• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
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 
29 #include <net/sock.h>
30 #include <net/tcp.h>
31 #include <net/smc.h>
32 #include <asm/ioctls.h>
33 
34 #include <net/net_namespace.h>
35 #include <net/netns/generic.h>
36 #include "smc_netns.h"
37 
38 #include "smc.h"
39 #include "smc_clc.h"
40 #include "smc_llc.h"
41 #include "smc_cdc.h"
42 #include "smc_core.h"
43 #include "smc_ib.h"
44 #include "smc_ism.h"
45 #include "smc_pnet.h"
46 #include "smc_tx.h"
47 #include "smc_rx.h"
48 #include "smc_close.h"
49 
50 static DEFINE_MUTEX(smc_server_lgr_pending);	/* serialize link group
51 						 * creation on server
52 						 */
53 static DEFINE_MUTEX(smc_client_lgr_pending);	/* serialize link group
54 						 * creation on client
55 						 */
56 
57 static void smc_tcp_listen_work(struct work_struct *);
58 static void smc_connect_work(struct work_struct *);
59 
smc_set_keepalive(struct sock * sk,int val)60 static void smc_set_keepalive(struct sock *sk, int val)
61 {
62 	struct smc_sock *smc = smc_sk(sk);
63 
64 	smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
65 }
66 
67 static struct smc_hashinfo smc_v4_hashinfo = {
68 	.lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
69 };
70 
71 static struct smc_hashinfo smc_v6_hashinfo = {
72 	.lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
73 };
74 
smc_hash_sk(struct sock * sk)75 int smc_hash_sk(struct sock *sk)
76 {
77 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
78 	struct hlist_head *head;
79 
80 	head = &h->ht;
81 
82 	write_lock_bh(&h->lock);
83 	sk_add_node(sk, head);
84 	sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
85 	write_unlock_bh(&h->lock);
86 
87 	return 0;
88 }
89 EXPORT_SYMBOL_GPL(smc_hash_sk);
90 
smc_unhash_sk(struct sock * sk)91 void smc_unhash_sk(struct sock *sk)
92 {
93 	struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
94 
95 	write_lock_bh(&h->lock);
96 	if (sk_del_node_init(sk))
97 		sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
98 	write_unlock_bh(&h->lock);
99 }
100 EXPORT_SYMBOL_GPL(smc_unhash_sk);
101 
102 struct proto smc_proto = {
103 	.name		= "SMC",
104 	.owner		= THIS_MODULE,
105 	.keepalive	= smc_set_keepalive,
106 	.hash		= smc_hash_sk,
107 	.unhash		= smc_unhash_sk,
108 	.obj_size	= sizeof(struct smc_sock),
109 	.h.smc_hash	= &smc_v4_hashinfo,
110 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
111 };
112 EXPORT_SYMBOL_GPL(smc_proto);
113 
114 struct proto smc_proto6 = {
115 	.name		= "SMC6",
116 	.owner		= THIS_MODULE,
117 	.keepalive	= smc_set_keepalive,
118 	.hash		= smc_hash_sk,
119 	.unhash		= smc_unhash_sk,
120 	.obj_size	= sizeof(struct smc_sock),
121 	.h.smc_hash	= &smc_v6_hashinfo,
122 	.slab_flags	= SLAB_TYPESAFE_BY_RCU,
123 };
124 EXPORT_SYMBOL_GPL(smc_proto6);
125 
smc_restore_fallback_changes(struct smc_sock * smc)126 static void smc_restore_fallback_changes(struct smc_sock *smc)
127 {
128 	smc->clcsock->file->private_data = smc->sk.sk_socket;
129 	smc->clcsock->file = NULL;
130 }
131 
__smc_release(struct smc_sock * smc)132 static int __smc_release(struct smc_sock *smc)
133 {
134 	struct sock *sk = &smc->sk;
135 	int rc = 0;
136 
137 	if (!smc->use_fallback) {
138 		rc = smc_close_active(smc);
139 		sock_set_flag(sk, SOCK_DEAD);
140 		sk->sk_shutdown |= SHUTDOWN_MASK;
141 	} else {
142 		if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT)
143 			sock_put(sk); /* passive closing */
144 		if (sk->sk_state == SMC_LISTEN) {
145 			/* wake up clcsock accept */
146 			rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR);
147 		}
148 		sk->sk_state = SMC_CLOSED;
149 		sk->sk_state_change(sk);
150 		smc_restore_fallback_changes(smc);
151 	}
152 
153 	sk->sk_prot->unhash(sk);
154 
155 	if (sk->sk_state == SMC_CLOSED) {
156 		if (smc->clcsock) {
157 			release_sock(sk);
158 			smc_clcsock_release(smc);
159 			lock_sock(sk);
160 		}
161 		if (!smc->use_fallback)
162 			smc_conn_free(&smc->conn);
163 	}
164 
165 	return rc;
166 }
167 
smc_release(struct socket * sock)168 static int smc_release(struct socket *sock)
169 {
170 	struct sock *sk = sock->sk;
171 	struct smc_sock *smc;
172 	int rc = 0;
173 
174 	if (!sk)
175 		goto out;
176 
177 	smc = smc_sk(sk);
178 
179 	/* cleanup for a dangling non-blocking connect */
180 	if (smc->connect_nonblock && sk->sk_state == SMC_INIT)
181 		tcp_abort(smc->clcsock->sk, ECONNABORTED);
182 	flush_work(&smc->connect_work);
183 
184 	if (sk->sk_state == SMC_LISTEN)
185 		/* smc_close_non_accepted() is called and acquires
186 		 * sock lock for child sockets again
187 		 */
188 		lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
189 	else
190 		lock_sock(sk);
191 
192 	rc = __smc_release(smc);
193 
194 	/* detach socket */
195 	sock_orphan(sk);
196 	sock->sk = NULL;
197 	release_sock(sk);
198 
199 	sock_put(sk); /* final sock_put */
200 out:
201 	return rc;
202 }
203 
smc_destruct(struct sock * sk)204 static void smc_destruct(struct sock *sk)
205 {
206 	if (sk->sk_state != SMC_CLOSED)
207 		return;
208 	if (!sock_flag(sk, SOCK_DEAD))
209 		return;
210 
211 	sk_refcnt_debug_dec(sk);
212 }
213 
smc_sock_alloc(struct net * net,struct socket * sock,int protocol)214 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
215 				   int protocol)
216 {
217 	struct smc_sock *smc;
218 	struct proto *prot;
219 	struct sock *sk;
220 
221 	prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
222 	sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
223 	if (!sk)
224 		return NULL;
225 
226 	sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
227 	sk->sk_state = SMC_INIT;
228 	sk->sk_destruct = smc_destruct;
229 	sk->sk_protocol = protocol;
230 	smc = smc_sk(sk);
231 	INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
232 	INIT_WORK(&smc->connect_work, smc_connect_work);
233 	INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
234 	INIT_LIST_HEAD(&smc->accept_q);
235 	spin_lock_init(&smc->accept_q_lock);
236 	spin_lock_init(&smc->conn.send_lock);
237 	sk->sk_prot->hash(sk);
238 	sk_refcnt_debug_inc(sk);
239 	mutex_init(&smc->clcsock_release_lock);
240 
241 	return sk;
242 }
243 
smc_bind(struct socket * sock,struct sockaddr * uaddr,int addr_len)244 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
245 		    int addr_len)
246 {
247 	struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
248 	struct sock *sk = sock->sk;
249 	struct smc_sock *smc;
250 	int rc;
251 
252 	smc = smc_sk(sk);
253 
254 	/* replicate tests from inet_bind(), to be safe wrt. future changes */
255 	rc = -EINVAL;
256 	if (addr_len < sizeof(struct sockaddr_in))
257 		goto out;
258 
259 	rc = -EAFNOSUPPORT;
260 	if (addr->sin_family != AF_INET &&
261 	    addr->sin_family != AF_INET6 &&
262 	    addr->sin_family != AF_UNSPEC)
263 		goto out;
264 	/* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
265 	if (addr->sin_family == AF_UNSPEC &&
266 	    addr->sin_addr.s_addr != htonl(INADDR_ANY))
267 		goto out;
268 
269 	lock_sock(sk);
270 
271 	/* Check if socket is already active */
272 	rc = -EINVAL;
273 	if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
274 		goto out_rel;
275 
276 	smc->clcsock->sk->sk_reuse = sk->sk_reuse;
277 	rc = kernel_bind(smc->clcsock, uaddr, addr_len);
278 
279 out_rel:
280 	release_sock(sk);
281 out:
282 	return rc;
283 }
284 
smc_copy_sock_settings(struct sock * nsk,struct sock * osk,unsigned long mask)285 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
286 				   unsigned long mask)
287 {
288 	/* options we don't get control via setsockopt for */
289 	nsk->sk_type = osk->sk_type;
290 	nsk->sk_sndbuf = osk->sk_sndbuf;
291 	nsk->sk_rcvbuf = osk->sk_rcvbuf;
292 	nsk->sk_sndtimeo = osk->sk_sndtimeo;
293 	nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
294 	nsk->sk_mark = osk->sk_mark;
295 	nsk->sk_priority = osk->sk_priority;
296 	nsk->sk_rcvlowat = osk->sk_rcvlowat;
297 	nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
298 	nsk->sk_err = osk->sk_err;
299 
300 	nsk->sk_flags &= ~mask;
301 	nsk->sk_flags |= osk->sk_flags & mask;
302 }
303 
304 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
305 			     (1UL << SOCK_KEEPOPEN) | \
306 			     (1UL << SOCK_LINGER) | \
307 			     (1UL << SOCK_BROADCAST) | \
308 			     (1UL << SOCK_TIMESTAMP) | \
309 			     (1UL << SOCK_DBG) | \
310 			     (1UL << SOCK_RCVTSTAMP) | \
311 			     (1UL << SOCK_RCVTSTAMPNS) | \
312 			     (1UL << SOCK_LOCALROUTE) | \
313 			     (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
314 			     (1UL << SOCK_RXQ_OVFL) | \
315 			     (1UL << SOCK_WIFI_STATUS) | \
316 			     (1UL << SOCK_NOFCS) | \
317 			     (1UL << SOCK_FILTER_LOCKED) | \
318 			     (1UL << SOCK_TSTAMP_NEW))
319 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
320  * clc socket (since smc is not called for these options from net/core)
321  */
smc_copy_sock_settings_to_clc(struct smc_sock * smc)322 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
323 {
324 	smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
325 }
326 
327 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
328 			     (1UL << SOCK_KEEPOPEN) | \
329 			     (1UL << SOCK_LINGER) | \
330 			     (1UL << SOCK_DBG))
331 /* copy only settings and flags relevant for smc from clc to smc socket */
smc_copy_sock_settings_to_smc(struct smc_sock * smc)332 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
333 {
334 	smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
335 }
336 
337 /* register a new rmb, send confirm_rkey msg to register with peer */
smc_reg_rmb(struct smc_link * link,struct smc_buf_desc * rmb_desc,bool conf_rkey)338 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc,
339 		       bool conf_rkey)
340 {
341 	if (!rmb_desc->wr_reg) {
342 		/* register memory region for new rmb */
343 		if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) {
344 			rmb_desc->regerr = 1;
345 			return -EFAULT;
346 		}
347 		rmb_desc->wr_reg = 1;
348 	}
349 	if (!conf_rkey)
350 		return 0;
351 	/* exchange confirm_rkey msg with peer */
352 	if (smc_llc_do_confirm_rkey(link, rmb_desc)) {
353 		rmb_desc->regerr = 1;
354 		return -EFAULT;
355 	}
356 	return 0;
357 }
358 
smc_clnt_conf_first_link(struct smc_sock * smc)359 static int smc_clnt_conf_first_link(struct smc_sock *smc)
360 {
361 	struct net *net = sock_net(smc->clcsock->sk);
362 	struct smc_link_group *lgr = smc->conn.lgr;
363 	struct smc_link *link;
364 	int rest;
365 	int rc;
366 
367 	link = &lgr->lnk[SMC_SINGLE_LINK];
368 	/* receive CONFIRM LINK request from server over RoCE fabric */
369 	rest = wait_for_completion_interruptible_timeout(
370 		&link->llc_confirm,
371 		SMC_LLC_WAIT_FIRST_TIME);
372 	if (rest <= 0) {
373 		struct smc_clc_msg_decline dclc;
374 
375 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
376 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
377 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
378 	}
379 
380 	if (link->llc_confirm_rc)
381 		return SMC_CLC_DECL_RMBE_EC;
382 
383 	rc = smc_ib_modify_qp_rts(link);
384 	if (rc)
385 		return SMC_CLC_DECL_ERR_RDYLNK;
386 
387 	smc_wr_remember_qp_attr(link);
388 
389 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
390 		return SMC_CLC_DECL_ERR_REGRMB;
391 
392 	/* send CONFIRM LINK response over RoCE fabric */
393 	rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
394 	if (rc < 0)
395 		return SMC_CLC_DECL_TIMEOUT_CL;
396 
397 	/* receive ADD LINK request from server over RoCE fabric */
398 	rest = wait_for_completion_interruptible_timeout(&link->llc_add,
399 							 SMC_LLC_WAIT_TIME);
400 	if (rest <= 0) {
401 		struct smc_clc_msg_decline dclc;
402 
403 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
404 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
405 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
406 	}
407 
408 	/* send add link reject message, only one link supported for now */
409 	rc = smc_llc_send_add_link(link,
410 				   link->smcibdev->mac[link->ibport - 1],
411 				   link->gid, SMC_LLC_RESP);
412 	if (rc < 0)
413 		return SMC_CLC_DECL_TIMEOUT_AL;
414 
415 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
416 
417 	return 0;
418 }
419 
smcr_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)420 static void smcr_conn_save_peer_info(struct smc_sock *smc,
421 				     struct smc_clc_msg_accept_confirm *clc)
422 {
423 	int bufsize = smc_uncompress_bufsize(clc->rmbe_size);
424 
425 	smc->conn.peer_rmbe_idx = clc->rmbe_idx;
426 	smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
427 	smc->conn.peer_rmbe_size = bufsize;
428 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
429 	smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
430 }
431 
smcd_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)432 static void smcd_conn_save_peer_info(struct smc_sock *smc,
433 				     struct smc_clc_msg_accept_confirm *clc)
434 {
435 	int bufsize = smc_uncompress_bufsize(clc->dmbe_size);
436 
437 	smc->conn.peer_rmbe_idx = clc->dmbe_idx;
438 	smc->conn.peer_token = clc->token;
439 	/* msg header takes up space in the buffer */
440 	smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
441 	atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
442 	smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
443 }
444 
smc_conn_save_peer_info(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc)445 static void smc_conn_save_peer_info(struct smc_sock *smc,
446 				    struct smc_clc_msg_accept_confirm *clc)
447 {
448 	if (smc->conn.lgr->is_smcd)
449 		smcd_conn_save_peer_info(smc, clc);
450 	else
451 		smcr_conn_save_peer_info(smc, clc);
452 }
453 
smc_link_save_peer_info(struct smc_link * link,struct smc_clc_msg_accept_confirm * clc)454 static void smc_link_save_peer_info(struct smc_link *link,
455 				    struct smc_clc_msg_accept_confirm *clc)
456 {
457 	link->peer_qpn = ntoh24(clc->qpn);
458 	memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
459 	memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
460 	link->peer_psn = ntoh24(clc->psn);
461 	link->peer_mtu = clc->qp_mtu;
462 }
463 
smc_switch_to_fallback(struct smc_sock * smc)464 static void smc_switch_to_fallback(struct smc_sock *smc)
465 {
466 	smc->use_fallback = true;
467 	if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
468 		smc->clcsock->file = smc->sk.sk_socket->file;
469 		smc->clcsock->file->private_data = smc->clcsock;
470 	}
471 }
472 
473 /* fall back during connect */
smc_connect_fallback(struct smc_sock * smc,int reason_code)474 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
475 {
476 	smc_switch_to_fallback(smc);
477 	smc->fallback_rsn = reason_code;
478 	smc_copy_sock_settings_to_clc(smc);
479 	smc->connect_nonblock = 0;
480 	if (smc->sk.sk_state == SMC_INIT)
481 		smc->sk.sk_state = SMC_ACTIVE;
482 	return 0;
483 }
484 
485 /* decline and fall back during connect */
smc_connect_decline_fallback(struct smc_sock * smc,int reason_code)486 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code)
487 {
488 	int rc;
489 
490 	if (reason_code < 0) { /* error, fallback is not possible */
491 		if (smc->sk.sk_state == SMC_INIT)
492 			sock_put(&smc->sk); /* passive closing */
493 		return reason_code;
494 	}
495 	if (reason_code != SMC_CLC_DECL_PEERDECL) {
496 		rc = smc_clc_send_decline(smc, reason_code);
497 		if (rc < 0) {
498 			if (smc->sk.sk_state == SMC_INIT)
499 				sock_put(&smc->sk); /* passive closing */
500 			return rc;
501 		}
502 	}
503 	return smc_connect_fallback(smc, reason_code);
504 }
505 
506 /* abort connecting */
smc_connect_abort(struct smc_sock * smc,int reason_code,int local_contact)507 static int smc_connect_abort(struct smc_sock *smc, int reason_code,
508 			     int local_contact)
509 {
510 	if (local_contact == SMC_FIRST_CONTACT)
511 		smc_lgr_forget(smc->conn.lgr);
512 	if (smc->conn.lgr->is_smcd)
513 		/* there is only one lgr role for SMC-D; use server lock */
514 		mutex_unlock(&smc_server_lgr_pending);
515 	else
516 		mutex_unlock(&smc_client_lgr_pending);
517 
518 	smc_conn_free(&smc->conn);
519 	smc->connect_nonblock = 0;
520 	return reason_code;
521 }
522 
523 /* check if there is a rdma device available for this connection. */
524 /* called for connect and listen */
smc_find_rdma_device(struct smc_sock * smc,struct smc_init_info * ini)525 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
526 {
527 	/* PNET table look up: search active ib_device and port
528 	 * within same PNETID that also contains the ethernet device
529 	 * used for the internal TCP socket
530 	 */
531 	smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
532 	if (!ini->ib_dev)
533 		return SMC_CLC_DECL_NOSMCRDEV;
534 	return 0;
535 }
536 
537 /* check if there is an ISM device available for this connection. */
538 /* called for connect and listen */
smc_find_ism_device(struct smc_sock * smc,struct smc_init_info * ini)539 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
540 {
541 	/* Find ISM device with same PNETID as connecting interface  */
542 	smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
543 	if (!ini->ism_dev)
544 		return SMC_CLC_DECL_NOSMCDDEV;
545 	return 0;
546 }
547 
548 /* 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)549 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
550 				      struct smc_init_info *ini)
551 {
552 	if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id))
553 		return SMC_CLC_DECL_ISMVLANERR;
554 	return 0;
555 }
556 
557 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
558  * used, the VLAN ID will be registered again during the connection setup.
559  */
smc_connect_ism_vlan_cleanup(struct smc_sock * smc,bool is_smcd,struct smc_init_info * ini)560 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd,
561 					struct smc_init_info *ini)
562 {
563 	if (!is_smcd)
564 		return 0;
565 	if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id))
566 		return SMC_CLC_DECL_CNFERR;
567 	return 0;
568 }
569 
570 /* CLC handshake during connect */
smc_connect_clc(struct smc_sock * smc,int smc_type,struct smc_clc_msg_accept_confirm * aclc,struct smc_init_info * ini)571 static int smc_connect_clc(struct smc_sock *smc, int smc_type,
572 			   struct smc_clc_msg_accept_confirm *aclc,
573 			   struct smc_init_info *ini)
574 {
575 	int rc = 0;
576 
577 	/* do inband token exchange */
578 	rc = smc_clc_send_proposal(smc, smc_type, ini);
579 	if (rc)
580 		return rc;
581 	/* receive SMC Accept CLC message */
582 	return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT,
583 				CLC_WAIT_TIME);
584 }
585 
586 /* 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)587 static int smc_connect_rdma(struct smc_sock *smc,
588 			    struct smc_clc_msg_accept_confirm *aclc,
589 			    struct smc_init_info *ini)
590 {
591 	struct smc_link *link;
592 	int reason_code = 0;
593 
594 	ini->is_smcd = false;
595 	ini->ib_lcl = &aclc->lcl;
596 	ini->ib_clcqpn = ntoh24(aclc->qpn);
597 	ini->srv_first_contact = aclc->hdr.flag;
598 
599 	mutex_lock(&smc_client_lgr_pending);
600 	reason_code = smc_conn_create(smc, ini);
601 	if (reason_code) {
602 		mutex_unlock(&smc_client_lgr_pending);
603 		return reason_code;
604 	}
605 	link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
606 
607 	smc_conn_save_peer_info(smc, aclc);
608 
609 	/* create send buffer and rmb */
610 	if (smc_buf_create(smc, false))
611 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
612 					 ini->cln_first_contact);
613 
614 	if (ini->cln_first_contact == SMC_FIRST_CONTACT)
615 		smc_link_save_peer_info(link, aclc);
616 
617 	if (smc_rmb_rtoken_handling(&smc->conn, aclc))
618 		return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK,
619 					 ini->cln_first_contact);
620 
621 	smc_close_init(smc);
622 	smc_rx_init(smc);
623 
624 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
625 		if (smc_ib_ready_link(link))
626 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK,
627 						 ini->cln_first_contact);
628 	} else {
629 		if (smc_reg_rmb(link, smc->conn.rmb_desc, true))
630 			return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB,
631 						 ini->cln_first_contact);
632 	}
633 	smc_rmb_sync_sg_for_device(&smc->conn);
634 
635 	reason_code = smc_clc_send_confirm(smc);
636 	if (reason_code)
637 		return smc_connect_abort(smc, reason_code,
638 					 ini->cln_first_contact);
639 
640 	smc_tx_init(smc);
641 
642 	if (ini->cln_first_contact == SMC_FIRST_CONTACT) {
643 		/* QP confirmation over RoCE fabric */
644 		reason_code = smc_clnt_conf_first_link(smc);
645 		if (reason_code)
646 			return smc_connect_abort(smc, reason_code,
647 						 ini->cln_first_contact);
648 	}
649 	mutex_unlock(&smc_client_lgr_pending);
650 
651 	smc_copy_sock_settings_to_clc(smc);
652 	smc->connect_nonblock = 0;
653 	if (smc->sk.sk_state == SMC_INIT)
654 		smc->sk.sk_state = SMC_ACTIVE;
655 
656 	return 0;
657 }
658 
659 /* 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)660 static int smc_connect_ism(struct smc_sock *smc,
661 			   struct smc_clc_msg_accept_confirm *aclc,
662 			   struct smc_init_info *ini)
663 {
664 	int rc = 0;
665 
666 	ini->is_smcd = true;
667 	ini->ism_gid = aclc->gid;
668 	ini->srv_first_contact = aclc->hdr.flag;
669 
670 	/* there is only one lgr role for SMC-D; use server lock */
671 	mutex_lock(&smc_server_lgr_pending);
672 	rc = smc_conn_create(smc, ini);
673 	if (rc) {
674 		mutex_unlock(&smc_server_lgr_pending);
675 		return rc;
676 	}
677 
678 	/* Create send and receive buffers */
679 	if (smc_buf_create(smc, true))
680 		return smc_connect_abort(smc, SMC_CLC_DECL_MEM,
681 					 ini->cln_first_contact);
682 
683 	smc_conn_save_peer_info(smc, aclc);
684 	smc_close_init(smc);
685 	smc_rx_init(smc);
686 	smc_tx_init(smc);
687 
688 	rc = smc_clc_send_confirm(smc);
689 	if (rc)
690 		return smc_connect_abort(smc, rc, ini->cln_first_contact);
691 	mutex_unlock(&smc_server_lgr_pending);
692 
693 	smc_copy_sock_settings_to_clc(smc);
694 	smc->connect_nonblock = 0;
695 	if (smc->sk.sk_state == SMC_INIT)
696 		smc->sk.sk_state = SMC_ACTIVE;
697 
698 	return 0;
699 }
700 
701 /* perform steps before actually connecting */
__smc_connect(struct smc_sock * smc)702 static int __smc_connect(struct smc_sock *smc)
703 {
704 	bool ism_supported = false, rdma_supported = false;
705 	struct smc_clc_msg_accept_confirm aclc;
706 	struct smc_init_info ini = {0};
707 	int smc_type;
708 	int rc = 0;
709 
710 	if (smc->use_fallback)
711 		return smc_connect_fallback(smc, smc->fallback_rsn);
712 
713 	/* if peer has not signalled SMC-capability, fall back */
714 	if (!tcp_sk(smc->clcsock->sk)->syn_smc)
715 		return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
716 
717 	/* IPSec connections opt out of SMC-R optimizations */
718 	if (using_ipsec(smc))
719 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC);
720 
721 	/* get vlan id from IP device */
722 	if (smc_vlan_by_tcpsk(smc->clcsock, &ini))
723 		return smc_connect_decline_fallback(smc,
724 						    SMC_CLC_DECL_GETVLANERR);
725 
726 	/* check if there is an ism device available */
727 	if (!smc_find_ism_device(smc, &ini) &&
728 	    !smc_connect_ism_vlan_setup(smc, &ini)) {
729 		/* ISM is supported for this connection */
730 		ism_supported = true;
731 		smc_type = SMC_TYPE_D;
732 	}
733 
734 	/* check if there is a rdma device available */
735 	if (!smc_find_rdma_device(smc, &ini)) {
736 		/* RDMA is supported for this connection */
737 		rdma_supported = true;
738 		if (ism_supported)
739 			smc_type = SMC_TYPE_B; /* both */
740 		else
741 			smc_type = SMC_TYPE_R; /* only RDMA */
742 	}
743 
744 	/* if neither ISM nor RDMA are supported, fallback */
745 	if (!rdma_supported && !ism_supported)
746 		return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV);
747 
748 	/* perform CLC handshake */
749 	rc = smc_connect_clc(smc, smc_type, &aclc, &ini);
750 	if (rc) {
751 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
752 		return smc_connect_decline_fallback(smc, rc);
753 	}
754 
755 	/* depending on previous steps, connect using rdma or ism */
756 	if (rdma_supported && aclc.hdr.path == SMC_TYPE_R)
757 		rc = smc_connect_rdma(smc, &aclc, &ini);
758 	else if (ism_supported && aclc.hdr.path == SMC_TYPE_D)
759 		rc = smc_connect_ism(smc, &aclc, &ini);
760 	else
761 		rc = SMC_CLC_DECL_MODEUNSUPP;
762 	if (rc) {
763 		smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
764 		return smc_connect_decline_fallback(smc, rc);
765 	}
766 
767 	smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini);
768 	return 0;
769 }
770 
smc_connect_work(struct work_struct * work)771 static void smc_connect_work(struct work_struct *work)
772 {
773 	struct smc_sock *smc = container_of(work, struct smc_sock,
774 					    connect_work);
775 	long timeo = smc->sk.sk_sndtimeo;
776 	int rc = 0;
777 
778 	if (!timeo)
779 		timeo = MAX_SCHEDULE_TIMEOUT;
780 	lock_sock(smc->clcsock->sk);
781 	if (smc->clcsock->sk->sk_err) {
782 		smc->sk.sk_err = smc->clcsock->sk->sk_err;
783 	} else if ((1 << smc->clcsock->sk->sk_state) &
784 					(TCPF_SYN_SENT | TCP_SYN_RECV)) {
785 		rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
786 		if ((rc == -EPIPE) &&
787 		    ((1 << smc->clcsock->sk->sk_state) &
788 					(TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
789 			rc = 0;
790 	}
791 	release_sock(smc->clcsock->sk);
792 	lock_sock(&smc->sk);
793 	if (rc != 0 || smc->sk.sk_err) {
794 		smc->sk.sk_state = SMC_CLOSED;
795 		if (rc == -EPIPE || rc == -EAGAIN)
796 			smc->sk.sk_err = EPIPE;
797 		else if (signal_pending(current))
798 			smc->sk.sk_err = -sock_intr_errno(timeo);
799 		sock_put(&smc->sk); /* passive closing */
800 		goto out;
801 	}
802 
803 	rc = __smc_connect(smc);
804 	if (rc < 0)
805 		smc->sk.sk_err = -rc;
806 
807 out:
808 	if (!sock_flag(&smc->sk, SOCK_DEAD)) {
809 		if (smc->sk.sk_err) {
810 			smc->sk.sk_state_change(&smc->sk);
811 		} else { /* allow polling before and after fallback decision */
812 			smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
813 			smc->sk.sk_write_space(&smc->sk);
814 		}
815 	}
816 	release_sock(&smc->sk);
817 }
818 
smc_connect(struct socket * sock,struct sockaddr * addr,int alen,int flags)819 static int smc_connect(struct socket *sock, struct sockaddr *addr,
820 		       int alen, int flags)
821 {
822 	struct sock *sk = sock->sk;
823 	struct smc_sock *smc;
824 	int rc = -EINVAL;
825 
826 	smc = smc_sk(sk);
827 
828 	/* separate smc parameter checking to be safe */
829 	if (alen < sizeof(addr->sa_family))
830 		goto out_err;
831 	if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
832 		goto out_err;
833 
834 	lock_sock(sk);
835 	switch (sk->sk_state) {
836 	default:
837 		goto out;
838 	case SMC_ACTIVE:
839 		rc = -EISCONN;
840 		goto out;
841 	case SMC_INIT:
842 		rc = 0;
843 		break;
844 	}
845 
846 	smc_copy_sock_settings_to_clc(smc);
847 	tcp_sk(smc->clcsock->sk)->syn_smc = 1;
848 	if (smc->connect_nonblock) {
849 		rc = -EALREADY;
850 		goto out;
851 	}
852 	rc = kernel_connect(smc->clcsock, addr, alen, flags);
853 	if (rc && rc != -EINPROGRESS)
854 		goto out;
855 
856 	sock_hold(&smc->sk); /* sock put in passive closing */
857 	if (smc->use_fallback)
858 		goto out;
859 	if (flags & O_NONBLOCK) {
860 		if (schedule_work(&smc->connect_work))
861 			smc->connect_nonblock = 1;
862 		rc = -EINPROGRESS;
863 	} else {
864 		rc = __smc_connect(smc);
865 		if (rc < 0)
866 			goto out;
867 		else
868 			rc = 0; /* success cases including fallback */
869 	}
870 
871 out:
872 	release_sock(sk);
873 out_err:
874 	return rc;
875 }
876 
smc_clcsock_accept(struct smc_sock * lsmc,struct smc_sock ** new_smc)877 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
878 {
879 	struct socket *new_clcsock = NULL;
880 	struct sock *lsk = &lsmc->sk;
881 	struct sock *new_sk;
882 	int rc = -EINVAL;
883 
884 	release_sock(lsk);
885 	new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
886 	if (!new_sk) {
887 		rc = -ENOMEM;
888 		lsk->sk_err = ENOMEM;
889 		*new_smc = NULL;
890 		lock_sock(lsk);
891 		goto out;
892 	}
893 	*new_smc = smc_sk(new_sk);
894 
895 	mutex_lock(&lsmc->clcsock_release_lock);
896 	if (lsmc->clcsock)
897 		rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
898 	mutex_unlock(&lsmc->clcsock_release_lock);
899 	lock_sock(lsk);
900 	if  (rc < 0)
901 		lsk->sk_err = -rc;
902 	if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
903 		new_sk->sk_prot->unhash(new_sk);
904 		if (new_clcsock)
905 			sock_release(new_clcsock);
906 		new_sk->sk_state = SMC_CLOSED;
907 		sock_set_flag(new_sk, SOCK_DEAD);
908 		sock_put(new_sk); /* final */
909 		*new_smc = NULL;
910 		goto out;
911 	}
912 
913 	(*new_smc)->clcsock = new_clcsock;
914 out:
915 	return rc;
916 }
917 
918 /* add a just created sock to the accept queue of the listen sock as
919  * candidate for a following socket accept call from user space
920  */
smc_accept_enqueue(struct sock * parent,struct sock * sk)921 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
922 {
923 	struct smc_sock *par = smc_sk(parent);
924 
925 	sock_hold(sk); /* sock_put in smc_accept_unlink () */
926 	spin_lock(&par->accept_q_lock);
927 	list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
928 	spin_unlock(&par->accept_q_lock);
929 	sk_acceptq_added(parent);
930 }
931 
932 /* remove a socket from the accept queue of its parental listening socket */
smc_accept_unlink(struct sock * sk)933 static void smc_accept_unlink(struct sock *sk)
934 {
935 	struct smc_sock *par = smc_sk(sk)->listen_smc;
936 
937 	spin_lock(&par->accept_q_lock);
938 	list_del_init(&smc_sk(sk)->accept_q);
939 	spin_unlock(&par->accept_q_lock);
940 	sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
941 	sock_put(sk); /* sock_hold in smc_accept_enqueue */
942 }
943 
944 /* remove a sock from the accept queue to bind it to a new socket created
945  * for a socket accept call from user space
946  */
smc_accept_dequeue(struct sock * parent,struct socket * new_sock)947 struct sock *smc_accept_dequeue(struct sock *parent,
948 				struct socket *new_sock)
949 {
950 	struct smc_sock *isk, *n;
951 	struct sock *new_sk;
952 
953 	list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
954 		new_sk = (struct sock *)isk;
955 
956 		smc_accept_unlink(new_sk);
957 		if (new_sk->sk_state == SMC_CLOSED) {
958 			new_sk->sk_prot->unhash(new_sk);
959 			if (isk->clcsock) {
960 				sock_release(isk->clcsock);
961 				isk->clcsock = NULL;
962 			}
963 			sock_put(new_sk); /* final */
964 			continue;
965 		}
966 		if (new_sock) {
967 			sock_graft(new_sk, new_sock);
968 			if (isk->use_fallback) {
969 				smc_sk(new_sk)->clcsock->file = new_sock->file;
970 				isk->clcsock->file->private_data = isk->clcsock;
971 			}
972 		}
973 		return new_sk;
974 	}
975 	return NULL;
976 }
977 
978 /* clean up for a created but never accepted sock */
smc_close_non_accepted(struct sock * sk)979 void smc_close_non_accepted(struct sock *sk)
980 {
981 	struct smc_sock *smc = smc_sk(sk);
982 
983 	lock_sock(sk);
984 	if (!sk->sk_lingertime)
985 		/* wait for peer closing */
986 		sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
987 	__smc_release(smc);
988 	release_sock(sk);
989 	sock_put(sk); /* final sock_put */
990 }
991 
smc_serv_conf_first_link(struct smc_sock * smc)992 static int smc_serv_conf_first_link(struct smc_sock *smc)
993 {
994 	struct net *net = sock_net(smc->clcsock->sk);
995 	struct smc_link_group *lgr = smc->conn.lgr;
996 	struct smc_link *link;
997 	int rest;
998 	int rc;
999 
1000 	link = &lgr->lnk[SMC_SINGLE_LINK];
1001 
1002 	if (smc_reg_rmb(link, smc->conn.rmb_desc, false))
1003 		return SMC_CLC_DECL_ERR_REGRMB;
1004 
1005 	/* send CONFIRM LINK request to client over the RoCE fabric */
1006 	rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1007 	if (rc < 0)
1008 		return SMC_CLC_DECL_TIMEOUT_CL;
1009 
1010 	/* receive CONFIRM LINK response from client over the RoCE fabric */
1011 	rest = wait_for_completion_interruptible_timeout(
1012 		&link->llc_confirm_resp,
1013 		SMC_LLC_WAIT_FIRST_TIME);
1014 	if (rest <= 0) {
1015 		struct smc_clc_msg_decline dclc;
1016 
1017 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1018 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1019 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1020 	}
1021 
1022 	if (link->llc_confirm_resp_rc)
1023 		return SMC_CLC_DECL_RMBE_EC;
1024 
1025 	/* send ADD LINK request to client over the RoCE fabric */
1026 	rc = smc_llc_send_add_link(link,
1027 				   link->smcibdev->mac[link->ibport - 1],
1028 				   link->gid, SMC_LLC_REQ);
1029 	if (rc < 0)
1030 		return SMC_CLC_DECL_TIMEOUT_AL;
1031 
1032 	/* receive ADD LINK response from client over the RoCE fabric */
1033 	rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp,
1034 							 SMC_LLC_WAIT_TIME);
1035 	if (rest <= 0) {
1036 		struct smc_clc_msg_decline dclc;
1037 
1038 		rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1039 				      SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1040 		return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc;
1041 	}
1042 
1043 	smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time);
1044 
1045 	return 0;
1046 }
1047 
1048 /* listen worker: finish */
smc_listen_out(struct smc_sock * new_smc)1049 static void smc_listen_out(struct smc_sock *new_smc)
1050 {
1051 	struct smc_sock *lsmc = new_smc->listen_smc;
1052 	struct sock *newsmcsk = &new_smc->sk;
1053 
1054 	if (lsmc->sk.sk_state == SMC_LISTEN) {
1055 		lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1056 		smc_accept_enqueue(&lsmc->sk, newsmcsk);
1057 		release_sock(&lsmc->sk);
1058 	} else { /* no longer listening */
1059 		smc_close_non_accepted(newsmcsk);
1060 	}
1061 
1062 	/* Wake up accept */
1063 	lsmc->sk.sk_data_ready(&lsmc->sk);
1064 	sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1065 }
1066 
1067 /* listen worker: finish in state connected */
smc_listen_out_connected(struct smc_sock * new_smc)1068 static void smc_listen_out_connected(struct smc_sock *new_smc)
1069 {
1070 	struct sock *newsmcsk = &new_smc->sk;
1071 
1072 	sk_refcnt_debug_inc(newsmcsk);
1073 	if (newsmcsk->sk_state == SMC_INIT)
1074 		newsmcsk->sk_state = SMC_ACTIVE;
1075 
1076 	smc_listen_out(new_smc);
1077 }
1078 
1079 /* listen worker: finish in error state */
smc_listen_out_err(struct smc_sock * new_smc)1080 static void smc_listen_out_err(struct smc_sock *new_smc)
1081 {
1082 	struct sock *newsmcsk = &new_smc->sk;
1083 
1084 	if (newsmcsk->sk_state == SMC_INIT)
1085 		sock_put(&new_smc->sk); /* passive closing */
1086 	newsmcsk->sk_state = SMC_CLOSED;
1087 	smc_conn_free(&new_smc->conn);
1088 
1089 	smc_listen_out(new_smc);
1090 }
1091 
1092 /* listen worker: decline and fall back if possible */
smc_listen_decline(struct smc_sock * new_smc,int reason_code,int local_contact)1093 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1094 			       int local_contact)
1095 {
1096 	/* RDMA setup failed, switch back to TCP */
1097 	if (local_contact == SMC_FIRST_CONTACT)
1098 		smc_lgr_forget(new_smc->conn.lgr);
1099 	if (reason_code < 0) { /* error, no fallback possible */
1100 		smc_listen_out_err(new_smc);
1101 		return;
1102 	}
1103 	smc_conn_free(&new_smc->conn);
1104 	smc_switch_to_fallback(new_smc);
1105 	new_smc->fallback_rsn = reason_code;
1106 	if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1107 		if (smc_clc_send_decline(new_smc, reason_code) < 0) {
1108 			smc_listen_out_err(new_smc);
1109 			return;
1110 		}
1111 	}
1112 	smc_listen_out_connected(new_smc);
1113 }
1114 
1115 /* listen worker: check prefixes */
smc_listen_prfx_check(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc)1116 static int smc_listen_prfx_check(struct smc_sock *new_smc,
1117 				 struct smc_clc_msg_proposal *pclc)
1118 {
1119 	struct smc_clc_msg_proposal_prefix *pclc_prfx;
1120 	struct socket *newclcsock = new_smc->clcsock;
1121 
1122 	pclc_prfx = smc_clc_proposal_get_prefix(pclc);
1123 	if (smc_clc_prfx_match(newclcsock, pclc_prfx))
1124 		return SMC_CLC_DECL_DIFFPREFIX;
1125 
1126 	return 0;
1127 }
1128 
1129 /* listen worker: initialize connection and buffers */
smc_listen_rdma_init(struct smc_sock * new_smc,struct smc_init_info * ini)1130 static int smc_listen_rdma_init(struct smc_sock *new_smc,
1131 				struct smc_init_info *ini)
1132 {
1133 	int rc;
1134 
1135 	/* allocate connection / link group */
1136 	rc = smc_conn_create(new_smc, ini);
1137 	if (rc)
1138 		return rc;
1139 
1140 	/* create send buffer and rmb */
1141 	if (smc_buf_create(new_smc, false))
1142 		return SMC_CLC_DECL_MEM;
1143 
1144 	return 0;
1145 }
1146 
1147 /* listen worker: initialize connection and buffers for SMC-D */
smc_listen_ism_init(struct smc_sock * new_smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)1148 static int smc_listen_ism_init(struct smc_sock *new_smc,
1149 			       struct smc_clc_msg_proposal *pclc,
1150 			       struct smc_init_info *ini)
1151 {
1152 	struct smc_clc_msg_smcd *pclc_smcd;
1153 	int rc;
1154 
1155 	pclc_smcd = smc_get_clc_msg_smcd(pclc);
1156 	ini->ism_gid = pclc_smcd->gid;
1157 	rc = smc_conn_create(new_smc, ini);
1158 	if (rc)
1159 		return rc;
1160 
1161 	/* Check if peer can be reached via ISM device */
1162 	if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid,
1163 			    new_smc->conn.lgr->vlan_id,
1164 			    new_smc->conn.lgr->smcd)) {
1165 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1166 			smc_lgr_forget(new_smc->conn.lgr);
1167 		smc_conn_free(&new_smc->conn);
1168 		return SMC_CLC_DECL_SMCDNOTALK;
1169 	}
1170 
1171 	/* Create send and receive buffers */
1172 	if (smc_buf_create(new_smc, true)) {
1173 		if (ini->cln_first_contact == SMC_FIRST_CONTACT)
1174 			smc_lgr_forget(new_smc->conn.lgr);
1175 		smc_conn_free(&new_smc->conn);
1176 		return SMC_CLC_DECL_MEM;
1177 	}
1178 
1179 	return 0;
1180 }
1181 
1182 /* listen worker: register buffers */
smc_listen_rdma_reg(struct smc_sock * new_smc,int local_contact)1183 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact)
1184 {
1185 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1186 
1187 	if (local_contact != SMC_FIRST_CONTACT) {
1188 		if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true))
1189 			return SMC_CLC_DECL_ERR_REGRMB;
1190 	}
1191 	smc_rmb_sync_sg_for_device(&new_smc->conn);
1192 
1193 	return 0;
1194 }
1195 
1196 /* listen worker: finish RDMA setup */
smc_listen_rdma_finish(struct smc_sock * new_smc,struct smc_clc_msg_accept_confirm * cclc,int local_contact)1197 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
1198 				  struct smc_clc_msg_accept_confirm *cclc,
1199 				  int local_contact)
1200 {
1201 	struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
1202 	int reason_code = 0;
1203 
1204 	if (local_contact == SMC_FIRST_CONTACT)
1205 		smc_link_save_peer_info(link, cclc);
1206 
1207 	if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) {
1208 		reason_code = SMC_CLC_DECL_ERR_RTOK;
1209 		goto decline;
1210 	}
1211 
1212 	if (local_contact == SMC_FIRST_CONTACT) {
1213 		if (smc_ib_ready_link(link)) {
1214 			reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1215 			goto decline;
1216 		}
1217 		/* QP confirmation over RoCE fabric */
1218 		reason_code = smc_serv_conf_first_link(new_smc);
1219 		if (reason_code)
1220 			goto decline;
1221 	}
1222 	return 0;
1223 
1224 decline:
1225 	smc_listen_decline(new_smc, reason_code, local_contact);
1226 	return reason_code;
1227 }
1228 
1229 /* setup for RDMA connection of server */
smc_listen_work(struct work_struct * work)1230 static void smc_listen_work(struct work_struct *work)
1231 {
1232 	struct smc_sock *new_smc = container_of(work, struct smc_sock,
1233 						smc_listen_work);
1234 	struct socket *newclcsock = new_smc->clcsock;
1235 	struct smc_clc_msg_accept_confirm cclc;
1236 	struct smc_clc_msg_proposal *pclc;
1237 	struct smc_init_info ini = {0};
1238 	bool ism_supported = false;
1239 	u8 buf[SMC_CLC_MAX_LEN];
1240 	int rc = 0;
1241 
1242 	if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
1243 		return smc_listen_out_err(new_smc);
1244 
1245 	if (new_smc->use_fallback) {
1246 		smc_listen_out_connected(new_smc);
1247 		return;
1248 	}
1249 
1250 	/* check if peer is smc capable */
1251 	if (!tcp_sk(newclcsock->sk)->syn_smc) {
1252 		smc_switch_to_fallback(new_smc);
1253 		new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC;
1254 		smc_listen_out_connected(new_smc);
1255 		return;
1256 	}
1257 
1258 	/* do inband token exchange -
1259 	 * wait for and receive SMC Proposal CLC message
1260 	 */
1261 	pclc = (struct smc_clc_msg_proposal *)&buf;
1262 	rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN,
1263 			      SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
1264 	if (rc)
1265 		goto out_decl;
1266 
1267 	/* IPSec connections opt out of SMC-R optimizations */
1268 	if (using_ipsec(new_smc)) {
1269 		rc = SMC_CLC_DECL_IPSEC;
1270 		goto out_decl;
1271 	}
1272 
1273 	/* check for matching IP prefix and subnet length */
1274 	rc = smc_listen_prfx_check(new_smc, pclc);
1275 	if (rc)
1276 		goto out_decl;
1277 
1278 	/* get vlan id from IP device */
1279 	if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) {
1280 		rc = SMC_CLC_DECL_GETVLANERR;
1281 		goto out_decl;
1282 	}
1283 
1284 	mutex_lock(&smc_server_lgr_pending);
1285 	smc_close_init(new_smc);
1286 	smc_rx_init(new_smc);
1287 	smc_tx_init(new_smc);
1288 
1289 	/* check if ISM is available */
1290 	if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) {
1291 		ini.is_smcd = true; /* prepare ISM check */
1292 		rc = smc_find_ism_device(new_smc, &ini);
1293 		if (!rc)
1294 			rc = smc_listen_ism_init(new_smc, pclc, &ini);
1295 		if (!rc)
1296 			ism_supported = true;
1297 		else if (pclc->hdr.path == SMC_TYPE_D)
1298 			goto out_unlock; /* skip RDMA and decline */
1299 	}
1300 
1301 	/* check if RDMA is available */
1302 	if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */
1303 		/* prepare RDMA check */
1304 		ini.is_smcd = false;
1305 		ini.ism_dev = NULL;
1306 		ini.ib_lcl = &pclc->lcl;
1307 		rc = smc_find_rdma_device(new_smc, &ini);
1308 		if (rc) {
1309 			/* no RDMA device found */
1310 			if (pclc->hdr.path == SMC_TYPE_B)
1311 				/* neither ISM nor RDMA device found */
1312 				rc = SMC_CLC_DECL_NOSMCDEV;
1313 			goto out_unlock;
1314 		}
1315 		rc = smc_listen_rdma_init(new_smc, &ini);
1316 		if (rc)
1317 			goto out_unlock;
1318 		rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact);
1319 		if (rc)
1320 			goto out_unlock;
1321 	}
1322 
1323 	/* send SMC Accept CLC message */
1324 	rc = smc_clc_send_accept(new_smc, ini.cln_first_contact);
1325 	if (rc)
1326 		goto out_unlock;
1327 
1328 	/* SMC-D does not need this lock any more */
1329 	if (ism_supported)
1330 		mutex_unlock(&smc_server_lgr_pending);
1331 
1332 	/* receive SMC Confirm CLC message */
1333 	rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
1334 			      SMC_CLC_CONFIRM, CLC_WAIT_TIME);
1335 	if (rc) {
1336 		if (!ism_supported)
1337 			goto out_unlock;
1338 		goto out_decl;
1339 	}
1340 
1341 	/* finish worker */
1342 	if (!ism_supported) {
1343 		rc = smc_listen_rdma_finish(new_smc, &cclc,
1344 					    ini.cln_first_contact);
1345 		mutex_unlock(&smc_server_lgr_pending);
1346 		if (rc)
1347 			return;
1348 	}
1349 	smc_conn_save_peer_info(new_smc, &cclc);
1350 	smc_listen_out_connected(new_smc);
1351 	return;
1352 
1353 out_unlock:
1354 	mutex_unlock(&smc_server_lgr_pending);
1355 out_decl:
1356 	smc_listen_decline(new_smc, rc, ini.cln_first_contact);
1357 }
1358 
smc_tcp_listen_work(struct work_struct * work)1359 static void smc_tcp_listen_work(struct work_struct *work)
1360 {
1361 	struct smc_sock *lsmc = container_of(work, struct smc_sock,
1362 					     tcp_listen_work);
1363 	struct sock *lsk = &lsmc->sk;
1364 	struct smc_sock *new_smc;
1365 	int rc = 0;
1366 
1367 	lock_sock(lsk);
1368 	while (lsk->sk_state == SMC_LISTEN) {
1369 		rc = smc_clcsock_accept(lsmc, &new_smc);
1370 		if (rc)
1371 			goto out;
1372 		if (!new_smc)
1373 			continue;
1374 
1375 		new_smc->listen_smc = lsmc;
1376 		new_smc->use_fallback = lsmc->use_fallback;
1377 		new_smc->fallback_rsn = lsmc->fallback_rsn;
1378 		sock_hold(lsk); /* sock_put in smc_listen_work */
1379 		INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
1380 		smc_copy_sock_settings_to_smc(new_smc);
1381 		new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf;
1382 		new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf;
1383 		sock_hold(&new_smc->sk); /* sock_put in passive closing */
1384 		if (!schedule_work(&new_smc->smc_listen_work))
1385 			sock_put(&new_smc->sk);
1386 	}
1387 
1388 out:
1389 	release_sock(lsk);
1390 	sock_put(&lsmc->sk); /* sock_hold in smc_listen */
1391 }
1392 
smc_listen(struct socket * sock,int backlog)1393 static int smc_listen(struct socket *sock, int backlog)
1394 {
1395 	struct sock *sk = sock->sk;
1396 	struct smc_sock *smc;
1397 	int rc;
1398 
1399 	smc = smc_sk(sk);
1400 	lock_sock(sk);
1401 
1402 	rc = -EINVAL;
1403 	if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
1404 	    smc->connect_nonblock)
1405 		goto out;
1406 
1407 	rc = 0;
1408 	if (sk->sk_state == SMC_LISTEN) {
1409 		sk->sk_max_ack_backlog = backlog;
1410 		goto out;
1411 	}
1412 	/* some socket options are handled in core, so we could not apply
1413 	 * them to the clc socket -- copy smc socket options to clc socket
1414 	 */
1415 	smc_copy_sock_settings_to_clc(smc);
1416 	if (!smc->use_fallback)
1417 		tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1418 
1419 	rc = kernel_listen(smc->clcsock, backlog);
1420 	if (rc)
1421 		goto out;
1422 	sk->sk_max_ack_backlog = backlog;
1423 	sk->sk_ack_backlog = 0;
1424 	sk->sk_state = SMC_LISTEN;
1425 	sock_hold(sk); /* sock_hold in tcp_listen_worker */
1426 	if (!schedule_work(&smc->tcp_listen_work))
1427 		sock_put(sk);
1428 
1429 out:
1430 	release_sock(sk);
1431 	return rc;
1432 }
1433 
smc_accept(struct socket * sock,struct socket * new_sock,int flags,bool kern)1434 static int smc_accept(struct socket *sock, struct socket *new_sock,
1435 		      int flags, bool kern)
1436 {
1437 	struct sock *sk = sock->sk, *nsk;
1438 	DECLARE_WAITQUEUE(wait, current);
1439 	struct smc_sock *lsmc;
1440 	long timeo;
1441 	int rc = 0;
1442 
1443 	lsmc = smc_sk(sk);
1444 	sock_hold(sk); /* sock_put below */
1445 	lock_sock(sk);
1446 
1447 	if (lsmc->sk.sk_state != SMC_LISTEN) {
1448 		rc = -EINVAL;
1449 		release_sock(sk);
1450 		goto out;
1451 	}
1452 
1453 	/* Wait for an incoming connection */
1454 	timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1455 	add_wait_queue_exclusive(sk_sleep(sk), &wait);
1456 	while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1457 		set_current_state(TASK_INTERRUPTIBLE);
1458 		if (!timeo) {
1459 			rc = -EAGAIN;
1460 			break;
1461 		}
1462 		release_sock(sk);
1463 		timeo = schedule_timeout(timeo);
1464 		/* wakeup by sk_data_ready in smc_listen_work() */
1465 		sched_annotate_sleep();
1466 		lock_sock(sk);
1467 		if (signal_pending(current)) {
1468 			rc = sock_intr_errno(timeo);
1469 			break;
1470 		}
1471 	}
1472 	set_current_state(TASK_RUNNING);
1473 	remove_wait_queue(sk_sleep(sk), &wait);
1474 
1475 	if (!rc)
1476 		rc = sock_error(nsk);
1477 	release_sock(sk);
1478 	if (rc)
1479 		goto out;
1480 
1481 	if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
1482 		/* wait till data arrives on the socket */
1483 		timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
1484 								MSEC_PER_SEC);
1485 		if (smc_sk(nsk)->use_fallback) {
1486 			struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
1487 
1488 			lock_sock(clcsk);
1489 			if (skb_queue_empty(&clcsk->sk_receive_queue))
1490 				sk_wait_data(clcsk, &timeo, NULL);
1491 			release_sock(clcsk);
1492 		} else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
1493 			lock_sock(nsk);
1494 			smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
1495 			release_sock(nsk);
1496 		}
1497 	}
1498 
1499 out:
1500 	sock_put(sk); /* sock_hold above */
1501 	return rc;
1502 }
1503 
smc_getname(struct socket * sock,struct sockaddr * addr,int peer)1504 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1505 		       int peer)
1506 {
1507 	struct smc_sock *smc;
1508 
1509 	if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1510 	    (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1511 		return -ENOTCONN;
1512 
1513 	smc = smc_sk(sock->sk);
1514 
1515 	return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1516 }
1517 
smc_sendmsg(struct socket * sock,struct msghdr * msg,size_t len)1518 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1519 {
1520 	struct sock *sk = sock->sk;
1521 	struct smc_sock *smc;
1522 	int rc = -EPIPE;
1523 
1524 	smc = smc_sk(sk);
1525 	lock_sock(sk);
1526 	if ((sk->sk_state != SMC_ACTIVE) &&
1527 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1528 	    (sk->sk_state != SMC_INIT))
1529 		goto out;
1530 
1531 	if (msg->msg_flags & MSG_FASTOPEN) {
1532 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1533 			smc_switch_to_fallback(smc);
1534 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1535 		} else {
1536 			rc = -EINVAL;
1537 			goto out;
1538 		}
1539 	}
1540 
1541 	if (smc->use_fallback)
1542 		rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1543 	else
1544 		rc = smc_tx_sendmsg(smc, msg, len);
1545 out:
1546 	release_sock(sk);
1547 	return rc;
1548 }
1549 
smc_recvmsg(struct socket * sock,struct msghdr * msg,size_t len,int flags)1550 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1551 		       int flags)
1552 {
1553 	struct sock *sk = sock->sk;
1554 	struct smc_sock *smc;
1555 	int rc = -ENOTCONN;
1556 
1557 	smc = smc_sk(sk);
1558 	lock_sock(sk);
1559 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1560 		/* socket was connected before, no more data to read */
1561 		rc = 0;
1562 		goto out;
1563 	}
1564 	if ((sk->sk_state == SMC_INIT) ||
1565 	    (sk->sk_state == SMC_LISTEN) ||
1566 	    (sk->sk_state == SMC_CLOSED))
1567 		goto out;
1568 
1569 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1570 		rc = 0;
1571 		goto out;
1572 	}
1573 
1574 	if (smc->use_fallback) {
1575 		rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1576 	} else {
1577 		msg->msg_namelen = 0;
1578 		rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
1579 	}
1580 
1581 out:
1582 	release_sock(sk);
1583 	return rc;
1584 }
1585 
smc_accept_poll(struct sock * parent)1586 static __poll_t smc_accept_poll(struct sock *parent)
1587 {
1588 	struct smc_sock *isk = smc_sk(parent);
1589 	__poll_t mask = 0;
1590 
1591 	spin_lock(&isk->accept_q_lock);
1592 	if (!list_empty(&isk->accept_q))
1593 		mask = EPOLLIN | EPOLLRDNORM;
1594 	spin_unlock(&isk->accept_q_lock);
1595 
1596 	return mask;
1597 }
1598 
smc_poll(struct file * file,struct socket * sock,poll_table * wait)1599 static __poll_t smc_poll(struct file *file, struct socket *sock,
1600 			     poll_table *wait)
1601 {
1602 	struct sock *sk = sock->sk;
1603 	struct smc_sock *smc;
1604 	__poll_t mask = 0;
1605 
1606 	if (!sk)
1607 		return EPOLLNVAL;
1608 
1609 	smc = smc_sk(sock->sk);
1610 	if (smc->use_fallback) {
1611 		/* delegate to CLC child sock */
1612 		mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1613 		sk->sk_err = smc->clcsock->sk->sk_err;
1614 	} else {
1615 		if (sk->sk_state != SMC_CLOSED)
1616 			sock_poll_wait(file, sock, wait);
1617 		if (sk->sk_err)
1618 			mask |= EPOLLERR;
1619 		if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1620 		    (sk->sk_state == SMC_CLOSED))
1621 			mask |= EPOLLHUP;
1622 		if (sk->sk_state == SMC_LISTEN) {
1623 			/* woken up by sk_data_ready in smc_listen_work() */
1624 			mask |= smc_accept_poll(sk);
1625 		} else if (smc->use_fallback) { /* as result of connect_work()*/
1626 			mask |= smc->clcsock->ops->poll(file, smc->clcsock,
1627 							   wait);
1628 			sk->sk_err = smc->clcsock->sk->sk_err;
1629 		} else {
1630 			if ((sk->sk_state != SMC_INIT &&
1631 			     atomic_read(&smc->conn.sndbuf_space)) ||
1632 			    sk->sk_shutdown & SEND_SHUTDOWN) {
1633 				mask |= EPOLLOUT | EPOLLWRNORM;
1634 			} else {
1635 				sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1636 				set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1637 			}
1638 			if (atomic_read(&smc->conn.bytes_to_rcv))
1639 				mask |= EPOLLIN | EPOLLRDNORM;
1640 			if (sk->sk_shutdown & RCV_SHUTDOWN)
1641 				mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1642 			if (sk->sk_state == SMC_APPCLOSEWAIT1)
1643 				mask |= EPOLLIN;
1644 			if (smc->conn.urg_state == SMC_URG_VALID)
1645 				mask |= EPOLLPRI;
1646 		}
1647 	}
1648 
1649 	return mask;
1650 }
1651 
smc_shutdown(struct socket * sock,int how)1652 static int smc_shutdown(struct socket *sock, int how)
1653 {
1654 	struct sock *sk = sock->sk;
1655 	struct smc_sock *smc;
1656 	int rc = -EINVAL;
1657 	int rc1 = 0;
1658 
1659 	smc = smc_sk(sk);
1660 
1661 	if ((how < SHUT_RD) || (how > SHUT_RDWR))
1662 		return rc;
1663 
1664 	lock_sock(sk);
1665 
1666 	rc = -ENOTCONN;
1667 	if ((sk->sk_state != SMC_ACTIVE) &&
1668 	    (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1669 	    (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1670 	    (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1671 	    (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1672 	    (sk->sk_state != SMC_APPFINCLOSEWAIT))
1673 		goto out;
1674 	if (smc->use_fallback) {
1675 		rc = kernel_sock_shutdown(smc->clcsock, how);
1676 		sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1677 		if (sk->sk_shutdown == SHUTDOWN_MASK)
1678 			sk->sk_state = SMC_CLOSED;
1679 		goto out;
1680 	}
1681 	switch (how) {
1682 	case SHUT_RDWR:		/* shutdown in both directions */
1683 		rc = smc_close_active(smc);
1684 		break;
1685 	case SHUT_WR:
1686 		rc = smc_close_shutdown_write(smc);
1687 		break;
1688 	case SHUT_RD:
1689 		rc = 0;
1690 		/* nothing more to do because peer is not involved */
1691 		break;
1692 	}
1693 	if (smc->clcsock)
1694 		rc1 = kernel_sock_shutdown(smc->clcsock, how);
1695 	/* map sock_shutdown_cmd constants to sk_shutdown value range */
1696 	sk->sk_shutdown |= how + 1;
1697 
1698 out:
1699 	release_sock(sk);
1700 	return rc ? rc : rc1;
1701 }
1702 
smc_setsockopt(struct socket * sock,int level,int optname,char __user * optval,unsigned int optlen)1703 static int smc_setsockopt(struct socket *sock, int level, int optname,
1704 			  char __user *optval, unsigned int optlen)
1705 {
1706 	struct sock *sk = sock->sk;
1707 	struct smc_sock *smc;
1708 	int val, rc;
1709 
1710 	smc = smc_sk(sk);
1711 
1712 	/* generic setsockopts reaching us here always apply to the
1713 	 * CLC socket
1714 	 */
1715 	rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1716 					   optval, optlen);
1717 	if (smc->clcsock->sk->sk_err) {
1718 		sk->sk_err = smc->clcsock->sk->sk_err;
1719 		sk->sk_error_report(sk);
1720 	}
1721 
1722 	if (optlen < sizeof(int))
1723 		return -EINVAL;
1724 	if (get_user(val, (int __user *)optval))
1725 		return -EFAULT;
1726 
1727 	lock_sock(sk);
1728 	if (rc || smc->use_fallback)
1729 		goto out;
1730 	switch (optname) {
1731 	case TCP_ULP:
1732 	case TCP_FASTOPEN:
1733 	case TCP_FASTOPEN_CONNECT:
1734 	case TCP_FASTOPEN_KEY:
1735 	case TCP_FASTOPEN_NO_COOKIE:
1736 		/* option not supported by SMC */
1737 		if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
1738 			smc_switch_to_fallback(smc);
1739 			smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP;
1740 		} else {
1741 			rc = -EINVAL;
1742 		}
1743 		break;
1744 	case TCP_NODELAY:
1745 		if (sk->sk_state != SMC_INIT &&
1746 		    sk->sk_state != SMC_LISTEN &&
1747 		    sk->sk_state != SMC_CLOSED) {
1748 			if (val)
1749 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1750 						 0);
1751 		}
1752 		break;
1753 	case TCP_CORK:
1754 		if (sk->sk_state != SMC_INIT &&
1755 		    sk->sk_state != SMC_LISTEN &&
1756 		    sk->sk_state != SMC_CLOSED) {
1757 			if (!val)
1758 				mod_delayed_work(system_wq, &smc->conn.tx_work,
1759 						 0);
1760 		}
1761 		break;
1762 	case TCP_DEFER_ACCEPT:
1763 		smc->sockopt_defer_accept = val;
1764 		break;
1765 	default:
1766 		break;
1767 	}
1768 out:
1769 	release_sock(sk);
1770 
1771 	return rc;
1772 }
1773 
smc_getsockopt(struct socket * sock,int level,int optname,char __user * optval,int __user * optlen)1774 static int smc_getsockopt(struct socket *sock, int level, int optname,
1775 			  char __user *optval, int __user *optlen)
1776 {
1777 	struct smc_sock *smc;
1778 
1779 	smc = smc_sk(sock->sk);
1780 	/* socket options apply to the CLC socket */
1781 	return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1782 					     optval, optlen);
1783 }
1784 
smc_ioctl(struct socket * sock,unsigned int cmd,unsigned long arg)1785 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1786 		     unsigned long arg)
1787 {
1788 	union smc_host_cursor cons, urg;
1789 	struct smc_connection *conn;
1790 	struct smc_sock *smc;
1791 	int answ;
1792 
1793 	smc = smc_sk(sock->sk);
1794 	conn = &smc->conn;
1795 	lock_sock(&smc->sk);
1796 	if (smc->use_fallback) {
1797 		if (!smc->clcsock) {
1798 			release_sock(&smc->sk);
1799 			return -EBADF;
1800 		}
1801 		answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1802 		release_sock(&smc->sk);
1803 		return answ;
1804 	}
1805 	switch (cmd) {
1806 	case SIOCINQ: /* same as FIONREAD */
1807 		if (smc->sk.sk_state == SMC_LISTEN) {
1808 			release_sock(&smc->sk);
1809 			return -EINVAL;
1810 		}
1811 		if (smc->sk.sk_state == SMC_INIT ||
1812 		    smc->sk.sk_state == SMC_CLOSED)
1813 			answ = 0;
1814 		else
1815 			answ = atomic_read(&smc->conn.bytes_to_rcv);
1816 		break;
1817 	case SIOCOUTQ:
1818 		/* output queue size (not send + not acked) */
1819 		if (smc->sk.sk_state == SMC_LISTEN) {
1820 			release_sock(&smc->sk);
1821 			return -EINVAL;
1822 		}
1823 		if (smc->sk.sk_state == SMC_INIT ||
1824 		    smc->sk.sk_state == SMC_CLOSED)
1825 			answ = 0;
1826 		else
1827 			answ = smc->conn.sndbuf_desc->len -
1828 					atomic_read(&smc->conn.sndbuf_space);
1829 		break;
1830 	case SIOCOUTQNSD:
1831 		/* output queue size (not send only) */
1832 		if (smc->sk.sk_state == SMC_LISTEN) {
1833 			release_sock(&smc->sk);
1834 			return -EINVAL;
1835 		}
1836 		if (smc->sk.sk_state == SMC_INIT ||
1837 		    smc->sk.sk_state == SMC_CLOSED)
1838 			answ = 0;
1839 		else
1840 			answ = smc_tx_prepared_sends(&smc->conn);
1841 		break;
1842 	case SIOCATMARK:
1843 		if (smc->sk.sk_state == SMC_LISTEN) {
1844 			release_sock(&smc->sk);
1845 			return -EINVAL;
1846 		}
1847 		if (smc->sk.sk_state == SMC_INIT ||
1848 		    smc->sk.sk_state == SMC_CLOSED) {
1849 			answ = 0;
1850 		} else {
1851 			smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
1852 			smc_curs_copy(&urg, &conn->urg_curs, conn);
1853 			answ = smc_curs_diff(conn->rmb_desc->len,
1854 					     &cons, &urg) == 1;
1855 		}
1856 		break;
1857 	default:
1858 		release_sock(&smc->sk);
1859 		return -ENOIOCTLCMD;
1860 	}
1861 	release_sock(&smc->sk);
1862 
1863 	return put_user(answ, (int __user *)arg);
1864 }
1865 
smc_sendpage(struct socket * sock,struct page * page,int offset,size_t size,int flags)1866 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1867 			    int offset, size_t size, int flags)
1868 {
1869 	struct sock *sk = sock->sk;
1870 	struct smc_sock *smc;
1871 	int rc = -EPIPE;
1872 
1873 	smc = smc_sk(sk);
1874 	lock_sock(sk);
1875 	if (sk->sk_state != SMC_ACTIVE) {
1876 		release_sock(sk);
1877 		goto out;
1878 	}
1879 	release_sock(sk);
1880 	if (smc->use_fallback)
1881 		rc = kernel_sendpage(smc->clcsock, page, offset,
1882 				     size, flags);
1883 	else
1884 		rc = sock_no_sendpage(sock, page, offset, size, flags);
1885 
1886 out:
1887 	return rc;
1888 }
1889 
1890 /* Map the affected portions of the rmbe into an spd, note the number of bytes
1891  * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
1892  * updates till whenever a respective page has been fully processed.
1893  * Note that subsequent recv() calls have to wait till all splice() processing
1894  * completed.
1895  */
smc_splice_read(struct socket * sock,loff_t * ppos,struct pipe_inode_info * pipe,size_t len,unsigned int flags)1896 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1897 			       struct pipe_inode_info *pipe, size_t len,
1898 			       unsigned int flags)
1899 {
1900 	struct sock *sk = sock->sk;
1901 	struct smc_sock *smc;
1902 	int rc = -ENOTCONN;
1903 
1904 	smc = smc_sk(sk);
1905 	lock_sock(sk);
1906 	if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
1907 		/* socket was connected before, no more data to read */
1908 		rc = 0;
1909 		goto out;
1910 	}
1911 	if (sk->sk_state == SMC_INIT ||
1912 	    sk->sk_state == SMC_LISTEN ||
1913 	    sk->sk_state == SMC_CLOSED)
1914 		goto out;
1915 
1916 	if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1917 		rc = 0;
1918 		goto out;
1919 	}
1920 
1921 	if (smc->use_fallback) {
1922 		rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1923 						    pipe, len, flags);
1924 	} else {
1925 		if (*ppos) {
1926 			rc = -ESPIPE;
1927 			goto out;
1928 		}
1929 		if (flags & SPLICE_F_NONBLOCK)
1930 			flags = MSG_DONTWAIT;
1931 		else
1932 			flags = 0;
1933 		rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
1934 	}
1935 out:
1936 	release_sock(sk);
1937 
1938 	return rc;
1939 }
1940 
1941 /* must look like tcp */
1942 static const struct proto_ops smc_sock_ops = {
1943 	.family		= PF_SMC,
1944 	.owner		= THIS_MODULE,
1945 	.release	= smc_release,
1946 	.bind		= smc_bind,
1947 	.connect	= smc_connect,
1948 	.socketpair	= sock_no_socketpair,
1949 	.accept		= smc_accept,
1950 	.getname	= smc_getname,
1951 	.poll		= smc_poll,
1952 	.ioctl		= smc_ioctl,
1953 	.listen		= smc_listen,
1954 	.shutdown	= smc_shutdown,
1955 	.setsockopt	= smc_setsockopt,
1956 	.getsockopt	= smc_getsockopt,
1957 	.sendmsg	= smc_sendmsg,
1958 	.recvmsg	= smc_recvmsg,
1959 	.mmap		= sock_no_mmap,
1960 	.sendpage	= smc_sendpage,
1961 	.splice_read	= smc_splice_read,
1962 };
1963 
smc_create(struct net * net,struct socket * sock,int protocol,int kern)1964 static int smc_create(struct net *net, struct socket *sock, int protocol,
1965 		      int kern)
1966 {
1967 	int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
1968 	struct smc_sock *smc;
1969 	struct sock *sk;
1970 	int rc;
1971 
1972 	rc = -ESOCKTNOSUPPORT;
1973 	if (sock->type != SOCK_STREAM)
1974 		goto out;
1975 
1976 	rc = -EPROTONOSUPPORT;
1977 	if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
1978 		goto out;
1979 
1980 	rc = -ENOBUFS;
1981 	sock->ops = &smc_sock_ops;
1982 	sk = smc_sock_alloc(net, sock, protocol);
1983 	if (!sk)
1984 		goto out;
1985 
1986 	/* create internal TCP socket for CLC handshake and fallback */
1987 	smc = smc_sk(sk);
1988 	smc->use_fallback = false; /* assume rdma capability first */
1989 	smc->fallback_rsn = 0;
1990 	rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
1991 			      &smc->clcsock);
1992 	if (rc) {
1993 		sk_common_release(sk);
1994 		goto out;
1995 	}
1996 	smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1997 	smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1998 
1999 out:
2000 	return rc;
2001 }
2002 
2003 static const struct net_proto_family smc_sock_family_ops = {
2004 	.family	= PF_SMC,
2005 	.owner	= THIS_MODULE,
2006 	.create	= smc_create,
2007 };
2008 
2009 unsigned int smc_net_id;
2010 
smc_net_init(struct net * net)2011 static __net_init int smc_net_init(struct net *net)
2012 {
2013 	return smc_pnet_net_init(net);
2014 }
2015 
smc_net_exit(struct net * net)2016 static void __net_exit smc_net_exit(struct net *net)
2017 {
2018 	smc_pnet_net_exit(net);
2019 }
2020 
2021 static struct pernet_operations smc_net_ops = {
2022 	.init = smc_net_init,
2023 	.exit = smc_net_exit,
2024 	.id   = &smc_net_id,
2025 	.size = sizeof(struct smc_net),
2026 };
2027 
smc_init(void)2028 static int __init smc_init(void)
2029 {
2030 	int rc;
2031 
2032 	rc = register_pernet_subsys(&smc_net_ops);
2033 	if (rc)
2034 		return rc;
2035 
2036 	rc = smc_pnet_init();
2037 	if (rc)
2038 		goto out_pernet_subsys;
2039 
2040 	rc = smc_llc_init();
2041 	if (rc) {
2042 		pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
2043 		goto out_pnet;
2044 	}
2045 
2046 	rc = smc_cdc_init();
2047 	if (rc) {
2048 		pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
2049 		goto out_pnet;
2050 	}
2051 
2052 	rc = proto_register(&smc_proto, 1);
2053 	if (rc) {
2054 		pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
2055 		goto out_pnet;
2056 	}
2057 
2058 	rc = proto_register(&smc_proto6, 1);
2059 	if (rc) {
2060 		pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
2061 		goto out_proto;
2062 	}
2063 
2064 	rc = sock_register(&smc_sock_family_ops);
2065 	if (rc) {
2066 		pr_err("%s: sock_register fails with %d\n", __func__, rc);
2067 		goto out_proto6;
2068 	}
2069 	INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
2070 	INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
2071 
2072 	rc = smc_ib_register_client();
2073 	if (rc) {
2074 		pr_err("%s: ib_register fails with %d\n", __func__, rc);
2075 		goto out_sock;
2076 	}
2077 
2078 	static_branch_enable(&tcp_have_smc);
2079 	return 0;
2080 
2081 out_sock:
2082 	sock_unregister(PF_SMC);
2083 out_proto6:
2084 	proto_unregister(&smc_proto6);
2085 out_proto:
2086 	proto_unregister(&smc_proto);
2087 out_pnet:
2088 	smc_pnet_exit();
2089 out_pernet_subsys:
2090 	unregister_pernet_subsys(&smc_net_ops);
2091 
2092 	return rc;
2093 }
2094 
smc_exit(void)2095 static void __exit smc_exit(void)
2096 {
2097 	smc_core_exit();
2098 	static_branch_disable(&tcp_have_smc);
2099 	smc_ib_unregister_client();
2100 	sock_unregister(PF_SMC);
2101 	proto_unregister(&smc_proto6);
2102 	proto_unregister(&smc_proto);
2103 	smc_pnet_exit();
2104 	unregister_pernet_subsys(&smc_net_ops);
2105 }
2106 
2107 module_init(smc_init);
2108 module_exit(smc_exit);
2109 
2110 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
2111 MODULE_DESCRIPTION("smc socket address family");
2112 MODULE_LICENSE("GPL");
2113 MODULE_ALIAS_NETPROTO(PF_SMC);
2114