xref: /linux/net/smc/af_smc.c (revision 8faabc041a001140564f718dabe37753e88b37fa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
4  *
5  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
6  *  applies to SOCK_STREAM sockets only
7  *  offers an alternative communication option for TCP-protocol sockets
8  *  applicable with RoCE-cards only
9  *
10  *  Initial restrictions:
11  *    - support for alternate links postponed
12  *
13  *  Copyright IBM Corp. 2016, 2018
14  *
15  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
16  *              based on prototype from Frank Blaschka
17  */
18 
19 #define KMSG_COMPONENT "smc"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21 
22 #include <linux/module.h>
23 #include <linux/socket.h>
24 #include <linux/workqueue.h>
25 #include <linux/in.h>
26 #include <linux/sched/signal.h>
27 #include <linux/if_vlan.h>
28 #include <linux/rcupdate_wait.h>
29 #include <linux/ctype.h>
30 #include <linux/splice.h>
31 
32 #include <net/sock.h>
33 #include <net/tcp.h>
34 #include <net/smc.h>
35 #include <asm/ioctls.h>
36 
37 #include <net/net_namespace.h>
38 #include <net/netns/generic.h>
39 #include "smc_netns.h"
40 
41 #include "smc.h"
42 #include "smc_clc.h"
43 #include "smc_llc.h"
44 #include "smc_cdc.h"
45 #include "smc_core.h"
46 #include "smc_ib.h"
47 #include "smc_ism.h"
48 #include "smc_pnet.h"
49 #include "smc_netlink.h"
50 #include "smc_tx.h"
51 #include "smc_rx.h"
52 #include "smc_close.h"
53 #include "smc_stats.h"
54 #include "smc_tracepoint.h"
55 #include "smc_sysctl.h"
56 #include "smc_loopback.h"
57 #include "smc_inet.h"
58 
59 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
60 						 * creation on server
61 						 */
62 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
63 						 * creation on client
64 						 */
65 
66 static struct workqueue_struct	*smc_tcp_ls_wq;	/* wq for tcp listen work */
67 struct workqueue_struct	*smc_hs_wq;	/* wq for handshake work */
68 struct workqueue_struct	*smc_close_wq;	/* wq for close work */
69 
70 static void smc_tcp_listen_work(struct work_struct *);
71 static void smc_connect_work(struct work_struct *);
72 
smc_nl_dump_hs_limitation(struct sk_buff * skb,struct netlink_callback * cb)73 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb)
74 {
75 	struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
76 	void *hdr;
77 
78 	if (cb_ctx->pos[0])
79 		goto out;
80 
81 	hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
82 			  &smc_gen_nl_family, NLM_F_MULTI,
83 			  SMC_NETLINK_DUMP_HS_LIMITATION);
84 	if (!hdr)
85 		return -ENOMEM;
86 
87 	if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED,
88 		       sock_net(skb->sk)->smc.limit_smc_hs))
89 		goto err;
90 
91 	genlmsg_end(skb, hdr);
92 	cb_ctx->pos[0] = 1;
93 out:
94 	return skb->len;
95 err:
96 	genlmsg_cancel(skb, hdr);
97 	return -EMSGSIZE;
98 }
99 
smc_nl_enable_hs_limitation(struct sk_buff * skb,struct genl_info * info)100 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
101 {
102 	sock_net(skb->sk)->smc.limit_smc_hs = true;
103 	return 0;
104 }
105 
smc_nl_disable_hs_limitation(struct sk_buff * skb,struct genl_info * info)106 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
107 {
108 	sock_net(skb->sk)->smc.limit_smc_hs = false;
109 	return 0;
110 }
111 
smc_set_keepalive(struct sock * sk,int val)112 static void smc_set_keepalive(struct sock *sk, int val)
113 {
114 	struct smc_sock *smc = smc_sk(sk);
115 
116 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
117 }
118 
smc_tcp_syn_recv_sock(const struct sock * sk,struct sk_buff * skb,struct request_sock * req,struct dst_entry * dst,struct request_sock * req_unhash,bool * own_req)119 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk,
120 					  struct sk_buff *skb,
121 					  struct request_sock *req,
122 					  struct dst_entry *dst,
123 					  struct request_sock *req_unhash,
124 					  bool *own_req)
125 {
126 	struct smc_sock *smc;
127 	struct sock *child;
128 
129 	smc = smc_clcsock_user_data(sk);
130 
131 	if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) >
132 				sk->sk_max_ack_backlog)
133 		goto drop;
134 
135 	if (sk_acceptq_is_full(&smc->sk)) {
136 		NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
137 		goto drop;
138 	}
139 
140 	/* passthrough to original syn recv sock fct */
141 	child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash,
142 					       own_req);
143 	/* child must not inherit smc or its ops */
144 	if (child) {
145 		rcu_assign_sk_user_data(child, NULL);
146 
147 		/* v4-mapped sockets don't inherit parent ops. Don't restore. */
148 		if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops)
149 			inet_csk(child)->icsk_af_ops = smc->ori_af_ops;
150 	}
151 	return child;
152 
153 drop:
154 	dst_release(dst);
155 	tcp_listendrop(sk);
156 	return NULL;
157 }
158 
smc_hs_congested(const struct sock * sk)159 static bool smc_hs_congested(const struct sock *sk)
160 {
161 	const struct smc_sock *smc;
162 
163 	smc = smc_clcsock_user_data(sk);
164 
165 	if (!smc)
166 		return true;
167 
168 	if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq))
169 		return true;
170 
171 	return false;
172 }
173 
174 struct smc_hashinfo smc_v4_hashinfo = {
175 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
176 };
177 
178 struct smc_hashinfo smc_v6_hashinfo = {
179 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
180 };
181 
smc_hash_sk(struct sock * sk)182 int smc_hash_sk(struct sock *sk)
183 {
184 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
185 	struct hlist_head *head;
186 
187 	head = &h->ht;
188 
189 	write_lock_bh(&h->lock);
190 	sk_add_node(sk, head);
191 	write_unlock_bh(&h->lock);
192 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
193 
194 	return 0;
195 }
196 
smc_unhash_sk(struct sock * sk)197 void smc_unhash_sk(struct sock *sk)
198 {
199 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
200 
201 	write_lock_bh(&h->lock);
202 	if (sk_del_node_init(sk))
203 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
204 	write_unlock_bh(&h->lock);
205 }
206 
207 /* This will be called before user really release sock_lock. So do the
208  * work which we didn't do because of user hold the sock_lock in the
209  * BH context
210  */
smc_release_cb(struct sock * sk)211 void smc_release_cb(struct sock *sk)
212 {
213 	struct smc_sock *smc = smc_sk(sk);
214 
215 	if (smc->conn.tx_in_release_sock) {
216 		smc_tx_pending(&smc->conn);
217 		smc->conn.tx_in_release_sock = false;
218 	}
219 }
220 
221 struct proto smc_proto = {
222 	.name		= "SMC",
223 	.owner		= THIS_MODULE,
224 	.keepalive	= smc_set_keepalive,
225 	.hash		= smc_hash_sk,
226 	.unhash		= smc_unhash_sk,
227 	.release_cb	= smc_release_cb,
228 	.obj_size	= sizeof(struct smc_sock),
229 	.h.smc_hash	= &smc_v4_hashinfo,
230 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
231 };
232 EXPORT_SYMBOL_GPL(smc_proto);
233 
234 struct proto smc_proto6 = {
235 	.name		= "SMC6",
236 	.owner		= THIS_MODULE,
237 	.keepalive	= smc_set_keepalive,
238 	.hash		= smc_hash_sk,
239 	.unhash		= smc_unhash_sk,
240 	.release_cb	= smc_release_cb,
241 	.obj_size	= sizeof(struct smc_sock),
242 	.h.smc_hash	= &smc_v6_hashinfo,
243 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
244 };
245 EXPORT_SYMBOL_GPL(smc_proto6);
246 
smc_fback_restore_callbacks(struct smc_sock * smc)247 static void smc_fback_restore_callbacks(struct smc_sock *smc)
248 {
249 	struct sock *clcsk = smc->clcsock->sk;
250 
251 	write_lock_bh(&clcsk->sk_callback_lock);
252 	clcsk->sk_user_data = NULL;
253 
254 	smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change);
255 	smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready);
256 	smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space);
257 	smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report);
258 
259 	write_unlock_bh(&clcsk->sk_callback_lock);
260 }
261 
smc_restore_fallback_changes(struct smc_sock * smc)262 static void smc_restore_fallback_changes(struct smc_sock *smc)
263 {
264 	if (smc->clcsock->file) { /* non-accepted sockets have no file yet */
265 		smc->clcsock->file->private_data = smc->sk.sk_socket;
266 		smc->clcsock->file = NULL;
267 		smc_fback_restore_callbacks(smc);
268 	}
269 }
270 
__smc_release(struct smc_sock * smc)271 static int __smc_release(struct smc_sock *smc)
272 {
273 	struct sock *sk = &smc->sk;
274 	int rc = 0;
275 
276 	if (!smc->use_fallback) {
277 		rc = smc_close_active(smc);
278 		smc_sock_set_flag(sk, SOCK_DEAD);
279 		sk->sk_shutdown |= SHUTDOWN_MASK;
280 	} else {
281 		if (sk->sk_state != SMC_CLOSED) {
282 			if (sk->sk_state != SMC_LISTEN &&
283 			    sk->sk_state != SMC_INIT)
284 				sock_put(sk); /* passive closing */
285 			if (sk->sk_state == SMC_LISTEN) {
286 				/* wake up clcsock accept */
287 				rc = kernel_sock_shutdown(smc->clcsock,
288 							  SHUT_RDWR);
289 			}
290 			sk->sk_state = SMC_CLOSED;
291 			sk->sk_state_change(sk);
292 		}
293 		smc_restore_fallback_changes(smc);
294 	}
295 
296 	sk->sk_prot->unhash(sk);
297 
298 	if (sk->sk_state == SMC_CLOSED) {
299 		if (smc->clcsock) {
300 			release_sock(sk);
301 			smc_clcsock_release(smc);
302 			lock_sock(sk);
303 		}
304 		if (!smc->use_fallback)
305 			smc_conn_free(&smc->conn);
306 	}
307 
308 	return rc;
309 }
310 
smc_release(struct socket * sock)311 int smc_release(struct socket *sock)
312 {
313 	struct sock *sk = sock->sk;
314 	struct smc_sock *smc;
315 	int old_state, rc = 0;
316 
317 	if (!sk)
318 		goto out;
319 
320 	sock_hold(sk); /* sock_put below */
321 	smc = smc_sk(sk);
322 
323 	old_state = sk->sk_state;
324 
325 	/* cleanup for a dangling non-blocking connect */
326 	if (smc->connect_nonblock && old_state == SMC_INIT)
327 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
328 
329 	if (cancel_work_sync(&smc->connect_work))
330 		sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */
331 
332 	if (sk->sk_state == SMC_LISTEN)
333 		/* smc_close_non_accepted() is called and acquires
334 		 * sock lock for child sockets again
335 		 */
336 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
337 	else
338 		lock_sock(sk);
339 
340 	if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE &&
341 	    !smc->use_fallback)
342 		smc_close_active_abort(smc);
343 
344 	rc = __smc_release(smc);
345 
346 	/* detach socket */
347 	sock_orphan(sk);
348 	sock->sk = NULL;
349 	release_sock(sk);
350 
351 	sock_put(sk); /* sock_hold above */
352 	sock_put(sk); /* final sock_put */
353 out:
354 	return rc;
355 }
356 
smc_destruct(struct sock * sk)357 static void smc_destruct(struct sock *sk)
358 {
359 	if (sk->sk_state != SMC_CLOSED)
360 		return;
361 	if (!sock_flag(sk, SOCK_DEAD))
362 		return;
363 }
364 
smc_sk_init(struct net * net,struct sock * sk,int protocol)365 void smc_sk_init(struct net *net, struct sock *sk, int protocol)
366 {
367 	struct smc_sock *smc = smc_sk(sk);
368 
369 	sk->sk_state = SMC_INIT;
370 	sk->sk_destruct = smc_destruct;
371 	sk->sk_protocol = protocol;
372 	WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem));
373 	WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem));
374 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
375 	INIT_WORK(&smc->connect_work, smc_connect_work);
376 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
377 	INIT_LIST_HEAD(&smc->accept_q);
378 	spin_lock_init(&smc->accept_q_lock);
379 	spin_lock_init(&smc->conn.send_lock);
380 	sk->sk_prot->hash(sk);
381 	mutex_init(&smc->clcsock_release_lock);
382 	smc_init_saved_callbacks(smc);
383 	smc->limit_smc_hs = net->smc.limit_smc_hs;
384 	smc->use_fallback = false; /* assume rdma capability first */
385 	smc->fallback_rsn = 0;
386 	smc_close_init(smc);
387 }
388 
smc_sock_alloc(struct net * net,struct socket * sock,int protocol)389 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
390 				   int protocol)
391 {
392 	struct proto *prot;
393 	struct sock *sk;
394 
395 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
396 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
397 	if (!sk)
398 		return NULL;
399 
400 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
401 	smc_sk_init(net, sk, protocol);
402 
403 	return sk;
404 }
405 
smc_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)406 int smc_bind(struct socket *sock, struct sockaddr *uaddr,
407 	     int addr_len)
408 {
409 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
410 	struct sock *sk = sock->sk;
411 	struct smc_sock *smc;
412 	int rc;
413 
414 	smc = smc_sk(sk);
415 
416 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
417 	rc = -EINVAL;
418 	if (addr_len < sizeof(struct sockaddr_in))
419 		goto out;
420 
421 	rc = -EAFNOSUPPORT;
422 	if (addr->sin_family != AF_INET &&
423 	    addr->sin_family != AF_INET6 &&
424 	    addr->sin_family != AF_UNSPEC)
425 		goto out;
426 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
427 	if (addr->sin_family == AF_UNSPEC &&
428 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
429 		goto out;
430 
431 	lock_sock(sk);
432 
433 	/* Check if socket is already active */
434 	rc = -EINVAL;
435 	if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
436 		goto out_rel;
437 
438 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
439 	smc->clcsock->sk->sk_reuseport = sk->sk_reuseport;
440 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
441 
442 out_rel:
443 	release_sock(sk);
444 out:
445 	return rc;
446 }
447 
448 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
449  * clc socket (since smc is not called for these options from net/core)
450  */
451 
452 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
453 			     (1UL << SOCK_KEEPOPEN) | \
454 			     (1UL << SOCK_LINGER) | \
455 			     (1UL << SOCK_BROADCAST) | \
456 			     (1UL << SOCK_TIMESTAMP) | \
457 			     (1UL << SOCK_DBG) | \
458 			     (1UL << SOCK_RCVTSTAMP) | \
459 			     (1UL << SOCK_RCVTSTAMPNS) | \
460 			     (1UL << SOCK_LOCALROUTE) | \
461 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
462 			     (1UL << SOCK_RXQ_OVFL) | \
463 			     (1UL << SOCK_WIFI_STATUS) | \
464 			     (1UL << SOCK_NOFCS) | \
465 			     (1UL << SOCK_FILTER_LOCKED) | \
466 			     (1UL << SOCK_TSTAMP_NEW))
467 
468 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */
smc_adjust_sock_bufsizes(struct sock * nsk,struct sock * osk,unsigned long mask)469 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk,
470 				     unsigned long mask)
471 {
472 	nsk->sk_userlocks = osk->sk_userlocks;
473 	if (osk->sk_userlocks & SOCK_SNDBUF_LOCK)
474 		nsk->sk_sndbuf = osk->sk_sndbuf;
475 	if (osk->sk_userlocks & SOCK_RCVBUF_LOCK)
476 		nsk->sk_rcvbuf = osk->sk_rcvbuf;
477 }
478 
smc_copy_sock_settings(struct sock * nsk,struct sock * osk,unsigned long mask)479 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
480 				   unsigned long mask)
481 {
482 	/* options we don't get control via setsockopt for */
483 	nsk->sk_type = osk->sk_type;
484 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
485 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
486 	nsk->sk_mark = READ_ONCE(osk->sk_mark);
487 	nsk->sk_priority = READ_ONCE(osk->sk_priority);
488 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
489 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
490 	nsk->sk_err = osk->sk_err;
491 
492 	nsk->sk_flags &= ~mask;
493 	nsk->sk_flags |= osk->sk_flags & mask;
494 
495 	smc_adjust_sock_bufsizes(nsk, osk, mask);
496 }
497 
smc_copy_sock_settings_to_clc(struct smc_sock * smc)498 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
499 {
500 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
501 }
502 
503 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
504 			     (1UL << SOCK_KEEPOPEN) | \
505 			     (1UL << SOCK_LINGER) | \
506 			     (1UL << SOCK_DBG))
507 /* copy only settings and flags relevant for smc from clc to smc socket */
smc_copy_sock_settings_to_smc(struct smc_sock * smc)508 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
509 {
510 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
511 }
512 
513 /* register the new vzalloced sndbuf on all links */
smcr_lgr_reg_sndbufs(struct smc_link * link,struct smc_buf_desc * snd_desc)514 static int smcr_lgr_reg_sndbufs(struct smc_link *link,
515 				struct smc_buf_desc *snd_desc)
516 {
517 	struct smc_link_group *lgr = link->lgr;
518 	int i, rc = 0;
519 
520 	if (!snd_desc->is_vm)
521 		return -EINVAL;
522 
523 	/* protect against parallel smcr_link_reg_buf() */
524 	down_write(&lgr->llc_conf_mutex);
525 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
526 		if (!smc_link_active(&lgr->lnk[i]))
527 			continue;
528 		rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc);
529 		if (rc)
530 			break;
531 	}
532 	up_write(&lgr->llc_conf_mutex);
533 	return rc;
534 }
535 
536 /* register the new rmb on all links */
smcr_lgr_reg_rmbs(struct smc_link * link,struct smc_buf_desc * rmb_desc)537 static int smcr_lgr_reg_rmbs(struct smc_link *link,
538 			     struct smc_buf_desc *rmb_desc)
539 {
540 	struct smc_link_group *lgr = link->lgr;
541 	bool do_slow = false;
542 	int i, rc = 0;
543 
544 	rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
545 	if (rc)
546 		return rc;
547 
548 	down_read(&lgr->llc_conf_mutex);
549 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
550 		if (!smc_link_active(&lgr->lnk[i]))
551 			continue;
552 		if (!rmb_desc->is_reg_mr[link->link_idx]) {
553 			up_read(&lgr->llc_conf_mutex);
554 			goto slow_path;
555 		}
556 	}
557 	/* mr register already */
558 	goto fast_path;
559 slow_path:
560 	do_slow = true;
561 	/* protect against parallel smc_llc_cli_rkey_exchange() and
562 	 * parallel smcr_link_reg_buf()
563 	 */
564 	down_write(&lgr->llc_conf_mutex);
565 	for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
566 		if (!smc_link_active(&lgr->lnk[i]))
567 			continue;
568 		rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc);
569 		if (rc)
570 			goto out;
571 	}
572 fast_path:
573 	/* exchange confirm_rkey msg with peer */
574 	rc = smc_llc_do_confirm_rkey(link, rmb_desc);
575 	if (rc) {
576 		rc = -EFAULT;
577 		goto out;
578 	}
579 	rmb_desc->is_conf_rkey = true;
580 out:
581 	do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex);
582 	smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
583 	return rc;
584 }
585 
smcr_clnt_conf_first_link(struct smc_sock * smc)586 static int smcr_clnt_conf_first_link(struct smc_sock *smc)
587 {
588 	struct smc_link *link = smc->conn.lnk;
589 	struct smc_llc_qentry *qentry;
590 	int rc;
591 
592 	/* Receive CONFIRM LINK request from server over RoCE fabric.
593 	 * Increasing the client's timeout by twice as much as the server's
594 	 * timeout by default can temporarily avoid decline messages of
595 	 * both sides crossing or colliding
596 	 */
597 	qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME,
598 			      SMC_LLC_CONFIRM_LINK);
599 	if (!qentry) {
600 		struct smc_clc_msg_decline dclc;
601 
602 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
603 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
604 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
605 	}
606 	smc_llc_save_peer_uid(qentry);
607 	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
608 	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
609 	if (rc)
610 		return SMC_CLC_DECL_RMBE_EC;
611 
612 	rc = smc_ib_modify_qp_rts(link);
613 	if (rc)
614 		return SMC_CLC_DECL_ERR_RDYLNK;
615 
616 	smc_wr_remember_qp_attr(link);
617 
618 	/* reg the sndbuf if it was vzalloced */
619 	if (smc->conn.sndbuf_desc->is_vm) {
620 		if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
621 			return SMC_CLC_DECL_ERR_REGBUF;
622 	}
623 
624 	/* reg the rmb */
625 	if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
626 		return SMC_CLC_DECL_ERR_REGBUF;
627 
628 	/* confirm_rkey is implicit on 1st contact */
629 	smc->conn.rmb_desc->is_conf_rkey = true;
630 
631 	/* send CONFIRM LINK response over RoCE fabric */
632 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
633 	if (rc < 0)
634 		return SMC_CLC_DECL_TIMEOUT_CL;
635 
636 	smc_llc_link_active(link);
637 	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
638 
639 	if (link->lgr->max_links > 1) {
640 		/* optional 2nd link, receive ADD LINK request from server */
641 		qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
642 				      SMC_LLC_ADD_LINK);
643 		if (!qentry) {
644 			struct smc_clc_msg_decline dclc;
645 
646 			rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
647 					      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
648 			if (rc == -EAGAIN)
649 				rc = 0; /* no DECLINE received, go with one link */
650 			return rc;
651 		}
652 		smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
653 		smc_llc_cli_add_link(link, qentry);
654 	}
655 	return 0;
656 }
657 
smc_isascii(char * hostname)658 static bool smc_isascii(char *hostname)
659 {
660 	int i;
661 
662 	for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
663 		if (!isascii(hostname[i]))
664 			return false;
665 	return true;
666 }
667 
smc_conn_save_peer_info_fce(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)668 static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
669 					struct smc_clc_msg_accept_confirm *clc)
670 {
671 	struct smc_clc_first_contact_ext *fce;
672 	int clc_v2_len;
673 
674 	if (clc->hdr.version == SMC_V1 ||
675 	    !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
676 		return;
677 
678 	if (smc->conn.lgr->is_smcd) {
679 		memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid,
680 		       SMC_MAX_EID_LEN);
681 		clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1);
682 	} else {
683 		memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid,
684 		       SMC_MAX_EID_LEN);
685 		clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1);
686 	}
687 	fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len);
688 	smc->conn.lgr->peer_os = fce->os_type;
689 	smc->conn.lgr->peer_smc_release = fce->release;
690 	if (smc_isascii(fce->hostname))
691 		memcpy(smc->conn.lgr->peer_hostname, fce->hostname,
692 		       SMC_MAX_HOSTNAME_LEN);
693 }
694 
smcr_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)695 static void smcr_conn_save_peer_info(struct smc_sock *smc,
696 				     struct smc_clc_msg_accept_confirm *clc)
697 {
698 	int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
699 
700 	smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx;
701 	smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token);
702 	smc->conn.peer_rmbe_size = bufsize;
703 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
704 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
705 }
706 
smcd_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)707 static void smcd_conn_save_peer_info(struct smc_sock *smc,
708 				     struct smc_clc_msg_accept_confirm *clc)
709 {
710 	int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
711 
712 	smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
713 	smc->conn.peer_token = ntohll(clc->d0.token);
714 	/* msg header takes up space in the buffer */
715 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
716 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
717 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
718 }
719 
smc_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)720 static void smc_conn_save_peer_info(struct smc_sock *smc,
721 				    struct smc_clc_msg_accept_confirm *clc)
722 {
723 	if (smc->conn.lgr->is_smcd)
724 		smcd_conn_save_peer_info(smc, clc);
725 	else
726 		smcr_conn_save_peer_info(smc, clc);
727 	smc_conn_save_peer_info_fce(smc, clc);
728 }
729 
smc_link_save_peer_info(struct smc_link * link,struct smc_clc_msg_accept_confirm * clc,struct smc_init_info * ini)730 static void smc_link_save_peer_info(struct smc_link *link,
731 				    struct smc_clc_msg_accept_confirm *clc,
732 				    struct smc_init_info *ini)
733 {
734 	link->peer_qpn = ntoh24(clc->r0.qpn);
735 	memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE);
736 	memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac));
737 	link->peer_psn = ntoh24(clc->r0.psn);
738 	link->peer_mtu = clc->r0.qp_mtu;
739 }
740 
smc_stat_inc_fback_rsn_cnt(struct smc_sock * smc,struct smc_stats_fback * fback_arr)741 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
742 				       struct smc_stats_fback *fback_arr)
743 {
744 	int cnt;
745 
746 	for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) {
747 		if (fback_arr[cnt].fback_code == smc->fallback_rsn) {
748 			fback_arr[cnt].count++;
749 			break;
750 		}
751 		if (!fback_arr[cnt].fback_code) {
752 			fback_arr[cnt].fback_code = smc->fallback_rsn;
753 			fback_arr[cnt].count++;
754 			break;
755 		}
756 	}
757 }
758 
smc_stat_fallback(struct smc_sock * smc)759 static void smc_stat_fallback(struct smc_sock *smc)
760 {
761 	struct net *net = sock_net(&smc->sk);
762 
763 	mutex_lock(&net->smc.mutex_fback_rsn);
764 	if (smc->listen_smc) {
765 		smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv);
766 		net->smc.fback_rsn->srv_fback_cnt++;
767 	} else {
768 		smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
769 		net->smc.fback_rsn->clnt_fback_cnt++;
770 	}
771 	mutex_unlock(&net->smc.mutex_fback_rsn);
772 }
773 
774 /* must be called under rcu read lock */
smc_fback_wakeup_waitqueue(struct smc_sock * smc,void * key)775 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
776 {
777 	struct socket_wq *wq;
778 	__poll_t flags;
779 
780 	wq = rcu_dereference(smc->sk.sk_wq);
781 	if (!skwq_has_sleeper(wq))
782 		return;
783 
784 	/* wake up smc sk->sk_wq */
785 	if (!key) {
786 		/* sk_state_change */
787 		wake_up_interruptible_all(&wq->wait);
788 	} else {
789 		flags = key_to_poll(key);
790 		if (flags & (EPOLLIN | EPOLLOUT))
791 			/* sk_data_ready or sk_write_space */
792 			wake_up_interruptible_sync_poll(&wq->wait, flags);
793 		else if (flags & EPOLLERR)
794 			/* sk_error_report */
795 			wake_up_interruptible_poll(&wq->wait, flags);
796 	}
797 }
798 
smc_fback_mark_woken(wait_queue_entry_t * wait,unsigned int mode,int sync,void * key)799 static int smc_fback_mark_woken(wait_queue_entry_t *wait,
800 				unsigned int mode, int sync, void *key)
801 {
802 	struct smc_mark_woken *mark =
803 		container_of(wait, struct smc_mark_woken, wait_entry);
804 
805 	mark->woken = true;
806 	mark->key = key;
807 	return 0;
808 }
809 
smc_fback_forward_wakeup(struct smc_sock * smc,struct sock * clcsk,void (* clcsock_callback)(struct sock * sk))810 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
811 				     void (*clcsock_callback)(struct sock *sk))
812 {
813 	struct smc_mark_woken mark = { .woken = false };
814 	struct socket_wq *wq;
815 
816 	init_waitqueue_func_entry(&mark.wait_entry,
817 				  smc_fback_mark_woken);
818 	rcu_read_lock();
819 	wq = rcu_dereference(clcsk->sk_wq);
820 	if (!wq)
821 		goto out;
822 	add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
823 	clcsock_callback(clcsk);
824 	remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
825 
826 	if (mark.woken)
827 		smc_fback_wakeup_waitqueue(smc, mark.key);
828 out:
829 	rcu_read_unlock();
830 }
831 
smc_fback_state_change(struct sock * clcsk)832 static void smc_fback_state_change(struct sock *clcsk)
833 {
834 	struct smc_sock *smc;
835 
836 	read_lock_bh(&clcsk->sk_callback_lock);
837 	smc = smc_clcsock_user_data(clcsk);
838 	if (smc)
839 		smc_fback_forward_wakeup(smc, clcsk,
840 					 smc->clcsk_state_change);
841 	read_unlock_bh(&clcsk->sk_callback_lock);
842 }
843 
smc_fback_data_ready(struct sock * clcsk)844 static void smc_fback_data_ready(struct sock *clcsk)
845 {
846 	struct smc_sock *smc;
847 
848 	read_lock_bh(&clcsk->sk_callback_lock);
849 	smc = smc_clcsock_user_data(clcsk);
850 	if (smc)
851 		smc_fback_forward_wakeup(smc, clcsk,
852 					 smc->clcsk_data_ready);
853 	read_unlock_bh(&clcsk->sk_callback_lock);
854 }
855 
smc_fback_write_space(struct sock * clcsk)856 static void smc_fback_write_space(struct sock *clcsk)
857 {
858 	struct smc_sock *smc;
859 
860 	read_lock_bh(&clcsk->sk_callback_lock);
861 	smc = smc_clcsock_user_data(clcsk);
862 	if (smc)
863 		smc_fback_forward_wakeup(smc, clcsk,
864 					 smc->clcsk_write_space);
865 	read_unlock_bh(&clcsk->sk_callback_lock);
866 }
867 
smc_fback_error_report(struct sock * clcsk)868 static void smc_fback_error_report(struct sock *clcsk)
869 {
870 	struct smc_sock *smc;
871 
872 	read_lock_bh(&clcsk->sk_callback_lock);
873 	smc = smc_clcsock_user_data(clcsk);
874 	if (smc)
875 		smc_fback_forward_wakeup(smc, clcsk,
876 					 smc->clcsk_error_report);
877 	read_unlock_bh(&clcsk->sk_callback_lock);
878 }
879 
smc_fback_replace_callbacks(struct smc_sock * smc)880 static void smc_fback_replace_callbacks(struct smc_sock *smc)
881 {
882 	struct sock *clcsk = smc->clcsock->sk;
883 
884 	write_lock_bh(&clcsk->sk_callback_lock);
885 	clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
886 
887 	smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change,
888 			       &smc->clcsk_state_change);
889 	smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready,
890 			       &smc->clcsk_data_ready);
891 	smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space,
892 			       &smc->clcsk_write_space);
893 	smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report,
894 			       &smc->clcsk_error_report);
895 
896 	write_unlock_bh(&clcsk->sk_callback_lock);
897 }
898 
smc_switch_to_fallback(struct smc_sock * smc,int reason_code)899 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
900 {
901 	int rc = 0;
902 
903 	mutex_lock(&smc->clcsock_release_lock);
904 	if (!smc->clcsock) {
905 		rc = -EBADF;
906 		goto out;
907 	}
908 
909 	smc->use_fallback = true;
910 	smc->fallback_rsn = reason_code;
911 	smc_stat_fallback(smc);
912 	trace_smc_switch_to_fallback(smc, reason_code);
913 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
914 		smc->clcsock->file = smc->sk.sk_socket->file;
915 		smc->clcsock->file->private_data = smc->clcsock;
916 		smc->clcsock->wq.fasync_list =
917 			smc->sk.sk_socket->wq.fasync_list;
918 		smc->sk.sk_socket->wq.fasync_list = NULL;
919 
920 		/* There might be some wait entries remaining
921 		 * in smc sk->sk_wq and they should be woken up
922 		 * as clcsock's wait queue is woken up.
923 		 */
924 		smc_fback_replace_callbacks(smc);
925 	}
926 out:
927 	mutex_unlock(&smc->clcsock_release_lock);
928 	return rc;
929 }
930 
931 /* fall back during connect */
smc_connect_fallback(struct smc_sock * smc,int reason_code)932 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
933 {
934 	struct net *net = sock_net(&smc->sk);
935 	int rc = 0;
936 
937 	rc = smc_switch_to_fallback(smc, reason_code);
938 	if (rc) { /* fallback fails */
939 		this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
940 		if (smc->sk.sk_state == SMC_INIT)
941 			sock_put(&smc->sk); /* passive closing */
942 		return rc;
943 	}
944 	smc_copy_sock_settings_to_clc(smc);
945 	smc->connect_nonblock = 0;
946 	if (smc->sk.sk_state == SMC_INIT)
947 		smc->sk.sk_state = SMC_ACTIVE;
948 	return 0;
949 }
950 
951 /* decline and fall back during connect */
smc_connect_decline_fallback(struct smc_sock * smc,int reason_code,u8 version)952 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
953 					u8 version)
954 {
955 	struct net *net = sock_net(&smc->sk);
956 	int rc;
957 
958 	if (reason_code < 0) { /* error, fallback is not possible */
959 		this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
960 		if (smc->sk.sk_state == SMC_INIT)
961 			sock_put(&smc->sk); /* passive closing */
962 		return reason_code;
963 	}
964 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
965 		rc = smc_clc_send_decline(smc, reason_code, version);
966 		if (rc < 0) {
967 			this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
968 			if (smc->sk.sk_state == SMC_INIT)
969 				sock_put(&smc->sk); /* passive closing */
970 			return rc;
971 		}
972 	}
973 	return smc_connect_fallback(smc, reason_code);
974 }
975 
smc_conn_abort(struct smc_sock * smc,int local_first)976 static void smc_conn_abort(struct smc_sock *smc, int local_first)
977 {
978 	struct smc_connection *conn = &smc->conn;
979 	struct smc_link_group *lgr = conn->lgr;
980 	bool lgr_valid = false;
981 
982 	if (smc_conn_lgr_valid(conn))
983 		lgr_valid = true;
984 
985 	smc_conn_free(conn);
986 	if (local_first && lgr_valid)
987 		smc_lgr_cleanup_early(lgr);
988 }
989 
990 /* check if there is a rdma device available for this connection. */
991 /* called for connect and listen */
smc_find_rdma_device(struct smc_sock * smc,struct smc_init_info * ini)992 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
993 {
994 	/* PNET table look up: search active ib_device and port
995 	 * within same PNETID that also contains the ethernet device
996 	 * used for the internal TCP socket
997 	 */
998 	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
999 	if (!ini->check_smcrv2 && !ini->ib_dev)
1000 		return SMC_CLC_DECL_NOSMCRDEV;
1001 	if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1002 		return SMC_CLC_DECL_NOSMCRDEV;
1003 	return 0;
1004 }
1005 
1006 /* check if there is an ISM device available for this connection. */
1007 /* called for connect and listen */
smc_find_ism_device(struct smc_sock * smc,struct smc_init_info * ini)1008 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1009 {
1010 	/* Find ISM device with same PNETID as connecting interface  */
1011 	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1012 	if (!ini->ism_dev[0])
1013 		return SMC_CLC_DECL_NOSMCDDEV;
1014 	else
1015 		ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1016 	return 0;
1017 }
1018 
1019 /* is chid unique for the ism devices that are already determined? */
smc_find_ism_v2_is_unique_chid(u16 chid,struct smc_init_info * ini,int cnt)1020 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1021 					   int cnt)
1022 {
1023 	int i = (!ini->ism_dev[0]) ? 1 : 0;
1024 
1025 	for (; i < cnt; i++)
1026 		if (ini->ism_chid[i] == chid)
1027 			return false;
1028 	return true;
1029 }
1030 
1031 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1032  * PNETID matching net_device)
1033  */
smc_find_ism_v2_device_clnt(struct smc_sock * smc,struct smc_init_info * ini)1034 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1035 				       struct smc_init_info *ini)
1036 {
1037 	int rc = SMC_CLC_DECL_NOSMCDDEV;
1038 	struct smcd_dev *smcd;
1039 	int i = 1, entry = 1;
1040 	bool is_emulated;
1041 	u16 chid;
1042 
1043 	if (smcd_indicated(ini->smc_type_v1))
1044 		rc = 0;		/* already initialized for V1 */
1045 	mutex_lock(&smcd_dev_list.mutex);
1046 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1047 		if (smcd->going_away || smcd == ini->ism_dev[0])
1048 			continue;
1049 		chid = smc_ism_get_chid(smcd);
1050 		if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1051 			continue;
1052 		is_emulated = __smc_ism_is_emulated(chid);
1053 		if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1054 		    smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1055 			if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES)
1056 				/* It's the last GID-CHID entry left in CLC
1057 				 * Proposal SMC-Dv2 extension, but an Emulated-
1058 				 * ISM device will take two entries. So give
1059 				 * up it and try the next potential ISM device.
1060 				 */
1061 				continue;
1062 			ini->ism_dev[i] = smcd;
1063 			ini->ism_chid[i] = chid;
1064 			ini->is_smcd = true;
1065 			rc = 0;
1066 			i++;
1067 			entry = is_emulated ? entry + 2 : entry + 1;
1068 			if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES)
1069 				break;
1070 		}
1071 	}
1072 	mutex_unlock(&smcd_dev_list.mutex);
1073 	ini->ism_offered_cnt = i - 1;
1074 	if (!ini->ism_dev[0] && !ini->ism_dev[1])
1075 		ini->smcd_version = 0;
1076 
1077 	return rc;
1078 }
1079 
1080 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
smc_connect_ism_vlan_setup(struct smc_sock * smc,struct smc_init_info * ini)1081 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
1082 				      struct smc_init_info *ini)
1083 {
1084 	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1085 		return SMC_CLC_DECL_ISMVLANERR;
1086 	return 0;
1087 }
1088 
smc_find_proposal_devices(struct smc_sock * smc,struct smc_init_info * ini)1089 static int smc_find_proposal_devices(struct smc_sock *smc,
1090 				     struct smc_init_info *ini)
1091 {
1092 	int rc = 0;
1093 
1094 	/* check if there is an ism device available */
1095 	if (!(ini->smcd_version & SMC_V1) ||
1096 	    smc_find_ism_device(smc, ini) ||
1097 	    smc_connect_ism_vlan_setup(smc, ini))
1098 		ini->smcd_version &= ~SMC_V1;
1099 	/* else ISM V1 is supported for this connection */
1100 
1101 	/* check if there is an rdma device available */
1102 	if (!(ini->smcr_version & SMC_V1) ||
1103 	    smc_find_rdma_device(smc, ini))
1104 		ini->smcr_version &= ~SMC_V1;
1105 	/* else RDMA is supported for this connection */
1106 
1107 	ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1108 					      ini->smcr_version & SMC_V1);
1109 
1110 	/* check if there is an ism v2 device available */
1111 	if (!(ini->smcd_version & SMC_V2) ||
1112 	    !smc_ism_is_v2_capable() ||
1113 	    smc_find_ism_v2_device_clnt(smc, ini))
1114 		ini->smcd_version &= ~SMC_V2;
1115 
1116 	/* check if there is an rdma v2 device available */
1117 	ini->check_smcrv2 = true;
1118 	ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1119 	if (!(ini->smcr_version & SMC_V2) ||
1120 	    smc->clcsock->sk->sk_family != AF_INET ||
1121 	    !smc_clc_ueid_count() ||
1122 	    smc_find_rdma_device(smc, ini))
1123 		ini->smcr_version &= ~SMC_V2;
1124 	ini->check_smcrv2 = false;
1125 
1126 	ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1127 					      ini->smcr_version & SMC_V2);
1128 
1129 	/* if neither ISM nor RDMA are supported, fallback */
1130 	if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1131 		rc = SMC_CLC_DECL_NOSMCDEV;
1132 
1133 	return rc;
1134 }
1135 
1136 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1137  * used, the VLAN ID will be registered again during the connection setup.
1138  */
smc_connect_ism_vlan_cleanup(struct smc_sock * smc,struct smc_init_info * ini)1139 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc,
1140 					struct smc_init_info *ini)
1141 {
1142 	if (!smcd_indicated(ini->smc_type_v1))
1143 		return 0;
1144 	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1145 		return SMC_CLC_DECL_CNFERR;
1146 	return 0;
1147 }
1148 
1149 #define SMC_CLC_MAX_ACCEPT_LEN \
1150 	(sizeof(struct smc_clc_msg_accept_confirm) + \
1151 	 sizeof(struct smc_clc_first_contact_ext_v2x) + \
1152 	 sizeof(struct smc_clc_msg_trail))
1153 
1154 /* CLC handshake during connect */
smc_connect_clc(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)1155 static int smc_connect_clc(struct smc_sock *smc,
1156 			   struct smc_clc_msg_accept_confirm *aclc,
1157 			   struct smc_init_info *ini)
1158 {
1159 	int rc = 0;
1160 
1161 	/* do inband token exchange */
1162 	rc = smc_clc_send_proposal(smc, ini);
1163 	if (rc)
1164 		return rc;
1165 	/* receive SMC Accept CLC message */
1166 	return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN,
1167 				SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1168 }
1169 
smc_fill_gid_list(struct smc_link_group * lgr,struct smc_gidlist * gidlist,struct smc_ib_device * known_dev,u8 * known_gid)1170 void smc_fill_gid_list(struct smc_link_group *lgr,
1171 		       struct smc_gidlist *gidlist,
1172 		       struct smc_ib_device *known_dev, u8 *known_gid)
1173 {
1174 	struct smc_init_info *alt_ini = NULL;
1175 
1176 	memset(gidlist, 0, sizeof(*gidlist));
1177 	memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1178 
1179 	alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1180 	if (!alt_ini)
1181 		goto out;
1182 
1183 	alt_ini->vlan_id = lgr->vlan_id;
1184 	alt_ini->check_smcrv2 = true;
1185 	alt_ini->smcrv2.saddr = lgr->saddr;
1186 	smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1187 
1188 	if (!alt_ini->smcrv2.ib_dev_v2)
1189 		goto out;
1190 
1191 	memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1192 	       SMC_GID_SIZE);
1193 
1194 out:
1195 	kfree(alt_ini);
1196 }
1197 
smc_connect_rdma_v2_prepare(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)1198 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1199 				       struct smc_clc_msg_accept_confirm *aclc,
1200 				       struct smc_init_info *ini)
1201 {
1202 	struct smc_clc_first_contact_ext *fce =
1203 		smc_get_clc_first_contact_ext(aclc, false);
1204 	struct net *net = sock_net(&smc->sk);
1205 	int rc;
1206 
1207 	if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1208 		return 0;
1209 
1210 	if (fce->v2_direct) {
1211 		memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1212 		ini->smcrv2.uses_gateway = false;
1213 	} else {
1214 		if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr,
1215 				      smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1216 				      ini->smcrv2.nexthop_mac,
1217 				      &ini->smcrv2.uses_gateway))
1218 			return SMC_CLC_DECL_NOROUTE;
1219 		if (!ini->smcrv2.uses_gateway) {
1220 			/* mismatch: peer claims indirect, but its direct */
1221 			return SMC_CLC_DECL_NOINDIRECT;
1222 		}
1223 	}
1224 
1225 	ini->release_nr = fce->release;
1226 	rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1227 	if (rc)
1228 		return rc;
1229 
1230 	return 0;
1231 }
1232 
1233 /* setup for RDMA connection of client */
smc_connect_rdma(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)1234 static int smc_connect_rdma(struct smc_sock *smc,
1235 			    struct smc_clc_msg_accept_confirm *aclc,
1236 			    struct smc_init_info *ini)
1237 {
1238 	int i, reason_code = 0;
1239 	struct smc_link *link;
1240 	u8 *eid = NULL;
1241 
1242 	ini->is_smcd = false;
1243 	ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1244 	ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1245 	memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1246 	memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1247 	memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1248 	ini->max_conns = SMC_CONN_PER_LGR_MAX;
1249 	ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1250 
1251 	reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1252 	if (reason_code)
1253 		return reason_code;
1254 
1255 	mutex_lock(&smc_client_lgr_pending);
1256 	reason_code = smc_conn_create(smc, ini);
1257 	if (reason_code) {
1258 		mutex_unlock(&smc_client_lgr_pending);
1259 		return reason_code;
1260 	}
1261 
1262 	smc_conn_save_peer_info(smc, aclc);
1263 
1264 	if (ini->first_contact_local) {
1265 		link = smc->conn.lnk;
1266 	} else {
1267 		/* set link that was assigned by server */
1268 		link = NULL;
1269 		for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1270 			struct smc_link *l = &smc->conn.lgr->lnk[i];
1271 
1272 			if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1273 			    !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1274 				    SMC_GID_SIZE) &&
1275 			    (aclc->hdr.version > SMC_V1 ||
1276 			     !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1277 				     sizeof(l->peer_mac)))) {
1278 				link = l;
1279 				break;
1280 			}
1281 		}
1282 		if (!link) {
1283 			reason_code = SMC_CLC_DECL_NOSRVLINK;
1284 			goto connect_abort;
1285 		}
1286 		smc_switch_link_and_count(&smc->conn, link);
1287 	}
1288 
1289 	/* create send buffer and rmb */
1290 	if (smc_buf_create(smc, false)) {
1291 		reason_code = SMC_CLC_DECL_MEM;
1292 		goto connect_abort;
1293 	}
1294 
1295 	if (ini->first_contact_local)
1296 		smc_link_save_peer_info(link, aclc, ini);
1297 
1298 	if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1299 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1300 		goto connect_abort;
1301 	}
1302 
1303 	smc_rx_init(smc);
1304 
1305 	if (ini->first_contact_local) {
1306 		if (smc_ib_ready_link(link)) {
1307 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1308 			goto connect_abort;
1309 		}
1310 	} else {
1311 		/* reg sendbufs if they were vzalloced */
1312 		if (smc->conn.sndbuf_desc->is_vm) {
1313 			if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) {
1314 				reason_code = SMC_CLC_DECL_ERR_REGBUF;
1315 				goto connect_abort;
1316 			}
1317 		}
1318 		if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1319 			reason_code = SMC_CLC_DECL_ERR_REGBUF;
1320 			goto connect_abort;
1321 		}
1322 	}
1323 
1324 	if (aclc->hdr.version > SMC_V1) {
1325 		eid = aclc->r1.eid;
1326 		if (ini->first_contact_local)
1327 			smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1328 					  link->smcibdev, link->gid);
1329 	}
1330 
1331 	reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1332 					   aclc->hdr.version, eid, ini);
1333 	if (reason_code)
1334 		goto connect_abort;
1335 
1336 	smc_tx_init(smc);
1337 
1338 	if (ini->first_contact_local) {
1339 		/* QP confirmation over RoCE fabric */
1340 		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1341 		reason_code = smcr_clnt_conf_first_link(smc);
1342 		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1343 		if (reason_code)
1344 			goto connect_abort;
1345 	}
1346 	mutex_unlock(&smc_client_lgr_pending);
1347 
1348 	smc_copy_sock_settings_to_clc(smc);
1349 	smc->connect_nonblock = 0;
1350 	if (smc->sk.sk_state == SMC_INIT)
1351 		smc->sk.sk_state = SMC_ACTIVE;
1352 
1353 	return 0;
1354 connect_abort:
1355 	smc_conn_abort(smc, ini->first_contact_local);
1356 	mutex_unlock(&smc_client_lgr_pending);
1357 	smc->connect_nonblock = 0;
1358 
1359 	return reason_code;
1360 }
1361 
1362 /* The server has chosen one of the proposed ISM devices for the communication.
1363  * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1364  */
1365 static int
smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)1366 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc,
1367 			       struct smc_init_info *ini)
1368 {
1369 	int i;
1370 
1371 	for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1372 		if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1373 			ini->ism_selected = i;
1374 			return 0;
1375 		}
1376 	}
1377 
1378 	return -EPROTO;
1379 }
1380 
1381 /* setup for ISM connection of client */
smc_connect_ism(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)1382 static int smc_connect_ism(struct smc_sock *smc,
1383 			   struct smc_clc_msg_accept_confirm *aclc,
1384 			   struct smc_init_info *ini)
1385 {
1386 	u8 *eid = NULL;
1387 	int rc = 0;
1388 
1389 	ini->is_smcd = true;
1390 	ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1391 
1392 	if (aclc->hdr.version == SMC_V2) {
1393 		if (ini->first_contact_peer) {
1394 			struct smc_clc_first_contact_ext *fce =
1395 				smc_get_clc_first_contact_ext(aclc, true);
1396 
1397 			ini->release_nr = fce->release;
1398 			rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1399 			if (rc)
1400 				return rc;
1401 		}
1402 
1403 		rc = smc_v2_determine_accepted_chid(aclc, ini);
1404 		if (rc)
1405 			return rc;
1406 
1407 		if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected]))
1408 			ini->ism_peer_gid[ini->ism_selected].gid_ext =
1409 						ntohll(aclc->d1.gid_ext);
1410 		/* for non-Emulated-ISM devices, peer gid_ext remains 0. */
1411 	}
1412 	ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid);
1413 
1414 	/* there is only one lgr role for SMC-D; use server lock */
1415 	mutex_lock(&smc_server_lgr_pending);
1416 	rc = smc_conn_create(smc, ini);
1417 	if (rc) {
1418 		mutex_unlock(&smc_server_lgr_pending);
1419 		return rc;
1420 	}
1421 
1422 	/* Create send and receive buffers */
1423 	rc = smc_buf_create(smc, true);
1424 	if (rc) {
1425 		rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1426 		goto connect_abort;
1427 	}
1428 
1429 	smc_conn_save_peer_info(smc, aclc);
1430 
1431 	if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) {
1432 		rc = smcd_buf_attach(smc);
1433 		if (rc) {
1434 			rc = SMC_CLC_DECL_MEM;	/* try to fallback */
1435 			goto connect_abort;
1436 		}
1437 	}
1438 	smc_rx_init(smc);
1439 	smc_tx_init(smc);
1440 
1441 	if (aclc->hdr.version > SMC_V1)
1442 		eid = aclc->d1.eid;
1443 
1444 	rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1445 				  aclc->hdr.version, eid, ini);
1446 	if (rc)
1447 		goto connect_abort;
1448 	mutex_unlock(&smc_server_lgr_pending);
1449 
1450 	smc_copy_sock_settings_to_clc(smc);
1451 	smc->connect_nonblock = 0;
1452 	if (smc->sk.sk_state == SMC_INIT)
1453 		smc->sk.sk_state = SMC_ACTIVE;
1454 
1455 	return 0;
1456 connect_abort:
1457 	smc_conn_abort(smc, ini->first_contact_local);
1458 	mutex_unlock(&smc_server_lgr_pending);
1459 	smc->connect_nonblock = 0;
1460 
1461 	return rc;
1462 }
1463 
1464 /* check if received accept type and version matches a proposed one */
smc_connect_check_aclc(struct smc_init_info * ini,struct smc_clc_msg_accept_confirm * aclc)1465 static int smc_connect_check_aclc(struct smc_init_info *ini,
1466 				  struct smc_clc_msg_accept_confirm *aclc)
1467 {
1468 	if (aclc->hdr.version >= SMC_V2) {
1469 		if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1470 		     !smcr_indicated(ini->smc_type_v2)) ||
1471 		    (aclc->hdr.typev1 == SMC_TYPE_D &&
1472 		     !smcd_indicated(ini->smc_type_v2)))
1473 			return SMC_CLC_DECL_MODEUNSUPP;
1474 	} else {
1475 		if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1476 		     !smcr_indicated(ini->smc_type_v1)) ||
1477 		    (aclc->hdr.typev1 == SMC_TYPE_D &&
1478 		     !smcd_indicated(ini->smc_type_v1)))
1479 			return SMC_CLC_DECL_MODEUNSUPP;
1480 	}
1481 
1482 	return 0;
1483 }
1484 
1485 /* perform steps before actually connecting */
__smc_connect(struct smc_sock * smc)1486 static int __smc_connect(struct smc_sock *smc)
1487 {
1488 	u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1489 	struct smc_clc_msg_accept_confirm *aclc;
1490 	struct smc_init_info *ini = NULL;
1491 	u8 *buf = NULL;
1492 	int rc = 0;
1493 
1494 	if (smc->use_fallback)
1495 		return smc_connect_fallback(smc, smc->fallback_rsn);
1496 
1497 	/* if peer has not signalled SMC-capability, fall back */
1498 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1499 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1500 
1501 	/* IPSec connections opt out of SMC optimizations */
1502 	if (using_ipsec(smc))
1503 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1504 						    version);
1505 
1506 	ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1507 	if (!ini)
1508 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1509 						    version);
1510 
1511 	ini->smcd_version = SMC_V1 | SMC_V2;
1512 	ini->smcr_version = SMC_V1 | SMC_V2;
1513 	ini->smc_type_v1 = SMC_TYPE_B;
1514 	ini->smc_type_v2 = SMC_TYPE_B;
1515 
1516 	/* get vlan id from IP device */
1517 	if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1518 		ini->smcd_version &= ~SMC_V1;
1519 		ini->smcr_version = 0;
1520 		ini->smc_type_v1 = SMC_TYPE_N;
1521 	}
1522 
1523 	rc = smc_find_proposal_devices(smc, ini);
1524 	if (rc)
1525 		goto fallback;
1526 
1527 	buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1528 	if (!buf) {
1529 		rc = SMC_CLC_DECL_MEM;
1530 		goto fallback;
1531 	}
1532 	aclc = (struct smc_clc_msg_accept_confirm *)buf;
1533 
1534 	/* perform CLC handshake */
1535 	rc = smc_connect_clc(smc, aclc, ini);
1536 	if (rc) {
1537 		/* -EAGAIN on timeout, see tcp_recvmsg() */
1538 		if (rc == -EAGAIN) {
1539 			rc = -ETIMEDOUT;
1540 			smc->sk.sk_err = ETIMEDOUT;
1541 		}
1542 		goto vlan_cleanup;
1543 	}
1544 
1545 	/* check if smc modes and versions of CLC proposal and accept match */
1546 	rc = smc_connect_check_aclc(ini, aclc);
1547 	version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1548 	if (rc)
1549 		goto vlan_cleanup;
1550 
1551 	/* depending on previous steps, connect using rdma or ism */
1552 	if (aclc->hdr.typev1 == SMC_TYPE_R) {
1553 		ini->smcr_version = version;
1554 		rc = smc_connect_rdma(smc, aclc, ini);
1555 	} else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1556 		ini->smcd_version = version;
1557 		rc = smc_connect_ism(smc, aclc, ini);
1558 	}
1559 	if (rc)
1560 		goto vlan_cleanup;
1561 
1562 	SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1563 	smc_connect_ism_vlan_cleanup(smc, ini);
1564 	kfree(buf);
1565 	kfree(ini);
1566 	return 0;
1567 
1568 vlan_cleanup:
1569 	smc_connect_ism_vlan_cleanup(smc, ini);
1570 	kfree(buf);
1571 fallback:
1572 	kfree(ini);
1573 	return smc_connect_decline_fallback(smc, rc, version);
1574 }
1575 
smc_connect_work(struct work_struct * work)1576 static void smc_connect_work(struct work_struct *work)
1577 {
1578 	struct smc_sock *smc = container_of(work, struct smc_sock,
1579 					    connect_work);
1580 	long timeo = smc->sk.sk_sndtimeo;
1581 	int rc = 0;
1582 
1583 	if (!timeo)
1584 		timeo = MAX_SCHEDULE_TIMEOUT;
1585 	lock_sock(smc->clcsock->sk);
1586 	if (smc->clcsock->sk->sk_err) {
1587 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
1588 	} else if ((1 << smc->clcsock->sk->sk_state) &
1589 					(TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1590 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1591 		if ((rc == -EPIPE) &&
1592 		    ((1 << smc->clcsock->sk->sk_state) &
1593 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1594 			rc = 0;
1595 	}
1596 	release_sock(smc->clcsock->sk);
1597 	lock_sock(&smc->sk);
1598 	if (rc != 0 || smc->sk.sk_err) {
1599 		smc->sk.sk_state = SMC_CLOSED;
1600 		if (rc == -EPIPE || rc == -EAGAIN)
1601 			smc->sk.sk_err = EPIPE;
1602 		else if (rc == -ECONNREFUSED)
1603 			smc->sk.sk_err = ECONNREFUSED;
1604 		else if (signal_pending(current))
1605 			smc->sk.sk_err = -sock_intr_errno(timeo);
1606 		sock_put(&smc->sk); /* passive closing */
1607 		goto out;
1608 	}
1609 
1610 	rc = __smc_connect(smc);
1611 	if (rc < 0)
1612 		smc->sk.sk_err = -rc;
1613 
1614 out:
1615 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1616 		if (smc->sk.sk_err) {
1617 			smc->sk.sk_state_change(&smc->sk);
1618 		} else { /* allow polling before and after fallback decision */
1619 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1620 			smc->sk.sk_write_space(&smc->sk);
1621 		}
1622 	}
1623 	release_sock(&smc->sk);
1624 }
1625 
smc_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)1626 int smc_connect(struct socket *sock, struct sockaddr *addr,
1627 		int alen, int flags)
1628 {
1629 	struct sock *sk = sock->sk;
1630 	struct smc_sock *smc;
1631 	int rc = -EINVAL;
1632 
1633 	smc = smc_sk(sk);
1634 
1635 	/* separate smc parameter checking to be safe */
1636 	if (alen < sizeof(addr->sa_family))
1637 		goto out_err;
1638 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1639 		goto out_err;
1640 
1641 	lock_sock(sk);
1642 	switch (sock->state) {
1643 	default:
1644 		rc = -EINVAL;
1645 		goto out;
1646 	case SS_CONNECTED:
1647 		rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1648 		goto out;
1649 	case SS_CONNECTING:
1650 		if (sk->sk_state == SMC_ACTIVE)
1651 			goto connected;
1652 		break;
1653 	case SS_UNCONNECTED:
1654 		sock->state = SS_CONNECTING;
1655 		break;
1656 	}
1657 
1658 	switch (sk->sk_state) {
1659 	default:
1660 		goto out;
1661 	case SMC_CLOSED:
1662 		rc = sock_error(sk) ? : -ECONNABORTED;
1663 		sock->state = SS_UNCONNECTED;
1664 		goto out;
1665 	case SMC_ACTIVE:
1666 		rc = -EISCONN;
1667 		goto out;
1668 	case SMC_INIT:
1669 		break;
1670 	}
1671 
1672 	smc_copy_sock_settings_to_clc(smc);
1673 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1674 	if (smc->connect_nonblock) {
1675 		rc = -EALREADY;
1676 		goto out;
1677 	}
1678 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
1679 	if (rc && rc != -EINPROGRESS)
1680 		goto out;
1681 
1682 	if (smc->use_fallback) {
1683 		sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1684 		goto out;
1685 	}
1686 	sock_hold(&smc->sk); /* sock put in passive closing */
1687 	if (flags & O_NONBLOCK) {
1688 		if (queue_work(smc_hs_wq, &smc->connect_work))
1689 			smc->connect_nonblock = 1;
1690 		rc = -EINPROGRESS;
1691 		goto out;
1692 	} else {
1693 		rc = __smc_connect(smc);
1694 		if (rc < 0)
1695 			goto out;
1696 	}
1697 
1698 connected:
1699 	rc = 0;
1700 	sock->state = SS_CONNECTED;
1701 out:
1702 	release_sock(sk);
1703 out_err:
1704 	return rc;
1705 }
1706 
smc_clcsock_accept(struct smc_sock * lsmc,struct smc_sock ** new_smc)1707 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1708 {
1709 	struct socket *new_clcsock = NULL;
1710 	struct sock *lsk = &lsmc->sk;
1711 	struct sock *new_sk;
1712 	int rc = -EINVAL;
1713 
1714 	release_sock(lsk);
1715 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1716 	if (!new_sk) {
1717 		rc = -ENOMEM;
1718 		lsk->sk_err = ENOMEM;
1719 		*new_smc = NULL;
1720 		lock_sock(lsk);
1721 		goto out;
1722 	}
1723 	*new_smc = smc_sk(new_sk);
1724 
1725 	mutex_lock(&lsmc->clcsock_release_lock);
1726 	if (lsmc->clcsock)
1727 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1728 	mutex_unlock(&lsmc->clcsock_release_lock);
1729 	lock_sock(lsk);
1730 	if  (rc < 0 && rc != -EAGAIN)
1731 		lsk->sk_err = -rc;
1732 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1733 		new_sk->sk_prot->unhash(new_sk);
1734 		if (new_clcsock)
1735 			sock_release(new_clcsock);
1736 		new_sk->sk_state = SMC_CLOSED;
1737 		smc_sock_set_flag(new_sk, SOCK_DEAD);
1738 		sock_put(new_sk); /* final */
1739 		*new_smc = NULL;
1740 		goto out;
1741 	}
1742 
1743 	/* new clcsock has inherited the smc listen-specific sk_data_ready
1744 	 * function; switch it back to the original sk_data_ready function
1745 	 */
1746 	new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1747 
1748 	/* if new clcsock has also inherited the fallback-specific callback
1749 	 * functions, switch them back to the original ones.
1750 	 */
1751 	if (lsmc->use_fallback) {
1752 		if (lsmc->clcsk_state_change)
1753 			new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1754 		if (lsmc->clcsk_write_space)
1755 			new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1756 		if (lsmc->clcsk_error_report)
1757 			new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1758 	}
1759 
1760 	(*new_smc)->clcsock = new_clcsock;
1761 out:
1762 	return rc;
1763 }
1764 
1765 /* add a just created sock to the accept queue of the listen sock as
1766  * candidate for a following socket accept call from user space
1767  */
smc_accept_enqueue(struct sock * parent,struct sock * sk)1768 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1769 {
1770 	struct smc_sock *par = smc_sk(parent);
1771 
1772 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
1773 	spin_lock(&par->accept_q_lock);
1774 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1775 	spin_unlock(&par->accept_q_lock);
1776 	sk_acceptq_added(parent);
1777 }
1778 
1779 /* remove a socket from the accept queue of its parental listening socket */
smc_accept_unlink(struct sock * sk)1780 static void smc_accept_unlink(struct sock *sk)
1781 {
1782 	struct smc_sock *par = smc_sk(sk)->listen_smc;
1783 
1784 	spin_lock(&par->accept_q_lock);
1785 	list_del_init(&smc_sk(sk)->accept_q);
1786 	spin_unlock(&par->accept_q_lock);
1787 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1788 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
1789 }
1790 
1791 /* remove a sock from the accept queue to bind it to a new socket created
1792  * for a socket accept call from user space
1793  */
smc_accept_dequeue(struct sock * parent,struct socket * new_sock)1794 struct sock *smc_accept_dequeue(struct sock *parent,
1795 				struct socket *new_sock)
1796 {
1797 	struct smc_sock *isk, *n;
1798 	struct sock *new_sk;
1799 
1800 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1801 		new_sk = (struct sock *)isk;
1802 
1803 		smc_accept_unlink(new_sk);
1804 		if (new_sk->sk_state == SMC_CLOSED) {
1805 			new_sk->sk_prot->unhash(new_sk);
1806 			if (isk->clcsock) {
1807 				sock_release(isk->clcsock);
1808 				isk->clcsock = NULL;
1809 			}
1810 			sock_put(new_sk); /* final */
1811 			continue;
1812 		}
1813 		if (new_sock) {
1814 			sock_graft(new_sk, new_sock);
1815 			new_sock->state = SS_CONNECTED;
1816 			if (isk->use_fallback) {
1817 				smc_sk(new_sk)->clcsock->file = new_sock->file;
1818 				isk->clcsock->file->private_data = isk->clcsock;
1819 			}
1820 		}
1821 		return new_sk;
1822 	}
1823 	return NULL;
1824 }
1825 
1826 /* clean up for a created but never accepted sock */
smc_close_non_accepted(struct sock * sk)1827 void smc_close_non_accepted(struct sock *sk)
1828 {
1829 	struct smc_sock *smc = smc_sk(sk);
1830 
1831 	sock_hold(sk); /* sock_put below */
1832 	lock_sock(sk);
1833 	if (!sk->sk_lingertime)
1834 		/* wait for peer closing */
1835 		WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT);
1836 	__smc_release(smc);
1837 	release_sock(sk);
1838 	sock_put(sk); /* sock_hold above */
1839 	sock_put(sk); /* final sock_put */
1840 }
1841 
smcr_serv_conf_first_link(struct smc_sock * smc)1842 static int smcr_serv_conf_first_link(struct smc_sock *smc)
1843 {
1844 	struct smc_link *link = smc->conn.lnk;
1845 	struct smc_llc_qentry *qentry;
1846 	int rc;
1847 
1848 	/* reg the sndbuf if it was vzalloced*/
1849 	if (smc->conn.sndbuf_desc->is_vm) {
1850 		if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1851 			return SMC_CLC_DECL_ERR_REGBUF;
1852 	}
1853 
1854 	/* reg the rmb */
1855 	if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1856 		return SMC_CLC_DECL_ERR_REGBUF;
1857 
1858 	/* send CONFIRM LINK request to client over the RoCE fabric */
1859 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1860 	if (rc < 0)
1861 		return SMC_CLC_DECL_TIMEOUT_CL;
1862 
1863 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1864 	qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1865 			      SMC_LLC_CONFIRM_LINK);
1866 	if (!qentry) {
1867 		struct smc_clc_msg_decline dclc;
1868 
1869 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1870 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1871 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1872 	}
1873 	smc_llc_save_peer_uid(qentry);
1874 	rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1875 	smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1876 	if (rc)
1877 		return SMC_CLC_DECL_RMBE_EC;
1878 
1879 	/* confirm_rkey is implicit on 1st contact */
1880 	smc->conn.rmb_desc->is_conf_rkey = true;
1881 
1882 	smc_llc_link_active(link);
1883 	smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1884 
1885 	if (link->lgr->max_links > 1) {
1886 		down_write(&link->lgr->llc_conf_mutex);
1887 		/* initial contact - try to establish second link */
1888 		smc_llc_srv_add_link(link, NULL);
1889 		up_write(&link->lgr->llc_conf_mutex);
1890 	}
1891 	return 0;
1892 }
1893 
1894 /* listen worker: finish */
smc_listen_out(struct smc_sock * new_smc)1895 static void smc_listen_out(struct smc_sock *new_smc)
1896 {
1897 	struct smc_sock *lsmc = new_smc->listen_smc;
1898 	struct sock *newsmcsk = &new_smc->sk;
1899 
1900 	if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1901 		atomic_dec(&lsmc->queued_smc_hs);
1902 
1903 	release_sock(newsmcsk); /* lock in smc_listen_work() */
1904 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1905 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1906 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1907 		release_sock(&lsmc->sk);
1908 	} else { /* no longer listening */
1909 		smc_close_non_accepted(newsmcsk);
1910 	}
1911 
1912 	/* Wake up accept */
1913 	lsmc->sk.sk_data_ready(&lsmc->sk);
1914 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1915 }
1916 
1917 /* listen worker: finish in state connected */
smc_listen_out_connected(struct smc_sock * new_smc)1918 static void smc_listen_out_connected(struct smc_sock *new_smc)
1919 {
1920 	struct sock *newsmcsk = &new_smc->sk;
1921 
1922 	if (newsmcsk->sk_state == SMC_INIT)
1923 		newsmcsk->sk_state = SMC_ACTIVE;
1924 
1925 	smc_listen_out(new_smc);
1926 }
1927 
1928 /* listen worker: finish in error state */
smc_listen_out_err(struct smc_sock * new_smc)1929 static void smc_listen_out_err(struct smc_sock *new_smc)
1930 {
1931 	struct sock *newsmcsk = &new_smc->sk;
1932 	struct net *net = sock_net(newsmcsk);
1933 
1934 	this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1935 	if (newsmcsk->sk_state == SMC_INIT)
1936 		sock_put(&new_smc->sk); /* passive closing */
1937 	newsmcsk->sk_state = SMC_CLOSED;
1938 
1939 	smc_listen_out(new_smc);
1940 }
1941 
1942 /* listen worker: decline and fall back if possible */
smc_listen_decline(struct smc_sock * new_smc,int reason_code,int local_first,u8 version)1943 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1944 			       int local_first, u8 version)
1945 {
1946 	/* RDMA setup failed, switch back to TCP */
1947 	smc_conn_abort(new_smc, local_first);
1948 	if (reason_code < 0 ||
1949 	    smc_switch_to_fallback(new_smc, reason_code)) {
1950 		/* error, no fallback possible */
1951 		smc_listen_out_err(new_smc);
1952 		return;
1953 	}
1954 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1955 		if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1956 			smc_listen_out_err(new_smc);
1957 			return;
1958 		}
1959 	}
1960 	smc_listen_out_connected(new_smc);
1961 }
1962 
1963 /* listen worker: version checking */
smc_listen_v2_check(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)1964 static int smc_listen_v2_check(struct smc_sock *new_smc,
1965 			       struct smc_clc_msg_proposal *pclc,
1966 			       struct smc_init_info *ini)
1967 {
1968 	struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
1969 	struct smc_clc_v2_extension *pclc_v2_ext;
1970 	int rc = SMC_CLC_DECL_PEERNOSMC;
1971 
1972 	ini->smc_type_v1 = pclc->hdr.typev1;
1973 	ini->smc_type_v2 = pclc->hdr.typev2;
1974 	ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1975 	ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1976 	if (pclc->hdr.version > SMC_V1) {
1977 		if (smcd_indicated(ini->smc_type_v2))
1978 			ini->smcd_version |= SMC_V2;
1979 		if (smcr_indicated(ini->smc_type_v2))
1980 			ini->smcr_version |= SMC_V2;
1981 	}
1982 	if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
1983 		rc = SMC_CLC_DECL_PEERNOSMC;
1984 		goto out;
1985 	}
1986 	pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1987 	if (!pclc_v2_ext) {
1988 		ini->smcd_version &= ~SMC_V2;
1989 		ini->smcr_version &= ~SMC_V2;
1990 		rc = SMC_CLC_DECL_NOV2EXT;
1991 		goto out;
1992 	}
1993 	pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
1994 	if (ini->smcd_version & SMC_V2) {
1995 		if (!smc_ism_is_v2_capable()) {
1996 			ini->smcd_version &= ~SMC_V2;
1997 			rc = SMC_CLC_DECL_NOISM2SUPP;
1998 		} else if (!pclc_smcd_v2_ext) {
1999 			ini->smcd_version &= ~SMC_V2;
2000 			rc = SMC_CLC_DECL_NOV2DEXT;
2001 		} else if (!pclc_v2_ext->hdr.eid_cnt &&
2002 			   !pclc_v2_ext->hdr.flag.seid) {
2003 			ini->smcd_version &= ~SMC_V2;
2004 			rc = SMC_CLC_DECL_NOUEID;
2005 		}
2006 	}
2007 	if (ini->smcr_version & SMC_V2) {
2008 		if (!pclc_v2_ext->hdr.eid_cnt) {
2009 			ini->smcr_version &= ~SMC_V2;
2010 			rc = SMC_CLC_DECL_NOUEID;
2011 		}
2012 	}
2013 
2014 	ini->release_nr = pclc_v2_ext->hdr.flag.release;
2015 	if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE)
2016 		ini->release_nr = SMC_RELEASE;
2017 
2018 out:
2019 	if (!ini->smcd_version && !ini->smcr_version)
2020 		return rc;
2021 
2022 	return 0;
2023 }
2024 
2025 /* listen worker: check prefixes */
smc_listen_prfx_check(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc)2026 static int smc_listen_prfx_check(struct smc_sock *new_smc,
2027 				 struct smc_clc_msg_proposal *pclc)
2028 {
2029 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
2030 	struct socket *newclcsock = new_smc->clcsock;
2031 
2032 	if (pclc->hdr.typev1 == SMC_TYPE_N)
2033 		return 0;
2034 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2035 	if (!pclc_prfx)
2036 		return -EPROTO;
2037 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2038 		return SMC_CLC_DECL_DIFFPREFIX;
2039 
2040 	return 0;
2041 }
2042 
2043 /* listen worker: initialize connection and buffers */
smc_listen_rdma_init(struct smc_sock * new_smc,struct smc_init_info * ini)2044 static int smc_listen_rdma_init(struct smc_sock *new_smc,
2045 				struct smc_init_info *ini)
2046 {
2047 	int rc;
2048 
2049 	/* allocate connection / link group */
2050 	rc = smc_conn_create(new_smc, ini);
2051 	if (rc)
2052 		return rc;
2053 
2054 	/* create send buffer and rmb */
2055 	if (smc_buf_create(new_smc, false)) {
2056 		smc_conn_abort(new_smc, ini->first_contact_local);
2057 		return SMC_CLC_DECL_MEM;
2058 	}
2059 
2060 	return 0;
2061 }
2062 
2063 /* listen worker: initialize connection and buffers for SMC-D */
smc_listen_ism_init(struct smc_sock * new_smc,struct smc_init_info * ini)2064 static int smc_listen_ism_init(struct smc_sock *new_smc,
2065 			       struct smc_init_info *ini)
2066 {
2067 	int rc;
2068 
2069 	rc = smc_conn_create(new_smc, ini);
2070 	if (rc)
2071 		return rc;
2072 
2073 	/* Create send and receive buffers */
2074 	rc = smc_buf_create(new_smc, true);
2075 	if (rc) {
2076 		smc_conn_abort(new_smc, ini->first_contact_local);
2077 		return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2078 					 SMC_CLC_DECL_MEM;
2079 	}
2080 
2081 	return 0;
2082 }
2083 
smc_is_already_selected(struct smcd_dev * smcd,struct smc_init_info * ini,int matches)2084 static bool smc_is_already_selected(struct smcd_dev *smcd,
2085 				    struct smc_init_info *ini,
2086 				    int matches)
2087 {
2088 	int i;
2089 
2090 	for (i = 0; i < matches; i++)
2091 		if (smcd == ini->ism_dev[i])
2092 			return true;
2093 
2094 	return false;
2095 }
2096 
2097 /* check for ISM devices matching proposed ISM devices */
smc_check_ism_v2_match(struct smc_init_info * ini,u16 proposed_chid,struct smcd_gid * proposed_gid,unsigned int * matches)2098 static void smc_check_ism_v2_match(struct smc_init_info *ini,
2099 				   u16 proposed_chid,
2100 				   struct smcd_gid *proposed_gid,
2101 				   unsigned int *matches)
2102 {
2103 	struct smcd_dev *smcd;
2104 
2105 	list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2106 		if (smcd->going_away)
2107 			continue;
2108 		if (smc_is_already_selected(smcd, ini, *matches))
2109 			continue;
2110 		if (smc_ism_get_chid(smcd) == proposed_chid &&
2111 		    !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2112 			ini->ism_peer_gid[*matches].gid = proposed_gid->gid;
2113 			if (__smc_ism_is_emulated(proposed_chid))
2114 				ini->ism_peer_gid[*matches].gid_ext =
2115 							proposed_gid->gid_ext;
2116 				/* non-Emulated-ISM's peer gid_ext remains 0. */
2117 			ini->ism_dev[*matches] = smcd;
2118 			(*matches)++;
2119 			break;
2120 		}
2121 	}
2122 }
2123 
smc_find_ism_store_rc(u32 rc,struct smc_init_info * ini)2124 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini)
2125 {
2126 	if (!ini->rc)
2127 		ini->rc = rc;
2128 }
2129 
smc_find_ism_v2_device_serv(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)2130 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc,
2131 					struct smc_clc_msg_proposal *pclc,
2132 					struct smc_init_info *ini)
2133 {
2134 	struct smc_clc_smcd_v2_extension *smcd_v2_ext;
2135 	struct smc_clc_v2_extension *smc_v2_ext;
2136 	struct smc_clc_msg_smcd *pclc_smcd;
2137 	unsigned int matches = 0;
2138 	struct smcd_gid smcd_gid;
2139 	u8 smcd_version;
2140 	u8 *eid = NULL;
2141 	int i, rc;
2142 	u16 chid;
2143 
2144 	if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2145 		goto not_found;
2146 
2147 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
2148 	smc_v2_ext = smc_get_clc_v2_ext(pclc);
2149 	smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2150 	if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext)
2151 		goto not_found;
2152 
2153 	mutex_lock(&smcd_dev_list.mutex);
2154 	if (pclc_smcd->ism.chid) {
2155 		/* check for ISM device matching proposed native ISM device */
2156 		smcd_gid.gid = ntohll(pclc_smcd->ism.gid);
2157 		smcd_gid.gid_ext = 0;
2158 		smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2159 				       &smcd_gid, &matches);
2160 	}
2161 	for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) {
2162 		/* check for ISM devices matching proposed non-native ISM
2163 		 * devices
2164 		 */
2165 		smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid);
2166 		smcd_gid.gid_ext = 0;
2167 		chid = ntohs(smcd_v2_ext->gidchid[i].chid);
2168 		if (__smc_ism_is_emulated(chid)) {
2169 			if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt ||
2170 			    chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid))
2171 				/* each Emulated-ISM device takes two GID-CHID
2172 				 * entries and CHID of the second entry repeats
2173 				 * that of the first entry.
2174 				 *
2175 				 * So check if the next GID-CHID entry exists
2176 				 * and both two entries' CHIDs are the same.
2177 				 */
2178 				continue;
2179 			smcd_gid.gid_ext =
2180 				ntohll(smcd_v2_ext->gidchid[++i].gid);
2181 		}
2182 		smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches);
2183 	}
2184 	mutex_unlock(&smcd_dev_list.mutex);
2185 
2186 	if (!ini->ism_dev[0]) {
2187 		smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2188 		goto not_found;
2189 	}
2190 
2191 	smc_ism_get_system_eid(&eid);
2192 	if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2193 			       smcd_v2_ext->system_eid, eid))
2194 		goto not_found;
2195 
2196 	/* separate - outside the smcd_dev_list.lock */
2197 	smcd_version = ini->smcd_version;
2198 	for (i = 0; i < matches; i++) {
2199 		ini->smcd_version = SMC_V2;
2200 		ini->is_smcd = true;
2201 		ini->ism_selected = i;
2202 		rc = smc_listen_ism_init(new_smc, ini);
2203 		if (rc) {
2204 			smc_find_ism_store_rc(rc, ini);
2205 			/* try next active ISM device */
2206 			continue;
2207 		}
2208 		return; /* matching and usable V2 ISM device found */
2209 	}
2210 	/* no V2 ISM device could be initialized */
2211 	ini->smcd_version = smcd_version;	/* restore original value */
2212 	ini->negotiated_eid[0] = 0;
2213 
2214 not_found:
2215 	ini->smcd_version &= ~SMC_V2;
2216 	ini->ism_dev[0] = NULL;
2217 	ini->is_smcd = false;
2218 }
2219 
smc_find_ism_v1_device_serv(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)2220 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2221 					struct smc_clc_msg_proposal *pclc,
2222 					struct smc_init_info *ini)
2223 {
2224 	struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2225 	int rc = 0;
2226 
2227 	/* check if ISM V1 is available */
2228 	if (!(ini->smcd_version & SMC_V1) ||
2229 	    !smcd_indicated(ini->smc_type_v1) ||
2230 	    !pclc_smcd)
2231 		goto not_found;
2232 	ini->is_smcd = true; /* prepare ISM check */
2233 	ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid);
2234 	ini->ism_peer_gid[0].gid_ext = 0;
2235 	rc = smc_find_ism_device(new_smc, ini);
2236 	if (rc)
2237 		goto not_found;
2238 	ini->ism_selected = 0;
2239 	rc = smc_listen_ism_init(new_smc, ini);
2240 	if (!rc)
2241 		return;		/* V1 ISM device found */
2242 
2243 not_found:
2244 	smc_find_ism_store_rc(rc, ini);
2245 	ini->smcd_version &= ~SMC_V1;
2246 	ini->ism_dev[0] = NULL;
2247 	ini->is_smcd = false;
2248 }
2249 
2250 /* listen worker: register buffers */
smc_listen_rdma_reg(struct smc_sock * new_smc,bool local_first)2251 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2252 {
2253 	struct smc_connection *conn = &new_smc->conn;
2254 
2255 	if (!local_first) {
2256 		/* reg sendbufs if they were vzalloced */
2257 		if (conn->sndbuf_desc->is_vm) {
2258 			if (smcr_lgr_reg_sndbufs(conn->lnk,
2259 						 conn->sndbuf_desc))
2260 				return SMC_CLC_DECL_ERR_REGBUF;
2261 		}
2262 		if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2263 			return SMC_CLC_DECL_ERR_REGBUF;
2264 	}
2265 
2266 	return 0;
2267 }
2268 
smc_find_rdma_v2_device_serv(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)2269 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2270 					 struct smc_clc_msg_proposal *pclc,
2271 					 struct smc_init_info *ini)
2272 {
2273 	struct smc_clc_v2_extension *smc_v2_ext;
2274 	u8 smcr_version;
2275 	int rc;
2276 
2277 	if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2278 		goto not_found;
2279 
2280 	smc_v2_ext = smc_get_clc_v2_ext(pclc);
2281 	if (!smc_v2_ext ||
2282 	    !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2283 		goto not_found;
2284 
2285 	/* prepare RDMA check */
2286 	memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2287 	memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2288 	memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2289 	ini->check_smcrv2 = true;
2290 	ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2291 	ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2292 	ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2293 	rc = smc_find_rdma_device(new_smc, ini);
2294 	if (rc) {
2295 		smc_find_ism_store_rc(rc, ini);
2296 		goto not_found;
2297 	}
2298 	if (!ini->smcrv2.uses_gateway)
2299 		memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2300 
2301 	smcr_version = ini->smcr_version;
2302 	ini->smcr_version = SMC_V2;
2303 	rc = smc_listen_rdma_init(new_smc, ini);
2304 	if (!rc) {
2305 		rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2306 		if (rc)
2307 			smc_conn_abort(new_smc, ini->first_contact_local);
2308 	}
2309 	if (!rc)
2310 		return;
2311 	ini->smcr_version = smcr_version;
2312 	smc_find_ism_store_rc(rc, ini);
2313 
2314 not_found:
2315 	ini->smcr_version &= ~SMC_V2;
2316 	ini->smcrv2.ib_dev_v2 = NULL;
2317 	ini->check_smcrv2 = false;
2318 }
2319 
smc_find_rdma_v1_device_serv(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)2320 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2321 					struct smc_clc_msg_proposal *pclc,
2322 					struct smc_init_info *ini)
2323 {
2324 	int rc;
2325 
2326 	if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2327 		return SMC_CLC_DECL_NOSMCDEV;
2328 
2329 	/* prepare RDMA check */
2330 	memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2331 	memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2332 	memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2333 	rc = smc_find_rdma_device(new_smc, ini);
2334 	if (rc) {
2335 		/* no RDMA device found */
2336 		return SMC_CLC_DECL_NOSMCDEV;
2337 	}
2338 	rc = smc_listen_rdma_init(new_smc, ini);
2339 	if (rc)
2340 		return rc;
2341 	return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2342 }
2343 
2344 /* determine the local device matching to proposal */
smc_listen_find_device(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)2345 static int smc_listen_find_device(struct smc_sock *new_smc,
2346 				  struct smc_clc_msg_proposal *pclc,
2347 				  struct smc_init_info *ini)
2348 {
2349 	int prfx_rc;
2350 
2351 	/* check for ISM device matching V2 proposed device */
2352 	smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2353 	if (ini->ism_dev[0])
2354 		return 0;
2355 
2356 	/* check for matching IP prefix and subnet length (V1) */
2357 	prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2358 	if (prfx_rc)
2359 		smc_find_ism_store_rc(prfx_rc, ini);
2360 
2361 	/* get vlan id from IP device */
2362 	if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2363 		return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2364 
2365 	/* check for ISM device matching V1 proposed device */
2366 	if (!prfx_rc)
2367 		smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2368 	if (ini->ism_dev[0])
2369 		return 0;
2370 
2371 	if (!smcr_indicated(pclc->hdr.typev1) &&
2372 	    !smcr_indicated(pclc->hdr.typev2))
2373 		/* skip RDMA and decline */
2374 		return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2375 
2376 	/* check if RDMA V2 is available */
2377 	smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2378 	if (ini->smcrv2.ib_dev_v2)
2379 		return 0;
2380 
2381 	/* check if RDMA V1 is available */
2382 	if (!prfx_rc) {
2383 		int rc;
2384 
2385 		rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2386 		smc_find_ism_store_rc(rc, ini);
2387 		return (!rc) ? 0 : ini->rc;
2388 	}
2389 	return prfx_rc;
2390 }
2391 
2392 /* listen worker: finish RDMA setup */
smc_listen_rdma_finish(struct smc_sock * new_smc,struct smc_clc_msg_accept_confirm * cclc,bool local_first,struct smc_init_info * ini)2393 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2394 				  struct smc_clc_msg_accept_confirm *cclc,
2395 				  bool local_first,
2396 				  struct smc_init_info *ini)
2397 {
2398 	struct smc_link *link = new_smc->conn.lnk;
2399 	int reason_code = 0;
2400 
2401 	if (local_first)
2402 		smc_link_save_peer_info(link, cclc, ini);
2403 
2404 	if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2405 		return SMC_CLC_DECL_ERR_RTOK;
2406 
2407 	if (local_first) {
2408 		if (smc_ib_ready_link(link))
2409 			return SMC_CLC_DECL_ERR_RDYLNK;
2410 		/* QP confirmation over RoCE fabric */
2411 		smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2412 		reason_code = smcr_serv_conf_first_link(new_smc);
2413 		smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2414 	}
2415 	return reason_code;
2416 }
2417 
2418 /* setup for connection of server */
smc_listen_work(struct work_struct * work)2419 static void smc_listen_work(struct work_struct *work)
2420 {
2421 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
2422 						smc_listen_work);
2423 	struct socket *newclcsock = new_smc->clcsock;
2424 	struct smc_clc_msg_accept_confirm *cclc;
2425 	struct smc_clc_msg_proposal_area *buf;
2426 	struct smc_clc_msg_proposal *pclc;
2427 	struct smc_init_info *ini = NULL;
2428 	u8 proposal_version = SMC_V1;
2429 	u8 accept_version;
2430 	int rc = 0;
2431 
2432 	lock_sock(&new_smc->sk); /* release in smc_listen_out() */
2433 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2434 		return smc_listen_out_err(new_smc);
2435 
2436 	if (new_smc->use_fallback) {
2437 		smc_listen_out_connected(new_smc);
2438 		return;
2439 	}
2440 
2441 	/* check if peer is smc capable */
2442 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
2443 		rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2444 		if (rc)
2445 			smc_listen_out_err(new_smc);
2446 		else
2447 			smc_listen_out_connected(new_smc);
2448 		return;
2449 	}
2450 
2451 	/* do inband token exchange -
2452 	 * wait for and receive SMC Proposal CLC message
2453 	 */
2454 	buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2455 	if (!buf) {
2456 		rc = SMC_CLC_DECL_MEM;
2457 		goto out_decl;
2458 	}
2459 	pclc = (struct smc_clc_msg_proposal *)buf;
2460 	rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2461 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2462 	if (rc)
2463 		goto out_decl;
2464 
2465 	if (pclc->hdr.version > SMC_V1)
2466 		proposal_version = SMC_V2;
2467 
2468 	/* IPSec connections opt out of SMC optimizations */
2469 	if (using_ipsec(new_smc)) {
2470 		rc = SMC_CLC_DECL_IPSEC;
2471 		goto out_decl;
2472 	}
2473 
2474 	ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2475 	if (!ini) {
2476 		rc = SMC_CLC_DECL_MEM;
2477 		goto out_decl;
2478 	}
2479 
2480 	/* initial version checking */
2481 	rc = smc_listen_v2_check(new_smc, pclc, ini);
2482 	if (rc)
2483 		goto out_decl;
2484 
2485 	rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini);
2486 	if (rc)
2487 		goto out_decl;
2488 
2489 	mutex_lock(&smc_server_lgr_pending);
2490 	smc_rx_init(new_smc);
2491 	smc_tx_init(new_smc);
2492 
2493 	/* determine ISM or RoCE device used for connection */
2494 	rc = smc_listen_find_device(new_smc, pclc, ini);
2495 	if (rc)
2496 		goto out_unlock;
2497 
2498 	/* send SMC Accept CLC message */
2499 	accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2500 	rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2501 				 accept_version, ini->negotiated_eid, ini);
2502 	if (rc)
2503 		goto out_unlock;
2504 
2505 	/* SMC-D does not need this lock any more */
2506 	if (ini->is_smcd)
2507 		mutex_unlock(&smc_server_lgr_pending);
2508 
2509 	/* receive SMC Confirm CLC message */
2510 	memset(buf, 0, sizeof(*buf));
2511 	cclc = (struct smc_clc_msg_accept_confirm *)buf;
2512 	rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2513 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2514 	if (rc) {
2515 		if (!ini->is_smcd)
2516 			goto out_unlock;
2517 		goto out_decl;
2518 	}
2519 
2520 	rc = smc_clc_v2x_features_confirm_check(cclc, ini);
2521 	if (rc) {
2522 		if (!ini->is_smcd)
2523 			goto out_unlock;
2524 		goto out_decl;
2525 	}
2526 
2527 	/* fce smc release version is needed in smc_listen_rdma_finish,
2528 	 * so save fce info here.
2529 	 */
2530 	smc_conn_save_peer_info_fce(new_smc, cclc);
2531 
2532 	/* finish worker */
2533 	if (!ini->is_smcd) {
2534 		rc = smc_listen_rdma_finish(new_smc, cclc,
2535 					    ini->first_contact_local, ini);
2536 		if (rc)
2537 			goto out_unlock;
2538 		mutex_unlock(&smc_server_lgr_pending);
2539 	}
2540 	smc_conn_save_peer_info(new_smc, cclc);
2541 
2542 	if (ini->is_smcd &&
2543 	    smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) {
2544 		rc = smcd_buf_attach(new_smc);
2545 		if (rc)
2546 			goto out_decl;
2547 	}
2548 
2549 	smc_listen_out_connected(new_smc);
2550 	SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2551 	goto out_free;
2552 
2553 out_unlock:
2554 	mutex_unlock(&smc_server_lgr_pending);
2555 out_decl:
2556 	smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2557 			   proposal_version);
2558 out_free:
2559 	kfree(ini);
2560 	kfree(buf);
2561 }
2562 
smc_tcp_listen_work(struct work_struct * work)2563 static void smc_tcp_listen_work(struct work_struct *work)
2564 {
2565 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
2566 					     tcp_listen_work);
2567 	struct sock *lsk = &lsmc->sk;
2568 	struct smc_sock *new_smc;
2569 	int rc = 0;
2570 
2571 	lock_sock(lsk);
2572 	while (lsk->sk_state == SMC_LISTEN) {
2573 		rc = smc_clcsock_accept(lsmc, &new_smc);
2574 		if (rc) /* clcsock accept queue empty or error */
2575 			goto out;
2576 		if (!new_smc)
2577 			continue;
2578 
2579 		if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2580 			atomic_inc(&lsmc->queued_smc_hs);
2581 
2582 		new_smc->listen_smc = lsmc;
2583 		new_smc->use_fallback = lsmc->use_fallback;
2584 		new_smc->fallback_rsn = lsmc->fallback_rsn;
2585 		sock_hold(lsk); /* sock_put in smc_listen_work */
2586 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2587 		smc_copy_sock_settings_to_smc(new_smc);
2588 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
2589 		if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2590 			sock_put(&new_smc->sk);
2591 	}
2592 
2593 out:
2594 	release_sock(lsk);
2595 	sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2596 }
2597 
smc_clcsock_data_ready(struct sock * listen_clcsock)2598 static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2599 {
2600 	struct smc_sock *lsmc;
2601 
2602 	read_lock_bh(&listen_clcsock->sk_callback_lock);
2603 	lsmc = smc_clcsock_user_data(listen_clcsock);
2604 	if (!lsmc)
2605 		goto out;
2606 	lsmc->clcsk_data_ready(listen_clcsock);
2607 	if (lsmc->sk.sk_state == SMC_LISTEN) {
2608 		sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2609 		if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2610 			sock_put(&lsmc->sk);
2611 	}
2612 out:
2613 	read_unlock_bh(&listen_clcsock->sk_callback_lock);
2614 }
2615 
smc_listen(struct socket * sock,int backlog)2616 int smc_listen(struct socket *sock, int backlog)
2617 {
2618 	struct sock *sk = sock->sk;
2619 	struct smc_sock *smc;
2620 	int rc;
2621 
2622 	smc = smc_sk(sk);
2623 	lock_sock(sk);
2624 
2625 	rc = -EINVAL;
2626 	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2627 	    smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2628 		goto out;
2629 
2630 	rc = 0;
2631 	if (sk->sk_state == SMC_LISTEN) {
2632 		sk->sk_max_ack_backlog = backlog;
2633 		goto out;
2634 	}
2635 	/* some socket options are handled in core, so we could not apply
2636 	 * them to the clc socket -- copy smc socket options to clc socket
2637 	 */
2638 	smc_copy_sock_settings_to_clc(smc);
2639 	if (!smc->use_fallback)
2640 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2641 
2642 	/* save original sk_data_ready function and establish
2643 	 * smc-specific sk_data_ready function
2644 	 */
2645 	write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2646 	smc->clcsock->sk->sk_user_data =
2647 		(void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
2648 	smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2649 			       smc_clcsock_data_ready, &smc->clcsk_data_ready);
2650 	write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2651 
2652 	/* save original ops */
2653 	smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2654 
2655 	smc->af_ops = *smc->ori_af_ops;
2656 	smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2657 
2658 	inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2659 
2660 	if (smc->limit_smc_hs)
2661 		tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2662 
2663 	rc = kernel_listen(smc->clcsock, backlog);
2664 	if (rc) {
2665 		write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2666 		smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2667 				       &smc->clcsk_data_ready);
2668 		smc->clcsock->sk->sk_user_data = NULL;
2669 		write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2670 		goto out;
2671 	}
2672 	sk->sk_max_ack_backlog = backlog;
2673 	sk->sk_ack_backlog = 0;
2674 	sk->sk_state = SMC_LISTEN;
2675 
2676 out:
2677 	release_sock(sk);
2678 	return rc;
2679 }
2680 
smc_accept(struct socket * sock,struct socket * new_sock,struct proto_accept_arg * arg)2681 int smc_accept(struct socket *sock, struct socket *new_sock,
2682 	       struct proto_accept_arg *arg)
2683 {
2684 	struct sock *sk = sock->sk, *nsk;
2685 	DECLARE_WAITQUEUE(wait, current);
2686 	struct smc_sock *lsmc;
2687 	long timeo;
2688 	int rc = 0;
2689 
2690 	lsmc = smc_sk(sk);
2691 	sock_hold(sk); /* sock_put below */
2692 	lock_sock(sk);
2693 
2694 	if (lsmc->sk.sk_state != SMC_LISTEN) {
2695 		rc = -EINVAL;
2696 		release_sock(sk);
2697 		goto out;
2698 	}
2699 
2700 	/* Wait for an incoming connection */
2701 	timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
2702 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
2703 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2704 		set_current_state(TASK_INTERRUPTIBLE);
2705 		if (!timeo) {
2706 			rc = -EAGAIN;
2707 			break;
2708 		}
2709 		release_sock(sk);
2710 		timeo = schedule_timeout(timeo);
2711 		/* wakeup by sk_data_ready in smc_listen_work() */
2712 		sched_annotate_sleep();
2713 		lock_sock(sk);
2714 		if (signal_pending(current)) {
2715 			rc = sock_intr_errno(timeo);
2716 			break;
2717 		}
2718 	}
2719 	set_current_state(TASK_RUNNING);
2720 	remove_wait_queue(sk_sleep(sk), &wait);
2721 
2722 	if (!rc)
2723 		rc = sock_error(nsk);
2724 	release_sock(sk);
2725 	if (rc)
2726 		goto out;
2727 
2728 	if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) {
2729 		/* wait till data arrives on the socket */
2730 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
2731 								MSEC_PER_SEC);
2732 		if (smc_sk(nsk)->use_fallback) {
2733 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2734 
2735 			lock_sock(clcsk);
2736 			if (skb_queue_empty(&clcsk->sk_receive_queue))
2737 				sk_wait_data(clcsk, &timeo, NULL);
2738 			release_sock(clcsk);
2739 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2740 			lock_sock(nsk);
2741 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
2742 			release_sock(nsk);
2743 		}
2744 	}
2745 
2746 out:
2747 	sock_put(sk); /* sock_hold above */
2748 	return rc;
2749 }
2750 
smc_getname(struct socket * sock,struct sockaddr * addr,int peer)2751 int smc_getname(struct socket *sock, struct sockaddr *addr,
2752 		int peer)
2753 {
2754 	struct smc_sock *smc;
2755 
2756 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2757 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2758 		return -ENOTCONN;
2759 
2760 	smc = smc_sk(sock->sk);
2761 
2762 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2763 }
2764 
smc_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)2765 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2766 {
2767 	struct sock *sk = sock->sk;
2768 	struct smc_sock *smc;
2769 	int rc;
2770 
2771 	smc = smc_sk(sk);
2772 	lock_sock(sk);
2773 
2774 	/* SMC does not support connect with fastopen */
2775 	if (msg->msg_flags & MSG_FASTOPEN) {
2776 		/* not connected yet, fallback */
2777 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2778 			rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2779 			if (rc)
2780 				goto out;
2781 		} else {
2782 			rc = -EINVAL;
2783 			goto out;
2784 		}
2785 	} else if ((sk->sk_state != SMC_ACTIVE) &&
2786 		   (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2787 		   (sk->sk_state != SMC_INIT)) {
2788 		rc = -EPIPE;
2789 		goto out;
2790 	}
2791 
2792 	if (smc->use_fallback) {
2793 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2794 	} else {
2795 		rc = smc_tx_sendmsg(smc, msg, len);
2796 		SMC_STAT_TX_PAYLOAD(smc, len, rc);
2797 	}
2798 out:
2799 	release_sock(sk);
2800 	return rc;
2801 }
2802 
smc_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)2803 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2804 		int flags)
2805 {
2806 	struct sock *sk = sock->sk;
2807 	struct smc_sock *smc;
2808 	int rc = -ENOTCONN;
2809 
2810 	smc = smc_sk(sk);
2811 	lock_sock(sk);
2812 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2813 		/* socket was connected before, no more data to read */
2814 		rc = 0;
2815 		goto out;
2816 	}
2817 	if ((sk->sk_state == SMC_INIT) ||
2818 	    (sk->sk_state == SMC_LISTEN) ||
2819 	    (sk->sk_state == SMC_CLOSED))
2820 		goto out;
2821 
2822 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2823 		rc = 0;
2824 		goto out;
2825 	}
2826 
2827 	if (smc->use_fallback) {
2828 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2829 	} else {
2830 		msg->msg_namelen = 0;
2831 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2832 		SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2833 	}
2834 
2835 out:
2836 	release_sock(sk);
2837 	return rc;
2838 }
2839 
smc_accept_poll(struct sock * parent)2840 static __poll_t smc_accept_poll(struct sock *parent)
2841 {
2842 	struct smc_sock *isk = smc_sk(parent);
2843 	__poll_t mask = 0;
2844 
2845 	spin_lock(&isk->accept_q_lock);
2846 	if (!list_empty(&isk->accept_q))
2847 		mask = EPOLLIN | EPOLLRDNORM;
2848 	spin_unlock(&isk->accept_q_lock);
2849 
2850 	return mask;
2851 }
2852 
smc_poll(struct file * file,struct socket * sock,poll_table * wait)2853 __poll_t smc_poll(struct file *file, struct socket *sock,
2854 		  poll_table *wait)
2855 {
2856 	struct sock *sk = sock->sk;
2857 	struct smc_sock *smc;
2858 	__poll_t mask = 0;
2859 
2860 	if (!sk)
2861 		return EPOLLNVAL;
2862 
2863 	smc = smc_sk(sock->sk);
2864 	if (smc->use_fallback) {
2865 		/* delegate to CLC child sock */
2866 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2867 		sk->sk_err = smc->clcsock->sk->sk_err;
2868 	} else {
2869 		if (sk->sk_state != SMC_CLOSED)
2870 			sock_poll_wait(file, sock, wait);
2871 		if (sk->sk_err)
2872 			mask |= EPOLLERR;
2873 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2874 		    (sk->sk_state == SMC_CLOSED))
2875 			mask |= EPOLLHUP;
2876 		if (sk->sk_state == SMC_LISTEN) {
2877 			/* woken up by sk_data_ready in smc_listen_work() */
2878 			mask |= smc_accept_poll(sk);
2879 		} else if (smc->use_fallback) { /* as result of connect_work()*/
2880 			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2881 							   wait);
2882 			sk->sk_err = smc->clcsock->sk->sk_err;
2883 		} else {
2884 			if ((sk->sk_state != SMC_INIT &&
2885 			     atomic_read(&smc->conn.sndbuf_space)) ||
2886 			    sk->sk_shutdown & SEND_SHUTDOWN) {
2887 				mask |= EPOLLOUT | EPOLLWRNORM;
2888 			} else {
2889 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2890 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2891 
2892 				if (sk->sk_state != SMC_INIT) {
2893 					/* Race breaker the same way as tcp_poll(). */
2894 					smp_mb__after_atomic();
2895 					if (atomic_read(&smc->conn.sndbuf_space))
2896 						mask |= EPOLLOUT | EPOLLWRNORM;
2897 				}
2898 			}
2899 			if (atomic_read(&smc->conn.bytes_to_rcv))
2900 				mask |= EPOLLIN | EPOLLRDNORM;
2901 			if (sk->sk_shutdown & RCV_SHUTDOWN)
2902 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2903 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
2904 				mask |= EPOLLIN;
2905 			if (smc->conn.urg_state == SMC_URG_VALID)
2906 				mask |= EPOLLPRI;
2907 		}
2908 	}
2909 
2910 	return mask;
2911 }
2912 
smc_shutdown(struct socket * sock,int how)2913 int smc_shutdown(struct socket *sock, int how)
2914 {
2915 	struct sock *sk = sock->sk;
2916 	bool do_shutdown = true;
2917 	struct smc_sock *smc;
2918 	int rc = -EINVAL;
2919 	int old_state;
2920 	int rc1 = 0;
2921 
2922 	smc = smc_sk(sk);
2923 
2924 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
2925 		return rc;
2926 
2927 	lock_sock(sk);
2928 
2929 	if (sock->state == SS_CONNECTING) {
2930 		if (sk->sk_state == SMC_ACTIVE)
2931 			sock->state = SS_CONNECTED;
2932 		else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2933 			 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2934 			 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2935 			 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2936 			 sk->sk_state == SMC_APPFINCLOSEWAIT)
2937 			sock->state = SS_DISCONNECTING;
2938 	}
2939 
2940 	rc = -ENOTCONN;
2941 	if ((sk->sk_state != SMC_ACTIVE) &&
2942 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2943 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2944 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2945 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2946 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
2947 		goto out;
2948 	if (smc->use_fallback) {
2949 		rc = kernel_sock_shutdown(smc->clcsock, how);
2950 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2951 		if (sk->sk_shutdown == SHUTDOWN_MASK) {
2952 			sk->sk_state = SMC_CLOSED;
2953 			sk->sk_socket->state = SS_UNCONNECTED;
2954 			sock_put(sk);
2955 		}
2956 		goto out;
2957 	}
2958 	switch (how) {
2959 	case SHUT_RDWR:		/* shutdown in both directions */
2960 		old_state = sk->sk_state;
2961 		rc = smc_close_active(smc);
2962 		if (old_state == SMC_ACTIVE &&
2963 		    sk->sk_state == SMC_PEERCLOSEWAIT1)
2964 			do_shutdown = false;
2965 		break;
2966 	case SHUT_WR:
2967 		rc = smc_close_shutdown_write(smc);
2968 		break;
2969 	case SHUT_RD:
2970 		rc = 0;
2971 		/* nothing more to do because peer is not involved */
2972 		break;
2973 	}
2974 	if (do_shutdown && smc->clcsock)
2975 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
2976 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
2977 	sk->sk_shutdown |= how + 1;
2978 
2979 	if (sk->sk_state == SMC_CLOSED)
2980 		sock->state = SS_UNCONNECTED;
2981 	else
2982 		sock->state = SS_DISCONNECTING;
2983 out:
2984 	release_sock(sk);
2985 	return rc ? rc : rc1;
2986 }
2987 
__smc_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)2988 static int __smc_getsockopt(struct socket *sock, int level, int optname,
2989 			    char __user *optval, int __user *optlen)
2990 {
2991 	struct smc_sock *smc;
2992 	int val, len;
2993 
2994 	smc = smc_sk(sock->sk);
2995 
2996 	if (get_user(len, optlen))
2997 		return -EFAULT;
2998 
2999 	len = min_t(int, len, sizeof(int));
3000 
3001 	if (len < 0)
3002 		return -EINVAL;
3003 
3004 	switch (optname) {
3005 	case SMC_LIMIT_HS:
3006 		val = smc->limit_smc_hs;
3007 		break;
3008 	default:
3009 		return -EOPNOTSUPP;
3010 	}
3011 
3012 	if (put_user(len, optlen))
3013 		return -EFAULT;
3014 	if (copy_to_user(optval, &val, len))
3015 		return -EFAULT;
3016 
3017 	return 0;
3018 }
3019 
__smc_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)3020 static int __smc_setsockopt(struct socket *sock, int level, int optname,
3021 			    sockptr_t optval, unsigned int optlen)
3022 {
3023 	struct sock *sk = sock->sk;
3024 	struct smc_sock *smc;
3025 	int val, rc;
3026 
3027 	smc = smc_sk(sk);
3028 
3029 	lock_sock(sk);
3030 	switch (optname) {
3031 	case SMC_LIMIT_HS:
3032 		if (optlen < sizeof(int)) {
3033 			rc = -EINVAL;
3034 			break;
3035 		}
3036 		if (copy_from_sockptr(&val, optval, sizeof(int))) {
3037 			rc = -EFAULT;
3038 			break;
3039 		}
3040 
3041 		smc->limit_smc_hs = !!val;
3042 		rc = 0;
3043 		break;
3044 	default:
3045 		rc = -EOPNOTSUPP;
3046 		break;
3047 	}
3048 	release_sock(sk);
3049 
3050 	return rc;
3051 }
3052 
smc_setsockopt(struct socket * sock,int level,int optname,sockptr_t optval,unsigned int optlen)3053 int smc_setsockopt(struct socket *sock, int level, int optname,
3054 		   sockptr_t optval, unsigned int optlen)
3055 {
3056 	struct sock *sk = sock->sk;
3057 	struct smc_sock *smc;
3058 	int val, rc;
3059 
3060 	if (level == SOL_TCP && optname == TCP_ULP)
3061 		return -EOPNOTSUPP;
3062 	else if (level == SOL_SMC)
3063 		return __smc_setsockopt(sock, level, optname, optval, optlen);
3064 
3065 	smc = smc_sk(sk);
3066 
3067 	/* generic setsockopts reaching us here always apply to the
3068 	 * CLC socket
3069 	 */
3070 	mutex_lock(&smc->clcsock_release_lock);
3071 	if (!smc->clcsock) {
3072 		mutex_unlock(&smc->clcsock_release_lock);
3073 		return -EBADF;
3074 	}
3075 	if (unlikely(!smc->clcsock->ops->setsockopt))
3076 		rc = -EOPNOTSUPP;
3077 	else
3078 		rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
3079 						   optval, optlen);
3080 	if (smc->clcsock->sk->sk_err) {
3081 		sk->sk_err = smc->clcsock->sk->sk_err;
3082 		sk_error_report(sk);
3083 	}
3084 	mutex_unlock(&smc->clcsock_release_lock);
3085 
3086 	if (optlen < sizeof(int))
3087 		return -EINVAL;
3088 	if (copy_from_sockptr(&val, optval, sizeof(int)))
3089 		return -EFAULT;
3090 
3091 	lock_sock(sk);
3092 	if (rc || smc->use_fallback)
3093 		goto out;
3094 	switch (optname) {
3095 	case TCP_FASTOPEN:
3096 	case TCP_FASTOPEN_CONNECT:
3097 	case TCP_FASTOPEN_KEY:
3098 	case TCP_FASTOPEN_NO_COOKIE:
3099 		/* option not supported by SMC */
3100 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3101 			rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3102 		} else {
3103 			rc = -EINVAL;
3104 		}
3105 		break;
3106 	case TCP_NODELAY:
3107 		if (sk->sk_state != SMC_INIT &&
3108 		    sk->sk_state != SMC_LISTEN &&
3109 		    sk->sk_state != SMC_CLOSED) {
3110 			if (val) {
3111 				SMC_STAT_INC(smc, ndly_cnt);
3112 				smc_tx_pending(&smc->conn);
3113 				cancel_delayed_work(&smc->conn.tx_work);
3114 			}
3115 		}
3116 		break;
3117 	case TCP_CORK:
3118 		if (sk->sk_state != SMC_INIT &&
3119 		    sk->sk_state != SMC_LISTEN &&
3120 		    sk->sk_state != SMC_CLOSED) {
3121 			if (!val) {
3122 				SMC_STAT_INC(smc, cork_cnt);
3123 				smc_tx_pending(&smc->conn);
3124 				cancel_delayed_work(&smc->conn.tx_work);
3125 			}
3126 		}
3127 		break;
3128 	case TCP_DEFER_ACCEPT:
3129 		smc->sockopt_defer_accept = val;
3130 		break;
3131 	default:
3132 		break;
3133 	}
3134 out:
3135 	release_sock(sk);
3136 
3137 	return rc;
3138 }
3139 
smc_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)3140 int smc_getsockopt(struct socket *sock, int level, int optname,
3141 		   char __user *optval, int __user *optlen)
3142 {
3143 	struct smc_sock *smc;
3144 	int rc;
3145 
3146 	if (level == SOL_SMC)
3147 		return __smc_getsockopt(sock, level, optname, optval, optlen);
3148 
3149 	smc = smc_sk(sock->sk);
3150 	mutex_lock(&smc->clcsock_release_lock);
3151 	if (!smc->clcsock) {
3152 		mutex_unlock(&smc->clcsock_release_lock);
3153 		return -EBADF;
3154 	}
3155 	/* socket options apply to the CLC socket */
3156 	if (unlikely(!smc->clcsock->ops->getsockopt)) {
3157 		mutex_unlock(&smc->clcsock_release_lock);
3158 		return -EOPNOTSUPP;
3159 	}
3160 	rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3161 					   optval, optlen);
3162 	mutex_unlock(&smc->clcsock_release_lock);
3163 	return rc;
3164 }
3165 
smc_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)3166 int smc_ioctl(struct socket *sock, unsigned int cmd,
3167 	      unsigned long arg)
3168 {
3169 	union smc_host_cursor cons, urg;
3170 	struct smc_connection *conn;
3171 	struct smc_sock *smc;
3172 	int answ;
3173 
3174 	smc = smc_sk(sock->sk);
3175 	conn = &smc->conn;
3176 	lock_sock(&smc->sk);
3177 	if (smc->use_fallback) {
3178 		if (!smc->clcsock) {
3179 			release_sock(&smc->sk);
3180 			return -EBADF;
3181 		}
3182 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3183 		release_sock(&smc->sk);
3184 		return answ;
3185 	}
3186 	switch (cmd) {
3187 	case SIOCINQ: /* same as FIONREAD */
3188 		if (smc->sk.sk_state == SMC_LISTEN) {
3189 			release_sock(&smc->sk);
3190 			return -EINVAL;
3191 		}
3192 		if (smc->sk.sk_state == SMC_INIT ||
3193 		    smc->sk.sk_state == SMC_CLOSED)
3194 			answ = 0;
3195 		else
3196 			answ = atomic_read(&smc->conn.bytes_to_rcv);
3197 		break;
3198 	case SIOCOUTQ:
3199 		/* output queue size (not send + not acked) */
3200 		if (smc->sk.sk_state == SMC_LISTEN) {
3201 			release_sock(&smc->sk);
3202 			return -EINVAL;
3203 		}
3204 		if (smc->sk.sk_state == SMC_INIT ||
3205 		    smc->sk.sk_state == SMC_CLOSED)
3206 			answ = 0;
3207 		else
3208 			answ = smc->conn.sndbuf_desc->len -
3209 					atomic_read(&smc->conn.sndbuf_space);
3210 		break;
3211 	case SIOCOUTQNSD:
3212 		/* output queue size (not send only) */
3213 		if (smc->sk.sk_state == SMC_LISTEN) {
3214 			release_sock(&smc->sk);
3215 			return -EINVAL;
3216 		}
3217 		if (smc->sk.sk_state == SMC_INIT ||
3218 		    smc->sk.sk_state == SMC_CLOSED)
3219 			answ = 0;
3220 		else
3221 			answ = smc_tx_prepared_sends(&smc->conn);
3222 		break;
3223 	case SIOCATMARK:
3224 		if (smc->sk.sk_state == SMC_LISTEN) {
3225 			release_sock(&smc->sk);
3226 			return -EINVAL;
3227 		}
3228 		if (smc->sk.sk_state == SMC_INIT ||
3229 		    smc->sk.sk_state == SMC_CLOSED) {
3230 			answ = 0;
3231 		} else {
3232 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3233 			smc_curs_copy(&urg, &conn->urg_curs, conn);
3234 			answ = smc_curs_diff(conn->rmb_desc->len,
3235 					     &cons, &urg) == 1;
3236 		}
3237 		break;
3238 	default:
3239 		release_sock(&smc->sk);
3240 		return -ENOIOCTLCMD;
3241 	}
3242 	release_sock(&smc->sk);
3243 
3244 	return put_user(answ, (int __user *)arg);
3245 }
3246 
3247 /* Map the affected portions of the rmbe into an spd, note the number of bytes
3248  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3249  * updates till whenever a respective page has been fully processed.
3250  * Note that subsequent recv() calls have to wait till all splice() processing
3251  * completed.
3252  */
smc_splice_read(struct socket * sock,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)3253 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3254 			struct pipe_inode_info *pipe, size_t len,
3255 			unsigned int flags)
3256 {
3257 	struct sock *sk = sock->sk;
3258 	struct smc_sock *smc;
3259 	int rc = -ENOTCONN;
3260 
3261 	smc = smc_sk(sk);
3262 	lock_sock(sk);
3263 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3264 		/* socket was connected before, no more data to read */
3265 		rc = 0;
3266 		goto out;
3267 	}
3268 	if (sk->sk_state == SMC_INIT ||
3269 	    sk->sk_state == SMC_LISTEN ||
3270 	    sk->sk_state == SMC_CLOSED)
3271 		goto out;
3272 
3273 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3274 		rc = 0;
3275 		goto out;
3276 	}
3277 
3278 	if (smc->use_fallback) {
3279 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3280 						    pipe, len, flags);
3281 	} else {
3282 		if (*ppos) {
3283 			rc = -ESPIPE;
3284 			goto out;
3285 		}
3286 		if (flags & SPLICE_F_NONBLOCK)
3287 			flags = MSG_DONTWAIT;
3288 		else
3289 			flags = 0;
3290 		SMC_STAT_INC(smc, splice_cnt);
3291 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3292 	}
3293 out:
3294 	release_sock(sk);
3295 
3296 	return rc;
3297 }
3298 
3299 /* must look like tcp */
3300 static const struct proto_ops smc_sock_ops = {
3301 	.family		= PF_SMC,
3302 	.owner		= THIS_MODULE,
3303 	.release	= smc_release,
3304 	.bind		= smc_bind,
3305 	.connect	= smc_connect,
3306 	.socketpair	= sock_no_socketpair,
3307 	.accept		= smc_accept,
3308 	.getname	= smc_getname,
3309 	.poll		= smc_poll,
3310 	.ioctl		= smc_ioctl,
3311 	.listen		= smc_listen,
3312 	.shutdown	= smc_shutdown,
3313 	.setsockopt	= smc_setsockopt,
3314 	.getsockopt	= smc_getsockopt,
3315 	.sendmsg	= smc_sendmsg,
3316 	.recvmsg	= smc_recvmsg,
3317 	.mmap		= sock_no_mmap,
3318 	.splice_read	= smc_splice_read,
3319 };
3320 
smc_create_clcsk(struct net * net,struct sock * sk,int family)3321 int smc_create_clcsk(struct net *net, struct sock *sk, int family)
3322 {
3323 	struct smc_sock *smc = smc_sk(sk);
3324 	int rc;
3325 
3326 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3327 			      &smc->clcsock);
3328 	if (rc)
3329 		return rc;
3330 
3331 	/* smc_clcsock_release() does not wait smc->clcsock->sk's
3332 	 * destruction;  its sk_state might not be TCP_CLOSE after
3333 	 * smc->sk is close()d, and TCP timers can be fired later,
3334 	 * which need net ref.
3335 	 */
3336 	sk = smc->clcsock->sk;
3337 	__netns_tracker_free(net, &sk->ns_tracker, false);
3338 	sk->sk_net_refcnt = 1;
3339 	get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
3340 	sock_inuse_add(net, 1);
3341 	return 0;
3342 }
3343 
__smc_create(struct net * net,struct socket * sock,int protocol,int kern,struct socket * clcsock)3344 static int __smc_create(struct net *net, struct socket *sock, int protocol,
3345 			int kern, struct socket *clcsock)
3346 {
3347 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3348 	struct smc_sock *smc;
3349 	struct sock *sk;
3350 	int rc;
3351 
3352 	rc = -ESOCKTNOSUPPORT;
3353 	if (sock->type != SOCK_STREAM)
3354 		goto out;
3355 
3356 	rc = -EPROTONOSUPPORT;
3357 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3358 		goto out;
3359 
3360 	rc = -ENOBUFS;
3361 	sock->ops = &smc_sock_ops;
3362 	sock->state = SS_UNCONNECTED;
3363 	sk = smc_sock_alloc(net, sock, protocol);
3364 	if (!sk)
3365 		goto out;
3366 
3367 	/* create internal TCP socket for CLC handshake and fallback */
3368 	smc = smc_sk(sk);
3369 
3370 	rc = 0;
3371 	if (clcsock)
3372 		smc->clcsock = clcsock;
3373 	else
3374 		rc = smc_create_clcsk(net, sk, family);
3375 
3376 	if (rc) {
3377 		sk_common_release(sk);
3378 		sock->sk = NULL;
3379 	}
3380 out:
3381 	return rc;
3382 }
3383 
smc_create(struct net * net,struct socket * sock,int protocol,int kern)3384 static int smc_create(struct net *net, struct socket *sock, int protocol,
3385 		      int kern)
3386 {
3387 	return __smc_create(net, sock, protocol, kern, NULL);
3388 }
3389 
3390 static const struct net_proto_family smc_sock_family_ops = {
3391 	.family	= PF_SMC,
3392 	.owner	= THIS_MODULE,
3393 	.create	= smc_create,
3394 };
3395 
smc_ulp_init(struct sock * sk)3396 static int smc_ulp_init(struct sock *sk)
3397 {
3398 	struct socket *tcp = sk->sk_socket;
3399 	struct net *net = sock_net(sk);
3400 	struct socket *smcsock;
3401 	int protocol, ret;
3402 
3403 	/* only TCP can be replaced */
3404 	if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP ||
3405 	    (sk->sk_family != AF_INET && sk->sk_family != AF_INET6))
3406 		return -ESOCKTNOSUPPORT;
3407 	/* don't handle wq now */
3408 	if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list)
3409 		return -ENOTCONN;
3410 
3411 	if (sk->sk_family == AF_INET)
3412 		protocol = SMCPROTO_SMC;
3413 	else
3414 		protocol = SMCPROTO_SMC6;
3415 
3416 	smcsock = sock_alloc();
3417 	if (!smcsock)
3418 		return -ENFILE;
3419 
3420 	smcsock->type = SOCK_STREAM;
3421 	__module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */
3422 	ret = __smc_create(net, smcsock, protocol, 1, tcp);
3423 	if (ret) {
3424 		sock_release(smcsock); /* module_put() which ops won't be NULL */
3425 		return ret;
3426 	}
3427 
3428 	/* replace tcp socket to smc */
3429 	smcsock->file = tcp->file;
3430 	smcsock->file->private_data = smcsock;
3431 	smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */
3432 	smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */
3433 	tcp->file = NULL;
3434 
3435 	return ret;
3436 }
3437 
smc_ulp_clone(const struct request_sock * req,struct sock * newsk,const gfp_t priority)3438 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk,
3439 			  const gfp_t priority)
3440 {
3441 	struct inet_connection_sock *icsk = inet_csk(newsk);
3442 
3443 	/* don't inherit ulp ops to child when listen */
3444 	icsk->icsk_ulp_ops = NULL;
3445 }
3446 
3447 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = {
3448 	.name		= "smc",
3449 	.owner		= THIS_MODULE,
3450 	.init		= smc_ulp_init,
3451 	.clone		= smc_ulp_clone,
3452 };
3453 
3454 unsigned int smc_net_id;
3455 
smc_net_init(struct net * net)3456 static __net_init int smc_net_init(struct net *net)
3457 {
3458 	int rc;
3459 
3460 	rc = smc_sysctl_net_init(net);
3461 	if (rc)
3462 		return rc;
3463 	return smc_pnet_net_init(net);
3464 }
3465 
smc_net_exit(struct net * net)3466 static void __net_exit smc_net_exit(struct net *net)
3467 {
3468 	smc_sysctl_net_exit(net);
3469 	smc_pnet_net_exit(net);
3470 }
3471 
smc_net_stat_init(struct net * net)3472 static __net_init int smc_net_stat_init(struct net *net)
3473 {
3474 	return smc_stats_init(net);
3475 }
3476 
smc_net_stat_exit(struct net * net)3477 static void __net_exit smc_net_stat_exit(struct net *net)
3478 {
3479 	smc_stats_exit(net);
3480 }
3481 
3482 static struct pernet_operations smc_net_ops = {
3483 	.init = smc_net_init,
3484 	.exit = smc_net_exit,
3485 	.id   = &smc_net_id,
3486 	.size = sizeof(struct smc_net),
3487 };
3488 
3489 static struct pernet_operations smc_net_stat_ops = {
3490 	.init = smc_net_stat_init,
3491 	.exit = smc_net_stat_exit,
3492 };
3493 
smc_init(void)3494 static int __init smc_init(void)
3495 {
3496 	int rc;
3497 
3498 	rc = register_pernet_subsys(&smc_net_ops);
3499 	if (rc)
3500 		return rc;
3501 
3502 	rc = register_pernet_subsys(&smc_net_stat_ops);
3503 	if (rc)
3504 		goto out_pernet_subsys;
3505 
3506 	rc = smc_ism_init();
3507 	if (rc)
3508 		goto out_pernet_subsys_stat;
3509 	smc_clc_init();
3510 
3511 	rc = smc_nl_init();
3512 	if (rc)
3513 		goto out_ism;
3514 
3515 	rc = smc_pnet_init();
3516 	if (rc)
3517 		goto out_nl;
3518 
3519 	rc = -ENOMEM;
3520 
3521 	smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0);
3522 	if (!smc_tcp_ls_wq)
3523 		goto out_pnet;
3524 
3525 	smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0);
3526 	if (!smc_hs_wq)
3527 		goto out_alloc_tcp_ls_wq;
3528 
3529 	smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0);
3530 	if (!smc_close_wq)
3531 		goto out_alloc_hs_wq;
3532 
3533 	rc = smc_core_init();
3534 	if (rc) {
3535 		pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3536 		goto out_alloc_wqs;
3537 	}
3538 
3539 	rc = smc_llc_init();
3540 	if (rc) {
3541 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3542 		goto out_core;
3543 	}
3544 
3545 	rc = smc_cdc_init();
3546 	if (rc) {
3547 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3548 		goto out_core;
3549 	}
3550 
3551 	rc = proto_register(&smc_proto, 1);
3552 	if (rc) {
3553 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3554 		goto out_core;
3555 	}
3556 
3557 	rc = proto_register(&smc_proto6, 1);
3558 	if (rc) {
3559 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3560 		goto out_proto;
3561 	}
3562 
3563 	rc = sock_register(&smc_sock_family_ops);
3564 	if (rc) {
3565 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
3566 		goto out_proto6;
3567 	}
3568 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3569 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3570 
3571 	rc = smc_ib_register_client();
3572 	if (rc) {
3573 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
3574 		goto out_sock;
3575 	}
3576 
3577 	rc = smc_loopback_init();
3578 	if (rc) {
3579 		pr_err("%s: smc_loopback_init fails with %d\n", __func__, rc);
3580 		goto out_ib;
3581 	}
3582 
3583 	rc = tcp_register_ulp(&smc_ulp_ops);
3584 	if (rc) {
3585 		pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc);
3586 		goto out_lo;
3587 	}
3588 	rc = smc_inet_init();
3589 	if (rc) {
3590 		pr_err("%s: smc_inet_init fails with %d\n", __func__, rc);
3591 		goto out_ulp;
3592 	}
3593 	static_branch_enable(&tcp_have_smc);
3594 	return 0;
3595 out_ulp:
3596 	tcp_unregister_ulp(&smc_ulp_ops);
3597 out_lo:
3598 	smc_loopback_exit();
3599 out_ib:
3600 	smc_ib_unregister_client();
3601 out_sock:
3602 	sock_unregister(PF_SMC);
3603 out_proto6:
3604 	proto_unregister(&smc_proto6);
3605 out_proto:
3606 	proto_unregister(&smc_proto);
3607 out_core:
3608 	smc_core_exit();
3609 out_alloc_wqs:
3610 	destroy_workqueue(smc_close_wq);
3611 out_alloc_hs_wq:
3612 	destroy_workqueue(smc_hs_wq);
3613 out_alloc_tcp_ls_wq:
3614 	destroy_workqueue(smc_tcp_ls_wq);
3615 out_pnet:
3616 	smc_pnet_exit();
3617 out_nl:
3618 	smc_nl_exit();
3619 out_ism:
3620 	smc_clc_exit();
3621 	smc_ism_exit();
3622 out_pernet_subsys_stat:
3623 	unregister_pernet_subsys(&smc_net_stat_ops);
3624 out_pernet_subsys:
3625 	unregister_pernet_subsys(&smc_net_ops);
3626 
3627 	return rc;
3628 }
3629 
smc_exit(void)3630 static void __exit smc_exit(void)
3631 {
3632 	static_branch_disable(&tcp_have_smc);
3633 	smc_inet_exit();
3634 	tcp_unregister_ulp(&smc_ulp_ops);
3635 	sock_unregister(PF_SMC);
3636 	smc_core_exit();
3637 	smc_loopback_exit();
3638 	smc_ib_unregister_client();
3639 	smc_ism_exit();
3640 	destroy_workqueue(smc_close_wq);
3641 	destroy_workqueue(smc_tcp_ls_wq);
3642 	destroy_workqueue(smc_hs_wq);
3643 	proto_unregister(&smc_proto6);
3644 	proto_unregister(&smc_proto);
3645 	smc_pnet_exit();
3646 	smc_nl_exit();
3647 	smc_clc_exit();
3648 	unregister_pernet_subsys(&smc_net_stat_ops);
3649 	unregister_pernet_subsys(&smc_net_ops);
3650 	rcu_barrier();
3651 }
3652 
3653 module_init(smc_init);
3654 module_exit(smc_exit);
3655 
3656 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3657 MODULE_DESCRIPTION("smc socket address family");
3658 MODULE_LICENSE("GPL");
3659 MODULE_ALIAS_NETPROTO(PF_SMC);
3660 MODULE_ALIAS_TCP_ULP("smc");
3661 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */
3662 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1);
3663 #if IS_ENABLED(CONFIG_IPV6)
3664 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1);
3665 #endif /* CONFIG_IPV6 */
3666 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);
3667