 06d5d6b7f9948a89543e1160ef852d57892c750d)
1 /*
2  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
3  *
4  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
5  *  applies to SOCK_STREAM sockets only
6  *  offers an alternative communication option for TCP-protocol sockets
7  *  applicable with RoCE-cards only
8  *
9  *  Initial restrictions:
10  *    - support for alternate links postponed
11  *
12  *  Copyright IBM Corp. 2016, 2018
13  *
14  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
15  *              based on prototype from Frank Blaschka
16  */
17 
18 #define KMSG_COMPONENT "smc"
19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
20 
21 #include <linux/module.h>
22 #include <linux/socket.h>
23 #include <linux/workqueue.h>
24 #include <linux/in.h>
25 #include <linux/sched/signal.h>
26 #include <linux/if_vlan.h>
27 
28 #include <net/sock.h>
29 #include <net/tcp.h>
30 #include <net/smc.h>
31 #include <asm/ioctls.h>
32 
33 #include <net/net_namespace.h>
34 #include <net/netns/generic.h>
35 #include "smc_netns.h"
36 
37 #include "smc.h"
38 #include "smc_clc.h"
39 #include "smc_llc.h"
40 #include "smc_cdc.h"
41 #include "smc_core.h"
42 #include "smc_ib.h"
43 #include "smc_ism.h"
44 #include "smc_pnet.h"
45 #include "smc_tx.h"
46 #include "smc_rx.h"
47 #include "smc_close.h"
48 
49 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
50 						 * creation on server
51 						 */
52 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
53 						 * creation on client
54 						 */
55 
56 static void smc_tcp_listen_work(struct work_struct *);
57 static void smc_connect_work(struct work_struct *);
58 
59 static void smc_set_keepalive(struct sock *sk, int val)
60 {
61 	struct smc_sock *smc = smc_sk(sk);
62 
63 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
64 }
65 
66 static struct smc_hashinfo smc_v4_hashinfo = {
67 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
68 };
69 
70 static struct smc_hashinfo smc_v6_hashinfo = {
71 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
72 };
73 
74 int smc_hash_sk(struct sock *sk)
75 {
76 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
77 	struct hlist_head *head;
78 
79 	head = &h->ht;
80 
81 	write_lock_bh(&h->lock);
82 	sk_add_node(sk, head);
83 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
84 	write_unlock_bh(&h->lock);
85 
86 	return 0;
87 }
88 EXPORT_SYMBOL_GPL(smc_hash_sk);
89 
90 void smc_unhash_sk(struct sock *sk)
91 {
92 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
93 
94 	write_lock_bh(&h->lock);
95 	if (sk_del_node_init(sk))
96 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
97 	write_unlock_bh(&h->lock);
98 }
99 EXPORT_SYMBOL_GPL(smc_unhash_sk);
100 
101 struct proto smc_proto = {
102 	.name		= "SMC",
103 	.owner		= THIS_MODULE,
104 	.keepalive	= smc_set_keepalive,
105 	.hash		= smc_hash_sk,
106 	.unhash		= smc_unhash_sk,
107 	.obj_size	= sizeof(struct smc_sock),
108 	.h.smc_hash	= &smc_v4_hashinfo,
109 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
110 };
111 EXPORT_SYMBOL_GPL(smc_proto);
112 
113 struct proto smc_proto6 = {
114 	.name		= "SMC6",
115 	.owner		= THIS_MODULE,
116 	.keepalive	= smc_set_keepalive,
117 	.hash		= smc_hash_sk,
118 	.unhash		= smc_unhash_sk,
119 	.obj_size	= sizeof(struct smc_sock),
120 	.h.smc_hash	= &smc_v6_hashinfo,
121 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
122 };
123 EXPORT_SYMBOL_GPL(smc_proto6);
124 
125 static int smc_release(struct socket *sock)
126 {
127 	struct sock *sk = sock->sk;
128 	struct smc_sock *smc;
129 	int rc = 0;
130 
131 	if (!sk)
132 		goto out;
133 
134 	smc = smc_sk(sk);
135 
136 	/* cleanup for a dangling non-blocking connect */
137 	if (smc->connect_info && sk->sk_state == SMC_INIT)
138 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
139 	flush_work(&smc->connect_work);
140 	kfree(smc->connect_info);
141 	smc->connect_info = NULL;
142 
143 	if (sk->sk_state == SMC_LISTEN)
144 		/* smc_close_non_accepted() is called and acquires
145 		 * sock lock for child sockets again
146 		 */
147 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
148 	else
149 		lock_sock(sk);
150 
151 	if (!smc->use_fallback) {
152 		rc = smc_close_active(smc);
153 		sock_set_flag(sk, SOCK_DEAD);
154 		sk->sk_shutdown |= SHUTDOWN_MASK;
155 	} else {
156 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
157 			sock_put(sk); /* passive closing */
158 		if (sk->sk_state == SMC_LISTEN) {
159 			/* wake up clcsock accept */
160 			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
161 		}
162 		sk->sk_state = SMC_CLOSED;
163 		sk->sk_state_change(sk);
164 	}
165 
166 	sk->sk_prot->unhash(sk);
167 
168 	if (sk->sk_state == SMC_CLOSED) {
169 		if (smc->clcsock) {
170 			release_sock(sk);
171 			smc_clcsock_release(smc);
172 			lock_sock(sk);
173 		}
174 		if (!smc->use_fallback)
175 			smc_conn_free(&smc->conn);
176 	}
177 
178 	/* detach socket */
179 	sock_orphan(sk);
180 	sock->sk = NULL;
181 	release_sock(sk);
182 
183 	sock_put(sk); /* final sock_put */
184 out:
185 	return rc;
186 }
187 
188 static void smc_destruct(struct sock *sk)
189 {
190 	if (sk->sk_state != SMC_CLOSED)
191 		return;
192 	if (!sock_flag(sk, SOCK_DEAD))
193 		return;
194 
195 	sk_refcnt_debug_dec(sk);
196 }
197 
198 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
199 				   int protocol)
200 {
201 	struct smc_sock *smc;
202 	struct proto *prot;
203 	struct sock *sk;
204 
205 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
206 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
207 	if (!sk)
208 		return NULL;
209 
210 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
211 	sk->sk_state = SMC_INIT;
212 	sk->sk_destruct = smc_destruct;
213 	sk->sk_protocol = protocol;
214 	smc = smc_sk(sk);
215 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
216 	INIT_WORK(&smc->connect_work, smc_connect_work);
217 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
218 	INIT_LIST_HEAD(&smc->accept_q);
219 	spin_lock_init(&smc->accept_q_lock);
220 	spin_lock_init(&smc->conn.send_lock);
221 	sk->sk_prot->hash(sk);
222 	sk_refcnt_debug_inc(sk);
223 	mutex_init(&smc->clcsock_release_lock);
224 
225 	return sk;
226 }
227 
228 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
229 		    int addr_len)
230 {
231 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
232 	struct sock *sk = sock->sk;
233 	struct smc_sock *smc;
234 	int rc;
235 
236 	smc = smc_sk(sk);
237 
238 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
239 	rc = -EINVAL;
240 	if (addr_len < sizeof(struct sockaddr_in))
241 		goto out;
242 
243 	rc = -EAFNOSUPPORT;
244 	if (addr->sin_family != AF_INET &&
245 	    addr->sin_family != AF_INET6 &&
246 	    addr->sin_family != AF_UNSPEC)
247 		goto out;
248 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
249 	if (addr->sin_family == AF_UNSPEC &&
250 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
251 		goto out;
252 
253 	lock_sock(sk);
254 
255 	/* Check if socket is already active */
256 	rc = -EINVAL;
257 	if (sk->sk_state != SMC_INIT)
258 		goto out_rel;
259 
260 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
261 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
262 
263 out_rel:
264 	release_sock(sk);
265 out:
266 	return rc;
267 }
268 
269 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
270 				   unsigned long mask)
271 {
272 	/* options we don't get control via setsockopt for */
273 	nsk->sk_type = osk->sk_type;
274 	nsk->sk_sndbuf = osk->sk_sndbuf;
275 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
276 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
277 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
278 	nsk->sk_mark = osk->sk_mark;
279 	nsk->sk_priority = osk->sk_priority;
280 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
281 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
282 	nsk->sk_err = osk->sk_err;
283 
284 	nsk->sk_flags &= ~mask;
285 	nsk->sk_flags |= osk->sk_flags & mask;
286 }
287 
288 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
289 			     (1UL << SOCK_KEEPOPEN) | \
290 			     (1UL << SOCK_LINGER) | \
291 			     (1UL << SOCK_BROADCAST) | \
292 			     (1UL << SOCK_TIMESTAMP) | \
293 			     (1UL << SOCK_DBG) | \
294 			     (1UL << SOCK_RCVTSTAMP) | \
295 			     (1UL << SOCK_RCVTSTAMPNS) | \
296 			     (1UL << SOCK_LOCALROUTE) | \
297 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
298 			     (1UL << SOCK_RXQ_OVFL) | \
299 			     (1UL << SOCK_WIFI_STATUS) | \
300 			     (1UL << SOCK_NOFCS) | \
301 			     (1UL << SOCK_FILTER_LOCKED) | \
302 			     (1UL << SOCK_TSTAMP_NEW))
303 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
304  * clc socket (since smc is not called for these options from net/core)
305  */
306 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
307 {
308 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
309 }
310 
311 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
312 			     (1UL << SOCK_KEEPOPEN) | \
313 			     (1UL << SOCK_LINGER) | \
314 			     (1UL << SOCK_DBG))
315 /* copy only settings and flags relevant for smc from clc to smc socket */
316 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
317 {
318 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
319 }
320 
321 /* register a new rmb, send confirm_rkey msg to register with peer */
322 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc,
323 		       bool conf_rkey)
324 {
325 	if (!rmb_desc->wr_reg) {
326 		/* register memory region for new rmb */
327 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
328 			rmb_desc->regerr = 1;
329 			return -EFAULT;
330 		}
331 		rmb_desc->wr_reg = 1;
332 	}
333 	if (!conf_rkey)
334 		return 0;
335 	/* exchange confirm_rkey msg with peer */
336 	if (smc_llc_do_confirm_rkey(link, rmb_desc)) {
337 		rmb_desc->regerr = 1;
338 		return -EFAULT;
339 	}
340 	return 0;
341 }
342 
343 static int smc_clnt_conf_first_link(struct smc_sock *smc)
344 {
345 	struct net *net = sock_net(smc->clcsock->sk);
346 	struct smc_link_group *lgr = smc->conn.lgr;
347 	struct smc_link *link;
348 	int rest;
349 	int rc;
350 
351 	link = &lgr->lnk[SMC_SINGLE_LINK];
352 	/* receive CONFIRM LINK request from server over RoCE fabric */
353 	rest = wait_for_completion_interruptible_timeout(
354 		&link->llc_confirm,
355 		SMC_LLC_WAIT_FIRST_TIME);
356 	if (rest <= 0) {
357 		struct smc_clc_msg_decline dclc;
358 
359 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
360 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
361 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
362 	}
363 
364 	if (link->llc_confirm_rc)
365 		return SMC_CLC_DECL_RMBE_EC;
366 
367 	rc = smc_ib_modify_qp_rts(link);
368 	if (rc)
369 		return SMC_CLC_DECL_ERR_RDYLNK;
370 
371 	smc_wr_remember_qp_attr(link);
372 
373 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
374 		return SMC_CLC_DECL_ERR_REGRMB;
375 
376 	/* send CONFIRM LINK response over RoCE fabric */
377 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
378 	if (rc < 0)
379 		return SMC_CLC_DECL_TIMEOUT_CL;
380 
381 	/* receive ADD LINK request from server over RoCE fabric */
382 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
383 							 SMC_LLC_WAIT_TIME);
384 	if (rest <= 0) {
385 		struct smc_clc_msg_decline dclc;
386 
387 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
388 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
389 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
390 	}
391 
392 	/* send add link reject message, only one link supported for now */
393 	rc = smc_llc_send_add_link(link,
394 				   link->smcibdev->mac[link->ibport - 1],
395 				   link->gid, SMC_LLC_RESP);
396 	if (rc < 0)
397 		return SMC_CLC_DECL_TIMEOUT_AL;
398 
399 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
400 
401 	return 0;
402 }
403 
404 static void smcr_conn_save_peer_info(struct smc_sock *smc,
405 				     struct smc_clc_msg_accept_confirm *clc)
406 {
407 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
408 
409 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
410 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
411 	smc->conn.peer_rmbe_size = bufsize;
412 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
413 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
414 }
415 
416 static void smcd_conn_save_peer_info(struct smc_sock *smc,
417 				     struct smc_clc_msg_accept_confirm *clc)
418 {
419 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
420 
421 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
422 	smc->conn.peer_token = clc->token;
423 	/* msg header takes up space in the buffer */
424 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
425 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
426 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
427 }
428 
429 static void smc_conn_save_peer_info(struct smc_sock *smc,
430 				    struct smc_clc_msg_accept_confirm *clc)
431 {
432 	if (smc->conn.lgr->is_smcd)
433 		smcd_conn_save_peer_info(smc, clc);
434 	else
435 		smcr_conn_save_peer_info(smc, clc);
436 }
437 
438 static void smc_link_save_peer_info(struct smc_link *link,
439 				    struct smc_clc_msg_accept_confirm *clc)
440 {
441 	link->peer_qpn = ntoh24(clc->qpn);
442 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
443 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
444 	link->peer_psn = ntoh24(clc->psn);
445 	link->peer_mtu = clc->qp_mtu;
446 }
447 
448 static void smc_switch_to_fallback(struct smc_sock *smc)
449 {
450 	smc->use_fallback = true;
451 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
452 		smc->clcsock->file = smc->sk.sk_socket->file;
453 		smc->clcsock->file->private_data = smc->clcsock;
454 	}
455 }
456 
457 /* fall back during connect */
458 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
459 {
460 	smc_switch_to_fallback(smc);
461 	smc->fallback_rsn = reason_code;
462 	smc_copy_sock_settings_to_clc(smc);
463 	if (smc->sk.sk_state == SMC_INIT)
464 		smc->sk.sk_state = SMC_ACTIVE;
465 	return 0;
466 }
467 
468 /* decline and fall back during connect */
469 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
470 {
471 	int rc;
472 
473 	if (reason_code < 0) { /* error, fallback is not possible */
474 		if (smc->sk.sk_state == SMC_INIT)
475 			sock_put(&smc->sk); /* passive closing */
476 		return reason_code;
477 	}
478 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
479 		rc = smc_clc_send_decline(smc, reason_code);
480 		if (rc < 0) {
481 			if (smc->sk.sk_state == SMC_INIT)
482 				sock_put(&smc->sk); /* passive closing */
483 			return rc;
484 		}
485 	}
486 	return smc_connect_fallback(smc, reason_code);
487 }
488 
489 /* abort connecting */
490 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
491 			     int local_contact)
492 {
493 	if (local_contact == SMC_FIRST_CONTACT)
494 		smc_lgr_forget(smc->conn.lgr);
495 	if (smc->conn.lgr->is_smcd)
496 		/* there is only one lgr role for SMC-D; use server lock */
497 		mutex_unlock(&smc_server_lgr_pending);
498 	else
499 		mutex_unlock(&smc_client_lgr_pending);
500 
501 	smc_conn_free(&smc->conn);
502 	return reason_code;
503 }
504 
505 /* check if there is a rdma device available for this connection. */
506 /* called for connect and listen */
507 static int smc_check_rdma(struct smc_sock *smc, struct smc_ib_device **ibdev,
508 			  u8 *ibport, unsigned short vlan_id, u8 gid[])
509 {
510 	int reason_code = 0;
511 
512 	/* PNET table look up: search active ib_device and port
513 	 * within same PNETID that also contains the ethernet device
514 	 * used for the internal TCP socket
515 	 */
516 	smc_pnet_find_roce_resource(smc->clcsock->sk, ibdev, ibport, vlan_id,
517 				    gid);
518 	if (!(*ibdev))
519 		reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
520 
521 	return reason_code;
522 }
523 
524 /* check if there is an ISM device available for this connection. */
525 /* called for connect and listen */
526 static int smc_check_ism(struct smc_sock *smc, struct smcd_dev **ismdev)
527 {
528 	/* Find ISM device with same PNETID as connecting interface  */
529 	smc_pnet_find_ism_resource(smc->clcsock->sk, ismdev);
530 	if (!(*ismdev))
531 		return SMC_CLC_DECL_CNFERR; /* configuration error */
532 	return 0;
533 }
534 
535 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
536 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
537 				      struct smcd_dev *ismdev,
538 				      unsigned short vlan_id)
539 {
540 	if (vlan_id && smc_ism_get_vlan(ismdev, vlan_id))
541 		return SMC_CLC_DECL_CNFERR;
542 	return 0;
543 }
544 
545 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
546  * used, the VLAN ID will be registered again during the connection setup.
547  */
548 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
549 					struct smcd_dev *ismdev,
550 					unsigned short vlan_id)
551 {
552 	if (!is_smcd)
553 		return 0;
554 	if (vlan_id && smc_ism_put_vlan(ismdev, vlan_id))
555 		return SMC_CLC_DECL_CNFERR;
556 	return 0;
557 }
558 
559 /* CLC handshake during connect */
560 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
561 			   struct smc_clc_msg_accept_confirm *aclc,
562 			   struct smc_ib_device *ibdev, u8 ibport,
563 			   u8 gid[], struct smcd_dev *ismdev)
564 {
565 	int rc = 0;
566 
567 	/* do inband token exchange */
568 	rc = smc_clc_send_proposal(smc, smc_type, ibdev, ibport, gid, ismdev);
569 	if (rc)
570 		return rc;
571 	/* receive SMC Accept CLC message */
572 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT,
573 				CLC_WAIT_TIME);
574 }
575 
576 /* setup for RDMA connection of client */
577 static int smc_connect_rdma(struct smc_sock *smc,
578 			    struct smc_clc_msg_accept_confirm *aclc,
579 			    struct smc_ib_device *ibdev, u8 ibport)
580 {
581 	int local_contact = SMC_FIRST_CONTACT;
582 	struct smc_link *link;
583 	int reason_code = 0;
584 
585 	mutex_lock(&smc_client_lgr_pending);
586 	local_contact = smc_conn_create(smc, false, aclc->hdr.flag, ibdev,
587 					ibport, ntoh24(aclc->qpn), &aclc->lcl,
588 					NULL, 0);
589 	if (local_contact < 0) {
590 		if (local_contact == -ENOMEM)
591 			reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
592 		else if (local_contact == -ENOLINK)
593 			reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
594 		else
595 			reason_code = SMC_CLC_DECL_INTERR; /* other error */
596 		mutex_unlock(&smc_client_lgr_pending);
597 		return reason_code;
598 	}
599 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
600 
601 	smc_conn_save_peer_info(smc, aclc);
602 
603 	/* create send buffer and rmb */
604 	if (smc_buf_create(smc, false))
605 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM, local_contact);
606 
607 	if (local_contact == SMC_FIRST_CONTACT)
608 		smc_link_save_peer_info(link, aclc);
609 
610 	if (smc_rmb_rtoken_handling(&smc->conn, aclc))
611 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
612 					 local_contact);
613 
614 	smc_close_init(smc);
615 	smc_rx_init(smc);
616 
617 	if (local_contact == SMC_FIRST_CONTACT) {
618 		if (smc_ib_ready_link(link))
619 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
620 						 local_contact);
621 	} else {
622 		if (smc_reg_rmb(link, smc->conn.rmb_desc, true))
623 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
624 						 local_contact);
625 	}
626 	smc_rmb_sync_sg_for_device(&smc->conn);
627 
628 	reason_code = smc_clc_send_confirm(smc);
629 	if (reason_code)
630 		return smc_connect_abort(smc, reason_code, local_contact);
631 
632 	smc_tx_init(smc);
633 
634 	if (local_contact == SMC_FIRST_CONTACT) {
635 		/* QP confirmation over RoCE fabric */
636 		reason_code = smc_clnt_conf_first_link(smc);
637 		if (reason_code)
638 			return smc_connect_abort(smc, reason_code,
639 						 local_contact);
640 	}
641 	mutex_unlock(&smc_client_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 /* setup for ISM connection of client */
651 static int smc_connect_ism(struct smc_sock *smc,
652 			   struct smc_clc_msg_accept_confirm *aclc,
653 			   struct smcd_dev *ismdev)
654 {
655 	int local_contact = SMC_FIRST_CONTACT;
656 	int rc = 0;
657 
658 	/* there is only one lgr role for SMC-D; use server lock */
659 	mutex_lock(&smc_server_lgr_pending);
660 	local_contact = smc_conn_create(smc, true, aclc->hdr.flag, NULL, 0, 0,
661 					NULL, ismdev, aclc->gid);
662 	if (local_contact < 0) {
663 		mutex_unlock(&smc_server_lgr_pending);
664 		return SMC_CLC_DECL_MEM;
665 	}
666 
667 	/* Create send and receive buffers */
668 	if (smc_buf_create(smc, true))
669 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM, local_contact);
670 
671 	smc_conn_save_peer_info(smc, aclc);
672 	smc_close_init(smc);
673 	smc_rx_init(smc);
674 	smc_tx_init(smc);
675 
676 	rc = smc_clc_send_confirm(smc);
677 	if (rc)
678 		return smc_connect_abort(smc, rc, local_contact);
679 	mutex_unlock(&smc_server_lgr_pending);
680 
681 	smc_copy_sock_settings_to_clc(smc);
682 	if (smc->sk.sk_state == SMC_INIT)
683 		smc->sk.sk_state = SMC_ACTIVE;
684 
685 	return 0;
686 }
687 
688 /* perform steps before actually connecting */
689 static int __smc_connect(struct smc_sock *smc)
690 {
691 	bool ism_supported = false, rdma_supported = false;
692 	struct smc_clc_msg_accept_confirm aclc;
693 	struct smc_ib_device *ibdev;
694 	struct smcd_dev *ismdev;
695 	u8 gid[SMC_GID_SIZE];
696 	unsigned short vlan;
697 	int smc_type;
698 	int rc = 0;
699 	u8 ibport;
700 
701 	sock_hold(&smc->sk); /* sock put in passive closing */
702 
703 	if (smc->use_fallback)
704 		return smc_connect_fallback(smc, smc->fallback_rsn);
705 
706 	/* if peer has not signalled SMC-capability, fall back */
707 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
708 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
709 
710 	/* IPSec connections opt out of SMC-R optimizations */
711 	if (using_ipsec(smc))
712 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
713 
714 	/* check for VLAN ID */
715 	if (smc_vlan_by_tcpsk(smc->clcsock, &vlan))
716 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_CNFERR);
717 
718 	/* check if there is an ism device available */
719 	if (!smc_check_ism(smc, &ismdev) &&
720 	    !smc_connect_ism_vlan_setup(smc, ismdev, vlan)) {
721 		/* ISM is supported for this connection */
722 		ism_supported = true;
723 		smc_type = SMC_TYPE_D;
724 	}
725 
726 	/* check if there is a rdma device available */
727 	if (!smc_check_rdma(smc, &ibdev, &ibport, vlan, gid)) {
728 		/* RDMA is supported for this connection */
729 		rdma_supported = true;
730 		if (ism_supported)
731 			smc_type = SMC_TYPE_B; /* both */
732 		else
733 			smc_type = SMC_TYPE_R; /* only RDMA */
734 	}
735 
736 	/* if neither ISM nor RDMA are supported, fallback */
737 	if (!rdma_supported && !ism_supported)
738 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
739 
740 	/* perform CLC handshake */
741 	rc = smc_connect_clc(smc, smc_type, &aclc, ibdev, ibport, gid, ismdev);
742 	if (rc) {
743 		smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
744 		return smc_connect_decline_fallback(smc, rc);
745 	}
746 
747 	/* depending on previous steps, connect using rdma or ism */
748 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
749 		rc = smc_connect_rdma(smc, &aclc, ibdev, ibport);
750 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
751 		rc = smc_connect_ism(smc, &aclc, ismdev);
752 	else
753 		rc = SMC_CLC_DECL_MODEUNSUPP;
754 	if (rc) {
755 		smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
756 		return smc_connect_decline_fallback(smc, rc);
757 	}
758 
759 	smc_connect_ism_vlan_cleanup(smc, ism_supported, ismdev, vlan);
760 	return 0;
761 }
762 
763 static void smc_connect_work(struct work_struct *work)
764 {
765 	struct smc_sock *smc = container_of(work, struct smc_sock,
766 					    connect_work);
767 	int rc;
768 
769 	lock_sock(&smc->sk);
770 	rc = kernel_connect(smc->clcsock, &smc->connect_info->addr,
771 			    smc->connect_info->alen, smc->connect_info->flags);
772 	if (smc->clcsock->sk->sk_err) {
773 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
774 		goto out;
775 	}
776 	if (rc < 0) {
777 		smc->sk.sk_err = -rc;
778 		goto out;
779 	}
780 
781 	rc = __smc_connect(smc);
782 	if (rc < 0)
783 		smc->sk.sk_err = -rc;
784 
785 out:
786 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
787 		if (smc->sk.sk_err) {
788 			smc->sk.sk_state_change(&smc->sk);
789 		} else { /* allow polling before and after fallback decision */
790 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
791 			smc->sk.sk_write_space(&smc->sk);
792 		}
793 	}
794 	kfree(smc->connect_info);
795 	smc->connect_info = NULL;
796 	release_sock(&smc->sk);
797 }
798 
799 static int smc_connect(struct socket *sock, struct sockaddr *addr,
800 		       int alen, int flags)
801 {
802 	struct sock *sk = sock->sk;
803 	struct smc_sock *smc;
804 	int rc = -EINVAL;
805 
806 	smc = smc_sk(sk);
807 
808 	/* separate smc parameter checking to be safe */
809 	if (alen < sizeof(addr->sa_family))
810 		goto out_err;
811 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
812 		goto out_err;
813 
814 	lock_sock(sk);
815 	switch (sk->sk_state) {
816 	default:
817 		goto out;
818 	case SMC_ACTIVE:
819 		rc = -EISCONN;
820 		goto out;
821 	case SMC_INIT:
822 		rc = 0;
823 		break;
824 	}
825 
826 	smc_copy_sock_settings_to_clc(smc);
827 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
828 	if (flags & O_NONBLOCK) {
829 		if (smc->connect_info) {
830 			rc = -EALREADY;
831 			goto out;
832 		}
833 		smc->connect_info = kzalloc(alen + 2 * sizeof(int), GFP_KERNEL);
834 		if (!smc->connect_info) {
835 			rc = -ENOMEM;
836 			goto out;
837 		}
838 		smc->connect_info->alen = alen;
839 		smc->connect_info->flags = flags ^ O_NONBLOCK;
840 		memcpy(&smc->connect_info->addr, addr, alen);
841 		schedule_work(&smc->connect_work);
842 		rc = -EINPROGRESS;
843 	} else {
844 		rc = kernel_connect(smc->clcsock, addr, alen, flags);
845 		if (rc)
846 			goto out;
847 
848 		rc = __smc_connect(smc);
849 		if (rc < 0)
850 			goto out;
851 		else
852 			rc = 0; /* success cases including fallback */
853 	}
854 
855 out:
856 	release_sock(sk);
857 out_err:
858 	return rc;
859 }
860 
861 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
862 {
863 	struct socket *new_clcsock = NULL;
864 	struct sock *lsk = &lsmc->sk;
865 	struct sock *new_sk;
866 	int rc = -EINVAL;
867 
868 	release_sock(lsk);
869 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
870 	if (!new_sk) {
871 		rc = -ENOMEM;
872 		lsk->sk_err = ENOMEM;
873 		*new_smc = NULL;
874 		lock_sock(lsk);
875 		goto out;
876 	}
877 	*new_smc = smc_sk(new_sk);
878 
879 	mutex_lock(&lsmc->clcsock_release_lock);
880 	if (lsmc->clcsock)
881 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
882 	mutex_unlock(&lsmc->clcsock_release_lock);
883 	lock_sock(lsk);
884 	if  (rc < 0)
885 		lsk->sk_err = -rc;
886 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
887 		new_sk->sk_prot->unhash(new_sk);
888 		if (new_clcsock)
889 			sock_release(new_clcsock);
890 		new_sk->sk_state = SMC_CLOSED;
891 		sock_set_flag(new_sk, SOCK_DEAD);
892 		sock_put(new_sk); /* final */
893 		*new_smc = NULL;
894 		goto out;
895 	}
896 
897 	(*new_smc)->clcsock = new_clcsock;
898 out:
899 	return rc;
900 }
901 
902 /* add a just created sock to the accept queue of the listen sock as
903  * candidate for a following socket accept call from user space
904  */
905 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
906 {
907 	struct smc_sock *par = smc_sk(parent);
908 
909 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
910 	spin_lock(&par->accept_q_lock);
911 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
912 	spin_unlock(&par->accept_q_lock);
913 	sk_acceptq_added(parent);
914 }
915 
916 /* remove a socket from the accept queue of its parental listening socket */
917 static void smc_accept_unlink(struct sock *sk)
918 {
919 	struct smc_sock *par = smc_sk(sk)->listen_smc;
920 
921 	spin_lock(&par->accept_q_lock);
922 	list_del_init(&smc_sk(sk)->accept_q);
923 	spin_unlock(&par->accept_q_lock);
924 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
925 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
926 }
927 
928 /* remove a sock from the accept queue to bind it to a new socket created
929  * for a socket accept call from user space
930  */
931 struct sock *smc_accept_dequeue(struct sock *parent,
932 				struct socket *new_sock)
933 {
934 	struct smc_sock *isk, *n;
935 	struct sock *new_sk;
936 
937 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
938 		new_sk = (struct sock *)isk;
939 
940 		smc_accept_unlink(new_sk);
941 		if (new_sk->sk_state == SMC_CLOSED) {
942 			new_sk->sk_prot->unhash(new_sk);
943 			if (isk->clcsock) {
944 				sock_release(isk->clcsock);
945 				isk->clcsock = NULL;
946 			}
947 			sock_put(new_sk); /* final */
948 			continue;
949 		}
950 		if (new_sock) {
951 			sock_graft(new_sk, new_sock);
952 			if (isk->use_fallback) {
953 				smc_sk(new_sk)->clcsock->file = new_sock->file;
954 				isk->clcsock->file->private_data = isk->clcsock;
955 			}
956 		}
957 		return new_sk;
958 	}
959 	return NULL;
960 }
961 
962 /* clean up for a created but never accepted sock */
963 void smc_close_non_accepted(struct sock *sk)
964 {
965 	struct smc_sock *smc = smc_sk(sk);
966 
967 	lock_sock(sk);
968 	if (!sk->sk_lingertime)
969 		/* wait for peer closing */
970 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
971 	if (!smc->use_fallback) {
972 		smc_close_active(smc);
973 		sock_set_flag(sk, SOCK_DEAD);
974 		sk->sk_shutdown |= SHUTDOWN_MASK;
975 	}
976 	sk->sk_prot->unhash(sk);
977 	if (smc->clcsock) {
978 		struct socket *tcp;
979 
980 		tcp = smc->clcsock;
981 		smc->clcsock = NULL;
982 		sock_release(tcp);
983 	}
984 	if (smc->use_fallback) {
985 		sock_put(sk); /* passive closing */
986 		sk->sk_state = SMC_CLOSED;
987 	} else {
988 		if (sk->sk_state == SMC_CLOSED)
989 			smc_conn_free(&smc->conn);
990 	}
991 	release_sock(sk);
992 	sock_put(sk); /* final sock_put */
993 }
994 
995 static int smc_serv_conf_first_link(struct smc_sock *smc)
996 {
997 	struct net *net = sock_net(smc->clcsock->sk);
998 	struct smc_link_group *lgr = smc->conn.lgr;
999 	struct smc_link *link;
1000 	int rest;
1001 	int rc;
1002 
1003 	link = &lgr->lnk[SMC_SINGLE_LINK];
1004 
1005 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
1006 		return SMC_CLC_DECL_ERR_REGRMB;
1007 
1008 	/* send CONFIRM LINK request to client over the RoCE fabric */
1009 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1010 	if (rc < 0)
1011 		return SMC_CLC_DECL_TIMEOUT_CL;
1012 
1013 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1014 	rest = wait_for_completion_interruptible_timeout(
1015 		&link->llc_confirm_resp,
1016 		SMC_LLC_WAIT_FIRST_TIME);
1017 	if (rest <= 0) {
1018 		struct smc_clc_msg_decline dclc;
1019 
1020 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1021 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1022 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1023 	}
1024 
1025 	if (link->llc_confirm_resp_rc)
1026 		return SMC_CLC_DECL_RMBE_EC;
1027 
1028 	/* send ADD LINK request to client over the RoCE fabric */
1029 	rc = smc_llc_send_add_link(link,
1030 				   link->smcibdev->mac[link->ibport - 1],
1031 				   link->gid, SMC_LLC_REQ);
1032 	if (rc < 0)
1033 		return SMC_CLC_DECL_TIMEOUT_AL;
1034 
1035 	/* receive ADD LINK response from client over the RoCE fabric */
1036 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
1037 							 SMC_LLC_WAIT_TIME);
1038 	if (rest <= 0) {
1039 		struct smc_clc_msg_decline dclc;
1040 
1041 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1042 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1043 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
1044 	}
1045 
1046 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
1047 
1048 	return 0;
1049 }
1050 
1051 /* listen worker: finish */
1052 static void smc_listen_out(struct smc_sock *new_smc)
1053 {
1054 	struct smc_sock *lsmc = new_smc->listen_smc;
1055 	struct sock *newsmcsk = &new_smc->sk;
1056 
1057 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1058 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1059 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1060 		release_sock(&lsmc->sk);
1061 	} else { /* no longer listening */
1062 		smc_close_non_accepted(newsmcsk);
1063 	}
1064 
1065 	/* Wake up accept */
1066 	lsmc->sk.sk_data_ready(&lsmc->sk);
1067 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1068 }
1069 
1070 /* listen worker: finish in state connected */
1071 static void smc_listen_out_connected(struct smc_sock *new_smc)
1072 {
1073 	struct sock *newsmcsk = &new_smc->sk;
1074 
1075 	sk_refcnt_debug_inc(newsmcsk);
1076 	if (newsmcsk->sk_state == SMC_INIT)
1077 		newsmcsk->sk_state = SMC_ACTIVE;
1078 
1079 	smc_listen_out(new_smc);
1080 }
1081 
1082 /* listen worker: finish in error state */
1083 static void smc_listen_out_err(struct smc_sock *new_smc)
1084 {
1085 	struct sock *newsmcsk = &new_smc->sk;
1086 
1087 	if (newsmcsk->sk_state == SMC_INIT)
1088 		sock_put(&new_smc->sk); /* passive closing */
1089 	newsmcsk->sk_state = SMC_CLOSED;
1090 	smc_conn_free(&new_smc->conn);
1091 
1092 	smc_listen_out(new_smc);
1093 }
1094 
1095 /* listen worker: decline and fall back if possible */
1096 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1097 			       int local_contact)
1098 {
1099 	/* RDMA setup failed, switch back to TCP */
1100 	if (local_contact == SMC_FIRST_CONTACT)
1101 		smc_lgr_forget(new_smc->conn.lgr);
1102 	if (reason_code < 0) { /* error, no fallback possible */
1103 		smc_listen_out_err(new_smc);
1104 		return;
1105 	}
1106 	smc_conn_free(&new_smc->conn);
1107 	smc_switch_to_fallback(new_smc);
1108 	new_smc->fallback_rsn = reason_code;
1109 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1110 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1111 			smc_listen_out_err(new_smc);
1112 			return;
1113 		}
1114 	}
1115 	smc_listen_out_connected(new_smc);
1116 }
1117 
1118 /* listen worker: check prefixes */
1119 static int smc_listen_rdma_check(struct smc_sock *new_smc,
1120 				 struct smc_clc_msg_proposal *pclc)
1121 {
1122 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1123 	struct socket *newclcsock = new_smc->clcsock;
1124 
1125 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1126 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1127 		return SMC_CLC_DECL_CNFERR;
1128 
1129 	return 0;
1130 }
1131 
1132 /* listen worker: initialize connection and buffers */
1133 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1134 				struct smc_clc_msg_proposal *pclc,
1135 				struct smc_ib_device *ibdev, u8 ibport,
1136 				int *local_contact)
1137 {
1138 	/* allocate connection / link group */
1139 	*local_contact = smc_conn_create(new_smc, false, 0, ibdev, ibport, 0,
1140 					 &pclc->lcl, NULL, 0);
1141 	if (*local_contact < 0) {
1142 		if (*local_contact == -ENOMEM)
1143 			return SMC_CLC_DECL_MEM;/* insufficient memory*/
1144 		return SMC_CLC_DECL_INTERR; /* other error */
1145 	}
1146 
1147 	/* create send buffer and rmb */
1148 	if (smc_buf_create(new_smc, false))
1149 		return SMC_CLC_DECL_MEM;
1150 
1151 	return 0;
1152 }
1153 
1154 /* listen worker: initialize connection and buffers for SMC-D */
1155 static int smc_listen_ism_init(struct smc_sock *new_smc,
1156 			       struct smc_clc_msg_proposal *pclc,
1157 			       struct smcd_dev *ismdev,
1158 			       int *local_contact)
1159 {
1160 	struct smc_clc_msg_smcd *pclc_smcd;
1161 
1162 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1163 	*local_contact = smc_conn_create(new_smc, true, 0, NULL, 0, 0, NULL,
1164 					 ismdev, pclc_smcd->gid);
1165 	if (*local_contact < 0) {
1166 		if (*local_contact == -ENOMEM)
1167 			return SMC_CLC_DECL_MEM;/* insufficient memory*/
1168 		return SMC_CLC_DECL_INTERR; /* other error */
1169 	}
1170 
1171 	/* Check if peer can be reached via ISM device */
1172 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1173 			    new_smc->conn.lgr->vlan_id,
1174 			    new_smc->conn.lgr->smcd)) {
1175 		if (*local_contact == SMC_FIRST_CONTACT)
1176 			smc_lgr_forget(new_smc->conn.lgr);
1177 		smc_conn_free(&new_smc->conn);
1178 		return SMC_CLC_DECL_CNFERR;
1179 	}
1180 
1181 	/* Create send and receive buffers */
1182 	if (smc_buf_create(new_smc, true)) {
1183 		if (*local_contact == SMC_FIRST_CONTACT)
1184 			smc_lgr_forget(new_smc->conn.lgr);
1185 		smc_conn_free(&new_smc->conn);
1186 		return SMC_CLC_DECL_MEM;
1187 	}
1188 
1189 	return 0;
1190 }
1191 
1192 /* listen worker: register buffers */
1193 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1194 {
1195 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1196 
1197 	if (local_contact != SMC_FIRST_CONTACT) {
1198 		if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true))
1199 			return SMC_CLC_DECL_ERR_REGRMB;
1200 	}
1201 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1202 
1203 	return 0;
1204 }
1205 
1206 /* listen worker: finish RDMA setup */
1207 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1208 				  struct smc_clc_msg_accept_confirm *cclc,
1209 				  int local_contact)
1210 {
1211 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1212 	int reason_code = 0;
1213 
1214 	if (local_contact == SMC_FIRST_CONTACT)
1215 		smc_link_save_peer_info(link, cclc);
1216 
1217 	if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) {
1218 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1219 		goto decline;
1220 	}
1221 
1222 	if (local_contact == SMC_FIRST_CONTACT) {
1223 		if (smc_ib_ready_link(link)) {
1224 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1225 			goto decline;
1226 		}
1227 		/* QP confirmation over RoCE fabric */
1228 		reason_code = smc_serv_conf_first_link(new_smc);
1229 		if (reason_code)
1230 			goto decline;
1231 	}
1232 	return 0;
1233 
1234 decline:
1235 	smc_listen_decline(new_smc, reason_code, local_contact);
1236 	return reason_code;
1237 }
1238 
1239 /* setup for RDMA connection of server */
1240 static void smc_listen_work(struct work_struct *work)
1241 {
1242 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1243 						smc_listen_work);
1244 	struct socket *newclcsock = new_smc->clcsock;
1245 	struct smc_clc_msg_accept_confirm cclc;
1246 	struct smc_clc_msg_proposal *pclc;
1247 	struct smc_ib_device *ibdev;
1248 	bool ism_supported = false;
1249 	struct smcd_dev *ismdev;
1250 	u8 buf[SMC_CLC_MAX_LEN];
1251 	int local_contact = 0;
1252 	unsigned short vlan;
1253 	int reason_code = 0;
1254 	int rc = 0;
1255 	u8 ibport;
1256 
1257 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1258 		return smc_listen_out_err(new_smc);
1259 
1260 	if (new_smc->use_fallback) {
1261 		smc_listen_out_connected(new_smc);
1262 		return;
1263 	}
1264 
1265 	/* check if peer is smc capable */
1266 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1267 		smc_switch_to_fallback(new_smc);
1268 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1269 		smc_listen_out_connected(new_smc);
1270 		return;
1271 	}
1272 
1273 	/* do inband token exchange -
1274 	 * wait for and receive SMC Proposal CLC message
1275 	 */
1276 	pclc = (struct smc_clc_msg_proposal *)&buf;
1277 	reason_code = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1278 				       SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1279 	if (reason_code) {
1280 		smc_listen_decline(new_smc, reason_code, 0);
1281 		return;
1282 	}
1283 
1284 	/* IPSec connections opt out of SMC-R optimizations */
1285 	if (using_ipsec(new_smc)) {
1286 		smc_listen_decline(new_smc, SMC_CLC_DECL_IPSEC, 0);
1287 		return;
1288 	}
1289 
1290 	mutex_lock(&smc_server_lgr_pending);
1291 	smc_close_init(new_smc);
1292 	smc_rx_init(new_smc);
1293 	smc_tx_init(new_smc);
1294 
1295 	/* check if ISM is available */
1296 	if ((pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) &&
1297 	    !smc_check_ism(new_smc, &ismdev) &&
1298 	    !smc_listen_ism_init(new_smc, pclc, ismdev, &local_contact)) {
1299 		ism_supported = true;
1300 	}
1301 
1302 	/* check if RDMA is available */
1303 	if (!ism_supported &&
1304 	    ((pclc->hdr.path != SMC_TYPE_R && pclc->hdr.path != SMC_TYPE_B) ||
1305 	     smc_vlan_by_tcpsk(new_smc->clcsock, &vlan) ||
1306 	     smc_check_rdma(new_smc, &ibdev, &ibport, vlan, NULL) ||
1307 	     smc_listen_rdma_check(new_smc, pclc) ||
1308 	     smc_listen_rdma_init(new_smc, pclc, ibdev, ibport,
1309 				  &local_contact) ||
1310 	     smc_listen_rdma_reg(new_smc, local_contact))) {
1311 		/* SMC not supported, decline */
1312 		mutex_unlock(&smc_server_lgr_pending);
1313 		smc_listen_decline(new_smc, SMC_CLC_DECL_MODEUNSUPP,
1314 				   local_contact);
1315 		return;
1316 	}
1317 
1318 	/* send SMC Accept CLC message */
1319 	rc = smc_clc_send_accept(new_smc, local_contact);
1320 	if (rc) {
1321 		mutex_unlock(&smc_server_lgr_pending);
1322 		smc_listen_decline(new_smc, rc, local_contact);
1323 		return;
1324 	}
1325 
1326 	/* SMC-D does not need this lock any more */
1327 	if (ism_supported)
1328 		mutex_unlock(&smc_server_lgr_pending);
1329 
1330 	/* receive SMC Confirm CLC message */
1331 	reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1332 				       SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1333 	if (reason_code) {
1334 		if (!ism_supported)
1335 			mutex_unlock(&smc_server_lgr_pending);
1336 		smc_listen_decline(new_smc, reason_code, local_contact);
1337 		return;
1338 	}
1339 
1340 	/* finish worker */
1341 	if (!ism_supported) {
1342 		rc = smc_listen_rdma_finish(new_smc, &cclc, local_contact);
1343 		mutex_unlock(&smc_server_lgr_pending);
1344 		if (rc)
1345 			return;
1346 	}
1347 	smc_conn_save_peer_info(new_smc, &cclc);
1348 	smc_listen_out_connected(new_smc);
1349 }
1350 
1351 static void smc_tcp_listen_work(struct work_struct *work)
1352 {
1353 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1354 					     tcp_listen_work);
1355 	struct sock *lsk = &lsmc->sk;
1356 	struct smc_sock *new_smc;
1357 	int rc = 0;
1358 
1359 	lock_sock(lsk);
1360 	while (lsk->sk_state == SMC_LISTEN) {
1361 		rc = smc_clcsock_accept(lsmc, &new_smc);
1362 		if (rc)
1363 			goto out;
1364 		if (!new_smc)
1365 			continue;
1366 
1367 		new_smc->listen_smc = lsmc;
1368 		new_smc->use_fallback = lsmc->use_fallback;
1369 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1370 		sock_hold(lsk); /* sock_put in smc_listen_work */
1371 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1372 		smc_copy_sock_settings_to_smc(new_smc);
1373 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1374 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1375 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1376 		if (!schedule_work(&new_smc->smc_listen_work))
1377 			sock_put(&new_smc->sk);
1378 	}
1379 
1380 out:
1381 	release_sock(lsk);
1382 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1383 }
1384 
1385 static int smc_listen(struct socket *sock, int backlog)
1386 {
1387 	struct sock *sk = sock->sk;
1388 	struct smc_sock *smc;
1389 	int rc;
1390 
1391 	smc = smc_sk(sk);
1392 	lock_sock(sk);
1393 
1394 	rc = -EINVAL;
1395 	if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
1396 		goto out;
1397 
1398 	rc = 0;
1399 	if (sk->sk_state == SMC_LISTEN) {
1400 		sk->sk_max_ack_backlog = backlog;
1401 		goto out;
1402 	}
1403 	/* some socket options are handled in core, so we could not apply
1404 	 * them to the clc socket -- copy smc socket options to clc socket
1405 	 */
1406 	smc_copy_sock_settings_to_clc(smc);
1407 	if (!smc->use_fallback)
1408 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1409 
1410 	rc = kernel_listen(smc->clcsock, backlog);
1411 	if (rc)
1412 		goto out;
1413 	sk->sk_max_ack_backlog = backlog;
1414 	sk->sk_ack_backlog = 0;
1415 	sk->sk_state = SMC_LISTEN;
1416 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1417 	if (!schedule_work(&smc->tcp_listen_work))
1418 		sock_put(sk);
1419 
1420 out:
1421 	release_sock(sk);
1422 	return rc;
1423 }
1424 
1425 static int smc_accept(struct socket *sock, struct socket *new_sock,
1426 		      int flags, bool kern)
1427 {
1428 	struct sock *sk = sock->sk, *nsk;
1429 	DECLARE_WAITQUEUE(wait, current);
1430 	struct smc_sock *lsmc;
1431 	long timeo;
1432 	int rc = 0;
1433 
1434 	lsmc = smc_sk(sk);
1435 	sock_hold(sk); /* sock_put below */
1436 	lock_sock(sk);
1437 
1438 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1439 		rc = -EINVAL;
1440 		release_sock(sk);
1441 		goto out;
1442 	}
1443 
1444 	/* Wait for an incoming connection */
1445 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1446 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1447 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1448 		set_current_state(TASK_INTERRUPTIBLE);
1449 		if (!timeo) {
1450 			rc = -EAGAIN;
1451 			break;
1452 		}
1453 		release_sock(sk);
1454 		timeo = schedule_timeout(timeo);
1455 		/* wakeup by sk_data_ready in smc_listen_work() */
1456 		sched_annotate_sleep();
1457 		lock_sock(sk);
1458 		if (signal_pending(current)) {
1459 			rc = sock_intr_errno(timeo);
1460 			break;
1461 		}
1462 	}
1463 	set_current_state(TASK_RUNNING);
1464 	remove_wait_queue(sk_sleep(sk), &wait);
1465 
1466 	if (!rc)
1467 		rc = sock_error(nsk);
1468 	release_sock(sk);
1469 	if (rc)
1470 		goto out;
1471 
1472 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1473 		/* wait till data arrives on the socket */
1474 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1475 								MSEC_PER_SEC);
1476 		if (smc_sk(nsk)->use_fallback) {
1477 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1478 
1479 			lock_sock(clcsk);
1480 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1481 				sk_wait_data(clcsk, &timeo, NULL);
1482 			release_sock(clcsk);
1483 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1484 			lock_sock(nsk);
1485 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1486 			release_sock(nsk);
1487 		}
1488 	}
1489 
1490 out:
1491 	sock_put(sk); /* sock_hold above */
1492 	return rc;
1493 }
1494 
1495 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1496 		       int peer)
1497 {
1498 	struct smc_sock *smc;
1499 
1500 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1501 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1502 		return -ENOTCONN;
1503 
1504 	smc = smc_sk(sock->sk);
1505 
1506 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1507 }
1508 
1509 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1510 {
1511 	struct sock *sk = sock->sk;
1512 	struct smc_sock *smc;
1513 	int rc = -EPIPE;
1514 
1515 	smc = smc_sk(sk);
1516 	lock_sock(sk);
1517 	if ((sk->sk_state != SMC_ACTIVE) &&
1518 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1519 	    (sk->sk_state != SMC_INIT))
1520 		goto out;
1521 
1522 	if (msg->msg_flags & MSG_FASTOPEN) {
1523 		if (sk->sk_state == SMC_INIT) {
1524 			smc_switch_to_fallback(smc);
1525 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1526 		} else {
1527 			rc = -EINVAL;
1528 			goto out;
1529 		}
1530 	}
1531 
1532 	if (smc->use_fallback)
1533 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1534 	else
1535 		rc = smc_tx_sendmsg(smc, msg, len);
1536 out:
1537 	release_sock(sk);
1538 	return rc;
1539 }
1540 
1541 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1542 		       int flags)
1543 {
1544 	struct sock *sk = sock->sk;
1545 	struct smc_sock *smc;
1546 	int rc = -ENOTCONN;
1547 
1548 	smc = smc_sk(sk);
1549 	lock_sock(sk);
1550 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1551 		/* socket was connected before, no more data to read */
1552 		rc = 0;
1553 		goto out;
1554 	}
1555 	if ((sk->sk_state == SMC_INIT) ||
1556 	    (sk->sk_state == SMC_LISTEN) ||
1557 	    (sk->sk_state == SMC_CLOSED))
1558 		goto out;
1559 
1560 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1561 		rc = 0;
1562 		goto out;
1563 	}
1564 
1565 	if (smc->use_fallback) {
1566 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1567 	} else {
1568 		msg->msg_namelen = 0;
1569 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1570 	}
1571 
1572 out:
1573 	release_sock(sk);
1574 	return rc;
1575 }
1576 
1577 static __poll_t smc_accept_poll(struct sock *parent)
1578 {
1579 	struct smc_sock *isk = smc_sk(parent);
1580 	__poll_t mask = 0;
1581 
1582 	spin_lock(&isk->accept_q_lock);
1583 	if (!list_empty(&isk->accept_q))
1584 		mask = EPOLLIN | EPOLLRDNORM;
1585 	spin_unlock(&isk->accept_q_lock);
1586 
1587 	return mask;
1588 }
1589 
1590 static __poll_t smc_poll(struct file *file, struct socket *sock,
1591 			     poll_table *wait)
1592 {
1593 	struct sock *sk = sock->sk;
1594 	__poll_t mask = 0;
1595 	struct smc_sock *smc;
1596 
1597 	if (!sk)
1598 		return EPOLLNVAL;
1599 
1600 	smc = smc_sk(sock->sk);
1601 	if (smc->use_fallback) {
1602 		/* delegate to CLC child sock */
1603 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1604 		sk->sk_err = smc->clcsock->sk->sk_err;
1605 		if (sk->sk_err)
1606 			mask |= EPOLLERR;
1607 	} else {
1608 		if (sk->sk_state != SMC_CLOSED)
1609 			sock_poll_wait(file, sock, wait);
1610 		if (sk->sk_err)
1611 			mask |= EPOLLERR;
1612 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1613 		    (sk->sk_state == SMC_CLOSED))
1614 			mask |= EPOLLHUP;
1615 		if (sk->sk_state == SMC_LISTEN) {
1616 			/* woken up by sk_data_ready in smc_listen_work() */
1617 			mask = smc_accept_poll(sk);
1618 		} else {
1619 			if (atomic_read(&smc->conn.sndbuf_space) ||
1620 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1621 				mask |= EPOLLOUT | EPOLLWRNORM;
1622 			} else {
1623 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1624 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1625 			}
1626 			if (atomic_read(&smc->conn.bytes_to_rcv))
1627 				mask |= EPOLLIN | EPOLLRDNORM;
1628 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1629 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1630 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1631 				mask |= EPOLLIN;
1632 			if (smc->conn.urg_state == SMC_URG_VALID)
1633 				mask |= EPOLLPRI;
1634 		}
1635 	}
1636 
1637 	return mask;
1638 }
1639 
1640 static int smc_shutdown(struct socket *sock, int how)
1641 {
1642 	struct sock *sk = sock->sk;
1643 	struct smc_sock *smc;
1644 	int rc = -EINVAL;
1645 	int rc1 = 0;
1646 
1647 	smc = smc_sk(sk);
1648 
1649 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1650 		return rc;
1651 
1652 	lock_sock(sk);
1653 
1654 	rc = -ENOTCONN;
1655 	if ((sk->sk_state != SMC_ACTIVE) &&
1656 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1657 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1658 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1659 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1660 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1661 		goto out;
1662 	if (smc->use_fallback) {
1663 		rc = kernel_sock_shutdown(smc->clcsock, how);
1664 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1665 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1666 			sk->sk_state = SMC_CLOSED;
1667 		goto out;
1668 	}
1669 	switch (how) {
1670 	case SHUT_RDWR:		/* shutdown in both directions */
1671 		rc = smc_close_active(smc);
1672 		break;
1673 	case SHUT_WR:
1674 		rc = smc_close_shutdown_write(smc);
1675 		break;
1676 	case SHUT_RD:
1677 		rc = 0;
1678 		/* nothing more to do because peer is not involved */
1679 		break;
1680 	}
1681 	if (smc->clcsock)
1682 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1683 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1684 	sk->sk_shutdown |= how + 1;
1685 
1686 out:
1687 	release_sock(sk);
1688 	return rc ? rc : rc1;
1689 }
1690 
1691 static int smc_setsockopt(struct socket *sock, int level, int optname,
1692 			  char __user *optval, unsigned int optlen)
1693 {
1694 	struct sock *sk = sock->sk;
1695 	struct smc_sock *smc;
1696 	int val, rc;
1697 
1698 	smc = smc_sk(sk);
1699 
1700 	/* generic setsockopts reaching us here always apply to the
1701 	 * CLC socket
1702 	 */
1703 	rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1704 					   optval, optlen);
1705 	if (smc->clcsock->sk->sk_err) {
1706 		sk->sk_err = smc->clcsock->sk->sk_err;
1707 		sk->sk_error_report(sk);
1708 	}
1709 	if (rc)
1710 		return rc;
1711 
1712 	if (optlen < sizeof(int))
1713 		return -EINVAL;
1714 	if (get_user(val, (int __user *)optval))
1715 		return -EFAULT;
1716 
1717 	lock_sock(sk);
1718 	switch (optname) {
1719 	case TCP_ULP:
1720 	case TCP_FASTOPEN:
1721 	case TCP_FASTOPEN_CONNECT:
1722 	case TCP_FASTOPEN_KEY:
1723 	case TCP_FASTOPEN_NO_COOKIE:
1724 		/* option not supported by SMC */
1725 		if (sk->sk_state == SMC_INIT) {
1726 			smc_switch_to_fallback(smc);
1727 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1728 		} else {
1729 			if (!smc->use_fallback)
1730 				rc = -EINVAL;
1731 		}
1732 		break;
1733 	case TCP_NODELAY:
1734 		if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
1735 			if (val && !smc->use_fallback)
1736 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1737 						 0);
1738 		}
1739 		break;
1740 	case TCP_CORK:
1741 		if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) {
1742 			if (!val && !smc->use_fallback)
1743 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1744 						 0);
1745 		}
1746 		break;
1747 	case TCP_DEFER_ACCEPT:
1748 		smc->sockopt_defer_accept = val;
1749 		break;
1750 	default:
1751 		break;
1752 	}
1753 	release_sock(sk);
1754 
1755 	return rc;
1756 }
1757 
1758 static int smc_getsockopt(struct socket *sock, int level, int optname,
1759 			  char __user *optval, int __user *optlen)
1760 {
1761 	struct smc_sock *smc;
1762 
1763 	smc = smc_sk(sock->sk);
1764 	/* socket options apply to the CLC socket */
1765 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1766 					     optval, optlen);
1767 }
1768 
1769 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1770 		     unsigned long arg)
1771 {
1772 	union smc_host_cursor cons, urg;
1773 	struct smc_connection *conn;
1774 	struct smc_sock *smc;
1775 	int answ;
1776 
1777 	smc = smc_sk(sock->sk);
1778 	conn = &smc->conn;
1779 	lock_sock(&smc->sk);
1780 	if (smc->use_fallback) {
1781 		if (!smc->clcsock) {
1782 			release_sock(&smc->sk);
1783 			return -EBADF;
1784 		}
1785 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1786 		release_sock(&smc->sk);
1787 		return answ;
1788 	}
1789 	switch (cmd) {
1790 	case SIOCINQ: /* same as FIONREAD */
1791 		if (smc->sk.sk_state == SMC_LISTEN) {
1792 			release_sock(&smc->sk);
1793 			return -EINVAL;
1794 		}
1795 		if (smc->sk.sk_state == SMC_INIT ||
1796 		    smc->sk.sk_state == SMC_CLOSED)
1797 			answ = 0;
1798 		else
1799 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1800 		break;
1801 	case SIOCOUTQ:
1802 		/* output queue size (not send + not acked) */
1803 		if (smc->sk.sk_state == SMC_LISTEN) {
1804 			release_sock(&smc->sk);
1805 			return -EINVAL;
1806 		}
1807 		if (smc->sk.sk_state == SMC_INIT ||
1808 		    smc->sk.sk_state == SMC_CLOSED)
1809 			answ = 0;
1810 		else
1811 			answ = smc->conn.sndbuf_desc->len -
1812 					atomic_read(&smc->conn.sndbuf_space);
1813 		break;
1814 	case SIOCOUTQNSD:
1815 		/* output queue size (not send only) */
1816 		if (smc->sk.sk_state == SMC_LISTEN) {
1817 			release_sock(&smc->sk);
1818 			return -EINVAL;
1819 		}
1820 		if (smc->sk.sk_state == SMC_INIT ||
1821 		    smc->sk.sk_state == SMC_CLOSED)
1822 			answ = 0;
1823 		else
1824 			answ = smc_tx_prepared_sends(&smc->conn);
1825 		break;
1826 	case SIOCATMARK:
1827 		if (smc->sk.sk_state == SMC_LISTEN) {
1828 			release_sock(&smc->sk);
1829 			return -EINVAL;
1830 		}
1831 		if (smc->sk.sk_state == SMC_INIT ||
1832 		    smc->sk.sk_state == SMC_CLOSED) {
1833 			answ = 0;
1834 		} else {
1835 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1836 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1837 			answ = smc_curs_diff(conn->rmb_desc->len,
1838 					     &cons, &urg) == 1;
1839 		}
1840 		break;
1841 	default:
1842 		release_sock(&smc->sk);
1843 		return -ENOIOCTLCMD;
1844 	}
1845 	release_sock(&smc->sk);
1846 
1847 	return put_user(answ, (int __user *)arg);
1848 }
1849 
1850 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1851 			    int offset, size_t size, int flags)
1852 {
1853 	struct sock *sk = sock->sk;
1854 	struct smc_sock *smc;
1855 	int rc = -EPIPE;
1856 
1857 	smc = smc_sk(sk);
1858 	lock_sock(sk);
1859 	if (sk->sk_state != SMC_ACTIVE) {
1860 		release_sock(sk);
1861 		goto out;
1862 	}
1863 	release_sock(sk);
1864 	if (smc->use_fallback)
1865 		rc = kernel_sendpage(smc->clcsock, page, offset,
1866 				     size, flags);
1867 	else
1868 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1869 
1870 out:
1871 	return rc;
1872 }
1873 
1874 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1875  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1876  * updates till whenever a respective page has been fully processed.
1877  * Note that subsequent recv() calls have to wait till all splice() processing
1878  * completed.
1879  */
1880 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1881 			       struct pipe_inode_info *pipe, size_t len,
1882 			       unsigned int flags)
1883 {
1884 	struct sock *sk = sock->sk;
1885 	struct smc_sock *smc;
1886 	int rc = -ENOTCONN;
1887 
1888 	smc = smc_sk(sk);
1889 	lock_sock(sk);
1890 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1891 		/* socket was connected before, no more data to read */
1892 		rc = 0;
1893 		goto out;
1894 	}
1895 	if (sk->sk_state == SMC_INIT ||
1896 	    sk->sk_state == SMC_LISTEN ||
1897 	    sk->sk_state == SMC_CLOSED)
1898 		goto out;
1899 
1900 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1901 		rc = 0;
1902 		goto out;
1903 	}
1904 
1905 	if (smc->use_fallback) {
1906 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1907 						    pipe, len, flags);
1908 	} else {
1909 		if (*ppos) {
1910 			rc = -ESPIPE;
1911 			goto out;
1912 		}
1913 		if (flags & SPLICE_F_NONBLOCK)
1914 			flags = MSG_DONTWAIT;
1915 		else
1916 			flags = 0;
1917 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1918 	}
1919 out:
1920 	release_sock(sk);
1921 
1922 	return rc;
1923 }
1924 
1925 /* must look like tcp */
1926 static const struct proto_ops smc_sock_ops = {
1927 	.family		= PF_SMC,
1928 	.owner		= THIS_MODULE,
1929 	.release	= smc_release,
1930 	.bind		= smc_bind,
1931 	.connect	= smc_connect,
1932 	.socketpair	= sock_no_socketpair,
1933 	.accept		= smc_accept,
1934 	.getname	= smc_getname,
1935 	.poll		= smc_poll,
1936 	.ioctl		= smc_ioctl,
1937 	.listen		= smc_listen,
1938 	.shutdown	= smc_shutdown,
1939 	.setsockopt	= smc_setsockopt,
1940 	.getsockopt	= smc_getsockopt,
1941 	.sendmsg	= smc_sendmsg,
1942 	.recvmsg	= smc_recvmsg,
1943 	.mmap		= sock_no_mmap,
1944 	.sendpage	= smc_sendpage,
1945 	.splice_read	= smc_splice_read,
1946 };
1947 
1948 static int smc_create(struct net *net, struct socket *sock, int protocol,
1949 		      int kern)
1950 {
1951 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1952 	struct smc_sock *smc;
1953 	struct sock *sk;
1954 	int rc;
1955 
1956 	rc = -ESOCKTNOSUPPORT;
1957 	if (sock->type != SOCK_STREAM)
1958 		goto out;
1959 
1960 	rc = -EPROTONOSUPPORT;
1961 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1962 		goto out;
1963 
1964 	rc = -ENOBUFS;
1965 	sock->ops = &smc_sock_ops;
1966 	sk = smc_sock_alloc(net, sock, protocol);
1967 	if (!sk)
1968 		goto out;
1969 
1970 	/* create internal TCP socket for CLC handshake and fallback */
1971 	smc = smc_sk(sk);
1972 	smc->use_fallback = false; /* assume rdma capability first */
1973 	smc->fallback_rsn = 0;
1974 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1975 			      &smc->clcsock);
1976 	if (rc) {
1977 		sk_common_release(sk);
1978 		goto out;
1979 	}
1980 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1981 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1982 
1983 out:
1984 	return rc;
1985 }
1986 
1987 static const struct net_proto_family smc_sock_family_ops = {
1988 	.family	= PF_SMC,
1989 	.owner	= THIS_MODULE,
1990 	.create	= smc_create,
1991 };
1992 
1993 unsigned int smc_net_id;
1994 
1995 static __net_init int smc_net_init(struct net *net)
1996 {
1997 	return smc_pnet_net_init(net);
1998 }
1999 
2000 static void __net_exit smc_net_exit(struct net *net)
2001 {
2002 	smc_pnet_net_exit(net);
2003 }
2004 
2005 static struct pernet_operations smc_net_ops = {
2006 	.init = smc_net_init,
2007 	.exit = smc_net_exit,
2008 	.id   = &smc_net_id,
2009 	.size = sizeof(struct smc_net),
2010 };
2011 
2012 static int __init smc_init(void)
2013 {
2014 	int rc;
2015 
2016 	rc = register_pernet_subsys(&smc_net_ops);
2017 	if (rc)
2018 		return rc;
2019 
2020 	rc = smc_pnet_init();
2021 	if (rc)
2022 		return rc;
2023 
2024 	rc = smc_llc_init();
2025 	if (rc) {
2026 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2027 		goto out_pnet;
2028 	}
2029 
2030 	rc = smc_cdc_init();
2031 	if (rc) {
2032 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2033 		goto out_pnet;
2034 	}
2035 
2036 	rc = proto_register(&smc_proto, 1);
2037 	if (rc) {
2038 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2039 		goto out_pnet;
2040 	}
2041 
2042 	rc = proto_register(&smc_proto6, 1);
2043 	if (rc) {
2044 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2045 		goto out_proto;
2046 	}
2047 
2048 	rc = sock_register(&smc_sock_family_ops);
2049 	if (rc) {
2050 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2051 		goto out_proto6;
2052 	}
2053 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2054 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2055 
2056 	rc = smc_ib_register_client();
2057 	if (rc) {
2058 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2059 		goto out_sock;
2060 	}
2061 
2062 	static_branch_enable(&tcp_have_smc);
2063 	return 0;
2064 
2065 out_sock:
2066 	sock_unregister(PF_SMC);
2067 out_proto6:
2068 	proto_unregister(&smc_proto6);
2069 out_proto:
2070 	proto_unregister(&smc_proto);
2071 out_pnet:
2072 	smc_pnet_exit();
2073 	return rc;
2074 }
2075 
2076 static void __exit smc_exit(void)
2077 {
2078 	smc_core_exit();
2079 	static_branch_disable(&tcp_have_smc);
2080 	smc_ib_unregister_client();
2081 	sock_unregister(PF_SMC);
2082 	proto_unregister(&smc_proto6);
2083 	proto_unregister(&smc_proto);
2084 	smc_pnet_exit();
2085 	unregister_pernet_subsys(&smc_net_ops);
2086 }
2087 
2088 module_init(smc_init);
2089 module_exit(smc_exit);
2090 
2091 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2092 MODULE_DESCRIPTION("smc socket address family");
2093 MODULE_LICENSE("GPL");
2094 MODULE_ALIAS_NETPROTO(PF_SMC);
2095