xref: /openbmc/linux/fs/dlm/lowcomms.c (revision 8dda2eac)
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 	void (*connect_action) (struct connection *);	/* What to do to connect */
88 	void (*shutdown_action)(struct connection *con); /* What to do to shutdown */
89 	bool (*eof_condition)(struct connection *con); /* What to do to eof check */
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 static struct listen_sock_callbacks {
149 	void (*sk_error_report)(struct sock *);
150 	void (*sk_data_ready)(struct sock *);
151 	void (*sk_state_change)(struct sock *);
152 	void (*sk_write_space)(struct sock *);
153 } listen_sock;
154 
155 static LIST_HEAD(dlm_node_addrs);
156 static DEFINE_SPINLOCK(dlm_node_addrs_spin);
157 
158 static struct listen_connection listen_con;
159 static struct sockaddr_storage *dlm_local_addr[DLM_MAX_ADDR_COUNT];
160 static int dlm_local_count;
161 int dlm_allow_conn;
162 
163 /* Work queues */
164 static struct workqueue_struct *recv_workqueue;
165 static struct workqueue_struct *send_workqueue;
166 
167 static struct hlist_head connection_hash[CONN_HASH_SIZE];
168 static DEFINE_SPINLOCK(connections_lock);
169 DEFINE_STATIC_SRCU(connections_srcu);
170 
171 static void process_recv_sockets(struct work_struct *work);
172 static void process_send_sockets(struct work_struct *work);
173 
174 static void sctp_connect_to_sock(struct connection *con);
175 static void tcp_connect_to_sock(struct connection *con);
176 static void dlm_tcp_shutdown(struct connection *con);
177 
178 static struct connection *__find_con(int nodeid, int r)
179 {
180 	struct connection *con;
181 
182 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
183 		if (con->nodeid == nodeid)
184 			return con;
185 	}
186 
187 	return NULL;
188 }
189 
190 static bool tcp_eof_condition(struct connection *con)
191 {
192 	return atomic_read(&con->writequeue_cnt);
193 }
194 
195 static int dlm_con_init(struct connection *con, int nodeid)
196 {
197 	con->rx_buflen = dlm_config.ci_buffer_size;
198 	con->rx_buf = kmalloc(con->rx_buflen, GFP_NOFS);
199 	if (!con->rx_buf)
200 		return -ENOMEM;
201 
202 	con->nodeid = nodeid;
203 	mutex_init(&con->sock_mutex);
204 	INIT_LIST_HEAD(&con->writequeue);
205 	spin_lock_init(&con->writequeue_lock);
206 	atomic_set(&con->writequeue_cnt, 0);
207 	INIT_WORK(&con->swork, process_send_sockets);
208 	INIT_WORK(&con->rwork, process_recv_sockets);
209 	init_waitqueue_head(&con->shutdown_wait);
210 
211 	switch (dlm_config.ci_protocol) {
212 	case DLM_PROTO_TCP:
213 		con->connect_action = tcp_connect_to_sock;
214 		con->shutdown_action = dlm_tcp_shutdown;
215 		con->eof_condition = tcp_eof_condition;
216 		break;
217 	case DLM_PROTO_SCTP:
218 		con->connect_action = sctp_connect_to_sock;
219 		break;
220 	default:
221 		kfree(con->rx_buf);
222 		return -EINVAL;
223 	}
224 
225 	return 0;
226 }
227 
228 /*
229  * If 'allocation' is zero then we don't attempt to create a new
230  * connection structure for this node.
231  */
232 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
233 {
234 	struct connection *con, *tmp;
235 	int r, ret;
236 
237 	r = nodeid_hash(nodeid);
238 	con = __find_con(nodeid, r);
239 	if (con || !alloc)
240 		return con;
241 
242 	con = kzalloc(sizeof(*con), alloc);
243 	if (!con)
244 		return NULL;
245 
246 	ret = dlm_con_init(con, nodeid);
247 	if (ret) {
248 		kfree(con);
249 		return NULL;
250 	}
251 
252 	spin_lock(&connections_lock);
253 	/* Because multiple workqueues/threads calls this function it can
254 	 * race on multiple cpu's. Instead of locking hot path __find_con()
255 	 * we just check in rare cases of recently added nodes again
256 	 * under protection of connections_lock. If this is the case we
257 	 * abort our connection creation and return the existing connection.
258 	 */
259 	tmp = __find_con(nodeid, r);
260 	if (tmp) {
261 		spin_unlock(&connections_lock);
262 		kfree(con->rx_buf);
263 		kfree(con);
264 		return tmp;
265 	}
266 
267 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
268 	spin_unlock(&connections_lock);
269 
270 	return con;
271 }
272 
273 /* Loop round all connections */
274 static void foreach_conn(void (*conn_func)(struct connection *c))
275 {
276 	int i;
277 	struct connection *con;
278 
279 	for (i = 0; i < CONN_HASH_SIZE; i++) {
280 		hlist_for_each_entry_rcu(con, &connection_hash[i], list)
281 			conn_func(con);
282 	}
283 }
284 
285 static struct dlm_node_addr *find_node_addr(int nodeid)
286 {
287 	struct dlm_node_addr *na;
288 
289 	list_for_each_entry(na, &dlm_node_addrs, list) {
290 		if (na->nodeid == nodeid)
291 			return na;
292 	}
293 	return NULL;
294 }
295 
296 static int addr_compare(const struct sockaddr_storage *x,
297 			const struct sockaddr_storage *y)
298 {
299 	switch (x->ss_family) {
300 	case AF_INET: {
301 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
302 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
303 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
304 			return 0;
305 		if (sinx->sin_port != siny->sin_port)
306 			return 0;
307 		break;
308 	}
309 	case AF_INET6: {
310 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
311 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
312 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
313 			return 0;
314 		if (sinx->sin6_port != siny->sin6_port)
315 			return 0;
316 		break;
317 	}
318 	default:
319 		return 0;
320 	}
321 	return 1;
322 }
323 
324 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
325 			  struct sockaddr *sa_out, bool try_new_addr,
326 			  unsigned int *mark)
327 {
328 	struct sockaddr_storage sas;
329 	struct dlm_node_addr *na;
330 
331 	if (!dlm_local_count)
332 		return -1;
333 
334 	spin_lock(&dlm_node_addrs_spin);
335 	na = find_node_addr(nodeid);
336 	if (na && na->addr_count) {
337 		memcpy(&sas, na->addr[na->curr_addr_index],
338 		       sizeof(struct sockaddr_storage));
339 
340 		if (try_new_addr) {
341 			na->curr_addr_index++;
342 			if (na->curr_addr_index == na->addr_count)
343 				na->curr_addr_index = 0;
344 		}
345 	}
346 	spin_unlock(&dlm_node_addrs_spin);
347 
348 	if (!na)
349 		return -EEXIST;
350 
351 	if (!na->addr_count)
352 		return -ENOENT;
353 
354 	*mark = na->mark;
355 
356 	if (sas_out)
357 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
358 
359 	if (!sa_out)
360 		return 0;
361 
362 	if (dlm_local_addr[0]->ss_family == AF_INET) {
363 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
364 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
365 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
366 	} else {
367 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
368 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
369 		ret6->sin6_addr = in6->sin6_addr;
370 	}
371 
372 	return 0;
373 }
374 
375 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
376 			  unsigned int *mark)
377 {
378 	struct dlm_node_addr *na;
379 	int rv = -EEXIST;
380 	int addr_i;
381 
382 	spin_lock(&dlm_node_addrs_spin);
383 	list_for_each_entry(na, &dlm_node_addrs, list) {
384 		if (!na->addr_count)
385 			continue;
386 
387 		for (addr_i = 0; addr_i < na->addr_count; addr_i++) {
388 			if (addr_compare(na->addr[addr_i], addr)) {
389 				*nodeid = na->nodeid;
390 				*mark = na->mark;
391 				rv = 0;
392 				goto unlock;
393 			}
394 		}
395 	}
396 unlock:
397 	spin_unlock(&dlm_node_addrs_spin);
398 	return rv;
399 }
400 
401 /* caller need to held dlm_node_addrs_spin lock */
402 static bool dlm_lowcomms_na_has_addr(const struct dlm_node_addr *na,
403 				     const struct sockaddr_storage *addr)
404 {
405 	int i;
406 
407 	for (i = 0; i < na->addr_count; i++) {
408 		if (addr_compare(na->addr[i], addr))
409 			return true;
410 	}
411 
412 	return false;
413 }
414 
415 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
416 {
417 	struct sockaddr_storage *new_addr;
418 	struct dlm_node_addr *new_node, *na;
419 	bool ret;
420 
421 	new_node = kzalloc(sizeof(struct dlm_node_addr), GFP_NOFS);
422 	if (!new_node)
423 		return -ENOMEM;
424 
425 	new_addr = kzalloc(sizeof(struct sockaddr_storage), GFP_NOFS);
426 	if (!new_addr) {
427 		kfree(new_node);
428 		return -ENOMEM;
429 	}
430 
431 	memcpy(new_addr, addr, len);
432 
433 	spin_lock(&dlm_node_addrs_spin);
434 	na = find_node_addr(nodeid);
435 	if (!na) {
436 		new_node->nodeid = nodeid;
437 		new_node->addr[0] = new_addr;
438 		new_node->addr_count = 1;
439 		new_node->mark = dlm_config.ci_mark;
440 		list_add(&new_node->list, &dlm_node_addrs);
441 		spin_unlock(&dlm_node_addrs_spin);
442 		return 0;
443 	}
444 
445 	ret = dlm_lowcomms_na_has_addr(na, addr);
446 	if (ret) {
447 		spin_unlock(&dlm_node_addrs_spin);
448 		kfree(new_addr);
449 		kfree(new_node);
450 		return -EEXIST;
451 	}
452 
453 	if (na->addr_count >= DLM_MAX_ADDR_COUNT) {
454 		spin_unlock(&dlm_node_addrs_spin);
455 		kfree(new_addr);
456 		kfree(new_node);
457 		return -ENOSPC;
458 	}
459 
460 	na->addr[na->addr_count++] = new_addr;
461 	spin_unlock(&dlm_node_addrs_spin);
462 	kfree(new_node);
463 	return 0;
464 }
465 
466 /* Data available on socket or listen socket received a connect */
467 static void lowcomms_data_ready(struct sock *sk)
468 {
469 	struct connection *con;
470 
471 	read_lock_bh(&sk->sk_callback_lock);
472 	con = sock2con(sk);
473 	if (con && !test_and_set_bit(CF_READ_PENDING, &con->flags))
474 		queue_work(recv_workqueue, &con->rwork);
475 	read_unlock_bh(&sk->sk_callback_lock);
476 }
477 
478 static void lowcomms_listen_data_ready(struct sock *sk)
479 {
480 	if (!dlm_allow_conn)
481 		return;
482 
483 	queue_work(recv_workqueue, &listen_con.rwork);
484 }
485 
486 static void lowcomms_write_space(struct sock *sk)
487 {
488 	struct connection *con;
489 
490 	read_lock_bh(&sk->sk_callback_lock);
491 	con = sock2con(sk);
492 	if (!con)
493 		goto out;
494 
495 	if (!test_and_set_bit(CF_CONNECTED, &con->flags)) {
496 		log_print("successful connected to node %d", con->nodeid);
497 		queue_work(send_workqueue, &con->swork);
498 		goto out;
499 	}
500 
501 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
502 
503 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
504 		con->sock->sk->sk_write_pending--;
505 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
506 	}
507 
508 	queue_work(send_workqueue, &con->swork);
509 out:
510 	read_unlock_bh(&sk->sk_callback_lock);
511 }
512 
513 static inline void lowcomms_connect_sock(struct connection *con)
514 {
515 	if (test_bit(CF_CLOSE, &con->flags))
516 		return;
517 	queue_work(send_workqueue, &con->swork);
518 	cond_resched();
519 }
520 
521 static void lowcomms_state_change(struct sock *sk)
522 {
523 	/* SCTP layer is not calling sk_data_ready when the connection
524 	 * is done, so we catch the signal through here. Also, it
525 	 * doesn't switch socket state when entering shutdown, so we
526 	 * skip the write in that case.
527 	 */
528 	if (sk->sk_shutdown) {
529 		if (sk->sk_shutdown == RCV_SHUTDOWN)
530 			lowcomms_data_ready(sk);
531 	} else if (sk->sk_state == TCP_ESTABLISHED) {
532 		lowcomms_write_space(sk);
533 	}
534 }
535 
536 int dlm_lowcomms_connect_node(int nodeid)
537 {
538 	struct connection *con;
539 	int idx;
540 
541 	if (nodeid == dlm_our_nodeid())
542 		return 0;
543 
544 	idx = srcu_read_lock(&connections_srcu);
545 	con = nodeid2con(nodeid, GFP_NOFS);
546 	if (!con) {
547 		srcu_read_unlock(&connections_srcu, idx);
548 		return -ENOMEM;
549 	}
550 
551 	lowcomms_connect_sock(con);
552 	srcu_read_unlock(&connections_srcu, idx);
553 
554 	return 0;
555 }
556 
557 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
558 {
559 	struct dlm_node_addr *na;
560 
561 	spin_lock(&dlm_node_addrs_spin);
562 	na = find_node_addr(nodeid);
563 	if (!na) {
564 		spin_unlock(&dlm_node_addrs_spin);
565 		return -ENOENT;
566 	}
567 
568 	na->mark = mark;
569 	spin_unlock(&dlm_node_addrs_spin);
570 
571 	return 0;
572 }
573 
574 static void lowcomms_error_report(struct sock *sk)
575 {
576 	struct connection *con;
577 	struct sockaddr_storage saddr;
578 	void (*orig_report)(struct sock *) = NULL;
579 
580 	read_lock_bh(&sk->sk_callback_lock);
581 	con = sock2con(sk);
582 	if (con == NULL)
583 		goto out;
584 
585 	orig_report = listen_sock.sk_error_report;
586 	if (con->sock == NULL ||
587 	    kernel_getpeername(con->sock, (struct sockaddr *)&saddr) < 0) {
588 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
589 				   "sending to node %d, port %d, "
590 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
591 				   con->nodeid, dlm_config.ci_tcp_port,
592 				   sk->sk_err, sk->sk_err_soft);
593 	} else if (saddr.ss_family == AF_INET) {
594 		struct sockaddr_in *sin4 = (struct sockaddr_in *)&saddr;
595 
596 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
597 				   "sending to node %d at %pI4, port %d, "
598 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
599 				   con->nodeid, &sin4->sin_addr.s_addr,
600 				   dlm_config.ci_tcp_port, sk->sk_err,
601 				   sk->sk_err_soft);
602 	} else {
603 		struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&saddr;
604 
605 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
606 				   "sending to node %d at %u.%u.%u.%u, "
607 				   "port %d, sk_err=%d/%d\n", dlm_our_nodeid(),
608 				   con->nodeid, sin6->sin6_addr.s6_addr32[0],
609 				   sin6->sin6_addr.s6_addr32[1],
610 				   sin6->sin6_addr.s6_addr32[2],
611 				   sin6->sin6_addr.s6_addr32[3],
612 				   dlm_config.ci_tcp_port, sk->sk_err,
613 				   sk->sk_err_soft);
614 	}
615 
616 	/* below sendcon only handling */
617 	if (test_bit(CF_IS_OTHERCON, &con->flags))
618 		con = con->sendcon;
619 
620 	switch (sk->sk_err) {
621 	case ECONNREFUSED:
622 		set_bit(CF_DELAY_CONNECT, &con->flags);
623 		break;
624 	default:
625 		break;
626 	}
627 
628 	if (!test_and_set_bit(CF_RECONNECT, &con->flags))
629 		queue_work(send_workqueue, &con->swork);
630 
631 out:
632 	read_unlock_bh(&sk->sk_callback_lock);
633 	if (orig_report)
634 		orig_report(sk);
635 }
636 
637 /* Note: sk_callback_lock must be locked before calling this function. */
638 static void save_listen_callbacks(struct socket *sock)
639 {
640 	struct sock *sk = sock->sk;
641 
642 	listen_sock.sk_data_ready = sk->sk_data_ready;
643 	listen_sock.sk_state_change = sk->sk_state_change;
644 	listen_sock.sk_write_space = sk->sk_write_space;
645 	listen_sock.sk_error_report = sk->sk_error_report;
646 }
647 
648 static void restore_callbacks(struct socket *sock)
649 {
650 	struct sock *sk = sock->sk;
651 
652 	write_lock_bh(&sk->sk_callback_lock);
653 	sk->sk_user_data = NULL;
654 	sk->sk_data_ready = listen_sock.sk_data_ready;
655 	sk->sk_state_change = listen_sock.sk_state_change;
656 	sk->sk_write_space = listen_sock.sk_write_space;
657 	sk->sk_error_report = listen_sock.sk_error_report;
658 	write_unlock_bh(&sk->sk_callback_lock);
659 }
660 
661 static void add_listen_sock(struct socket *sock, struct listen_connection *con)
662 {
663 	struct sock *sk = sock->sk;
664 
665 	write_lock_bh(&sk->sk_callback_lock);
666 	save_listen_callbacks(sock);
667 	con->sock = sock;
668 
669 	sk->sk_user_data = con;
670 	sk->sk_allocation = GFP_NOFS;
671 	/* Install a data_ready callback */
672 	sk->sk_data_ready = lowcomms_listen_data_ready;
673 	write_unlock_bh(&sk->sk_callback_lock);
674 }
675 
676 /* Make a socket active */
677 static void add_sock(struct socket *sock, struct connection *con)
678 {
679 	struct sock *sk = sock->sk;
680 
681 	write_lock_bh(&sk->sk_callback_lock);
682 	con->sock = sock;
683 
684 	sk->sk_user_data = con;
685 	/* Install a data_ready callback */
686 	sk->sk_data_ready = lowcomms_data_ready;
687 	sk->sk_write_space = lowcomms_write_space;
688 	sk->sk_state_change = lowcomms_state_change;
689 	sk->sk_allocation = GFP_NOFS;
690 	sk->sk_error_report = lowcomms_error_report;
691 	write_unlock_bh(&sk->sk_callback_lock);
692 }
693 
694 /* Add the port number to an IPv6 or 4 sockaddr and return the address
695    length */
696 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
697 			  int *addr_len)
698 {
699 	saddr->ss_family =  dlm_local_addr[0]->ss_family;
700 	if (saddr->ss_family == AF_INET) {
701 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
702 		in4_addr->sin_port = cpu_to_be16(port);
703 		*addr_len = sizeof(struct sockaddr_in);
704 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
705 	} else {
706 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
707 		in6_addr->sin6_port = cpu_to_be16(port);
708 		*addr_len = sizeof(struct sockaddr_in6);
709 	}
710 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
711 }
712 
713 static void dlm_page_release(struct kref *kref)
714 {
715 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
716 						  ref);
717 
718 	__free_page(e->page);
719 	kfree(e);
720 }
721 
722 static void dlm_msg_release(struct kref *kref)
723 {
724 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
725 
726 	kref_put(&msg->entry->ref, dlm_page_release);
727 	kfree(msg);
728 }
729 
730 static void free_entry(struct writequeue_entry *e)
731 {
732 	struct dlm_msg *msg, *tmp;
733 
734 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
735 		if (msg->orig_msg) {
736 			msg->orig_msg->retransmit = false;
737 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
738 		}
739 
740 		list_del(&msg->list);
741 		kref_put(&msg->ref, dlm_msg_release);
742 	}
743 
744 	list_del(&e->list);
745 	atomic_dec(&e->con->writequeue_cnt);
746 	kref_put(&e->ref, dlm_page_release);
747 }
748 
749 static void dlm_close_sock(struct socket **sock)
750 {
751 	if (*sock) {
752 		restore_callbacks(*sock);
753 		sock_release(*sock);
754 		*sock = NULL;
755 	}
756 }
757 
758 /* Close a remote connection and tidy up */
759 static void close_connection(struct connection *con, bool and_other,
760 			     bool tx, bool rx)
761 {
762 	bool closing = test_and_set_bit(CF_CLOSING, &con->flags);
763 	struct writequeue_entry *e;
764 
765 	if (tx && !closing && cancel_work_sync(&con->swork)) {
766 		log_print("canceled swork for node %d", con->nodeid);
767 		clear_bit(CF_WRITE_PENDING, &con->flags);
768 	}
769 	if (rx && !closing && cancel_work_sync(&con->rwork)) {
770 		log_print("canceled rwork for node %d", con->nodeid);
771 		clear_bit(CF_READ_PENDING, &con->flags);
772 	}
773 
774 	mutex_lock(&con->sock_mutex);
775 	dlm_close_sock(&con->sock);
776 
777 	if (con->othercon && and_other) {
778 		/* Will only re-enter once. */
779 		close_connection(con->othercon, false, tx, rx);
780 	}
781 
782 	/* if we send a writequeue entry only a half way, we drop the
783 	 * whole entry because reconnection and that we not start of the
784 	 * middle of a msg which will confuse the other end.
785 	 *
786 	 * we can always drop messages because retransmits, but what we
787 	 * cannot allow is to transmit half messages which may be processed
788 	 * at the other side.
789 	 *
790 	 * our policy is to start on a clean state when disconnects, we don't
791 	 * know what's send/received on transport layer in this case.
792 	 */
793 	spin_lock(&con->writequeue_lock);
794 	if (!list_empty(&con->writequeue)) {
795 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
796 				     list);
797 		if (e->dirty)
798 			free_entry(e);
799 	}
800 	spin_unlock(&con->writequeue_lock);
801 
802 	con->rx_leftover = 0;
803 	con->retries = 0;
804 	clear_bit(CF_CONNECTED, &con->flags);
805 	clear_bit(CF_DELAY_CONNECT, &con->flags);
806 	clear_bit(CF_RECONNECT, &con->flags);
807 	clear_bit(CF_EOF, &con->flags);
808 	mutex_unlock(&con->sock_mutex);
809 	clear_bit(CF_CLOSING, &con->flags);
810 }
811 
812 static void shutdown_connection(struct connection *con)
813 {
814 	int ret;
815 
816 	flush_work(&con->swork);
817 
818 	mutex_lock(&con->sock_mutex);
819 	/* nothing to shutdown */
820 	if (!con->sock) {
821 		mutex_unlock(&con->sock_mutex);
822 		return;
823 	}
824 
825 	set_bit(CF_SHUTDOWN, &con->flags);
826 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
827 	mutex_unlock(&con->sock_mutex);
828 	if (ret) {
829 		log_print("Connection %p failed to shutdown: %d will force close",
830 			  con, ret);
831 		goto force_close;
832 	} else {
833 		ret = wait_event_timeout(con->shutdown_wait,
834 					 !test_bit(CF_SHUTDOWN, &con->flags),
835 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
836 		if (ret == 0) {
837 			log_print("Connection %p shutdown timed out, will force close",
838 				  con);
839 			goto force_close;
840 		}
841 	}
842 
843 	return;
844 
845 force_close:
846 	clear_bit(CF_SHUTDOWN, &con->flags);
847 	close_connection(con, false, true, true);
848 }
849 
850 static void dlm_tcp_shutdown(struct connection *con)
851 {
852 	if (con->othercon)
853 		shutdown_connection(con->othercon);
854 	shutdown_connection(con);
855 }
856 
857 static int con_realloc_receive_buf(struct connection *con, int newlen)
858 {
859 	unsigned char *newbuf;
860 
861 	newbuf = kmalloc(newlen, GFP_NOFS);
862 	if (!newbuf)
863 		return -ENOMEM;
864 
865 	/* copy any leftover from last receive */
866 	if (con->rx_leftover)
867 		memmove(newbuf, con->rx_buf, con->rx_leftover);
868 
869 	/* swap to new buffer space */
870 	kfree(con->rx_buf);
871 	con->rx_buflen = newlen;
872 	con->rx_buf = newbuf;
873 
874 	return 0;
875 }
876 
877 /* Data received from remote end */
878 static int receive_from_sock(struct connection *con)
879 {
880 	int call_again_soon = 0;
881 	struct msghdr msg;
882 	struct kvec iov;
883 	int ret, buflen;
884 
885 	mutex_lock(&con->sock_mutex);
886 
887 	if (con->sock == NULL) {
888 		ret = -EAGAIN;
889 		goto out_close;
890 	}
891 
892 	/* realloc if we get new buffer size to read out */
893 	buflen = dlm_config.ci_buffer_size;
894 	if (con->rx_buflen != buflen && con->rx_leftover <= buflen) {
895 		ret = con_realloc_receive_buf(con, buflen);
896 		if (ret < 0)
897 			goto out_resched;
898 	}
899 
900 	/* calculate new buffer parameter regarding last receive and
901 	 * possible leftover bytes
902 	 */
903 	iov.iov_base = con->rx_buf + con->rx_leftover;
904 	iov.iov_len = con->rx_buflen - con->rx_leftover;
905 
906 	memset(&msg, 0, sizeof(msg));
907 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
908 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
909 			     msg.msg_flags);
910 	if (ret <= 0)
911 		goto out_close;
912 	else if (ret == iov.iov_len)
913 		call_again_soon = 1;
914 
915 	/* new buflen according readed bytes and leftover from last receive */
916 	buflen = ret + con->rx_leftover;
917 	ret = dlm_process_incoming_buffer(con->nodeid, con->rx_buf, buflen);
918 	if (ret < 0)
919 		goto out_close;
920 
921 	/* calculate leftover bytes from process and put it into begin of
922 	 * the receive buffer, so next receive we have the full message
923 	 * at the start address of the receive buffer.
924 	 */
925 	con->rx_leftover = buflen - ret;
926 	if (con->rx_leftover) {
927 		memmove(con->rx_buf, con->rx_buf + ret,
928 			con->rx_leftover);
929 		call_again_soon = true;
930 	}
931 
932 	if (call_again_soon)
933 		goto out_resched;
934 
935 	mutex_unlock(&con->sock_mutex);
936 	return 0;
937 
938 out_resched:
939 	if (!test_and_set_bit(CF_READ_PENDING, &con->flags))
940 		queue_work(recv_workqueue, &con->rwork);
941 	mutex_unlock(&con->sock_mutex);
942 	return -EAGAIN;
943 
944 out_close:
945 	if (ret == 0) {
946 		log_print("connection %p got EOF from %d",
947 			  con, con->nodeid);
948 
949 		if (con->eof_condition && con->eof_condition(con)) {
950 			set_bit(CF_EOF, &con->flags);
951 			mutex_unlock(&con->sock_mutex);
952 		} else {
953 			mutex_unlock(&con->sock_mutex);
954 			close_connection(con, false, true, false);
955 
956 			/* handling for tcp shutdown */
957 			clear_bit(CF_SHUTDOWN, &con->flags);
958 			wake_up(&con->shutdown_wait);
959 		}
960 
961 		/* signal to breaking receive worker */
962 		ret = -1;
963 	} else {
964 		mutex_unlock(&con->sock_mutex);
965 	}
966 	return ret;
967 }
968 
969 /* Listening socket is busy, accept a connection */
970 static int accept_from_sock(struct listen_connection *con)
971 {
972 	int result;
973 	struct sockaddr_storage peeraddr;
974 	struct socket *newsock;
975 	int len, idx;
976 	int nodeid;
977 	struct connection *newcon;
978 	struct connection *addcon;
979 	unsigned int mark;
980 
981 	if (!con->sock)
982 		return -ENOTCONN;
983 
984 	result = kernel_accept(con->sock, &newsock, O_NONBLOCK);
985 	if (result < 0)
986 		goto accept_err;
987 
988 	/* Get the connected socket's peer */
989 	memset(&peeraddr, 0, sizeof(peeraddr));
990 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
991 	if (len < 0) {
992 		result = -ECONNABORTED;
993 		goto accept_err;
994 	}
995 
996 	/* Get the new node's NODEID */
997 	make_sockaddr(&peeraddr, 0, &len);
998 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
999 		unsigned char *b=(unsigned char *)&peeraddr;
1000 		log_print("connect from non cluster node");
1001 		print_hex_dump_bytes("ss: ", DUMP_PREFIX_NONE,
1002 				     b, sizeof(struct sockaddr_storage));
1003 		sock_release(newsock);
1004 		return -1;
1005 	}
1006 
1007 	log_print("got connection from %d", nodeid);
1008 
1009 	/*  Check to see if we already have a connection to this node. This
1010 	 *  could happen if the two nodes initiate a connection at roughly
1011 	 *  the same time and the connections cross on the wire.
1012 	 *  In this case we store the incoming one in "othercon"
1013 	 */
1014 	idx = srcu_read_lock(&connections_srcu);
1015 	newcon = nodeid2con(nodeid, GFP_NOFS);
1016 	if (!newcon) {
1017 		srcu_read_unlock(&connections_srcu, idx);
1018 		result = -ENOMEM;
1019 		goto accept_err;
1020 	}
1021 
1022 	sock_set_mark(newsock->sk, mark);
1023 
1024 	mutex_lock(&newcon->sock_mutex);
1025 	if (newcon->sock) {
1026 		struct connection *othercon = newcon->othercon;
1027 
1028 		if (!othercon) {
1029 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1030 			if (!othercon) {
1031 				log_print("failed to allocate incoming socket");
1032 				mutex_unlock(&newcon->sock_mutex);
1033 				srcu_read_unlock(&connections_srcu, idx);
1034 				result = -ENOMEM;
1035 				goto accept_err;
1036 			}
1037 
1038 			result = dlm_con_init(othercon, nodeid);
1039 			if (result < 0) {
1040 				kfree(othercon);
1041 				mutex_unlock(&newcon->sock_mutex);
1042 				srcu_read_unlock(&connections_srcu, idx);
1043 				goto accept_err;
1044 			}
1045 
1046 			lockdep_set_subclass(&othercon->sock_mutex, 1);
1047 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1048 			newcon->othercon = othercon;
1049 			othercon->sendcon = newcon;
1050 		} else {
1051 			/* close other sock con if we have something new */
1052 			close_connection(othercon, false, true, false);
1053 		}
1054 
1055 		mutex_lock(&othercon->sock_mutex);
1056 		add_sock(newsock, othercon);
1057 		addcon = othercon;
1058 		mutex_unlock(&othercon->sock_mutex);
1059 	}
1060 	else {
1061 		/* accept copies the sk after we've saved the callbacks, so we
1062 		   don't want to save them a second time or comm errors will
1063 		   result in calling sk_error_report recursively. */
1064 		add_sock(newsock, newcon);
1065 		addcon = newcon;
1066 	}
1067 
1068 	set_bit(CF_CONNECTED, &addcon->flags);
1069 	mutex_unlock(&newcon->sock_mutex);
1070 
1071 	/*
1072 	 * Add it to the active queue in case we got data
1073 	 * between processing the accept adding the socket
1074 	 * to the read_sockets list
1075 	 */
1076 	if (!test_and_set_bit(CF_READ_PENDING, &addcon->flags))
1077 		queue_work(recv_workqueue, &addcon->rwork);
1078 
1079 	srcu_read_unlock(&connections_srcu, idx);
1080 
1081 	return 0;
1082 
1083 accept_err:
1084 	if (newsock)
1085 		sock_release(newsock);
1086 
1087 	if (result != -EAGAIN)
1088 		log_print("error accepting connection from node: %d", result);
1089 	return result;
1090 }
1091 
1092 /*
1093  * writequeue_entry_complete - try to delete and free write queue entry
1094  * @e: write queue entry to try to delete
1095  * @completed: bytes completed
1096  *
1097  * writequeue_lock must be held.
1098  */
1099 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1100 {
1101 	e->offset += completed;
1102 	e->len -= completed;
1103 	/* signal that page was half way transmitted */
1104 	e->dirty = true;
1105 
1106 	if (e->len == 0 && e->users == 0)
1107 		free_entry(e);
1108 }
1109 
1110 /*
1111  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1112  */
1113 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1114 {
1115 	struct sockaddr_storage localaddr;
1116 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1117 	int i, addr_len, result = 0;
1118 
1119 	for (i = 0; i < dlm_local_count; i++) {
1120 		memcpy(&localaddr, dlm_local_addr[i], sizeof(localaddr));
1121 		make_sockaddr(&localaddr, port, &addr_len);
1122 
1123 		if (!i)
1124 			result = kernel_bind(sock, addr, addr_len);
1125 		else
1126 			result = sock_bind_add(sock->sk, addr, addr_len);
1127 
1128 		if (result < 0) {
1129 			log_print("Can't bind to %d addr number %d, %d.\n",
1130 				  port, i + 1, result);
1131 			break;
1132 		}
1133 	}
1134 	return result;
1135 }
1136 
1137 /* Initiate an SCTP association.
1138    This is a special case of send_to_sock() in that we don't yet have a
1139    peeled-off socket for this association, so we use the listening socket
1140    and add the primary IP address of the remote node.
1141  */
1142 static void sctp_connect_to_sock(struct connection *con)
1143 {
1144 	struct sockaddr_storage daddr;
1145 	int result;
1146 	int addr_len;
1147 	struct socket *sock;
1148 	unsigned int mark;
1149 
1150 	mutex_lock(&con->sock_mutex);
1151 
1152 	/* Some odd races can cause double-connects, ignore them */
1153 	if (con->retries++ > MAX_CONNECT_RETRIES)
1154 		goto out;
1155 
1156 	if (con->sock) {
1157 		log_print("node %d already connected.", con->nodeid);
1158 		goto out;
1159 	}
1160 
1161 	memset(&daddr, 0, sizeof(daddr));
1162 	result = nodeid_to_addr(con->nodeid, &daddr, NULL, true, &mark);
1163 	if (result < 0) {
1164 		log_print("no address for nodeid %d", con->nodeid);
1165 		goto out;
1166 	}
1167 
1168 	/* Create a socket to communicate with */
1169 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1170 				  SOCK_STREAM, IPPROTO_SCTP, &sock);
1171 	if (result < 0)
1172 		goto socket_err;
1173 
1174 	sock_set_mark(sock->sk, mark);
1175 
1176 	add_sock(sock, con);
1177 
1178 	/* Bind to all addresses. */
1179 	if (sctp_bind_addrs(con->sock, 0))
1180 		goto bind_err;
1181 
1182 	make_sockaddr(&daddr, dlm_config.ci_tcp_port, &addr_len);
1183 
1184 	log_print_ratelimited("connecting to %d", con->nodeid);
1185 
1186 	/* Turn off Nagle's algorithm */
1187 	sctp_sock_set_nodelay(sock->sk);
1188 
1189 	/*
1190 	 * Make sock->ops->connect() function return in specified time,
1191 	 * since O_NONBLOCK argument in connect() function does not work here,
1192 	 * then, we should restore the default value of this attribute.
1193 	 */
1194 	sock_set_sndtimeo(sock->sk, 5);
1195 	result = sock->ops->connect(sock, (struct sockaddr *)&daddr, addr_len,
1196 				   0);
1197 	sock_set_sndtimeo(sock->sk, 0);
1198 
1199 	if (result == -EINPROGRESS)
1200 		result = 0;
1201 	if (result == 0) {
1202 		if (!test_and_set_bit(CF_CONNECTED, &con->flags))
1203 			log_print("successful connected to node %d", con->nodeid);
1204 		goto out;
1205 	}
1206 
1207 bind_err:
1208 	con->sock = NULL;
1209 	sock_release(sock);
1210 
1211 socket_err:
1212 	/*
1213 	 * Some errors are fatal and this list might need adjusting. For other
1214 	 * errors we try again until the max number of retries is reached.
1215 	 */
1216 	if (result != -EHOSTUNREACH &&
1217 	    result != -ENETUNREACH &&
1218 	    result != -ENETDOWN &&
1219 	    result != -EINVAL &&
1220 	    result != -EPROTONOSUPPORT) {
1221 		log_print("connect %d try %d error %d", con->nodeid,
1222 			  con->retries, result);
1223 		mutex_unlock(&con->sock_mutex);
1224 		msleep(1000);
1225 		lowcomms_connect_sock(con);
1226 		return;
1227 	}
1228 
1229 out:
1230 	mutex_unlock(&con->sock_mutex);
1231 }
1232 
1233 /* Connect a new socket to its peer */
1234 static void tcp_connect_to_sock(struct connection *con)
1235 {
1236 	struct sockaddr_storage saddr, src_addr;
1237 	unsigned int mark;
1238 	int addr_len;
1239 	struct socket *sock = NULL;
1240 	int result;
1241 
1242 	mutex_lock(&con->sock_mutex);
1243 	if (con->retries++ > MAX_CONNECT_RETRIES)
1244 		goto out;
1245 
1246 	/* Some odd races can cause double-connects, ignore them */
1247 	if (con->sock)
1248 		goto out;
1249 
1250 	/* Create a socket to communicate with */
1251 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1252 				  SOCK_STREAM, IPPROTO_TCP, &sock);
1253 	if (result < 0)
1254 		goto out_err;
1255 
1256 	memset(&saddr, 0, sizeof(saddr));
1257 	result = nodeid_to_addr(con->nodeid, &saddr, NULL, false, &mark);
1258 	if (result < 0) {
1259 		log_print("no address for nodeid %d", con->nodeid);
1260 		goto out_err;
1261 	}
1262 
1263 	sock_set_mark(sock->sk, mark);
1264 
1265 	add_sock(sock, con);
1266 
1267 	/* Bind to our cluster-known address connecting to avoid
1268 	   routing problems */
1269 	memcpy(&src_addr, dlm_local_addr[0], sizeof(src_addr));
1270 	make_sockaddr(&src_addr, 0, &addr_len);
1271 	result = sock->ops->bind(sock, (struct sockaddr *) &src_addr,
1272 				 addr_len);
1273 	if (result < 0) {
1274 		log_print("could not bind for connect: %d", result);
1275 		/* This *may* not indicate a critical error */
1276 	}
1277 
1278 	make_sockaddr(&saddr, dlm_config.ci_tcp_port, &addr_len);
1279 
1280 	log_print_ratelimited("connecting to %d", con->nodeid);
1281 
1282 	/* Turn off Nagle's algorithm */
1283 	tcp_sock_set_nodelay(sock->sk);
1284 
1285 	result = sock->ops->connect(sock, (struct sockaddr *)&saddr, addr_len,
1286 				   O_NONBLOCK);
1287 	if (result == -EINPROGRESS)
1288 		result = 0;
1289 	if (result == 0)
1290 		goto out;
1291 
1292 out_err:
1293 	if (con->sock) {
1294 		sock_release(con->sock);
1295 		con->sock = NULL;
1296 	} else if (sock) {
1297 		sock_release(sock);
1298 	}
1299 	/*
1300 	 * Some errors are fatal and this list might need adjusting. For other
1301 	 * errors we try again until the max number of retries is reached.
1302 	 */
1303 	if (result != -EHOSTUNREACH &&
1304 	    result != -ENETUNREACH &&
1305 	    result != -ENETDOWN &&
1306 	    result != -EINVAL &&
1307 	    result != -EPROTONOSUPPORT) {
1308 		log_print("connect %d try %d error %d", con->nodeid,
1309 			  con->retries, result);
1310 		mutex_unlock(&con->sock_mutex);
1311 		msleep(1000);
1312 		lowcomms_connect_sock(con);
1313 		return;
1314 	}
1315 out:
1316 	mutex_unlock(&con->sock_mutex);
1317 	return;
1318 }
1319 
1320 /* On error caller must run dlm_close_sock() for the
1321  * listen connection socket.
1322  */
1323 static int tcp_create_listen_sock(struct listen_connection *con,
1324 				  struct sockaddr_storage *saddr)
1325 {
1326 	struct socket *sock = NULL;
1327 	int result = 0;
1328 	int addr_len;
1329 
1330 	if (dlm_local_addr[0]->ss_family == AF_INET)
1331 		addr_len = sizeof(struct sockaddr_in);
1332 	else
1333 		addr_len = sizeof(struct sockaddr_in6);
1334 
1335 	/* Create a socket to communicate with */
1336 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1337 				  SOCK_STREAM, IPPROTO_TCP, &sock);
1338 	if (result < 0) {
1339 		log_print("Can't create listening comms socket");
1340 		goto create_out;
1341 	}
1342 
1343 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1344 
1345 	/* Turn off Nagle's algorithm */
1346 	tcp_sock_set_nodelay(sock->sk);
1347 
1348 	sock_set_reuseaddr(sock->sk);
1349 
1350 	add_listen_sock(sock, con);
1351 
1352 	/* Bind to our port */
1353 	make_sockaddr(saddr, dlm_config.ci_tcp_port, &addr_len);
1354 	result = sock->ops->bind(sock, (struct sockaddr *) saddr, addr_len);
1355 	if (result < 0) {
1356 		log_print("Can't bind to port %d", dlm_config.ci_tcp_port);
1357 		goto create_out;
1358 	}
1359 	sock_set_keepalive(sock->sk);
1360 
1361 	result = sock->ops->listen(sock, 5);
1362 	if (result < 0) {
1363 		log_print("Can't listen on port %d", dlm_config.ci_tcp_port);
1364 		goto create_out;
1365 	}
1366 
1367 	return 0;
1368 
1369 create_out:
1370 	return result;
1371 }
1372 
1373 /* Get local addresses */
1374 static void init_local(void)
1375 {
1376 	struct sockaddr_storage sas, *addr;
1377 	int i;
1378 
1379 	dlm_local_count = 0;
1380 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1381 		if (dlm_our_addr(&sas, i))
1382 			break;
1383 
1384 		addr = kmemdup(&sas, sizeof(*addr), GFP_NOFS);
1385 		if (!addr)
1386 			break;
1387 		dlm_local_addr[dlm_local_count++] = addr;
1388 	}
1389 }
1390 
1391 static void deinit_local(void)
1392 {
1393 	int i;
1394 
1395 	for (i = 0; i < dlm_local_count; i++)
1396 		kfree(dlm_local_addr[i]);
1397 }
1398 
1399 /* Initialise SCTP socket and bind to all interfaces
1400  * On error caller must run dlm_close_sock() for the
1401  * listen connection socket.
1402  */
1403 static int sctp_listen_for_all(struct listen_connection *con)
1404 {
1405 	struct socket *sock = NULL;
1406 	int result = -EINVAL;
1407 
1408 	log_print("Using SCTP for communications");
1409 
1410 	result = sock_create_kern(&init_net, dlm_local_addr[0]->ss_family,
1411 				  SOCK_STREAM, IPPROTO_SCTP, &sock);
1412 	if (result < 0) {
1413 		log_print("Can't create comms socket, check SCTP is loaded");
1414 		goto out;
1415 	}
1416 
1417 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1418 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1419 	sctp_sock_set_nodelay(sock->sk);
1420 
1421 	add_listen_sock(sock, con);
1422 
1423 	/* Bind to all addresses. */
1424 	result = sctp_bind_addrs(con->sock, dlm_config.ci_tcp_port);
1425 	if (result < 0)
1426 		goto out;
1427 
1428 	result = sock->ops->listen(sock, 5);
1429 	if (result < 0) {
1430 		log_print("Can't set socket listening");
1431 		goto out;
1432 	}
1433 
1434 	return 0;
1435 
1436 out:
1437 	return result;
1438 }
1439 
1440 static int tcp_listen_for_all(void)
1441 {
1442 	/* We don't support multi-homed hosts */
1443 	if (dlm_local_count > 1) {
1444 		log_print("TCP protocol can't handle multi-homed hosts, "
1445 			  "try SCTP");
1446 		return -EINVAL;
1447 	}
1448 
1449 	log_print("Using TCP for communications");
1450 
1451 	return tcp_create_listen_sock(&listen_con, dlm_local_addr[0]);
1452 }
1453 
1454 
1455 
1456 static struct writequeue_entry *new_writequeue_entry(struct connection *con,
1457 						     gfp_t allocation)
1458 {
1459 	struct writequeue_entry *entry;
1460 
1461 	entry = kzalloc(sizeof(*entry), allocation);
1462 	if (!entry)
1463 		return NULL;
1464 
1465 	entry->page = alloc_page(allocation | __GFP_ZERO);
1466 	if (!entry->page) {
1467 		kfree(entry);
1468 		return NULL;
1469 	}
1470 
1471 	entry->con = con;
1472 	entry->users = 1;
1473 	kref_init(&entry->ref);
1474 	INIT_LIST_HEAD(&entry->msgs);
1475 
1476 	return entry;
1477 }
1478 
1479 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1480 					     gfp_t allocation, char **ppc,
1481 					     void (*cb)(struct dlm_mhandle *mh),
1482 					     struct dlm_mhandle *mh)
1483 {
1484 	struct writequeue_entry *e;
1485 
1486 	spin_lock(&con->writequeue_lock);
1487 	if (!list_empty(&con->writequeue)) {
1488 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1489 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1490 			kref_get(&e->ref);
1491 
1492 			*ppc = page_address(e->page) + e->end;
1493 			if (cb)
1494 				cb(mh);
1495 
1496 			e->end += len;
1497 			e->users++;
1498 			spin_unlock(&con->writequeue_lock);
1499 
1500 			return e;
1501 		}
1502 	}
1503 	spin_unlock(&con->writequeue_lock);
1504 
1505 	e = new_writequeue_entry(con, allocation);
1506 	if (!e)
1507 		return NULL;
1508 
1509 	kref_get(&e->ref);
1510 	*ppc = page_address(e->page);
1511 	e->end += len;
1512 	atomic_inc(&con->writequeue_cnt);
1513 
1514 	spin_lock(&con->writequeue_lock);
1515 	if (cb)
1516 		cb(mh);
1517 
1518 	list_add_tail(&e->list, &con->writequeue);
1519 	spin_unlock(&con->writequeue_lock);
1520 
1521 	return e;
1522 };
1523 
1524 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1525 						gfp_t allocation, char **ppc,
1526 						void (*cb)(struct dlm_mhandle *mh),
1527 						struct dlm_mhandle *mh)
1528 {
1529 	struct writequeue_entry *e;
1530 	struct dlm_msg *msg;
1531 
1532 	msg = kzalloc(sizeof(*msg), allocation);
1533 	if (!msg)
1534 		return NULL;
1535 
1536 	kref_init(&msg->ref);
1537 
1538 	e = new_wq_entry(con, len, allocation, ppc, cb, mh);
1539 	if (!e) {
1540 		kfree(msg);
1541 		return NULL;
1542 	}
1543 
1544 	msg->ppc = *ppc;
1545 	msg->len = len;
1546 	msg->entry = e;
1547 
1548 	return msg;
1549 }
1550 
1551 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1552 				     char **ppc, void (*cb)(struct dlm_mhandle *mh),
1553 				     struct dlm_mhandle *mh)
1554 {
1555 	struct connection *con;
1556 	struct dlm_msg *msg;
1557 	int idx;
1558 
1559 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1560 	    len < sizeof(struct dlm_header)) {
1561 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1562 		log_print("failed to allocate a buffer of size %d", len);
1563 		WARN_ON(1);
1564 		return NULL;
1565 	}
1566 
1567 	idx = srcu_read_lock(&connections_srcu);
1568 	con = nodeid2con(nodeid, allocation);
1569 	if (!con) {
1570 		srcu_read_unlock(&connections_srcu, idx);
1571 		return NULL;
1572 	}
1573 
1574 	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, mh);
1575 	if (!msg) {
1576 		srcu_read_unlock(&connections_srcu, idx);
1577 		return NULL;
1578 	}
1579 
1580 	/* we assume if successful commit must called */
1581 	msg->idx = idx;
1582 	return msg;
1583 }
1584 
1585 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1586 {
1587 	struct writequeue_entry *e = msg->entry;
1588 	struct connection *con = e->con;
1589 	int users;
1590 
1591 	spin_lock(&con->writequeue_lock);
1592 	kref_get(&msg->ref);
1593 	list_add(&msg->list, &e->msgs);
1594 
1595 	users = --e->users;
1596 	if (users)
1597 		goto out;
1598 
1599 	e->len = DLM_WQ_LENGTH_BYTES(e);
1600 	spin_unlock(&con->writequeue_lock);
1601 
1602 	queue_work(send_workqueue, &con->swork);
1603 	return;
1604 
1605 out:
1606 	spin_unlock(&con->writequeue_lock);
1607 	return;
1608 }
1609 
1610 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1611 {
1612 	_dlm_lowcomms_commit_msg(msg);
1613 	srcu_read_unlock(&connections_srcu, msg->idx);
1614 }
1615 
1616 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1617 {
1618 	kref_put(&msg->ref, dlm_msg_release);
1619 }
1620 
1621 /* does not held connections_srcu, usage workqueue only */
1622 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1623 {
1624 	struct dlm_msg *msg_resend;
1625 	char *ppc;
1626 
1627 	if (msg->retransmit)
1628 		return 1;
1629 
1630 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1631 					      GFP_ATOMIC, &ppc, NULL, NULL);
1632 	if (!msg_resend)
1633 		return -ENOMEM;
1634 
1635 	msg->retransmit = true;
1636 	kref_get(&msg->ref);
1637 	msg_resend->orig_msg = msg;
1638 
1639 	memcpy(ppc, msg->ppc, msg->len);
1640 	_dlm_lowcomms_commit_msg(msg_resend);
1641 	dlm_lowcomms_put_msg(msg_resend);
1642 
1643 	return 0;
1644 }
1645 
1646 /* Send a message */
1647 static void send_to_sock(struct connection *con)
1648 {
1649 	int ret = 0;
1650 	const int msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
1651 	struct writequeue_entry *e;
1652 	int len, offset;
1653 	int count = 0;
1654 
1655 	mutex_lock(&con->sock_mutex);
1656 	if (con->sock == NULL)
1657 		goto out_connect;
1658 
1659 	spin_lock(&con->writequeue_lock);
1660 	for (;;) {
1661 		if (list_empty(&con->writequeue))
1662 			break;
1663 
1664 		e = list_first_entry(&con->writequeue, struct writequeue_entry, list);
1665 		len = e->len;
1666 		offset = e->offset;
1667 		BUG_ON(len == 0 && e->users == 0);
1668 		spin_unlock(&con->writequeue_lock);
1669 
1670 		ret = 0;
1671 		if (len) {
1672 			ret = kernel_sendpage(con->sock, e->page, offset, len,
1673 					      msg_flags);
1674 			if (ret == -EAGAIN || ret == 0) {
1675 				if (ret == -EAGAIN &&
1676 				    test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1677 				    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1678 					/* Notify TCP that we're limited by the
1679 					 * application window size.
1680 					 */
1681 					set_bit(SOCK_NOSPACE, &con->sock->flags);
1682 					con->sock->sk->sk_write_pending++;
1683 				}
1684 				cond_resched();
1685 				goto out;
1686 			} else if (ret < 0)
1687 				goto out;
1688 		}
1689 
1690 		/* Don't starve people filling buffers */
1691 		if (++count >= MAX_SEND_MSG_COUNT) {
1692 			cond_resched();
1693 			count = 0;
1694 		}
1695 
1696 		spin_lock(&con->writequeue_lock);
1697 		writequeue_entry_complete(e, ret);
1698 	}
1699 	spin_unlock(&con->writequeue_lock);
1700 
1701 	/* close if we got EOF */
1702 	if (test_and_clear_bit(CF_EOF, &con->flags)) {
1703 		mutex_unlock(&con->sock_mutex);
1704 		close_connection(con, false, false, true);
1705 
1706 		/* handling for tcp shutdown */
1707 		clear_bit(CF_SHUTDOWN, &con->flags);
1708 		wake_up(&con->shutdown_wait);
1709 	} else {
1710 		mutex_unlock(&con->sock_mutex);
1711 	}
1712 
1713 	return;
1714 
1715 out:
1716 	mutex_unlock(&con->sock_mutex);
1717 	return;
1718 
1719 out_connect:
1720 	mutex_unlock(&con->sock_mutex);
1721 	queue_work(send_workqueue, &con->swork);
1722 	cond_resched();
1723 }
1724 
1725 static void clean_one_writequeue(struct connection *con)
1726 {
1727 	struct writequeue_entry *e, *safe;
1728 
1729 	spin_lock(&con->writequeue_lock);
1730 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1731 		free_entry(e);
1732 	}
1733 	spin_unlock(&con->writequeue_lock);
1734 }
1735 
1736 /* Called from recovery when it knows that a node has
1737    left the cluster */
1738 int dlm_lowcomms_close(int nodeid)
1739 {
1740 	struct connection *con;
1741 	struct dlm_node_addr *na;
1742 	int idx;
1743 
1744 	log_print("closing connection to node %d", nodeid);
1745 	idx = srcu_read_lock(&connections_srcu);
1746 	con = nodeid2con(nodeid, 0);
1747 	if (con) {
1748 		set_bit(CF_CLOSE, &con->flags);
1749 		close_connection(con, true, true, true);
1750 		clean_one_writequeue(con);
1751 		if (con->othercon)
1752 			clean_one_writequeue(con->othercon);
1753 	}
1754 	srcu_read_unlock(&connections_srcu, idx);
1755 
1756 	spin_lock(&dlm_node_addrs_spin);
1757 	na = find_node_addr(nodeid);
1758 	if (na) {
1759 		list_del(&na->list);
1760 		while (na->addr_count--)
1761 			kfree(na->addr[na->addr_count]);
1762 		kfree(na);
1763 	}
1764 	spin_unlock(&dlm_node_addrs_spin);
1765 
1766 	return 0;
1767 }
1768 
1769 /* Receive workqueue function */
1770 static void process_recv_sockets(struct work_struct *work)
1771 {
1772 	struct connection *con = container_of(work, struct connection, rwork);
1773 	int err;
1774 
1775 	clear_bit(CF_READ_PENDING, &con->flags);
1776 	do {
1777 		err = receive_from_sock(con);
1778 	} while (!err);
1779 }
1780 
1781 static void process_listen_recv_socket(struct work_struct *work)
1782 {
1783 	accept_from_sock(&listen_con);
1784 }
1785 
1786 /* Send workqueue function */
1787 static void process_send_sockets(struct work_struct *work)
1788 {
1789 	struct connection *con = container_of(work, struct connection, swork);
1790 
1791 	WARN_ON(test_bit(CF_IS_OTHERCON, &con->flags));
1792 
1793 	clear_bit(CF_WRITE_PENDING, &con->flags);
1794 
1795 	if (test_and_clear_bit(CF_RECONNECT, &con->flags)) {
1796 		close_connection(con, false, false, true);
1797 		dlm_midcomms_unack_msg_resend(con->nodeid);
1798 	}
1799 
1800 	if (con->sock == NULL) { /* not mutex protected so check it inside too */
1801 		if (test_and_clear_bit(CF_DELAY_CONNECT, &con->flags))
1802 			msleep(1000);
1803 		con->connect_action(con);
1804 	}
1805 	if (!list_empty(&con->writequeue))
1806 		send_to_sock(con);
1807 }
1808 
1809 static void work_stop(void)
1810 {
1811 	if (recv_workqueue) {
1812 		destroy_workqueue(recv_workqueue);
1813 		recv_workqueue = NULL;
1814 	}
1815 
1816 	if (send_workqueue) {
1817 		destroy_workqueue(send_workqueue);
1818 		send_workqueue = NULL;
1819 	}
1820 }
1821 
1822 static int work_start(void)
1823 {
1824 	recv_workqueue = alloc_ordered_workqueue("dlm_recv", WQ_MEM_RECLAIM);
1825 	if (!recv_workqueue) {
1826 		log_print("can't start dlm_recv");
1827 		return -ENOMEM;
1828 	}
1829 
1830 	send_workqueue = alloc_ordered_workqueue("dlm_send", WQ_MEM_RECLAIM);
1831 	if (!send_workqueue) {
1832 		log_print("can't start dlm_send");
1833 		destroy_workqueue(recv_workqueue);
1834 		recv_workqueue = NULL;
1835 		return -ENOMEM;
1836 	}
1837 
1838 	return 0;
1839 }
1840 
1841 static void shutdown_conn(struct connection *con)
1842 {
1843 	if (con->shutdown_action)
1844 		con->shutdown_action(con);
1845 }
1846 
1847 void dlm_lowcomms_shutdown(void)
1848 {
1849 	int idx;
1850 
1851 	/* Set all the flags to prevent any
1852 	 * socket activity.
1853 	 */
1854 	dlm_allow_conn = 0;
1855 
1856 	if (recv_workqueue)
1857 		flush_workqueue(recv_workqueue);
1858 	if (send_workqueue)
1859 		flush_workqueue(send_workqueue);
1860 
1861 	dlm_close_sock(&listen_con.sock);
1862 
1863 	idx = srcu_read_lock(&connections_srcu);
1864 	foreach_conn(shutdown_conn);
1865 	srcu_read_unlock(&connections_srcu, idx);
1866 }
1867 
1868 static void _stop_conn(struct connection *con, bool and_other)
1869 {
1870 	mutex_lock(&con->sock_mutex);
1871 	set_bit(CF_CLOSE, &con->flags);
1872 	set_bit(CF_READ_PENDING, &con->flags);
1873 	set_bit(CF_WRITE_PENDING, &con->flags);
1874 	if (con->sock && con->sock->sk) {
1875 		write_lock_bh(&con->sock->sk->sk_callback_lock);
1876 		con->sock->sk->sk_user_data = NULL;
1877 		write_unlock_bh(&con->sock->sk->sk_callback_lock);
1878 	}
1879 	if (con->othercon && and_other)
1880 		_stop_conn(con->othercon, false);
1881 	mutex_unlock(&con->sock_mutex);
1882 }
1883 
1884 static void stop_conn(struct connection *con)
1885 {
1886 	_stop_conn(con, true);
1887 }
1888 
1889 static void connection_release(struct rcu_head *rcu)
1890 {
1891 	struct connection *con = container_of(rcu, struct connection, rcu);
1892 
1893 	kfree(con->rx_buf);
1894 	kfree(con);
1895 }
1896 
1897 static void free_conn(struct connection *con)
1898 {
1899 	close_connection(con, true, true, true);
1900 	spin_lock(&connections_lock);
1901 	hlist_del_rcu(&con->list);
1902 	spin_unlock(&connections_lock);
1903 	if (con->othercon) {
1904 		clean_one_writequeue(con->othercon);
1905 		call_srcu(&connections_srcu, &con->othercon->rcu,
1906 			  connection_release);
1907 	}
1908 	clean_one_writequeue(con);
1909 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1910 }
1911 
1912 static void work_flush(void)
1913 {
1914 	int ok;
1915 	int i;
1916 	struct connection *con;
1917 
1918 	do {
1919 		ok = 1;
1920 		foreach_conn(stop_conn);
1921 		if (recv_workqueue)
1922 			flush_workqueue(recv_workqueue);
1923 		if (send_workqueue)
1924 			flush_workqueue(send_workqueue);
1925 		for (i = 0; i < CONN_HASH_SIZE && ok; i++) {
1926 			hlist_for_each_entry_rcu(con, &connection_hash[i],
1927 						 list) {
1928 				ok &= test_bit(CF_READ_PENDING, &con->flags);
1929 				ok &= test_bit(CF_WRITE_PENDING, &con->flags);
1930 				if (con->othercon) {
1931 					ok &= test_bit(CF_READ_PENDING,
1932 						       &con->othercon->flags);
1933 					ok &= test_bit(CF_WRITE_PENDING,
1934 						       &con->othercon->flags);
1935 				}
1936 			}
1937 		}
1938 	} while (!ok);
1939 }
1940 
1941 void dlm_lowcomms_stop(void)
1942 {
1943 	int idx;
1944 
1945 	idx = srcu_read_lock(&connections_srcu);
1946 	work_flush();
1947 	foreach_conn(free_conn);
1948 	srcu_read_unlock(&connections_srcu, idx);
1949 	work_stop();
1950 	deinit_local();
1951 }
1952 
1953 int dlm_lowcomms_start(void)
1954 {
1955 	int error = -EINVAL;
1956 	int i;
1957 
1958 	for (i = 0; i < CONN_HASH_SIZE; i++)
1959 		INIT_HLIST_HEAD(&connection_hash[i]);
1960 
1961 	init_local();
1962 	if (!dlm_local_count) {
1963 		error = -ENOTCONN;
1964 		log_print("no local IP address has been set");
1965 		goto fail;
1966 	}
1967 
1968 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1969 
1970 	error = work_start();
1971 	if (error)
1972 		goto fail_local;
1973 
1974 	dlm_allow_conn = 1;
1975 
1976 	/* Start listening */
1977 	switch (dlm_config.ci_protocol) {
1978 	case DLM_PROTO_TCP:
1979 		error = tcp_listen_for_all();
1980 		break;
1981 	case DLM_PROTO_SCTP:
1982 		error = sctp_listen_for_all(&listen_con);
1983 		break;
1984 	default:
1985 		log_print("Invalid protocol identifier %d set",
1986 			  dlm_config.ci_protocol);
1987 		error = -EINVAL;
1988 		break;
1989 	}
1990 	if (error)
1991 		goto fail_unlisten;
1992 
1993 	return 0;
1994 
1995 fail_unlisten:
1996 	dlm_allow_conn = 0;
1997 	dlm_close_sock(&listen_con.sock);
1998 	work_stop();
1999 fail_local:
2000 	deinit_local();
2001 fail:
2002 	return error;
2003 }
2004 
2005 void dlm_lowcomms_exit(void)
2006 {
2007 	struct dlm_node_addr *na, *safe;
2008 
2009 	spin_lock(&dlm_node_addrs_spin);
2010 	list_for_each_entry_safe(na, safe, &dlm_node_addrs, list) {
2011 		list_del(&na->list);
2012 		while (na->addr_count--)
2013 			kfree(na->addr[na->addr_count]);
2014 		kfree(na);
2015 	}
2016 	spin_unlock(&dlm_node_addrs_spin);
2017 }
2018