xref: /linux/fs/dlm/lowcomms.c (revision 4359a011e259a4608afc7fb3635370c9d4ba5943)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
4 **
5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
6 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
7 **
8 **
9 *******************************************************************************
10 ******************************************************************************/
11 
12 /*
13  * lowcomms.c
14  *
15  * This is the "low-level" comms layer.
16  *
17  * It is responsible for sending/receiving messages
18  * from other nodes in the cluster.
19  *
20  * Cluster nodes are referred to by their nodeids. nodeids are
21  * simply 32 bit numbers to the locking module - if they need to
22  * be expanded for the cluster infrastructure then that is its
23  * responsibility. It is this layer's
24  * responsibility to resolve these into IP address or
25  * whatever it needs for inter-node communication.
26  *
27  * The comms level is two kernel threads that deal mainly with
28  * the receiving of messages from other nodes and passing them
29  * up to the mid-level comms layer (which understands the
30  * message format) for execution by the locking core, and
31  * a send thread which does all the setting up of connections
32  * to remote nodes and the sending of data. Threads are not allowed
33  * to send their own data because it may cause them to wait in times
34  * of high load. Also, this way, the sending thread can collect together
35  * messages bound for one node and send them in one block.
36  *
37  * lowcomms will choose to use either TCP or SCTP as its transport layer
38  * depending on the configuration variable 'protocol'. This should be set
39  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
40  * cluster-wide mechanism as it must be the same on all nodes of the cluster
41  * for the DLM to function.
42  *
43  */
44 
45 #include <asm/ioctls.h>
46 #include <net/sock.h>
47 #include <net/tcp.h>
48 #include <linux/pagemap.h>
49 #include <linux/file.h>
50 #include <linux/mutex.h>
51 #include <linux/sctp.h>
52 #include <linux/slab.h>
53 #include <net/sctp/sctp.h>
54 #include <net/ipv6.h>
55 
56 #include <trace/events/dlm.h>
57 
58 #include "dlm_internal.h"
59 #include "lowcomms.h"
60 #include "midcomms.h"
61 #include "memory.h"
62 #include "config.h"
63 
64 #define NEEDED_RMEM (4*1024*1024)
65 
66 /* Number of messages to send before rescheduling */
67 #define MAX_SEND_MSG_COUNT 25
68 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
69 
70 struct connection {
71 	struct socket *sock;	/* NULL if not connected */
72 	uint32_t nodeid;	/* So we know who we are in the list */
73 	struct mutex sock_mutex;
74 	unsigned long flags;
75 #define CF_READ_PENDING 1
76 #define CF_WRITE_PENDING 2
77 #define CF_INIT_PENDING 4
78 #define CF_IS_OTHERCON 5
79 #define CF_CLOSE 6
80 #define CF_APP_LIMITED 7
81 #define CF_CLOSING 8
82 #define CF_SHUTDOWN 9
83 #define CF_CONNECTED 10
84 #define CF_RECONNECT 11
85 #define CF_DELAY_CONNECT 12
86 #define CF_EOF 13
87 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
88 	spinlock_t writequeue_lock;
89 	atomic_t writequeue_cnt;
90 	int retries;
91 #define MAX_CONNECT_RETRIES 3
92 	struct hlist_node list;
93 	struct connection *othercon;
94 	struct connection *sendcon;
95 	struct work_struct rwork; /* Receive workqueue */
96 	struct work_struct swork; /* Send workqueue */
97 	wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
98 	unsigned char *rx_buf;
99 	int rx_buflen;
100 	int rx_leftover;
101 	struct rcu_head rcu;
102 };
103 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
104 
105 struct listen_connection {
106 	struct socket *sock;
107 	struct work_struct rwork;
108 };
109 
110 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
111 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
112 
113 /* An entry waiting to be sent */
114 struct writequeue_entry {
115 	struct list_head list;
116 	struct page *page;
117 	int offset;
118 	int len;
119 	int end;
120 	int users;
121 	bool dirty;
122 	struct connection *con;
123 	struct list_head msgs;
124 	struct kref ref;
125 };
126 
127 struct dlm_msg {
128 	struct writequeue_entry *entry;
129 	struct dlm_msg *orig_msg;
130 	bool retransmit;
131 	void *ppc;
132 	int len;
133 	int idx; /* new()/commit() idx exchange */
134 
135 	struct list_head list;
136 	struct kref ref;
137 };
138 
139 struct dlm_node_addr {
140 	struct list_head list;
141 	int nodeid;
142 	int mark;
143 	int addr_count;
144 	int curr_addr_index;
145 	struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
146 };
147 
148 struct dlm_proto_ops {
149 	bool try_new_addr;
150 	const char *name;
151 	int proto;
152 
153 	int (*connect)(struct connection *con, struct socket *sock,
154 		       struct sockaddr *addr, int addr_len);
155 	void (*sockopts)(struct socket *sock);
156 	int (*bind)(struct socket *sock);
157 	int (*listen_validate)(void);
158 	void (*listen_sockopts)(struct socket *sock);
159 	int (*listen_bind)(struct socket *sock);
160 	/* What to do to shutdown */
161 	void (*shutdown_action)(struct connection *con);
162 	/* What to do to eof check */
163 	bool (*eof_condition)(struct connection *con);
164 };
165 
166 static struct listen_sock_callbacks {
167 	void (*sk_error_report)(struct sock *);
168 	void (*sk_data_ready)(struct sock *);
169 	void (*sk_state_change)(struct sock *);
170 	void (*sk_write_space)(struct sock *);
171 } listen_sock;
172 
173 static LIST_HEAD(dlm_node_addrs);
174 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
175 
176 static struct listen_connection listen_con;
177 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
178 static int dlm_local_count;
179 int dlm_allow_conn;
180 
181 /* Work queues */
182 static struct workqueue_struct *recv_workqueue;
183 static struct workqueue_struct *send_workqueue;
184 
185 static struct hlist_head connection_hash[CONN_HASH_SIZE];
186 static DEFINE_SPINLOCK(connections_lock);
187 DEFINE_STATIC_SRCU(connections_srcu);
188 
189 static const struct dlm_proto_ops *dlm_proto_ops;
190 
191 static void process_recv_sockets(struct work_struct *work);
192 static void process_send_sockets(struct work_struct *work);
193 
194 static void writequeue_entry_ctor(void *data)
195 {
196 	struct writequeue_entry *entry = data;
197 
198 	INIT_LIST_HEAD(&entry->msgs);
199 }
200 
201 struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void)
202 {
203 	return kmem_cache_create("dlm_writequeue", sizeof(struct writequeue_entry),
204 				 0, 0, writequeue_entry_ctor);
205 }
206 
207 struct kmem_cache *dlm_lowcomms_msg_cache_create(void)
208 {
209 	return kmem_cache_create("dlm_msg", sizeof(struct dlm_msg), 0, 0, NULL);
210 }
211 
212 /* need to held writequeue_lock */
213 static struct writequeue_entry *con_next_wq(struct connection *con)
214 {
215 	struct writequeue_entry *e;
216 
217 	if (list_empty(&con->writequeue))
218 		return NULL;
219 
220 	e = list_first_entry(&con->writequeue, struct writequeue_entry,
221 			     list);
222 	/* if len is zero nothing is to send, if there are users filling
223 	 * buffers we wait until the users are done so we can send more.
224 	 */
225 	if (e->users || e->len == 0)
226 		return NULL;
227 
228 	return e;
229 }
230 
231 static struct connection *__find_con(int nodeid, int r)
232 {
233 	struct connection *con;
234 
235 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
236 		if (con->nodeid == nodeid)
237 			return con;
238 	}
239 
240 	return NULL;
241 }
242 
243 static bool tcp_eof_condition(struct connection *con)
244 {
245 	return atomic_read(&con->writequeue_cnt);
246 }
247 
248 static int dlm_con_init(struct connection *con, int nodeid)
249 {
250 	con->rx_buflen = dlm_config.ci_buffer_size;
251 	con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
252 	if (!con->rx_buf)
253 		return -ENOMEM;
254 
255 	con->nodeid = nodeid;
256 	mutex_init(&con->sock_mutex);
257 	INIT_LIST_HEAD(&con->writequeue);
258 	spin_lock_init(&con->writequeue_lock);
259 	atomic_set(&con->writequeue_cnt, 0);
260 	INIT_WORK(&con->swork, process_send_sockets);
261 	INIT_WORK(&con->rwork, process_recv_sockets);
262 	init_waitqueue_head(&con->shutdown_wait);
263 
264 	return 0;
265 }
266 
267 /*
268  * If 'allocation' is zero then we don't attempt to create a new
269  * connection structure for this node.
270  */
271 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
272 {
273 	struct connection *con, *tmp;
274 	int r, ret;
275 
276 	r = nodeid_hash(nodeid);
277 	con = __find_con(nodeid, r);
278 	if (con || !alloc)
279 		return con;
280 
281 	con = kzalloc(sizeof(*con), alloc);
282 	if (!con)
283 		return NULL;
284 
285 	ret = dlm_con_init(con, nodeid);
286 	if (ret) {
287 		kfree(con);
288 		return NULL;
289 	}
290 
291 	spin_lock(&connections_lock);
292 	/* Because multiple workqueues/threads calls this function it can
293 	 * race on multiple cpu's. Instead of locking hot path __find_con()
294 	 * we just check in rare cases of recently added nodes again
295 	 * under protection of connections_lock. If this is the case we
296 	 * abort our connection creation and return the existing connection.
297 	 */
298 	tmp = __find_con(nodeid, r);
299 	if (tmp) {
300 		spin_unlock(&connections_lock);
301 		kfree(con->rx_buf);
302 		kfree(con);
303 		return tmp;
304 	}
305 
306 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
307 	spin_unlock(&connections_lock);
308 
309 	return con;
310 }
311 
312 /* Loop round all connections */
313 static void foreach_conn(void (*conn_func)(struct connection *c))
314 {
315 	int i;
316 	struct connection *con;
317 
318 	for (i = 0; i < CONN_HASH_SIZE; i++) {
319 		hlist_for_each_entry_rcu(con, &connection_hash[i], list)
320 			conn_func(con);
321 	}
322 }
323 
324 static struct dlm_node_addr *find_node_addr(int nodeid)
325 {
326 	struct dlm_node_addr *na;
327 
328 	list_for_each_entry(na, &dlm_node_addrs, list) {
329 		if (na->nodeid == nodeid)
330 			return na;
331 	}
332 	return NULL;
333 }
334 
335 static int addr_compare(const struct sockaddr_storage *x,
336 			const struct sockaddr_storage *y)
337 {
338 	switch (x->ss_family) {
339 	case AF_INET: {
340 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
341 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
342 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
343 			return 0;
344 		if (sinx->sin_port != siny->sin_port)
345 			return 0;
346 		break;
347 	}
348 	case AF_INET6: {
349 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
350 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
351 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
352 			return 0;
353 		if (sinx->sin6_port != siny->sin6_port)
354 			return 0;
355 		break;
356 	}
357 	default:
358 		return 0;
359 	}
360 	return 1;
361 }
362 
363 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
364 			  struct sockaddr *sa_out, bool try_new_addr,
365 			  unsigned int *mark)
366 {
367 	struct sockaddr_storage sas;
368 	struct dlm_node_addr *na;
369 
370 	if (!dlm_local_count)
371 		return -1;
372 
373 	spin_lock(&dlm_node_addrs_spin);
374 	na = find_node_addr(nodeid);
375 	if (na && na->addr_count) {
376 		memcpy(&sas, na->addr[na->curr_addr_index],
377 		       sizeof(struct sockaddr_storage));
378 
379 		if (try_new_addr) {
380 			na->curr_addr_index++;
381 			if (na->curr_addr_index == na->addr_count)
382 				na->curr_addr_index = 0;
383 		}
384 	}
385 	spin_unlock(&dlm_node_addrs_spin);
386 
387 	if (!na)
388 		return -EEXIST;
389 
390 	if (!na->addr_count)
391 		return -ENOENT;
392 
393 	*mark = na->mark;
394 
395 	if (sas_out)
396 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
397 
398 	if (!sa_out)
399 		return 0;
400 
401 	if (dlm_local_addr[0]->ss_family == AF_INET) {
402 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
403 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
404 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
405 	} else {
406 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
407 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
408 		ret6->sin6_addr = in6->sin6_addr;
409 	}
410 
411 	return 0;
412 }
413 
414 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
415 			  unsigned int *mark)
416 {
417 	struct dlm_node_addr *na;
418 	int rv = -EEXIST;
419 	int addr_i;
420 
421 	spin_lock(&dlm_node_addrs_spin);
422 	list_for_each_entry(na, &dlm_node_addrs, list) {
423 		if (!na->addr_count)
424 			continue;
425 
426 		for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
427 			if (addr_compare(na->addr[addr_i], addr)) {
428 				*nodeid = na->nodeid;
429 				*mark = na->mark;
430 				rv = 0;
431 				goto unlock;
432 			}
433 		}
434 	}
435 unlock:
436 	spin_unlock(&dlm_node_addrs_spin);
437 	return rv;
438 }
439 
440 /* caller need to held dlm_node_addrs_spin lock */
441 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
442 				     const struct sockaddr_storage *addr)
443 {
444 	int i;
445 
446 	for (i = 0; i < na->addr_count; i++) {
447 		if (addr_compare(na->addr[i], addr))
448 			return true;
449 	}
450 
451 	return false;
452 }
453 
454 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
455 {
456 	struct sockaddr_storage *new_addr;
457 	struct dlm_node_addr *new_node, *na;
458 	bool ret;
459 
460 	new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
461 	if (!new_node)
462 		return -ENOMEM;
463 
464 	new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
465 	if (!new_addr) {
466 		kfree(new_node);
467 		return -ENOMEM;
468 	}
469 
470 	memcpy(new_addr, addr, len);
471 
472 	spin_lock(&dlm_node_addrs_spin);
473 	na = find_node_addr(nodeid);
474 	if (!na) {
475 		new_node->nodeid = nodeid;
476 		new_node->addr[0] = new_addr;
477 		new_node->addr_count = 1;
478 		new_node->mark = dlm_config.ci_mark;
479 		list_add(&new_node->list, &dlm_node_addrs);
480 		spin_unlock(&dlm_node_addrs_spin);
481 		return 0;
482 	}
483 
484 	ret = dlm_lowcomms_na_has_addr(na, addr);
485 	if (ret) {
486 		spin_unlock(&dlm_node_addrs_spin);
487 		kfree(new_addr);
488 		kfree(new_node);
489 		return -EEXIST;
490 	}
491 
492 	if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
493 		spin_unlock(&dlm_node_addrs_spin);
494 		kfree(new_addr);
495 		kfree(new_node);
496 		return -ENOSPC;
497 	}
498 
499 	na->addr[na->addr_count++] = new_addr;
500 	spin_unlock(&dlm_node_addrs_spin);
501 	kfree(new_node);
502 	return 0;
503 }
504 
505 /* Data available on socket or listen socket received a connect */
506 static void lowcomms_data_ready(struct sock *sk)
507 {
508 	struct connection *con;
509 
510 	con = sock2con(sk);
511 	if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
512 		queue_work(recv_workqueue, &con->rwork);
513 }
514 
515 static void lowcomms_listen_data_ready(struct sock *sk)
516 {
517 	if (!dlm_allow_conn)
518 		return;
519 
520 	queue_work(recv_workqueue, &listen_con.rwork);
521 }
522 
523 static void lowcomms_write_space(struct sock *sk)
524 {
525 	struct connection *con;
526 
527 	con = sock2con(sk);
528 	if (!con)
529 		return;
530 
531 	if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
532 		log_print("connected to node %d", con->nodeid);
533 		queue_work(send_workqueue, &con->swork);
534 		return;
535 	}
536 
537 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
538 
539 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
540 		con->sock->sk->sk_write_pending--;
541 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
542 	}
543 
544 	queue_work(send_workqueue, &con->swork);
545 }
546 
547 static inline void lowcomms_connect_sock(struct connection *con)
548 {
549 	if (test_bit(CF_CLOSE, &con->flags))
550 		return;
551 	queue_work(send_workqueue, &con->swork);
552 	cond_resched();
553 }
554 
555 static void lowcomms_state_change(struct sock *sk)
556 {
557 	/* SCTP layer is not calling sk_data_ready when the connection
558 	 * is done, so we catch the signal through here. Also, it
559 	 * doesn't switch socket state when entering shutdown, so we
560 	 * skip the write in that case.
561 	 */
562 	if (sk->sk_shutdown) {
563 		if (sk->sk_shutdown == RCV_SHUTDOWN)
564 			lowcomms_data_ready(sk);
565 	} else if (sk->sk_state == TCP_ESTABLISHED) {
566 		lowcomms_write_space(sk);
567 	}
568 }
569 
570 int dlm_lowcomms_connect_node(int nodeid)
571 {
572 	struct connection *con;
573 	int idx;
574 
575 	if (nodeid == dlm_our_nodeid())
576 		return 0;
577 
578 	idx = srcu_read_lock(&connections_srcu);
579 	con = nodeid2con(nodeid, GFP_NOFS);
580 	if (!con) {
581 		srcu_read_unlock(&connections_srcu, idx);
582 		return -ENOMEM;
583 	}
584 
585 	lowcomms_connect_sock(con);
586 	srcu_read_unlock(&connections_srcu, idx);
587 
588 	return 0;
589 }
590 
591 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
592 {
593 	struct dlm_node_addr *na;
594 
595 	spin_lock(&dlm_node_addrs_spin);
596 	na = find_node_addr(nodeid);
597 	if (!na) {
598 		spin_unlock(&dlm_node_addrs_spin);
599 		return -ENOENT;
600 	}
601 
602 	na->mark = mark;
603 	spin_unlock(&dlm_node_addrs_spin);
604 
605 	return 0;
606 }
607 
608 static void lowcomms_error_report(struct sock *sk)
609 {
610 	struct connection *con;
611 	void (*orig_report)(struct sock *) = NULL;
612 	struct inet_sock *inet;
613 
614 	con = sock2con(sk);
615 	if (con == NULL)
616 		goto out;
617 
618 	orig_report = listen_sock.sk_error_report;
619 
620 	inet = inet_sk(sk);
621 	switch (sk->sk_family) {
622 	case AF_INET:
623 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
624 				   "sending to node %d at %pI4, dport %d, "
625 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
626 				   con->nodeid, &inet->inet_daddr,
627 				   ntohs(inet->inet_dport), sk->sk_err,
628 				   sk->sk_err_soft);
629 		break;
630 #if IS_ENABLED(CONFIG_IPV6)
631 	case AF_INET6:
632 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
633 				   "sending to node %d at %pI6c, "
634 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
635 				   con->nodeid, &sk->sk_v6_daddr,
636 				   ntohs(inet->inet_dport), sk->sk_err,
637 				   sk->sk_err_soft);
638 		break;
639 #endif
640 	default:
641 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
642 				   "invalid socket family %d set, "
643 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
644 				   sk->sk_family, sk->sk_err, sk->sk_err_soft);
645 		goto out;
646 	}
647 
648 	/* below sendcon only handling */
649 	if (test_bit(CF_IS_OTHERCON, &con->flags))
650 		con = con->sendcon;
651 
652 	switch (sk->sk_err) {
653 	case ECONNREFUSED:
654 		set_bit(CF_DELAY_CONNECT, &con->flags);
655 		break;
656 	default:
657 		break;
658 	}
659 
660 	if (!test_and_set_bit(CF_RECONNECT, &con->flags))
661 		queue_work(send_workqueue, &con->swork);
662 
663 out:
664 	if (orig_report)
665 		orig_report(sk);
666 }
667 
668 /* Note: sk_callback_lock must be locked before calling this function. */
669 static void save_listen_callbacks(struct socket *sock)
670 {
671 	struct sock *sk = sock->sk;
672 
673 	listen_sock.sk_data_ready = sk->sk_data_ready;
674 	listen_sock.sk_state_change = sk->sk_state_change;
675 	listen_sock.sk_write_space = sk->sk_write_space;
676 	listen_sock.sk_error_report = sk->sk_error_report;
677 }
678 
679 static void restore_callbacks(struct socket *sock)
680 {
681 	struct sock *sk = sock->sk;
682 
683 	lock_sock(sk);
684 	sk->sk_user_data = NULL;
685 	sk->sk_data_ready = listen_sock.sk_data_ready;
686 	sk->sk_state_change = listen_sock.sk_state_change;
687 	sk->sk_write_space = listen_sock.sk_write_space;
688 	sk->sk_error_report = listen_sock.sk_error_report;
689 	release_sock(sk);
690 }
691 
692 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
693 {
694 	struct sock *sk = sock->sk;
695 
696 	lock_sock(sk);
697 	save_listen_callbacks(sock);
698 	con->sock = sock;
699 
700 	sk->sk_user_data = con;
701 	sk->sk_allocation = GFP_NOFS;
702 	/* Install a data_ready callback */
703 	sk->sk_data_ready = lowcomms_listen_data_ready;
704 	release_sock(sk);
705 }
706 
707 /* Make a socket active */
708 static void add_sock(struct socket *sock, struct connection *con)
709 {
710 	struct sock *sk = sock->sk;
711 
712 	lock_sock(sk);
713 	con->sock = sock;
714 
715 	sk->sk_user_data = con;
716 	/* Install a data_ready callback */
717 	sk->sk_data_ready = lowcomms_data_ready;
718 	sk->sk_write_space = lowcomms_write_space;
719 	sk->sk_state_change = lowcomms_state_change;
720 	sk->sk_allocation = GFP_NOFS;
721 	sk->sk_error_report = lowcomms_error_report;
722 	release_sock(sk);
723 }
724 
725 /* Add the port number to an IPv6 or 4 sockaddr and return the address
726    length */
727 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
728 			  int *addr_len)
729 {
730 	saddr->ss_family =  dlm_local_addr[0]->ss_family;
731 	if (saddr->ss_family == AF_INET) {
732 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
733 		in4_addr->sin_port = cpu_to_be16(port);
734 		*addr_len = sizeof(struct sockaddr_in);
735 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
736 	} else {
737 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
738 		in6_addr->sin6_port = cpu_to_be16(port);
739 		*addr_len = sizeof(struct sockaddr_in6);
740 	}
741 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
742 }
743 
744 static void dlm_page_release(struct kref *kref)
745 {
746 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
747 						  ref);
748 
749 	__free_page(e->page);
750 	dlm_free_writequeue(e);
751 }
752 
753 static void dlm_msg_release(struct kref *kref)
754 {
755 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
756 
757 	kref_put(&msg->entry->ref, dlm_page_release);
758 	dlm_free_msg(msg);
759 }
760 
761 static void free_entry(struct writequeue_entry *e)
762 {
763 	struct dlm_msg *msg, *tmp;
764 
765 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
766 		if (msg->orig_msg) {
767 			msg->orig_msg->retransmit = false;
768 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
769 		}
770 
771 		list_del(&msg->list);
772 		kref_put(&msg->ref, dlm_msg_release);
773 	}
774 
775 	list_del(&e->list);
776 	atomic_dec(&e->con->writequeue_cnt);
777 	kref_put(&e->ref, dlm_page_release);
778 }
779 
780 static void dlm_close_sock(struct socket **sock)
781 {
782 	if (*sock) {
783 		restore_callbacks(*sock);
784 		sock_release(*sock);
785 		*sock = NULL;
786 	}
787 }
788 
789 /* Close a remote connection and tidy up */
790 static void close_connection(struct connection *con, bool and_other,
791 			     bool tx, bool rx)
792 {
793 	bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
794 	struct writequeue_entry *e;
795 
796 	if (tx && !closing && cancel_work_sync(&con->swork)) {
797 		log_print("canceled swork for node %d", con->nodeid);
798 		clear_bit(CF_WRITE_PENDING, &con->flags);
799 	}
800 	if (rx && !closing && cancel_work_sync(&con->rwork)) {
801 		log_print("canceled rwork for node %d", con->nodeid);
802 		clear_bit(CF_READ_PENDING, &con->flags);
803 	}
804 
805 	mutex_lock(&con->sock_mutex);
806 	dlm_close_sock(&con->sock);
807 
808 	if (con->othercon && and_other) {
809 		/* Will only re-enter once. */
810 		close_connection(con->othercon, false, tx, rx);
811 	}
812 
813 	/* if we send a writequeue entry only a half way, we drop the
814 	 * whole entry because reconnection and that we not start of the
815 	 * middle of a msg which will confuse the other end.
816 	 *
817 	 * we can always drop messages because retransmits, but what we
818 	 * cannot allow is to transmit half messages which may be processed
819 	 * at the other side.
820 	 *
821 	 * our policy is to start on a clean state when disconnects, we don't
822 	 * know what's send/received on transport layer in this case.
823 	 */
824 	spin_lock(&con->writequeue_lock);
825 	if (!list_empty(&con->writequeue)) {
826 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
827 				     list);
828 		if (e->dirty)
829 			free_entry(e);
830 	}
831 	spin_unlock(&con->writequeue_lock);
832 
833 	con->rx_leftover = 0;
834 	con->retries = 0;
835 	clear_bit(CF_APP_LIMITED, &con->flags);
836 	clear_bit(CF_CONNECTED, &con->flags);
837 	clear_bit(CF_DELAY_CONNECT, &con->flags);
838 	clear_bit(CF_RECONNECT, &con->flags);
839 	clear_bit(CF_EOF, &con->flags);
840 	mutex_unlock(&con->sock_mutex);
841 	clear_bit(CF_CLOSING, &con->flags);
842 }
843 
844 static void shutdown_connection(struct connection *con)
845 {
846 	int ret;
847 
848 	flush_work(&con->swork);
849 
850 	mutex_lock(&con->sock_mutex);
851 	/* nothing to shutdown */
852 	if (!con->sock) {
853 		mutex_unlock(&con->sock_mutex);
854 		return;
855 	}
856 
857 	set_bit(CF_SHUTDOWN, &con->flags);
858 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
859 	mutex_unlock(&con->sock_mutex);
860 	if (ret) {
861 		log_print("Connection %p failed to shutdown: %d will force close",
862 			  con, ret);
863 		goto force_close;
864 	} else {
865 		ret = wait_event_timeout(con->shutdown_wait,
866 					 !test_bit(CF_SHUTDOWN, &con->flags),
867 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
868 		if (ret == 0) {
869 			log_print("Connection %p shutdown timed out, will force close",
870 				  con);
871 			goto force_close;
872 		}
873 	}
874 
875 	return;
876 
877 force_close:
878 	clear_bit(CF_SHUTDOWN, &con->flags);
879 	close_connection(con, false, true, true);
880 }
881 
882 static void dlm_tcp_shutdown(struct connection *con)
883 {
884 	if (con->othercon)
885 		shutdown_connection(con->othercon);
886 	shutdown_connection(con);
887 }
888 
889 static int con_realloc_receive_buf(struct connection *con, int newlen)
890 {
891 	unsigned char *newbuf;
892 
893 	newbuf = kmalloc(newlen, GFP_NOFS);
894 	if (!newbuf)
895 		return -ENOMEM;
896 
897 	/* copy any leftover from last receive */
898 	if (con->rx_leftover)
899 		memmove(newbuf, con->rx_buf, con->rx_leftover);
900 
901 	/* swap to new buffer space */
902 	kfree(con->rx_buf);
903 	con->rx_buflen = newlen;
904 	con->rx_buf = newbuf;
905 
906 	return 0;
907 }
908 
909 /* Data received from remote end */
910 static int receive_from_sock(struct connection *con)
911 {
912 	struct msghdr msg;
913 	struct kvec iov;
914 	int ret, buflen;
915 
916 	mutex_lock(&con->sock_mutex);
917 
918 	if (con->sock == NULL) {
919 		ret = -EAGAIN;
920 		goto out_close;
921 	}
922 
923 	/* realloc if we get new buffer size to read out */
924 	buflen = dlm_config.ci_buffer_size;
925 	if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
926 		ret = con_realloc_receive_buf(con, buflen);
927 		if (ret < 0)
928 			goto out_resched;
929 	}
930 
931 	for (;;) {
932 		/* calculate new buffer parameter regarding last receive and
933 		 * possible leftover bytes
934 		 */
935 		iov.iov_base = con->rx_buf + con->rx_leftover;
936 		iov.iov_len = con->rx_buflen - con->rx_leftover;
937 
938 		memset(&msg, 0, sizeof(msg));
939 		msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
940 		ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
941 				     msg.msg_flags);
942 		trace_dlm_recv(con->nodeid, ret);
943 		if (ret == -EAGAIN)
944 			break;
945 		else if (ret <= 0)
946 			goto out_close;
947 
948 		/* new buflen according readed bytes and leftover from last receive */
949 		buflen = ret + con->rx_leftover;
950 		ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
951 		if (ret < 0)
952 			goto out_close;
953 
954 		/* calculate leftover bytes from process and put it into begin of
955 		 * the receive buffer, so next receive we have the full message
956 		 * at the start address of the receive buffer.
957 		 */
958 		con->rx_leftover = buflen - ret;
959 		if (con->rx_leftover) {
960 			memmove(con->rx_buf, con->rx_buf + ret,
961 				con->rx_leftover);
962 		}
963 	}
964 
965 	dlm_midcomms_receive_done(con->nodeid);
966 	mutex_unlock(&con->sock_mutex);
967 	return 0;
968 
969 out_resched:
970 	if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
971 		queue_work(recv_workqueue, &con->rwork);
972 	mutex_unlock(&con->sock_mutex);
973 	return -EAGAIN;
974 
975 out_close:
976 	if (ret == 0) {
977 		log_print("connection %p got EOF from %d",
978 			  con, con->nodeid);
979 
980 		if (dlm_proto_ops->eof_condition &&
981 		    dlm_proto_ops->eof_condition(con)) {
982 			set_bit(CF_EOF, &con->flags);
983 			mutex_unlock(&con->sock_mutex);
984 		} else {
985 			mutex_unlock(&con->sock_mutex);
986 			close_connection(con, false, true, false);
987 
988 			/* handling for tcp shutdown */
989 			clear_bit(CF_SHUTDOWN, &con->flags);
990 			wake_up(&con->shutdown_wait);
991 		}
992 
993 		/* signal to breaking receive worker */
994 		ret = -1;
995 	} else {
996 		mutex_unlock(&con->sock_mutex);
997 	}
998 	return ret;
999 }
1000 
1001 /* Listening socket is busy, accept a connection */
1002 static int accept_from_sock(struct listen_connection *con)
1003 {
1004 	int result;
1005 	struct sockaddr_storage peeraddr;
1006 	struct socket *newsock;
1007 	int len, idx;
1008 	int nodeid;
1009 	struct connection *newcon;
1010 	struct connection *addcon;
1011 	unsigned int mark;
1012 
1013 	if (!con->sock)
1014 		return -ENOTCONN;
1015 
1016 	result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
1017 	if (result < 0)
1018 		goto accept_err;
1019 
1020 	/* Get the connected socket's peer */
1021 	memset(&peeraddr, 0, sizeof(peeraddr));
1022 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1023 	if (len < 0) {
1024 		result = -ECONNABORTED;
1025 		goto accept_err;
1026 	}
1027 
1028 	/* Get the new node's NODEID */
1029 	make_sockaddr(&peeraddr, 0, &len);
1030 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1031 		switch (peeraddr.ss_family) {
1032 		case AF_INET: {
1033 			struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr;
1034 
1035 			log_print("connect from non cluster IPv4 node %pI4",
1036 				  &sin->sin_addr);
1037 			break;
1038 		}
1039 #if IS_ENABLED(CONFIG_IPV6)
1040 		case AF_INET6: {
1041 			struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr;
1042 
1043 			log_print("connect from non cluster IPv6 node %pI6c",
1044 				  &sin6->sin6_addr);
1045 			break;
1046 		}
1047 #endif
1048 		default:
1049 			log_print("invalid family from non cluster node");
1050 			break;
1051 		}
1052 
1053 		sock_release(newsock);
1054 		return -1;
1055 	}
1056 
1057 	log_print("got connection from %d", nodeid);
1058 
1059 	/*  Check to see if we already have a connection to this node. This
1060 	 *  could happen if the two nodes initiate a connection at roughly
1061 	 *  the same time and the connections cross on the wire.
1062 	 *  In this case we store the incoming one in "othercon"
1063 	 */
1064 	idx = srcu_read_lock(&connections_srcu);
1065 	newcon = nodeid2con(nodeid, GFP_NOFS);
1066 	if (!newcon) {
1067 		srcu_read_unlock(&connections_srcu, idx);
1068 		result = -ENOMEM;
1069 		goto accept_err;
1070 	}
1071 
1072 	sock_set_mark(newsock->sk, mark);
1073 
1074 	mutex_lock(&newcon->sock_mutex);
1075 	if (newcon->sock) {
1076 		struct connection *othercon = newcon->othercon;
1077 
1078 		if (!othercon) {
1079 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1080 			if (!othercon) {
1081 				log_print("failed to allocate incoming socket");
1082 				mutex_unlock(&newcon->sock_mutex);
1083 				srcu_read_unlock(&connections_srcu, idx);
1084 				result = -ENOMEM;
1085 				goto accept_err;
1086 			}
1087 
1088 			result = dlm_con_init(othercon, nodeid);
1089 			if (result < 0) {
1090 				kfree(othercon);
1091 				mutex_unlock(&newcon->sock_mutex);
1092 				srcu_read_unlock(&connections_srcu, idx);
1093 				goto accept_err;
1094 			}
1095 
1096 			lockdep_set_subclass(&othercon->sock_mutex, 1);
1097 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1098 			newcon->othercon = othercon;
1099 			othercon->sendcon = newcon;
1100 		} else {
1101 			/* close other sock con if we have something new */
1102 			close_connection(othercon, false, true, false);
1103 		}
1104 
1105 		mutex_lock(&othercon->sock_mutex);
1106 		add_sock(newsock, othercon);
1107 		addcon = othercon;
1108 		mutex_unlock(&othercon->sock_mutex);
1109 	}
1110 	else {
1111 		/* accept copies the sk after we've saved the callbacks, so we
1112 		   don't want to save them a second time or comm errors will
1113 		   result in calling sk_error_report recursively. */
1114 		add_sock(newsock, newcon);
1115 		addcon = newcon;
1116 	}
1117 
1118 	set_bit(CF_CONNECTED, &addcon->flags);
1119 	mutex_unlock(&newcon->sock_mutex);
1120 
1121 	/*
1122 	 * Add it to the active queue in case we got data
1123 	 * between processing the accept adding the socket
1124 	 * to the read_sockets list
1125 	 */
1126 	if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1127 		queue_work(recv_workqueue, &addcon->rwork);
1128 
1129 	srcu_read_unlock(&connections_srcu, idx);
1130 
1131 	return 0;
1132 
1133 accept_err:
1134 	if (newsock)
1135 		sock_release(newsock);
1136 
1137 	if (result != -EAGAIN)
1138 		log_print("error accepting connection from node: %d", result);
1139 	return result;
1140 }
1141 
1142 /*
1143  * writequeue_entry_complete - try to delete and free write queue entry
1144  * @e: write queue entry to try to delete
1145  * @completed: bytes completed
1146  *
1147  * writequeue_lock must be held.
1148  */
1149 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1150 {
1151 	e->offset += completed;
1152 	e->len -= completed;
1153 	/* signal that page was half way transmitted */
1154 	e->dirty = true;
1155 
1156 	if (e->len == 0 && e->users == 0)
1157 		free_entry(e);
1158 }
1159 
1160 /*
1161  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1162  */
1163 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1164 {
1165 	struct sockaddr_storage localaddr;
1166 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1167 	int i, addr_len, result = 0;
1168 
1169 	for (i = 0; i < dlm_local_count; i++) {
1170 		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1171 		make_sockaddr(&localaddr, port, &addr_len);
1172 
1173 		if (!i)
1174 			result = kernel_bind(sock, addr, addr_len);
1175 		else
1176 			result = sock_bind_add(sock->sk, addr, addr_len);
1177 
1178 		if (result < 0) {
1179 			log_print("Can't bind to %d addr number %d, %d.\n",
1180 				  port, i + 1, result);
1181 			break;
1182 		}
1183 	}
1184 	return result;
1185 }
1186 
1187 /* Get local addresses */
1188 static void init_local(void)
1189 {
1190 	struct sockaddr_storage sas, *addr;
1191 	int i;
1192 
1193 	dlm_local_count = 0;
1194 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1195 		if (dlm_our_addr(&sas, i))
1196 			break;
1197 
1198 		addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1199 		if (!addr)
1200 			break;
1201 		dlm_local_addr[dlm_local_count++] = addr;
1202 	}
1203 }
1204 
1205 static void deinit_local(void)
1206 {
1207 	int i;
1208 
1209 	for (i = 0; i < dlm_local_count; i++)
1210 		kfree(dlm_local_addr[i]);
1211 }
1212 
1213 static struct writequeue_entry *new_writequeue_entry(struct connection *con)
1214 {
1215 	struct writequeue_entry *entry;
1216 
1217 	entry = dlm_allocate_writequeue();
1218 	if (!entry)
1219 		return NULL;
1220 
1221 	entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
1222 	if (!entry->page) {
1223 		dlm_free_writequeue(entry);
1224 		return NULL;
1225 	}
1226 
1227 	entry->offset = 0;
1228 	entry->len = 0;
1229 	entry->end = 0;
1230 	entry->dirty = false;
1231 	entry->con = con;
1232 	entry->users = 1;
1233 	kref_init(&entry->ref);
1234 	return entry;
1235 }
1236 
1237 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1238 					     char **ppc, void (*cb)(void *data),
1239 					     void *data)
1240 {
1241 	struct writequeue_entry *e;
1242 
1243 	spin_lock(&con->writequeue_lock);
1244 	if (!list_empty(&con->writequeue)) {
1245 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1246 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1247 			kref_get(&e->ref);
1248 
1249 			*ppc = page_address(e->page) + e->end;
1250 			if (cb)
1251 				cb(data);
1252 
1253 			e->end += len;
1254 			e->users++;
1255 			goto out;
1256 		}
1257 	}
1258 
1259 	e = new_writequeue_entry(con);
1260 	if (!e)
1261 		goto out;
1262 
1263 	kref_get(&e->ref);
1264 	*ppc = page_address(e->page);
1265 	e->end += len;
1266 	atomic_inc(&con->writequeue_cnt);
1267 	if (cb)
1268 		cb(data);
1269 
1270 	list_add_tail(&e->list, &con->writequeue);
1271 
1272 out:
1273 	spin_unlock(&con->writequeue_lock);
1274 	return e;
1275 };
1276 
1277 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1278 						gfp_t allocation, char **ppc,
1279 						void (*cb)(void *data),
1280 						void *data)
1281 {
1282 	struct writequeue_entry *e;
1283 	struct dlm_msg *msg;
1284 
1285 	msg = dlm_allocate_msg(allocation);
1286 	if (!msg)
1287 		return NULL;
1288 
1289 	kref_init(&msg->ref);
1290 
1291 	e = new_wq_entry(con, len, ppc, cb, data);
1292 	if (!e) {
1293 		dlm_free_msg(msg);
1294 		return NULL;
1295 	}
1296 
1297 	msg->retransmit = false;
1298 	msg->orig_msg = NULL;
1299 	msg->ppc = *ppc;
1300 	msg->len = len;
1301 	msg->entry = e;
1302 
1303 	return msg;
1304 }
1305 
1306 /* avoid false positive for nodes_srcu, unlock happens in
1307  * dlm_lowcomms_commit_msg which is a must call if success
1308  */
1309 #ifndef __CHECKER__
1310 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1311 				     char **ppc, void (*cb)(void *data),
1312 				     void *data)
1313 {
1314 	struct connection *con;
1315 	struct dlm_msg *msg;
1316 	int idx;
1317 
1318 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1319 	    len < sizeof(struct dlm_header)) {
1320 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1321 		log_print("failed to allocate a buffer of size %d", len);
1322 		WARN_ON(1);
1323 		return NULL;
1324 	}
1325 
1326 	idx = srcu_read_lock(&connections_srcu);
1327 	con = nodeid2con(nodeid, allocation);
1328 	if (!con) {
1329 		srcu_read_unlock(&connections_srcu, idx);
1330 		return NULL;
1331 	}
1332 
1333 	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data);
1334 	if (!msg) {
1335 		srcu_read_unlock(&connections_srcu, idx);
1336 		return NULL;
1337 	}
1338 
1339 	/* we assume if successful commit must called */
1340 	msg->idx = idx;
1341 	return msg;
1342 }
1343 #endif
1344 
1345 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1346 {
1347 	struct writequeue_entry *e = msg->entry;
1348 	struct connection *con = e->con;
1349 	int users;
1350 
1351 	spin_lock(&con->writequeue_lock);
1352 	kref_get(&msg->ref);
1353 	list_add(&msg->list, &e->msgs);
1354 
1355 	users = --e->users;
1356 	if (users)
1357 		goto out;
1358 
1359 	e->len = DLM_WQ_LENGTH_BYTES(e);
1360 	spin_unlock(&con->writequeue_lock);
1361 
1362 	queue_work(send_workqueue, &con->swork);
1363 	return;
1364 
1365 out:
1366 	spin_unlock(&con->writequeue_lock);
1367 	return;
1368 }
1369 
1370 /* avoid false positive for nodes_srcu, lock was happen in
1371  * dlm_lowcomms_new_msg
1372  */
1373 #ifndef __CHECKER__
1374 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1375 {
1376 	_dlm_lowcomms_commit_msg(msg);
1377 	srcu_read_unlock(&connections_srcu, msg->idx);
1378 }
1379 #endif
1380 
1381 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1382 {
1383 	kref_put(&msg->ref, dlm_msg_release);
1384 }
1385 
1386 /* does not held connections_srcu, usage workqueue only */
1387 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1388 {
1389 	struct dlm_msg *msg_resend;
1390 	char *ppc;
1391 
1392 	if (msg->retransmit)
1393 		return 1;
1394 
1395 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1396 					      GFP_ATOMIC, &ppc, NULL, NULL);
1397 	if (!msg_resend)
1398 		return -ENOMEM;
1399 
1400 	msg->retransmit = true;
1401 	kref_get(&msg->ref);
1402 	msg_resend->orig_msg = msg;
1403 
1404 	memcpy(ppc, msg->ppc, msg->len);
1405 	_dlm_lowcomms_commit_msg(msg_resend);
1406 	dlm_lowcomms_put_msg(msg_resend);
1407 
1408 	return 0;
1409 }
1410 
1411 /* Send a message */
1412 static void send_to_sock(struct connection *con)
1413 {
1414 	const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1415 	struct writequeue_entry *e;
1416 	int len, offset, ret;
1417 	int count = 0;
1418 
1419 	mutex_lock(&con->sock_mutex);
1420 	if (con->sock == NULL)
1421 		goto out_connect;
1422 
1423 	spin_lock(&con->writequeue_lock);
1424 	for (;;) {
1425 		e = con_next_wq(con);
1426 		if (!e)
1427 			break;
1428 
1429 		len = e->len;
1430 		offset = e->offset;
1431 		BUG_ON(len == 0 && e->users == 0);
1432 		spin_unlock(&con->writequeue_lock);
1433 
1434 		ret = kernel_sendpage(con->sock, e->page, offset, len,
1435 				      msg_flags);
1436 		trace_dlm_send(con->nodeid, ret);
1437 		if (ret == -EAGAIN || ret == 0) {
1438 			if (ret == -EAGAIN &&
1439 			    test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1440 			    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1441 				/* Notify TCP that we're limited by the
1442 				 * application window size.
1443 				 */
1444 				set_bit(SOCK_NOSPACE, &con->sock->flags);
1445 				con->sock->sk->sk_write_pending++;
1446 			}
1447 			cond_resched();
1448 			goto out;
1449 		} else if (ret < 0)
1450 			goto out;
1451 
1452 		/* Don't starve people filling buffers */
1453 		if (++count >= MAX_SEND_MSG_COUNT) {
1454 			cond_resched();
1455 			count = 0;
1456 		}
1457 
1458 		spin_lock(&con->writequeue_lock);
1459 		writequeue_entry_complete(e, ret);
1460 	}
1461 	spin_unlock(&con->writequeue_lock);
1462 
1463 	/* close if we got EOF */
1464 	if (test_and_clear_bit(CF_EOF, &con->flags)) {
1465 		mutex_unlock(&con->sock_mutex);
1466 		close_connection(con, false, false, true);
1467 
1468 		/* handling for tcp shutdown */
1469 		clear_bit(CF_SHUTDOWN, &con->flags);
1470 		wake_up(&con->shutdown_wait);
1471 	} else {
1472 		mutex_unlock(&con->sock_mutex);
1473 	}
1474 
1475 	return;
1476 
1477 out:
1478 	mutex_unlock(&con->sock_mutex);
1479 	return;
1480 
1481 out_connect:
1482 	mutex_unlock(&con->sock_mutex);
1483 	queue_work(send_workqueue, &con->swork);
1484 	cond_resched();
1485 }
1486 
1487 static void clean_one_writequeue(struct connection *con)
1488 {
1489 	struct writequeue_entry *e, *safe;
1490 
1491 	spin_lock(&con->writequeue_lock);
1492 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1493 		free_entry(e);
1494 	}
1495 	spin_unlock(&con->writequeue_lock);
1496 }
1497 
1498 /* Called from recovery when it knows that a node has
1499    left the cluster */
1500 int dlm_lowcomms_close(int nodeid)
1501 {
1502 	struct connection *con;
1503 	struct dlm_node_addr *na;
1504 	int idx;
1505 
1506 	log_print("closing connection to node %d", nodeid);
1507 	idx = srcu_read_lock(&connections_srcu);
1508 	con = nodeid2con(nodeid, 0);
1509 	if (con) {
1510 		set_bit(CF_CLOSE, &con->flags);
1511 		close_connection(con, true, true, true);
1512 		clean_one_writequeue(con);
1513 		if (con->othercon)
1514 			clean_one_writequeue(con->othercon);
1515 	}
1516 	srcu_read_unlock(&connections_srcu, idx);
1517 
1518 	spin_lock(&dlm_node_addrs_spin);
1519 	na = find_node_addr(nodeid);
1520 	if (na) {
1521 		list_del(&na->list);
1522 		while (na->addr_count--)
1523 			kfree(na->addr[na->addr_count]);
1524 		kfree(na);
1525 	}
1526 	spin_unlock(&dlm_node_addrs_spin);
1527 
1528 	return 0;
1529 }
1530 
1531 /* Receive workqueue function */
1532 static void process_recv_sockets(struct work_struct *work)
1533 {
1534 	struct connection *con = container_of(work, struct connection, rwork);
1535 
1536 	clear_bit(CF_READ_PENDING, &con->flags);
1537 	receive_from_sock(con);
1538 }
1539 
1540 static void process_listen_recv_socket(struct work_struct *work)
1541 {
1542 	accept_from_sock(&listen_con);
1543 }
1544 
1545 static void dlm_connect(struct connection *con)
1546 {
1547 	struct sockaddr_storage addr;
1548 	int result, addr_len;
1549 	struct socket *sock;
1550 	unsigned int mark;
1551 
1552 	/* Some odd races can cause double-connects, ignore them */
1553 	if (con->retries++ > MAX_CONNECT_RETRIES)
1554 		return;
1555 
1556 	if (con->sock) {
1557 		log_print("node %d already connected.", con->nodeid);
1558 		return;
1559 	}
1560 
1561 	memset(&addr, 0, sizeof(addr));
1562 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1563 				dlm_proto_ops->try_new_addr, &mark);
1564 	if (result < 0) {
1565 		log_print("no address for nodeid %d", con->nodeid);
1566 		return;
1567 	}
1568 
1569 	/* Create a socket to communicate with */
1570 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1571 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1572 	if (result < 0)
1573 		goto socket_err;
1574 
1575 	sock_set_mark(sock->sk, mark);
1576 	dlm_proto_ops->sockopts(sock);
1577 
1578 	add_sock(sock, con);
1579 
1580 	result = dlm_proto_ops->bind(sock);
1581 	if (result < 0)
1582 		goto add_sock_err;
1583 
1584 	log_print_ratelimited("connecting to %d", con->nodeid);
1585 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1586 	result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1587 					addr_len);
1588 	if (result < 0)
1589 		goto add_sock_err;
1590 
1591 	return;
1592 
1593 add_sock_err:
1594 	dlm_close_sock(&con->sock);
1595 
1596 socket_err:
1597 	/*
1598 	 * Some errors are fatal and this list might need adjusting. For other
1599 	 * errors we try again until the max number of retries is reached.
1600 	 */
1601 	if (result != -EHOSTUNREACH &&
1602 	    result != -ENETUNREACH &&
1603 	    result != -ENETDOWN &&
1604 	    result != -EINVAL &&
1605 	    result != -EPROTONOSUPPORT) {
1606 		log_print("connect %d try %d error %d", con->nodeid,
1607 			  con->retries, result);
1608 		msleep(1000);
1609 		lowcomms_connect_sock(con);
1610 	}
1611 }
1612 
1613 /* Send workqueue function */
1614 static void process_send_sockets(struct work_struct *work)
1615 {
1616 	struct connection *con = container_of(work, struct connection, swork);
1617 
1618 	WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1619 
1620 	clear_bit(CF_WRITE_PENDING, &con->flags);
1621 
1622 	if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1623 		close_connection(con, false, false, true);
1624 		dlm_midcomms_unack_msg_resend(con->nodeid);
1625 	}
1626 
1627 	if (con->sock == NULL) {
1628 		if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1629 			msleep(1000);
1630 
1631 		mutex_lock(&con->sock_mutex);
1632 		dlm_connect(con);
1633 		mutex_unlock(&con->sock_mutex);
1634 	}
1635 
1636 	if (!list_empty(&con->writequeue))
1637 		send_to_sock(con);
1638 }
1639 
1640 static void work_stop(void)
1641 {
1642 	if (recv_workqueue) {
1643 		destroy_workqueue(recv_workqueue);
1644 		recv_workqueue = NULL;
1645 	}
1646 
1647 	if (send_workqueue) {
1648 		destroy_workqueue(send_workqueue);
1649 		send_workqueue = NULL;
1650 	}
1651 }
1652 
1653 static int work_start(void)
1654 {
1655 	recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1656 	if (!recv_workqueue) {
1657 		log_print("can't start dlm_recv");
1658 		return -ENOMEM;
1659 	}
1660 
1661 	send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1662 	if (!send_workqueue) {
1663 		log_print("can't start dlm_send");
1664 		destroy_workqueue(recv_workqueue);
1665 		recv_workqueue = NULL;
1666 		return -ENOMEM;
1667 	}
1668 
1669 	return 0;
1670 }
1671 
1672 static void shutdown_conn(struct connection *con)
1673 {
1674 	if (dlm_proto_ops->shutdown_action)
1675 		dlm_proto_ops->shutdown_action(con);
1676 }
1677 
1678 void dlm_lowcomms_shutdown(void)
1679 {
1680 	int idx;
1681 
1682 	/* Set all the flags to prevent any
1683 	 * socket activity.
1684 	 */
1685 	dlm_allow_conn = 0;
1686 
1687 	if (recv_workqueue)
1688 		flush_workqueue(recv_workqueue);
1689 	if (send_workqueue)
1690 		flush_workqueue(send_workqueue);
1691 
1692 	dlm_close_sock(&listen_con.sock);
1693 
1694 	idx = srcu_read_lock(&connections_srcu);
1695 	foreach_conn(shutdown_conn);
1696 	srcu_read_unlock(&connections_srcu, idx);
1697 }
1698 
1699 static void _stop_conn(struct connection *con, bool and_other)
1700 {
1701 	mutex_lock(&con->sock_mutex);
1702 	set_bit(CF_CLOSE, &con->flags);
1703 	set_bit(CF_READ_PENDING, &con->flags);
1704 	set_bit(CF_WRITE_PENDING, &con->flags);
1705 	if (con->sock && con->sock->sk) {
1706 		lock_sock(con->sock->sk);
1707 		con->sock->sk->sk_user_data = NULL;
1708 		release_sock(con->sock->sk);
1709 	}
1710 	if (con->othercon && and_other)
1711 		_stop_conn(con->othercon, false);
1712 	mutex_unlock(&con->sock_mutex);
1713 }
1714 
1715 static void stop_conn(struct connection *con)
1716 {
1717 	_stop_conn(con, true);
1718 }
1719 
1720 static void connection_release(struct rcu_head *rcu)
1721 {
1722 	struct connection *con = container_of(rcu, struct connection, rcu);
1723 
1724 	kfree(con->rx_buf);
1725 	kfree(con);
1726 }
1727 
1728 static void free_conn(struct connection *con)
1729 {
1730 	close_connection(con, true, true, true);
1731 	spin_lock(&connections_lock);
1732 	hlist_del_rcu(&con->list);
1733 	spin_unlock(&connections_lock);
1734 	if (con->othercon) {
1735 		clean_one_writequeue(con->othercon);
1736 		call_srcu(&connections_srcu, &con->othercon->rcu,
1737 			  connection_release);
1738 	}
1739 	clean_one_writequeue(con);
1740 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1741 }
1742 
1743 static void work_flush(void)
1744 {
1745 	int ok;
1746 	int i;
1747 	struct connection *con;
1748 
1749 	do {
1750 		ok = 1;
1751 		foreach_conn(stop_conn);
1752 		if (recv_workqueue)
1753 			flush_workqueue(recv_workqueue);
1754 		if (send_workqueue)
1755 			flush_workqueue(send_workqueue);
1756 		for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1757 			hlist_for_each_entry_rcu(con, &connection_hash[i],
1758 						 list) {
1759 				ok &= test_bit(CF_READ_PENDING, &con->flags);
1760 				ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1761 				if (con->othercon) {
1762 					ok &= test_bit(CF_READ_PENDING,
1763 						       &con->othercon->flags);
1764 					ok &= test_bit(CF_WRITE_PENDING,
1765 						       &con->othercon->flags);
1766 				}
1767 			}
1768 		}
1769 	} while (!ok);
1770 }
1771 
1772 void dlm_lowcomms_stop(void)
1773 {
1774 	int idx;
1775 
1776 	idx = srcu_read_lock(&connections_srcu);
1777 	work_flush();
1778 	foreach_conn(free_conn);
1779 	srcu_read_unlock(&connections_srcu, idx);
1780 	work_stop();
1781 	deinit_local();
1782 
1783 	dlm_proto_ops = NULL;
1784 }
1785 
1786 static int dlm_listen_for_all(void)
1787 {
1788 	struct socket *sock;
1789 	int result;
1790 
1791 	log_print("Using %s for communications",
1792 		  dlm_proto_ops->name);
1793 
1794 	result = dlm_proto_ops->listen_validate();
1795 	if (result < 0)
1796 		return result;
1797 
1798 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1799 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1800 	if (result < 0) {
1801 		log_print("Can't create comms socket: %d", result);
1802 		return result;
1803 	}
1804 
1805 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1806 	dlm_proto_ops->listen_sockopts(sock);
1807 
1808 	result = dlm_proto_ops->listen_bind(sock);
1809 	if (result < 0)
1810 		goto out;
1811 
1812 	save_listen_callbacks(sock);
1813 	add_listen_sock(sock, &listen_con);
1814 
1815 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1816 	result = sock->ops->listen(sock, 5);
1817 	if (result < 0) {
1818 		dlm_close_sock(&listen_con.sock);
1819 		goto out;
1820 	}
1821 
1822 	return 0;
1823 
1824 out:
1825 	sock_release(sock);
1826 	return result;
1827 }
1828 
1829 static int dlm_tcp_bind(struct socket *sock)
1830 {
1831 	struct sockaddr_storage src_addr;
1832 	int result, addr_len;
1833 
1834 	/* Bind to our cluster-known address connecting to avoid
1835 	 * routing problems.
1836 	 */
1837 	memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1838 	make_sockaddr(&src_addr, 0, &addr_len);
1839 
1840 	result = sock->ops->bind(sock, (struct sockaddr *)&src_addr,
1841 				 addr_len);
1842 	if (result < 0) {
1843 		/* This *may* not indicate a critical error */
1844 		log_print("could not bind for connect: %d", result);
1845 	}
1846 
1847 	return 0;
1848 }
1849 
1850 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1851 			   struct sockaddr *addr, int addr_len)
1852 {
1853 	int ret;
1854 
1855 	ret = sock->ops->connect(sock, addr, addr_len, O_NONBLOCK);
1856 	switch (ret) {
1857 	case -EINPROGRESS:
1858 		fallthrough;
1859 	case 0:
1860 		return 0;
1861 	}
1862 
1863 	return ret;
1864 }
1865 
1866 static int dlm_tcp_listen_validate(void)
1867 {
1868 	/* We don't support multi-homed hosts */
1869 	if (dlm_local_count > 1) {
1870 		log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1871 		return -EINVAL;
1872 	}
1873 
1874 	return 0;
1875 }
1876 
1877 static void dlm_tcp_sockopts(struct socket *sock)
1878 {
1879 	/* Turn off Nagle's algorithm */
1880 	tcp_sock_set_nodelay(sock->sk);
1881 }
1882 
1883 static void dlm_tcp_listen_sockopts(struct socket *sock)
1884 {
1885 	dlm_tcp_sockopts(sock);
1886 	sock_set_reuseaddr(sock->sk);
1887 }
1888 
1889 static int dlm_tcp_listen_bind(struct socket *sock)
1890 {
1891 	int addr_len;
1892 
1893 	/* Bind to our port */
1894 	make_sockaddr(dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1895 	return sock->ops->bind(sock, (struct sockaddr *)dlm_local_addr[0],
1896 			       addr_len);
1897 }
1898 
1899 static const struct dlm_proto_ops dlm_tcp_ops = {
1900 	.name = "TCP",
1901 	.proto = IPPROTO_TCP,
1902 	.connect = dlm_tcp_connect,
1903 	.sockopts = dlm_tcp_sockopts,
1904 	.bind = dlm_tcp_bind,
1905 	.listen_validate = dlm_tcp_listen_validate,
1906 	.listen_sockopts = dlm_tcp_listen_sockopts,
1907 	.listen_bind = dlm_tcp_listen_bind,
1908 	.shutdown_action = dlm_tcp_shutdown,
1909 	.eof_condition = tcp_eof_condition,
1910 };
1911 
1912 static int dlm_sctp_bind(struct socket *sock)
1913 {
1914 	return sctp_bind_addrs(sock, 0);
1915 }
1916 
1917 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1918 			    struct sockaddr *addr, int addr_len)
1919 {
1920 	int ret;
1921 
1922 	/*
1923 	 * Make sock->ops->connect() function return in specified time,
1924 	 * since O_NONBLOCK argument in connect() function does not work here,
1925 	 * then, we should restore the default value of this attribute.
1926 	 */
1927 	sock_set_sndtimeo(sock->sk, 5);
1928 	ret = sock->ops->connect(sock, addr, addr_len, 0);
1929 	sock_set_sndtimeo(sock->sk, 0);
1930 	if (ret < 0)
1931 		return ret;
1932 
1933 	if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1934 		log_print("connected to node %d", con->nodeid);
1935 
1936 	return 0;
1937 }
1938 
1939 static int dlm_sctp_listen_validate(void)
1940 {
1941 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1942 		log_print("SCTP is not enabled by this kernel");
1943 		return -EOPNOTSUPP;
1944 	}
1945 
1946 	request_module("sctp");
1947 	return 0;
1948 }
1949 
1950 static int dlm_sctp_bind_listen(struct socket *sock)
1951 {
1952 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1953 }
1954 
1955 static void dlm_sctp_sockopts(struct socket *sock)
1956 {
1957 	/* Turn off Nagle's algorithm */
1958 	sctp_sock_set_nodelay(sock->sk);
1959 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1960 }
1961 
1962 static const struct dlm_proto_ops dlm_sctp_ops = {
1963 	.name = "SCTP",
1964 	.proto = IPPROTO_SCTP,
1965 	.try_new_addr = true,
1966 	.connect = dlm_sctp_connect,
1967 	.sockopts = dlm_sctp_sockopts,
1968 	.bind = dlm_sctp_bind,
1969 	.listen_validate = dlm_sctp_listen_validate,
1970 	.listen_sockopts = dlm_sctp_sockopts,
1971 	.listen_bind = dlm_sctp_bind_listen,
1972 };
1973 
1974 int dlm_lowcomms_start(void)
1975 {
1976 	int error = -EINVAL;
1977 	int i;
1978 
1979 	for (i = 0; i < CONN_HASH_SIZE; i++)
1980 		INIT_HLIST_HEAD(&connection_hash[i]);
1981 
1982 	init_local();
1983 	if (!dlm_local_count) {
1984 		error = -ENOTCONN;
1985 		log_print("no local IP address has been set");
1986 		goto fail;
1987 	}
1988 
1989 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1990 
1991 	error = work_start();
1992 	if (error)
1993 		goto fail_local;
1994 
1995 	dlm_allow_conn = 1;
1996 
1997 	/* Start listening */
1998 	switch (dlm_config.ci_protocol) {
1999 	case DLM_PROTO_TCP:
2000 		dlm_proto_ops = &dlm_tcp_ops;
2001 		break;
2002 	case DLM_PROTO_SCTP:
2003 		dlm_proto_ops = &dlm_sctp_ops;
2004 		break;
2005 	default:
2006 		log_print("Invalid protocol identifier %d set",
2007 			  dlm_config.ci_protocol);
2008 		error = -EINVAL;
2009 		goto fail_proto_ops;
2010 	}
2011 
2012 	error = dlm_listen_for_all();
2013 	if (error)
2014 		goto fail_listen;
2015 
2016 	return 0;
2017 
2018 fail_listen:
2019 	dlm_proto_ops = NULL;
2020 fail_proto_ops:
2021 	dlm_allow_conn = 0;
2022 	dlm_close_sock(&listen_con.sock);
2023 	work_stop();
2024 fail_local:
2025 	deinit_local();
2026 fail:
2027 	return error;
2028 }
2029 
2030 void dlm_lowcomms_exit(void)
2031 {
2032 	struct dlm_node_addr *na, *safe;
2033 
2034 	spin_lock(&dlm_node_addrs_spin);
2035 	list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2036 		list_del(&na->list);
2037 		while (na->addr_count--)
2038 			kfree(na->addr[na->addr_count]);
2039 		kfree(na);
2040 	}
2041 	spin_unlock(&dlm_node_addrs_spin);
2042 }
2043