xref: /linux/fs/dlm/lowcomms.c (revision 0ad53fe3ae82443c74ff8cfd7bd13377cc1134a3)
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 "dlm_internal.h"
57 #include "lowcomms.h"
58 #include "midcomms.h"
59 #include "config.h"
60 
61 #define NEEDED_RMEM (4*1024*1024)
62 
63 /* Number of messages to send before rescheduling */
64 #define MAX_SEND_MSG_COUNT 25
65 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(10000)
66 
67 struct connection {
68 	struct socket *sock;	/* NULL if not connected */
69 	uint32_t nodeid;	/* So we know who we are in the list */
70 	struct mutex sock_mutex;
71 	unsigned long flags;
72 #define CF_READ_PENDING 1
73 #define CF_WRITE_PENDING 2
74 #define CF_INIT_PENDING 4
75 #define CF_IS_OTHERCON 5
76 #define CF_CLOSE 6
77 #define CF_APP_LIMITED 7
78 #define CF_CLOSING 8
79 #define CF_SHUTDOWN 9
80 #define CF_CONNECTED 10
81 #define CF_RECONNECT 11
82 #define CF_DELAY_CONNECT 12
83 #define CF_EOF 13
84 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
85 	spinlock_t writequeue_lock;
86 	atomic_t writequeue_cnt;
87 	struct mutex wq_alloc;
88 	int retries;
89 #define MAX_CONNECT_RETRIES 3
90 	struct hlist_node list;
91 	struct connection *othercon;
92 	struct connection *sendcon;
93 	struct work_struct rwork; /* Receive workqueue */
94 	struct work_struct swork; /* Send workqueue */
95 	wait_queue_head_t shutdown_wait; /* wait for graceful shutdown */
96 	unsigned char *rx_buf;
97 	int rx_buflen;
98 	int rx_leftover;
99 	struct rcu_head rcu;
100 };
101 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
102 
103 struct listen_connection {
104 	struct socket *sock;
105 	struct work_struct rwork;
106 };
107 
108 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
109 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
110 
111 /* An entry waiting to be sent */
112 struct writequeue_entry {
113 	struct list_head list;
114 	struct page *page;
115 	int offset;
116 	int len;
117 	int end;
118 	int users;
119 	bool dirty;
120 	struct connection *con;
121 	struct list_head msgs;
122 	struct kref ref;
123 };
124 
125 struct dlm_msg {
126 	struct writequeue_entry *entry;
127 	struct dlm_msg *orig_msg;
128 	bool retransmit;
129 	void *ppc;
130 	int len;
131 	int idx; /* new()/commit() idx exchange */
132 
133 	struct list_head list;
134 	struct kref ref;
135 };
136 
137 struct dlm_node_addr {
138 	struct list_head list;
139 	int nodeid;
140 	int mark;
141 	int addr_count;
142 	int curr_addr_index;
143 	struct sockaddr_storage *addr[DLM_MAX_ADDR_COUNT];
144 };
145 
146 struct dlm_proto_ops {
147 	bool try_new_addr;
148 	const char *name;
149 	int proto;
150 
151 	int (*connect)(struct connection *con, struct socket *sock,
152 		       struct sockaddr *addr, int addr_len);
153 	void (*sockopts)(struct socket *sock);
154 	int (*bind)(struct socket *sock);
155 	int (*listen_validate)(void);
156 	void (*listen_sockopts)(struct socket *sock);
157 	int (*listen_bind)(struct socket *sock);
158 	/* What to do to shutdown */
159 	void (*shutdown_action)(struct connection *con);
160 	/* What to do to eof check */
161 	bool (*eof_condition)(struct connection *con);
162 };
163 
164 static struct listen_sock_callbacks {
165 	void (*sk_error_report)(struct sock *);
166 	void (*sk_data_ready)(struct sock *);
167 	void (*sk_state_change)(struct sock *);
168 	void (*sk_write_space)(struct sock *);
169 } listen_sock;
170 
171 static LIST_HEAD(dlm_node_addrs);
172 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
173 
174 static struct listen_connection listen_con;
175 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
176 static int dlm_local_count;
177 int dlm_allow_conn;
178 
179 /* Work queues */
180 static struct workqueue_struct *recv_workqueue;
181 static struct workqueue_struct *send_workqueue;
182 
183 static struct hlist_head connection_hash[CONN_HASH_SIZE];
184 static DEFINE_SPINLOCK(connections_lock);
185 DEFINE_STATIC_SRCU(connections_srcu);
186 
187 static const struct dlm_proto_ops *dlm_proto_ops;
188 
189 static void process_recv_sockets(struct work_struct *work);
190 static void process_send_sockets(struct work_struct *work);
191 
192 /* need to held writequeue_lock */
193 static struct writequeue_entry *con_next_wq(struct connection *con)
194 {
195 	struct writequeue_entry *e;
196 
197 	if (list_empty(&con->writequeue))
198 		return NULL;
199 
200 	e = list_first_entry(&con->writequeue, struct writequeue_entry,
201 			     list);
202 	if (e->len == 0)
203 		return NULL;
204 
205 	return e;
206 }
207 
208 static struct connection *__find_con(int nodeid, int r)
209 {
210 	struct connection *con;
211 
212 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
213 		if (con->nodeid == nodeid)
214 			return con;
215 	}
216 
217 	return NULL;
218 }
219 
220 static bool tcp_eof_condition(struct connection *con)
221 {
222 	return atomic_read(&con->writequeue_cnt);
223 }
224 
225 static int dlm_con_init(struct connection *con, int nodeid)
226 {
227 	con->rx_buflen = dlm_config.ci_buffer_size;
228 	con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
229 	if (!con->rx_buf)
230 		return -ENOMEM;
231 
232 	con->nodeid = nodeid;
233 	mutex_init(&con->sock_mutex);
234 	INIT_LIST_HEAD(&con->writequeue);
235 	spin_lock_init(&con->writequeue_lock);
236 	atomic_set(&con->writequeue_cnt, 0);
237 	INIT_WORK(&con->swork, process_send_sockets);
238 	INIT_WORK(&con->rwork, process_recv_sockets);
239 	init_waitqueue_head(&con->shutdown_wait);
240 
241 	return 0;
242 }
243 
244 /*
245  * If 'allocation' is zero then we don't attempt to create a new
246  * connection structure for this node.
247  */
248 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
249 {
250 	struct connection *con, *tmp;
251 	int r, ret;
252 
253 	r = nodeid_hash(nodeid);
254 	con = __find_con(nodeid, r);
255 	if (con || !alloc)
256 		return con;
257 
258 	con = kzalloc(sizeof(*con), alloc);
259 	if (!con)
260 		return NULL;
261 
262 	ret = dlm_con_init(con, nodeid);
263 	if (ret) {
264 		kfree(con);
265 		return NULL;
266 	}
267 
268 	mutex_init(&con->wq_alloc);
269 
270 	spin_lock(&connections_lock);
271 	/* Because multiple workqueues/threads calls this function it can
272 	 * race on multiple cpu's. Instead of locking hot path __find_con()
273 	 * we just check in rare cases of recently added nodes again
274 	 * under protection of connections_lock. If this is the case we
275 	 * abort our connection creation and return the existing connection.
276 	 */
277 	tmp = __find_con(nodeid, r);
278 	if (tmp) {
279 		spin_unlock(&connections_lock);
280 		kfree(con->rx_buf);
281 		kfree(con);
282 		return tmp;
283 	}
284 
285 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
286 	spin_unlock(&connections_lock);
287 
288 	return con;
289 }
290 
291 /* Loop round all connections */
292 static void foreach_conn(void (*conn_func)(struct connection *c))
293 {
294 	int i;
295 	struct connection *con;
296 
297 	for (i = 0; i < CONN_HASH_SIZE; i++) {
298 		hlist_for_each_entry_rcu(con, &connection_hash[i], list)
299 			conn_func(con);
300 	}
301 }
302 
303 static struct dlm_node_addr *find_node_addr(int nodeid)
304 {
305 	struct dlm_node_addr *na;
306 
307 	list_for_each_entry(na, &dlm_node_addrs, list) {
308 		if (na->nodeid == nodeid)
309 			return na;
310 	}
311 	return NULL;
312 }
313 
314 static int addr_compare(const struct sockaddr_storage *x,
315 			const struct sockaddr_storage *y)
316 {
317 	switch (x->ss_family) {
318 	case AF_INET: {
319 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
320 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
321 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
322 			return 0;
323 		if (sinx->sin_port != siny->sin_port)
324 			return 0;
325 		break;
326 	}
327 	case AF_INET6: {
328 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
329 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
330 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
331 			return 0;
332 		if (sinx->sin6_port != siny->sin6_port)
333 			return 0;
334 		break;
335 	}
336 	default:
337 		return 0;
338 	}
339 	return 1;
340 }
341 
342 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
343 			  struct sockaddr *sa_out, bool try_new_addr,
344 			  unsigned int *mark)
345 {
346 	struct sockaddr_storage sas;
347 	struct dlm_node_addr *na;
348 
349 	if (!dlm_local_count)
350 		return -1;
351 
352 	spin_lock(&dlm_node_addrs_spin);
353 	na = find_node_addr(nodeid);
354 	if (na && na->addr_count) {
355 		memcpy(&sas, na->addr[na->curr_addr_index],
356 		       sizeof(struct sockaddr_storage));
357 
358 		if (try_new_addr) {
359 			na->curr_addr_index++;
360 			if (na->curr_addr_index == na->addr_count)
361 				na->curr_addr_index = 0;
362 		}
363 	}
364 	spin_unlock(&dlm_node_addrs_spin);
365 
366 	if (!na)
367 		return -EEXIST;
368 
369 	if (!na->addr_count)
370 		return -ENOENT;
371 
372 	*mark = na->mark;
373 
374 	if (sas_out)
375 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
376 
377 	if (!sa_out)
378 		return 0;
379 
380 	if (dlm_local_addr[0]->ss_family == AF_INET) {
381 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
382 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
383 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
384 	} else {
385 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
386 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
387 		ret6->sin6_addr = in6->sin6_addr;
388 	}
389 
390 	return 0;
391 }
392 
393 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
394 			  unsigned int *mark)
395 {
396 	struct dlm_node_addr *na;
397 	int rv = -EEXIST;
398 	int addr_i;
399 
400 	spin_lock(&dlm_node_addrs_spin);
401 	list_for_each_entry(na, &dlm_node_addrs, list) {
402 		if (!na->addr_count)
403 			continue;
404 
405 		for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
406 			if (addr_compare(na->addr[addr_i], addr)) {
407 				*nodeid = na->nodeid;
408 				*mark = na->mark;
409 				rv = 0;
410 				goto unlock;
411 			}
412 		}
413 	}
414 unlock:
415 	spin_unlock(&dlm_node_addrs_spin);
416 	return rv;
417 }
418 
419 /* caller need to held dlm_node_addrs_spin lock */
420 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
421 				     const struct sockaddr_storage *addr)
422 {
423 	int i;
424 
425 	for (i = 0; i < na->addr_count; i++) {
426 		if (addr_compare(na->addr[i], addr))
427 			return true;
428 	}
429 
430 	return false;
431 }
432 
433 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
434 {
435 	struct sockaddr_storage *new_addr;
436 	struct dlm_node_addr *new_node, *na;
437 	bool ret;
438 
439 	new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
440 	if (!new_node)
441 		return -ENOMEM;
442 
443 	new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
444 	if (!new_addr) {
445 		kfree(new_node);
446 		return -ENOMEM;
447 	}
448 
449 	memcpy(new_addr, addr, len);
450 
451 	spin_lock(&dlm_node_addrs_spin);
452 	na = find_node_addr(nodeid);
453 	if (!na) {
454 		new_node->nodeid = nodeid;
455 		new_node->addr[0] = new_addr;
456 		new_node->addr_count = 1;
457 		new_node->mark = dlm_config.ci_mark;
458 		list_add(&new_node->list, &dlm_node_addrs);
459 		spin_unlock(&dlm_node_addrs_spin);
460 		return 0;
461 	}
462 
463 	ret = dlm_lowcomms_na_has_addr(na, addr);
464 	if (ret) {
465 		spin_unlock(&dlm_node_addrs_spin);
466 		kfree(new_addr);
467 		kfree(new_node);
468 		return -EEXIST;
469 	}
470 
471 	if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
472 		spin_unlock(&dlm_node_addrs_spin);
473 		kfree(new_addr);
474 		kfree(new_node);
475 		return -ENOSPC;
476 	}
477 
478 	na->addr[na->addr_count++] = new_addr;
479 	spin_unlock(&dlm_node_addrs_spin);
480 	kfree(new_node);
481 	return 0;
482 }
483 
484 /* Data available on socket or listen socket received a connect */
485 static void lowcomms_data_ready(struct sock *sk)
486 {
487 	struct connection *con;
488 
489 	read_lock_bh(&sk->sk_callback_lock);
490 	con = sock2con(sk);
491 	if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
492 		queue_work(recv_workqueue, &con->rwork);
493 	read_unlock_bh(&sk->sk_callback_lock);
494 }
495 
496 static void lowcomms_listen_data_ready(struct sock *sk)
497 {
498 	if (!dlm_allow_conn)
499 		return;
500 
501 	queue_work(recv_workqueue, &listen_con.rwork);
502 }
503 
504 static void lowcomms_write_space(struct sock *sk)
505 {
506 	struct connection *con;
507 
508 	read_lock_bh(&sk->sk_callback_lock);
509 	con = sock2con(sk);
510 	if (!con)
511 		goto out;
512 
513 	if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
514 		log_print("successful connected to node %d", con->nodeid);
515 		queue_work(send_workqueue, &con->swork);
516 		goto out;
517 	}
518 
519 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
520 
521 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
522 		con->sock->sk->sk_write_pending--;
523 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
524 	}
525 
526 	queue_work(send_workqueue, &con->swork);
527 out:
528 	read_unlock_bh(&sk->sk_callback_lock);
529 }
530 
531 static inline void lowcomms_connect_sock(struct connection *con)
532 {
533 	if (test_bit(CF_CLOSE, &con->flags))
534 		return;
535 	queue_work(send_workqueue, &con->swork);
536 	cond_resched();
537 }
538 
539 static void lowcomms_state_change(struct sock *sk)
540 {
541 	/* SCTP layer is not calling sk_data_ready when the connection
542 	 * is done, so we catch the signal through here. Also, it
543 	 * doesn't switch socket state when entering shutdown, so we
544 	 * skip the write in that case.
545 	 */
546 	if (sk->sk_shutdown) {
547 		if (sk->sk_shutdown == RCV_SHUTDOWN)
548 			lowcomms_data_ready(sk);
549 	} else if (sk->sk_state == TCP_ESTABLISHED) {
550 		lowcomms_write_space(sk);
551 	}
552 }
553 
554 int dlm_lowcomms_connect_node(int nodeid)
555 {
556 	struct connection *con;
557 	int idx;
558 
559 	if (nodeid == dlm_our_nodeid())
560 		return 0;
561 
562 	idx = srcu_read_lock(&connections_srcu);
563 	con = nodeid2con(nodeid, GFP_NOFS);
564 	if (!con) {
565 		srcu_read_unlock(&connections_srcu, idx);
566 		return -ENOMEM;
567 	}
568 
569 	lowcomms_connect_sock(con);
570 	srcu_read_unlock(&connections_srcu, idx);
571 
572 	return 0;
573 }
574 
575 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
576 {
577 	struct dlm_node_addr *na;
578 
579 	spin_lock(&dlm_node_addrs_spin);
580 	na = find_node_addr(nodeid);
581 	if (!na) {
582 		spin_unlock(&dlm_node_addrs_spin);
583 		return -ENOENT;
584 	}
585 
586 	na->mark = mark;
587 	spin_unlock(&dlm_node_addrs_spin);
588 
589 	return 0;
590 }
591 
592 static void lowcomms_error_report(struct sock *sk)
593 {
594 	struct connection *con;
595 	struct sockaddr_storage saddr;
596 	void (*orig_report)(struct sock *) = NULL;
597 
598 	read_lock_bh(&sk->sk_callback_lock);
599 	con = sock2con(sk);
600 	if (con == NULL)
601 		goto out;
602 
603 	orig_report = listen_sock.sk_error_report;
604 	if (kernel_getpeername(sk->sk_socket, (struct sockaddr *)&saddr) < 0) {
605 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
606 				   "sending to node %d, port %d, "
607 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
608 				   con->nodeid, dlm_config.ci_tcp_port,
609 				   sk->sk_err, sk->sk_err_soft);
610 	} else if (saddr.ss_family == AF_INET) {
611 		struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
612 
613 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
614 				   "sending to node %d at %pI4, port %d, "
615 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
616 				   con->nodeid, &sin4->sin_addr.s_addr,
617 				   dlm_config.ci_tcp_port, sk->sk_err,
618 				   sk->sk_err_soft);
619 	} else {
620 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
621 
622 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
623 				   "sending to node %d at %u.%u.%u.%u, "
624 				   "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
625 				   con->nodeid, sin6->sin6_addr.s6_addr32[0],
626 				   sin6->sin6_addr.s6_addr32[1],
627 				   sin6->sin6_addr.s6_addr32[2],
628 				   sin6->sin6_addr.s6_addr32[3],
629 				   dlm_config.ci_tcp_port, sk->sk_err,
630 				   sk->sk_err_soft);
631 	}
632 
633 	/* below sendcon only handling */
634 	if (test_bit(CF_IS_OTHERCON, &con->flags))
635 		con = con->sendcon;
636 
637 	switch (sk->sk_err) {
638 	case ECONNREFUSED:
639 		set_bit(CF_DELAY_CONNECT, &con->flags);
640 		break;
641 	default:
642 		break;
643 	}
644 
645 	if (!test_and_set_bit(CF_RECONNECT, &con->flags))
646 		queue_work(send_workqueue, &con->swork);
647 
648 out:
649 	read_unlock_bh(&sk->sk_callback_lock);
650 	if (orig_report)
651 		orig_report(sk);
652 }
653 
654 /* Note: sk_callback_lock must be locked before calling this function. */
655 static void save_listen_callbacks(struct socket *sock)
656 {
657 	struct sock *sk = sock->sk;
658 
659 	listen_sock.sk_data_ready = sk->sk_data_ready;
660 	listen_sock.sk_state_change = sk->sk_state_change;
661 	listen_sock.sk_write_space = sk->sk_write_space;
662 	listen_sock.sk_error_report = sk->sk_error_report;
663 }
664 
665 static void restore_callbacks(struct socket *sock)
666 {
667 	struct sock *sk = sock->sk;
668 
669 	write_lock_bh(&sk->sk_callback_lock);
670 	sk->sk_user_data = NULL;
671 	sk->sk_data_ready = listen_sock.sk_data_ready;
672 	sk->sk_state_change = listen_sock.sk_state_change;
673 	sk->sk_write_space = listen_sock.sk_write_space;
674 	sk->sk_error_report = listen_sock.sk_error_report;
675 	write_unlock_bh(&sk->sk_callback_lock);
676 }
677 
678 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
679 {
680 	struct sock *sk = sock->sk;
681 
682 	write_lock_bh(&sk->sk_callback_lock);
683 	save_listen_callbacks(sock);
684 	con->sock = sock;
685 
686 	sk->sk_user_data = con;
687 	sk->sk_allocation = GFP_NOFS;
688 	/* Install a data_ready callback */
689 	sk->sk_data_ready = lowcomms_listen_data_ready;
690 	write_unlock_bh(&sk->sk_callback_lock);
691 }
692 
693 /* Make a socket active */
694 static void add_sock(struct socket *sock, struct connection *con)
695 {
696 	struct sock *sk = sock->sk;
697 
698 	write_lock_bh(&sk->sk_callback_lock);
699 	con->sock = sock;
700 
701 	sk->sk_user_data = con;
702 	/* Install a data_ready callback */
703 	sk->sk_data_ready = lowcomms_data_ready;
704 	sk->sk_write_space = lowcomms_write_space;
705 	sk->sk_state_change = lowcomms_state_change;
706 	sk->sk_allocation = GFP_NOFS;
707 	sk->sk_error_report = lowcomms_error_report;
708 	write_unlock_bh(&sk->sk_callback_lock);
709 }
710 
711 /* Add the port number to an IPv6 or 4 sockaddr and return the address
712    length */
713 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
714 			  int *addr_len)
715 {
716 	saddr->ss_family =  dlm_local_addr[0]->ss_family;
717 	if (saddr->ss_family == AF_INET) {
718 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
719 		in4_addr->sin_port = cpu_to_be16(port);
720 		*addr_len = sizeof(struct sockaddr_in);
721 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
722 	} else {
723 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
724 		in6_addr->sin6_port = cpu_to_be16(port);
725 		*addr_len = sizeof(struct sockaddr_in6);
726 	}
727 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
728 }
729 
730 static void dlm_page_release(struct kref *kref)
731 {
732 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
733 						  ref);
734 
735 	__free_page(e->page);
736 	kfree(e);
737 }
738 
739 static void dlm_msg_release(struct kref *kref)
740 {
741 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
742 
743 	kref_put(&msg->entry->ref, dlm_page_release);
744 	kfree(msg);
745 }
746 
747 static void free_entry(struct writequeue_entry *e)
748 {
749 	struct dlm_msg *msg, *tmp;
750 
751 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
752 		if (msg->orig_msg) {
753 			msg->orig_msg->retransmit = false;
754 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
755 		}
756 
757 		list_del(&msg->list);
758 		kref_put(&msg->ref, dlm_msg_release);
759 	}
760 
761 	list_del(&e->list);
762 	atomic_dec(&e->con->writequeue_cnt);
763 	kref_put(&e->ref, dlm_page_release);
764 }
765 
766 static void dlm_close_sock(struct socket **sock)
767 {
768 	if (*sock) {
769 		restore_callbacks(*sock);
770 		sock_release(*sock);
771 		*sock = NULL;
772 	}
773 }
774 
775 /* Close a remote connection and tidy up */
776 static void close_connection(struct connection *con, bool and_other,
777 			     bool tx, bool rx)
778 {
779 	bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
780 	struct writequeue_entry *e;
781 
782 	if (tx && !closing && cancel_work_sync(&con->swork)) {
783 		log_print("canceled swork for node %d", con->nodeid);
784 		clear_bit(CF_WRITE_PENDING, &con->flags);
785 	}
786 	if (rx && !closing && cancel_work_sync(&con->rwork)) {
787 		log_print("canceled rwork for node %d", con->nodeid);
788 		clear_bit(CF_READ_PENDING, &con->flags);
789 	}
790 
791 	mutex_lock(&con->sock_mutex);
792 	dlm_close_sock(&con->sock);
793 
794 	if (con->othercon && and_other) {
795 		/* Will only re-enter once. */
796 		close_connection(con->othercon, false, tx, rx);
797 	}
798 
799 	/* if we send a writequeue entry only a half way, we drop the
800 	 * whole entry because reconnection and that we not start of the
801 	 * middle of a msg which will confuse the other end.
802 	 *
803 	 * we can always drop messages because retransmits, but what we
804 	 * cannot allow is to transmit half messages which may be processed
805 	 * at the other side.
806 	 *
807 	 * our policy is to start on a clean state when disconnects, we don't
808 	 * know what's send/received on transport layer in this case.
809 	 */
810 	spin_lock(&con->writequeue_lock);
811 	if (!list_empty(&con->writequeue)) {
812 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
813 				     list);
814 		if (e->dirty)
815 			free_entry(e);
816 	}
817 	spin_unlock(&con->writequeue_lock);
818 
819 	con->rx_leftover = 0;
820 	con->retries = 0;
821 	clear_bit(CF_APP_LIMITED, &con->flags);
822 	clear_bit(CF_CONNECTED, &con->flags);
823 	clear_bit(CF_DELAY_CONNECT, &con->flags);
824 	clear_bit(CF_RECONNECT, &con->flags);
825 	clear_bit(CF_EOF, &con->flags);
826 	mutex_unlock(&con->sock_mutex);
827 	clear_bit(CF_CLOSING, &con->flags);
828 }
829 
830 static void shutdown_connection(struct connection *con)
831 {
832 	int ret;
833 
834 	flush_work(&con->swork);
835 
836 	mutex_lock(&con->sock_mutex);
837 	/* nothing to shutdown */
838 	if (!con->sock) {
839 		mutex_unlock(&con->sock_mutex);
840 		return;
841 	}
842 
843 	set_bit(CF_SHUTDOWN, &con->flags);
844 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
845 	mutex_unlock(&con->sock_mutex);
846 	if (ret) {
847 		log_print("Connection %p failed to shutdown: %d will force close",
848 			  con, ret);
849 		goto force_close;
850 	} else {
851 		ret = wait_event_timeout(con->shutdown_wait,
852 					 !test_bit(CF_SHUTDOWN, &con->flags),
853 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
854 		if (ret == 0) {
855 			log_print("Connection %p shutdown timed out, will force close",
856 				  con);
857 			goto force_close;
858 		}
859 	}
860 
861 	return;
862 
863 force_close:
864 	clear_bit(CF_SHUTDOWN, &con->flags);
865 	close_connection(con, false, true, true);
866 }
867 
868 static void dlm_tcp_shutdown(struct connection *con)
869 {
870 	if (con->othercon)
871 		shutdown_connection(con->othercon);
872 	shutdown_connection(con);
873 }
874 
875 static int con_realloc_receive_buf(struct connection *con, int newlen)
876 {
877 	unsigned char *newbuf;
878 
879 	newbuf = kmalloc(newlen, GFP_NOFS);
880 	if (!newbuf)
881 		return -ENOMEM;
882 
883 	/* copy any leftover from last receive */
884 	if (con->rx_leftover)
885 		memmove(newbuf, con->rx_buf, con->rx_leftover);
886 
887 	/* swap to new buffer space */
888 	kfree(con->rx_buf);
889 	con->rx_buflen = newlen;
890 	con->rx_buf = newbuf;
891 
892 	return 0;
893 }
894 
895 /* Data received from remote end */
896 static int receive_from_sock(struct connection *con)
897 {
898 	struct msghdr msg;
899 	struct kvec iov;
900 	int ret, buflen;
901 
902 	mutex_lock(&con->sock_mutex);
903 
904 	if (con->sock == NULL) {
905 		ret = -EAGAIN;
906 		goto out_close;
907 	}
908 
909 	/* realloc if we get new buffer size to read out */
910 	buflen = dlm_config.ci_buffer_size;
911 	if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
912 		ret = con_realloc_receive_buf(con, buflen);
913 		if (ret < 0)
914 			goto out_resched;
915 	}
916 
917 	for (;;) {
918 		/* calculate new buffer parameter regarding last receive and
919 		 * possible leftover bytes
920 		 */
921 		iov.iov_base = con->rx_buf + con->rx_leftover;
922 		iov.iov_len = con->rx_buflen - con->rx_leftover;
923 
924 		memset(&msg, 0, sizeof(msg));
925 		msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
926 		ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
927 				     msg.msg_flags);
928 		if (ret == -EAGAIN)
929 			break;
930 		else if (ret <= 0)
931 			goto out_close;
932 
933 		/* new buflen according readed bytes and leftover from last receive */
934 		buflen = ret + con->rx_leftover;
935 		ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
936 		if (ret < 0)
937 			goto out_close;
938 
939 		/* calculate leftover bytes from process and put it into begin of
940 		 * the receive buffer, so next receive we have the full message
941 		 * at the start address of the receive buffer.
942 		 */
943 		con->rx_leftover = buflen - ret;
944 		if (con->rx_leftover) {
945 			memmove(con->rx_buf, con->rx_buf + ret,
946 				con->rx_leftover);
947 		}
948 	}
949 
950 	dlm_midcomms_receive_done(con->nodeid);
951 	mutex_unlock(&con->sock_mutex);
952 	return 0;
953 
954 out_resched:
955 	if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
956 		queue_work(recv_workqueue, &con->rwork);
957 	mutex_unlock(&con->sock_mutex);
958 	return -EAGAIN;
959 
960 out_close:
961 	if (ret == 0) {
962 		log_print("connection %p got EOF from %d",
963 			  con, con->nodeid);
964 
965 		if (dlm_proto_ops->eof_condition &&
966 		    dlm_proto_ops->eof_condition(con)) {
967 			set_bit(CF_EOF, &con->flags);
968 			mutex_unlock(&con->sock_mutex);
969 		} else {
970 			mutex_unlock(&con->sock_mutex);
971 			close_connection(con, false, true, false);
972 
973 			/* handling for tcp shutdown */
974 			clear_bit(CF_SHUTDOWN, &con->flags);
975 			wake_up(&con->shutdown_wait);
976 		}
977 
978 		/* signal to breaking receive worker */
979 		ret = -1;
980 	} else {
981 		mutex_unlock(&con->sock_mutex);
982 	}
983 	return ret;
984 }
985 
986 /* Listening socket is busy, accept a connection */
987 static int accept_from_sock(struct listen_connection *con)
988 {
989 	int result;
990 	struct sockaddr_storage peeraddr;
991 	struct socket *newsock;
992 	int len, idx;
993 	int nodeid;
994 	struct connection *newcon;
995 	struct connection *addcon;
996 	unsigned int mark;
997 
998 	if (!con->sock)
999 		return -ENOTCONN;
1000 
1001 	result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
1002 	if (result < 0)
1003 		goto accept_err;
1004 
1005 	/* Get the connected socket's peer */
1006 	memset(&peeraddr, 0, sizeof(peeraddr));
1007 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1008 	if (len < 0) {
1009 		result = -ECONNABORTED;
1010 		goto accept_err;
1011 	}
1012 
1013 	/* Get the new node's NODEID */
1014 	make_sockaddr(&peeraddr, 0, &len);
1015 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1016 		unsigned char *b=(unsigned char *)&peeraddr;
1017 		log_print("connect from non cluster node");
1018 		print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
1019 				     b, sizeof(struct sockaddr_storage));
1020 		sock_release(newsock);
1021 		return -1;
1022 	}
1023 
1024 	log_print("got connection from %d", nodeid);
1025 
1026 	/*  Check to see if we already have a connection to this node. This
1027 	 *  could happen if the two nodes initiate a connection at roughly
1028 	 *  the same time and the connections cross on the wire.
1029 	 *  In this case we store the incoming one in "othercon"
1030 	 */
1031 	idx = srcu_read_lock(&connections_srcu);
1032 	newcon = nodeid2con(nodeid, GFP_NOFS);
1033 	if (!newcon) {
1034 		srcu_read_unlock(&connections_srcu, idx);
1035 		result = -ENOMEM;
1036 		goto accept_err;
1037 	}
1038 
1039 	sock_set_mark(newsock->sk, mark);
1040 
1041 	mutex_lock(&newcon->sock_mutex);
1042 	if (newcon->sock) {
1043 		struct connection *othercon = newcon->othercon;
1044 
1045 		if (!othercon) {
1046 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1047 			if (!othercon) {
1048 				log_print("failed to allocate incoming socket");
1049 				mutex_unlock(&newcon->sock_mutex);
1050 				srcu_read_unlock(&connections_srcu, idx);
1051 				result = -ENOMEM;
1052 				goto accept_err;
1053 			}
1054 
1055 			result = dlm_con_init(othercon, nodeid);
1056 			if (result < 0) {
1057 				kfree(othercon);
1058 				mutex_unlock(&newcon->sock_mutex);
1059 				srcu_read_unlock(&connections_srcu, idx);
1060 				goto accept_err;
1061 			}
1062 
1063 			lockdep_set_subclass(&othercon->sock_mutex, 1);
1064 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1065 			newcon->othercon = othercon;
1066 			othercon->sendcon = newcon;
1067 		} else {
1068 			/* close other sock con if we have something new */
1069 			close_connection(othercon, false, true, false);
1070 		}
1071 
1072 		mutex_lock(&othercon->sock_mutex);
1073 		add_sock(newsock, othercon);
1074 		addcon = othercon;
1075 		mutex_unlock(&othercon->sock_mutex);
1076 	}
1077 	else {
1078 		/* accept copies the sk after we've saved the callbacks, so we
1079 		   don't want to save them a second time or comm errors will
1080 		   result in calling sk_error_report recursively. */
1081 		add_sock(newsock, newcon);
1082 		addcon = newcon;
1083 	}
1084 
1085 	set_bit(CF_CONNECTED, &addcon->flags);
1086 	mutex_unlock(&newcon->sock_mutex);
1087 
1088 	/*
1089 	 * Add it to the active queue in case we got data
1090 	 * between processing the accept adding the socket
1091 	 * to the read_sockets list
1092 	 */
1093 	if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1094 		queue_work(recv_workqueue, &addcon->rwork);
1095 
1096 	srcu_read_unlock(&connections_srcu, idx);
1097 
1098 	return 0;
1099 
1100 accept_err:
1101 	if (newsock)
1102 		sock_release(newsock);
1103 
1104 	if (result != -EAGAIN)
1105 		log_print("error accepting connection from node: %d", result);
1106 	return result;
1107 }
1108 
1109 /*
1110  * writequeue_entry_complete - try to delete and free write queue entry
1111  * @e: write queue entry to try to delete
1112  * @completed: bytes completed
1113  *
1114  * writequeue_lock must be held.
1115  */
1116 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1117 {
1118 	e->offset += completed;
1119 	e->len -= completed;
1120 	/* signal that page was half way transmitted */
1121 	e->dirty = true;
1122 
1123 	if (e->len == 0 && e->users == 0)
1124 		free_entry(e);
1125 }
1126 
1127 /*
1128  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1129  */
1130 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1131 {
1132 	struct sockaddr_storage localaddr;
1133 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1134 	int i, addr_len, result = 0;
1135 
1136 	for (i = 0; i < dlm_local_count; i++) {
1137 		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1138 		make_sockaddr(&localaddr, port, &addr_len);
1139 
1140 		if (!i)
1141 			result = kernel_bind(sock, addr, addr_len);
1142 		else
1143 			result = sock_bind_add(sock->sk, addr, addr_len);
1144 
1145 		if (result < 0) {
1146 			log_print("Can't bind to %d addr number %d, %d.\n",
1147 				  port, i + 1, result);
1148 			break;
1149 		}
1150 	}
1151 	return result;
1152 }
1153 
1154 /* Get local addresses */
1155 static void init_local(void)
1156 {
1157 	struct sockaddr_storage sas, *addr;
1158 	int i;
1159 
1160 	dlm_local_count = 0;
1161 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1162 		if (dlm_our_addr(&sas, i))
1163 			break;
1164 
1165 		addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1166 		if (!addr)
1167 			break;
1168 		dlm_local_addr[dlm_local_count++] = addr;
1169 	}
1170 }
1171 
1172 static void deinit_local(void)
1173 {
1174 	int i;
1175 
1176 	for (i = 0; i < dlm_local_count; i++)
1177 		kfree(dlm_local_addr[i]);
1178 }
1179 
1180 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1181 						     gfp_t allocation)
1182 {
1183 	struct writequeue_entry *entry;
1184 
1185 	entry = kzalloc(sizeof(*entry), allocation);
1186 	if (!entry)
1187 		return NULL;
1188 
1189 	entry->page = alloc_page(allocation | __GFP_ZERO);
1190 	if (!entry->page) {
1191 		kfree(entry);
1192 		return NULL;
1193 	}
1194 
1195 	entry->con = con;
1196 	entry->users = 1;
1197 	kref_init(&entry->ref);
1198 	INIT_LIST_HEAD(&entry->msgs);
1199 
1200 	return entry;
1201 }
1202 
1203 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1204 					     gfp_t allocation, char **ppc,
1205 					     void (*cb)(struct dlm_mhandle *mh),
1206 					     struct dlm_mhandle *mh)
1207 {
1208 	struct writequeue_entry *e;
1209 
1210 	spin_lock(&con->writequeue_lock);
1211 	if (!list_empty(&con->writequeue)) {
1212 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1213 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1214 			kref_get(&e->ref);
1215 
1216 			*ppc = page_address(e->page) + e->end;
1217 			if (cb)
1218 				cb(mh);
1219 
1220 			e->end += len;
1221 			e->users++;
1222 			spin_unlock(&con->writequeue_lock);
1223 
1224 			return e;
1225 		}
1226 	}
1227 	spin_unlock(&con->writequeue_lock);
1228 
1229 	e = new_writequeue_entry(con, allocation);
1230 	if (!e)
1231 		return NULL;
1232 
1233 	kref_get(&e->ref);
1234 	*ppc = page_address(e->page);
1235 	e->end += len;
1236 	atomic_inc(&con->writequeue_cnt);
1237 
1238 	spin_lock(&con->writequeue_lock);
1239 	if (cb)
1240 		cb(mh);
1241 
1242 	list_add_tail(&e->list, &con->writequeue);
1243 	spin_unlock(&con->writequeue_lock);
1244 
1245 	return e;
1246 };
1247 
1248 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1249 						gfp_t allocation, char **ppc,
1250 						void (*cb)(struct dlm_mhandle *mh),
1251 						struct dlm_mhandle *mh)
1252 {
1253 	struct writequeue_entry *e;
1254 	struct dlm_msg *msg;
1255 	bool sleepable;
1256 
1257 	msg = kzalloc(sizeof(*msg), allocation);
1258 	if (!msg)
1259 		return NULL;
1260 
1261 	/* this mutex is being used as a wait to avoid multiple "fast"
1262 	 * new writequeue page list entry allocs in new_wq_entry in
1263 	 * normal operation which is sleepable context. Without it
1264 	 * we could end in multiple writequeue entries with one
1265 	 * dlm message because multiple callers were waiting at
1266 	 * the writequeue_lock in new_wq_entry().
1267 	 */
1268 	sleepable = gfpflags_normal_context(allocation);
1269 	if (sleepable)
1270 		mutex_lock(&con->wq_alloc);
1271 
1272 	kref_init(&msg->ref);
1273 
1274 	e = new_wq_entry(con, len, allocation, ppc, cb, mh);
1275 	if (!e) {
1276 		if (sleepable)
1277 			mutex_unlock(&con->wq_alloc);
1278 
1279 		kfree(msg);
1280 		return NULL;
1281 	}
1282 
1283 	if (sleepable)
1284 		mutex_unlock(&con->wq_alloc);
1285 
1286 	msg->ppc = *ppc;
1287 	msg->len = len;
1288 	msg->entry = e;
1289 
1290 	return msg;
1291 }
1292 
1293 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1294 				     char **ppc, void (*cb)(struct dlm_mhandle *mh),
1295 				     struct dlm_mhandle *mh)
1296 {
1297 	struct connection *con;
1298 	struct dlm_msg *msg;
1299 	int idx;
1300 
1301 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1302 	    len < sizeof(struct dlm_header)) {
1303 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1304 		log_print("failed to allocate a buffer of size %d", len);
1305 		WARN_ON(1);
1306 		return NULL;
1307 	}
1308 
1309 	idx = srcu_read_lock(&connections_srcu);
1310 	con = nodeid2con(nodeid, allocation);
1311 	if (!con) {
1312 		srcu_read_unlock(&connections_srcu, idx);
1313 		return NULL;
1314 	}
1315 
1316 	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, mh);
1317 	if (!msg) {
1318 		srcu_read_unlock(&connections_srcu, idx);
1319 		return NULL;
1320 	}
1321 
1322 	/* we assume if successful commit must called */
1323 	msg->idx = idx;
1324 	return msg;
1325 }
1326 
1327 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1328 {
1329 	struct writequeue_entry *e = msg->entry;
1330 	struct connection *con = e->con;
1331 	int users;
1332 
1333 	spin_lock(&con->writequeue_lock);
1334 	kref_get(&msg->ref);
1335 	list_add(&msg->list, &e->msgs);
1336 
1337 	users = --e->users;
1338 	if (users)
1339 		goto out;
1340 
1341 	e->len = DLM_WQ_LENGTH_BYTES(e);
1342 	spin_unlock(&con->writequeue_lock);
1343 
1344 	queue_work(send_workqueue, &con->swork);
1345 	return;
1346 
1347 out:
1348 	spin_unlock(&con->writequeue_lock);
1349 	return;
1350 }
1351 
1352 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1353 {
1354 	_dlm_lowcomms_commit_msg(msg);
1355 	srcu_read_unlock(&connections_srcu, msg->idx);
1356 }
1357 
1358 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1359 {
1360 	kref_put(&msg->ref, dlm_msg_release);
1361 }
1362 
1363 /* does not held connections_srcu, usage workqueue only */
1364 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1365 {
1366 	struct dlm_msg *msg_resend;
1367 	char *ppc;
1368 
1369 	if (msg->retransmit)
1370 		return 1;
1371 
1372 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1373 					      GFP_ATOMIC, &ppc, NULL, NULL);
1374 	if (!msg_resend)
1375 		return -ENOMEM;
1376 
1377 	msg->retransmit = true;
1378 	kref_get(&msg->ref);
1379 	msg_resend->orig_msg = msg;
1380 
1381 	memcpy(ppc, msg->ppc, msg->len);
1382 	_dlm_lowcomms_commit_msg(msg_resend);
1383 	dlm_lowcomms_put_msg(msg_resend);
1384 
1385 	return 0;
1386 }
1387 
1388 /* Send a message */
1389 static void send_to_sock(struct connection *con)
1390 {
1391 	const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1392 	struct writequeue_entry *e;
1393 	int len, offset, ret;
1394 	int count = 0;
1395 
1396 	mutex_lock(&con->sock_mutex);
1397 	if (con->sock == NULL)
1398 		goto out_connect;
1399 
1400 	spin_lock(&con->writequeue_lock);
1401 	for (;;) {
1402 		e = con_next_wq(con);
1403 		if (!e)
1404 			break;
1405 
1406 		e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1407 		len = e->len;
1408 		offset = e->offset;
1409 		BUG_ON(len == 0 && e->users == 0);
1410 		spin_unlock(&con->writequeue_lock);
1411 
1412 		ret = kernel_sendpage(con->sock, e->page, offset, len,
1413 				      msg_flags);
1414 		if (ret == -EAGAIN || ret == 0) {
1415 			if (ret == -EAGAIN &&
1416 			    test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1417 			    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1418 				/* Notify TCP that we're limited by the
1419 				 * application window size.
1420 				 */
1421 				set_bit(SOCK_NOSPACE, &con->sock->flags);
1422 				con->sock->sk->sk_write_pending++;
1423 			}
1424 			cond_resched();
1425 			goto out;
1426 		} else if (ret < 0)
1427 			goto out;
1428 
1429 		/* Don't starve people filling buffers */
1430 		if (++count >= MAX_SEND_MSG_COUNT) {
1431 			cond_resched();
1432 			count = 0;
1433 		}
1434 
1435 		spin_lock(&con->writequeue_lock);
1436 		writequeue_entry_complete(e, ret);
1437 	}
1438 	spin_unlock(&con->writequeue_lock);
1439 
1440 	/* close if we got EOF */
1441 	if (test_and_clear_bit(CF_EOF, &con->flags)) {
1442 		mutex_unlock(&con->sock_mutex);
1443 		close_connection(con, false, false, true);
1444 
1445 		/* handling for tcp shutdown */
1446 		clear_bit(CF_SHUTDOWN, &con->flags);
1447 		wake_up(&con->shutdown_wait);
1448 	} else {
1449 		mutex_unlock(&con->sock_mutex);
1450 	}
1451 
1452 	return;
1453 
1454 out:
1455 	mutex_unlock(&con->sock_mutex);
1456 	return;
1457 
1458 out_connect:
1459 	mutex_unlock(&con->sock_mutex);
1460 	queue_work(send_workqueue, &con->swork);
1461 	cond_resched();
1462 }
1463 
1464 static void clean_one_writequeue(struct connection *con)
1465 {
1466 	struct writequeue_entry *e, *safe;
1467 
1468 	spin_lock(&con->writequeue_lock);
1469 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1470 		free_entry(e);
1471 	}
1472 	spin_unlock(&con->writequeue_lock);
1473 }
1474 
1475 /* Called from recovery when it knows that a node has
1476    left the cluster */
1477 int dlm_lowcomms_close(int nodeid)
1478 {
1479 	struct connection *con;
1480 	struct dlm_node_addr *na;
1481 	int idx;
1482 
1483 	log_print("closing connection to node %d", nodeid);
1484 	idx = srcu_read_lock(&connections_srcu);
1485 	con = nodeid2con(nodeid, 0);
1486 	if (con) {
1487 		set_bit(CF_CLOSE, &con->flags);
1488 		close_connection(con, true, true, true);
1489 		clean_one_writequeue(con);
1490 		if (con->othercon)
1491 			clean_one_writequeue(con->othercon);
1492 	}
1493 	srcu_read_unlock(&connections_srcu, idx);
1494 
1495 	spin_lock(&dlm_node_addrs_spin);
1496 	na = find_node_addr(nodeid);
1497 	if (na) {
1498 		list_del(&na->list);
1499 		while (na->addr_count--)
1500 			kfree(na->addr[na->addr_count]);
1501 		kfree(na);
1502 	}
1503 	spin_unlock(&dlm_node_addrs_spin);
1504 
1505 	return 0;
1506 }
1507 
1508 /* Receive workqueue function */
1509 static void process_recv_sockets(struct work_struct *work)
1510 {
1511 	struct connection *con = container_of(work, struct connection, rwork);
1512 
1513 	clear_bit(CF_READ_PENDING, &con->flags);
1514 	receive_from_sock(con);
1515 }
1516 
1517 static void process_listen_recv_socket(struct work_struct *work)
1518 {
1519 	accept_from_sock(&listen_con);
1520 }
1521 
1522 static void dlm_connect(struct connection *con)
1523 {
1524 	struct sockaddr_storage addr;
1525 	int result, addr_len;
1526 	struct socket *sock;
1527 	unsigned int mark;
1528 
1529 	/* Some odd races can cause double-connects, ignore them */
1530 	if (con->retries++ > MAX_CONNECT_RETRIES)
1531 		return;
1532 
1533 	if (con->sock) {
1534 		log_print("node %d already connected.", con->nodeid);
1535 		return;
1536 	}
1537 
1538 	memset(&addr, 0, sizeof(addr));
1539 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1540 				dlm_proto_ops->try_new_addr, &mark);
1541 	if (result < 0) {
1542 		log_print("no address for nodeid %d", con->nodeid);
1543 		return;
1544 	}
1545 
1546 	/* Create a socket to communicate with */
1547 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1548 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1549 	if (result < 0)
1550 		goto socket_err;
1551 
1552 	sock_set_mark(sock->sk, mark);
1553 	dlm_proto_ops->sockopts(sock);
1554 
1555 	add_sock(sock, con);
1556 
1557 	result = dlm_proto_ops->bind(sock);
1558 	if (result < 0)
1559 		goto add_sock_err;
1560 
1561 	log_print_ratelimited("connecting to %d", con->nodeid);
1562 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1563 	result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1564 					addr_len);
1565 	if (result < 0)
1566 		goto add_sock_err;
1567 
1568 	return;
1569 
1570 add_sock_err:
1571 	dlm_close_sock(&con->sock);
1572 
1573 socket_err:
1574 	/*
1575 	 * Some errors are fatal and this list might need adjusting. For other
1576 	 * errors we try again until the max number of retries is reached.
1577 	 */
1578 	if (result != -EHOSTUNREACH &&
1579 	    result != -ENETUNREACH &&
1580 	    result != -ENETDOWN &&
1581 	    result != -EINVAL &&
1582 	    result != -EPROTONOSUPPORT) {
1583 		log_print("connect %d try %d error %d", con->nodeid,
1584 			  con->retries, result);
1585 		msleep(1000);
1586 		lowcomms_connect_sock(con);
1587 	}
1588 }
1589 
1590 /* Send workqueue function */
1591 static void process_send_sockets(struct work_struct *work)
1592 {
1593 	struct connection *con = container_of(work, struct connection, swork);
1594 
1595 	WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1596 
1597 	clear_bit(CF_WRITE_PENDING, &con->flags);
1598 
1599 	if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1600 		close_connection(con, false, false, true);
1601 		dlm_midcomms_unack_msg_resend(con->nodeid);
1602 	}
1603 
1604 	if (con->sock == NULL) {
1605 		if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1606 			msleep(1000);
1607 
1608 		mutex_lock(&con->sock_mutex);
1609 		dlm_connect(con);
1610 		mutex_unlock(&con->sock_mutex);
1611 	}
1612 
1613 	if (!list_empty(&con->writequeue))
1614 		send_to_sock(con);
1615 }
1616 
1617 static void work_stop(void)
1618 {
1619 	if (recv_workqueue) {
1620 		destroy_workqueue(recv_workqueue);
1621 		recv_workqueue = NULL;
1622 	}
1623 
1624 	if (send_workqueue) {
1625 		destroy_workqueue(send_workqueue);
1626 		send_workqueue = NULL;
1627 	}
1628 }
1629 
1630 static int work_start(void)
1631 {
1632 	recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1633 	if (!recv_workqueue) {
1634 		log_print("can't start dlm_recv");
1635 		return -ENOMEM;
1636 	}
1637 
1638 	send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1639 	if (!send_workqueue) {
1640 		log_print("can't start dlm_send");
1641 		destroy_workqueue(recv_workqueue);
1642 		recv_workqueue = NULL;
1643 		return -ENOMEM;
1644 	}
1645 
1646 	return 0;
1647 }
1648 
1649 static void shutdown_conn(struct connection *con)
1650 {
1651 	if (dlm_proto_ops->shutdown_action)
1652 		dlm_proto_ops->shutdown_action(con);
1653 }
1654 
1655 void dlm_lowcomms_shutdown(void)
1656 {
1657 	int idx;
1658 
1659 	/* Set all the flags to prevent any
1660 	 * socket activity.
1661 	 */
1662 	dlm_allow_conn = 0;
1663 
1664 	if (recv_workqueue)
1665 		flush_workqueue(recv_workqueue);
1666 	if (send_workqueue)
1667 		flush_workqueue(send_workqueue);
1668 
1669 	dlm_close_sock(&listen_con.sock);
1670 
1671 	idx = srcu_read_lock(&connections_srcu);
1672 	foreach_conn(shutdown_conn);
1673 	srcu_read_unlock(&connections_srcu, idx);
1674 }
1675 
1676 static void _stop_conn(struct connection *con, bool and_other)
1677 {
1678 	mutex_lock(&con->sock_mutex);
1679 	set_bit(CF_CLOSE, &con->flags);
1680 	set_bit(CF_READ_PENDING, &con->flags);
1681 	set_bit(CF_WRITE_PENDING, &con->flags);
1682 	if (con->sock && con->sock->sk) {
1683 		write_lock_bh(&con->sock->sk->sk_callback_lock);
1684 		con->sock->sk->sk_user_data = NULL;
1685 		write_unlock_bh(&con->sock->sk->sk_callback_lock);
1686 	}
1687 	if (con->othercon && and_other)
1688 		_stop_conn(con->othercon, false);
1689 	mutex_unlock(&con->sock_mutex);
1690 }
1691 
1692 static void stop_conn(struct connection *con)
1693 {
1694 	_stop_conn(con, true);
1695 }
1696 
1697 static void connection_release(struct rcu_head *rcu)
1698 {
1699 	struct connection *con = container_of(rcu, struct connection, rcu);
1700 
1701 	kfree(con->rx_buf);
1702 	kfree(con);
1703 }
1704 
1705 static void free_conn(struct connection *con)
1706 {
1707 	close_connection(con, true, true, true);
1708 	spin_lock(&connections_lock);
1709 	hlist_del_rcu(&con->list);
1710 	spin_unlock(&connections_lock);
1711 	if (con->othercon) {
1712 		clean_one_writequeue(con->othercon);
1713 		call_srcu(&connections_srcu, &con->othercon->rcu,
1714 			  connection_release);
1715 	}
1716 	clean_one_writequeue(con);
1717 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1718 }
1719 
1720 static void work_flush(void)
1721 {
1722 	int ok;
1723 	int i;
1724 	struct connection *con;
1725 
1726 	do {
1727 		ok = 1;
1728 		foreach_conn(stop_conn);
1729 		if (recv_workqueue)
1730 			flush_workqueue(recv_workqueue);
1731 		if (send_workqueue)
1732 			flush_workqueue(send_workqueue);
1733 		for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1734 			hlist_for_each_entry_rcu(con, &connection_hash[i],
1735 						 list) {
1736 				ok &= test_bit(CF_READ_PENDING, &con->flags);
1737 				ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1738 				if (con->othercon) {
1739 					ok &= test_bit(CF_READ_PENDING,
1740 						       &con->othercon->flags);
1741 					ok &= test_bit(CF_WRITE_PENDING,
1742 						       &con->othercon->flags);
1743 				}
1744 			}
1745 		}
1746 	} while (!ok);
1747 }
1748 
1749 void dlm_lowcomms_stop(void)
1750 {
1751 	int idx;
1752 
1753 	idx = srcu_read_lock(&connections_srcu);
1754 	work_flush();
1755 	foreach_conn(free_conn);
1756 	srcu_read_unlock(&connections_srcu, idx);
1757 	work_stop();
1758 	deinit_local();
1759 
1760 	dlm_proto_ops = NULL;
1761 }
1762 
1763 static int dlm_listen_for_all(void)
1764 {
1765 	struct socket *sock;
1766 	int result;
1767 
1768 	log_print("Using %s for communications",
1769 		  dlm_proto_ops->name);
1770 
1771 	result = dlm_proto_ops->listen_validate();
1772 	if (result < 0)
1773 		return result;
1774 
1775 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1776 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1777 	if (result < 0) {
1778 		log_print("Can't create comms socket, check SCTP is loaded");
1779 		goto out;
1780 	}
1781 
1782 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1783 	dlm_proto_ops->listen_sockopts(sock);
1784 
1785 	result = dlm_proto_ops->listen_bind(sock);
1786 	if (result < 0)
1787 		goto out;
1788 
1789 	save_listen_callbacks(sock);
1790 	add_listen_sock(sock, &listen_con);
1791 
1792 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1793 	result = sock->ops->listen(sock, 5);
1794 	if (result < 0) {
1795 		dlm_close_sock(&listen_con.sock);
1796 		goto out;
1797 	}
1798 
1799 	return 0;
1800 
1801 out:
1802 	sock_release(sock);
1803 	return result;
1804 }
1805 
1806 static int dlm_tcp_bind(struct socket *sock)
1807 {
1808 	struct sockaddr_storage src_addr;
1809 	int result, addr_len;
1810 
1811 	/* Bind to our cluster-known address connecting to avoid
1812 	 * routing problems.
1813 	 */
1814 	memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1815 	make_sockaddr(&src_addr, 0, &addr_len);
1816 
1817 	result = sock->ops->bind(sock, (struct sockaddr *)&src_addr,
1818 				 addr_len);
1819 	if (result < 0) {
1820 		/* This *may* not indicate a critical error */
1821 		log_print("could not bind for connect: %d", result);
1822 	}
1823 
1824 	return 0;
1825 }
1826 
1827 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1828 			   struct sockaddr *addr, int addr_len)
1829 {
1830 	int ret;
1831 
1832 	ret = sock->ops->connect(sock, addr, addr_len, O_NONBLOCK);
1833 	switch (ret) {
1834 	case -EINPROGRESS:
1835 		fallthrough;
1836 	case 0:
1837 		return 0;
1838 	}
1839 
1840 	return ret;
1841 }
1842 
1843 static int dlm_tcp_listen_validate(void)
1844 {
1845 	/* We don't support multi-homed hosts */
1846 	if (dlm_local_count > 1) {
1847 		log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1848 		return -EINVAL;
1849 	}
1850 
1851 	return 0;
1852 }
1853 
1854 static void dlm_tcp_sockopts(struct socket *sock)
1855 {
1856 	/* Turn off Nagle's algorithm */
1857 	tcp_sock_set_nodelay(sock->sk);
1858 }
1859 
1860 static void dlm_tcp_listen_sockopts(struct socket *sock)
1861 {
1862 	dlm_tcp_sockopts(sock);
1863 	sock_set_reuseaddr(sock->sk);
1864 }
1865 
1866 static int dlm_tcp_listen_bind(struct socket *sock)
1867 {
1868 	int addr_len;
1869 
1870 	/* Bind to our port */
1871 	make_sockaddr(dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1872 	return sock->ops->bind(sock, (struct sockaddr *)dlm_local_addr[0],
1873 			       addr_len);
1874 }
1875 
1876 static const struct dlm_proto_ops dlm_tcp_ops = {
1877 	.name = "TCP",
1878 	.proto = IPPROTO_TCP,
1879 	.connect = dlm_tcp_connect,
1880 	.sockopts = dlm_tcp_sockopts,
1881 	.bind = dlm_tcp_bind,
1882 	.listen_validate = dlm_tcp_listen_validate,
1883 	.listen_sockopts = dlm_tcp_listen_sockopts,
1884 	.listen_bind = dlm_tcp_listen_bind,
1885 	.shutdown_action = dlm_tcp_shutdown,
1886 	.eof_condition = tcp_eof_condition,
1887 };
1888 
1889 static int dlm_sctp_bind(struct socket *sock)
1890 {
1891 	return sctp_bind_addrs(sock, 0);
1892 }
1893 
1894 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1895 			    struct sockaddr *addr, int addr_len)
1896 {
1897 	int ret;
1898 
1899 	/*
1900 	 * Make sock->ops->connect() function return in specified time,
1901 	 * since O_NONBLOCK argument in connect() function does not work here,
1902 	 * then, we should restore the default value of this attribute.
1903 	 */
1904 	sock_set_sndtimeo(sock->sk, 5);
1905 	ret = sock->ops->connect(sock, addr, addr_len, 0);
1906 	sock_set_sndtimeo(sock->sk, 0);
1907 	if (ret < 0)
1908 		return ret;
1909 
1910 	if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1911 		log_print("successful connected to node %d", con->nodeid);
1912 
1913 	return 0;
1914 }
1915 
1916 static int dlm_sctp_listen_validate(void)
1917 {
1918 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1919 		log_print("SCTP is not enabled by this kernel");
1920 		return -EOPNOTSUPP;
1921 	}
1922 
1923 	request_module("sctp");
1924 	return 0;
1925 }
1926 
1927 static int dlm_sctp_bind_listen(struct socket *sock)
1928 {
1929 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1930 }
1931 
1932 static void dlm_sctp_sockopts(struct socket *sock)
1933 {
1934 	/* Turn off Nagle's algorithm */
1935 	sctp_sock_set_nodelay(sock->sk);
1936 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1937 }
1938 
1939 static const struct dlm_proto_ops dlm_sctp_ops = {
1940 	.name = "SCTP",
1941 	.proto = IPPROTO_SCTP,
1942 	.try_new_addr = true,
1943 	.connect = dlm_sctp_connect,
1944 	.sockopts = dlm_sctp_sockopts,
1945 	.bind = dlm_sctp_bind,
1946 	.listen_validate = dlm_sctp_listen_validate,
1947 	.listen_sockopts = dlm_sctp_sockopts,
1948 	.listen_bind = dlm_sctp_bind_listen,
1949 };
1950 
1951 int dlm_lowcomms_start(void)
1952 {
1953 	int error = -EINVAL;
1954 	int i;
1955 
1956 	for (i = 0; i < CONN_HASH_SIZE; i++)
1957 		INIT_HLIST_HEAD(&connection_hash[i]);
1958 
1959 	init_local();
1960 	if (!dlm_local_count) {
1961 		error = -ENOTCONN;
1962 		log_print("no local IP address has been set");
1963 		goto fail;
1964 	}
1965 
1966 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1967 
1968 	error = work_start();
1969 	if (error)
1970 		goto fail_local;
1971 
1972 	dlm_allow_conn = 1;
1973 
1974 	/* Start listening */
1975 	switch (dlm_config.ci_protocol) {
1976 	case DLM_PROTO_TCP:
1977 		dlm_proto_ops = &dlm_tcp_ops;
1978 		break;
1979 	case DLM_PROTO_SCTP:
1980 		dlm_proto_ops = &dlm_sctp_ops;
1981 		break;
1982 	default:
1983 		log_print("Invalid protocol identifier %d set",
1984 			  dlm_config.ci_protocol);
1985 		error = -EINVAL;
1986 		goto fail_proto_ops;
1987 	}
1988 
1989 	error = dlm_listen_for_all();
1990 	if (error)
1991 		goto fail_listen;
1992 
1993 	return 0;
1994 
1995 fail_listen:
1996 	dlm_proto_ops = NULL;
1997 fail_proto_ops:
1998 	dlm_allow_conn = 0;
1999 	dlm_close_sock(&listen_con.sock);
2000 	work_stop();
2001 fail_local:
2002 	deinit_local();
2003 fail:
2004 	return error;
2005 }
2006 
2007 void dlm_lowcomms_exit(void)
2008 {
2009 	struct dlm_node_addr *na, *safe;
2010 
2011 	spin_lock(&dlm_node_addrs_spin);
2012 	list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2013 		list_del(&na->list);
2014 		while (na->addr_count--)
2015 			kfree(na->addr[na->addr_count]);
2016 		kfree(na);
2017 	}
2018 	spin_unlock(&dlm_node_addrs_spin);
2019 }
2020