xref: /linux/fs/dlm/lowcomms.c (revision c5288cda69ee2d8607f5026bd599a5cebf0ee783)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
4 **
5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
6 **  Copyright (C) 2004-2009 Red Hat, Inc.  All rights reserved.
7 **
8 **
9 *******************************************************************************
10 ******************************************************************************/
11 
12 /*
13  * lowcomms.c
14  *
15  * This is the "low-level" comms layer.
16  *
17  * It is responsible for sending/receiving messages
18  * from other nodes in the cluster.
19  *
20  * Cluster nodes are referred to by their nodeids. nodeids are
21  * simply 32 bit numbers to the locking module - if they need to
22  * be expanded for the cluster infrastructure then that is its
23  * responsibility. It is this layer's
24  * responsibility to resolve these into IP address or
25  * whatever it needs for inter-node communication.
26  *
27  * The comms level is two kernel threads that deal mainly with
28  * the receiving of messages from other nodes and passing them
29  * up to the mid-level comms layer (which understands the
30  * message format) for execution by the locking core, and
31  * a send thread which does all the setting up of connections
32  * to remote nodes and the sending of data. Threads are not allowed
33  * to send their own data because it may cause them to wait in times
34  * of high load. Also, this way, the sending thread can collect together
35  * messages bound for one node and send them in one block.
36  *
37  * lowcomms will choose to use either TCP or SCTP as its transport layer
38  * depending on the configuration variable 'protocol'. This should be set
39  * to 0 (default) for TCP or 1 for SCTP. It should be configured using a
40  * cluster-wide mechanism as it must be the same on all nodes of the cluster
41  * for the DLM to function.
42  *
43  */
44 
45 #include <asm/ioctls.h>
46 #include <net/sock.h>
47 #include <net/tcp.h>
48 #include <linux/pagemap.h>
49 #include <linux/file.h>
50 #include <linux/mutex.h>
51 #include <linux/sctp.h>
52 #include <linux/slab.h>
53 #include <net/sctp/sctp.h>
54 #include <net/ipv6.h>
55 
56 #include <trace/events/dlm.h>
57 #include <trace/events/sock.h>
58 
59 #include "dlm_internal.h"
60 #include "lowcomms.h"
61 #include "midcomms.h"
62 #include "memory.h"
63 #include "config.h"
64 
65 #define DLM_SHUTDOWN_WAIT_TIMEOUT msecs_to_jiffies(5000)
66 #define DLM_MAX_PROCESS_BUFFERS 24
67 #define NEEDED_RMEM (4*1024*1024)
68 
69 struct connection {
70 	struct socket *sock;	/* NULL if not connected */
71 	uint32_t nodeid;	/* So we know who we are in the list */
72 	/* this semaphore is used to allow parallel recv/send in read
73 	 * lock mode. When we release a sock we need to held the write lock.
74 	 *
75 	 * However this is locking code and not nice. When we remove the
76 	 * othercon handling we can look into other mechanism to synchronize
77 	 * io handling to call sock_release() at the right time.
78 	 */
79 	struct rw_semaphore sock_lock;
80 	unsigned long flags;
81 #define CF_APP_LIMITED 0
82 #define CF_RECV_PENDING 1
83 #define CF_SEND_PENDING 2
84 #define CF_RECV_INTR 3
85 #define CF_IO_STOP 4
86 #define CF_IS_OTHERCON 5
87 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
88 	spinlock_t writequeue_lock;
89 	int retries;
90 	struct hlist_node list;
91 	/* due some connect()/accept() races we currently have this cross over
92 	 * connection attempt second connection for one node.
93 	 *
94 	 * There is a solution to avoid the race by introducing a connect
95 	 * rule as e.g. our_nodeid > nodeid_to_connect who is allowed to
96 	 * connect. Otherside can connect but will only be considered that
97 	 * the other side wants to have a reconnect.
98 	 *
99 	 * However changing to this behaviour will break backwards compatible.
100 	 * In a DLM protocol major version upgrade we should remove this!
101 	 */
102 	struct connection *othercon;
103 	struct work_struct rwork; /* receive worker */
104 	struct work_struct swork; /* send worker */
105 	wait_queue_head_t shutdown_wait;
106 	unsigned char rx_leftover_buf[DLM_MAX_SOCKET_BUFSIZE];
107 	int rx_leftover;
108 	int mark;
109 	int addr_count;
110 	int curr_addr_index;
111 	struct sockaddr_storage addr[DLM_MAX_ADDR_COUNT];
112 	spinlock_t addrs_lock;
113 	struct rcu_head rcu;
114 };
115 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
116 
117 struct listen_connection {
118 	struct socket *sock;
119 	struct work_struct rwork;
120 };
121 
122 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
123 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
124 
125 /* An entry waiting to be sent */
126 struct writequeue_entry {
127 	struct list_head list;
128 	struct page *page;
129 	int offset;
130 	int len;
131 	int end;
132 	int users;
133 	bool dirty;
134 	struct connection *con;
135 	struct list_head msgs;
136 	struct kref ref;
137 };
138 
139 struct dlm_msg {
140 	struct writequeue_entry *entry;
141 	struct dlm_msg *orig_msg;
142 	bool retransmit;
143 	void *ppc;
144 	int len;
145 	int idx; /* new()/commit() idx exchange */
146 
147 	struct list_head list;
148 	struct kref ref;
149 };
150 
151 struct processqueue_entry {
152 	unsigned char *buf;
153 	int nodeid;
154 	int buflen;
155 
156 	struct list_head list;
157 };
158 
159 struct dlm_proto_ops {
160 	bool try_new_addr;
161 	const char *name;
162 	int proto;
163 
164 	int (*connect)(struct connection *con, struct socket *sock,
165 		       struct sockaddr *addr, int addr_len);
166 	void (*sockopts)(struct socket *sock);
167 	int (*bind)(struct socket *sock);
168 	int (*listen_validate)(void);
169 	void (*listen_sockopts)(struct socket *sock);
170 	int (*listen_bind)(struct socket *sock);
171 };
172 
173 static struct listen_sock_callbacks {
174 	void (*sk_error_report)(struct sock *);
175 	void (*sk_data_ready)(struct sock *);
176 	void (*sk_state_change)(struct sock *);
177 	void (*sk_write_space)(struct sock *);
178 } listen_sock;
179 
180 static struct listen_connection listen_con;
181 static struct sockaddr_storage dlm_local_addr[DLM_MAX_ADDR_COUNT];
182 static int dlm_local_count;
183 
184 /* Work queues */
185 static struct workqueue_struct *io_workqueue;
186 static struct workqueue_struct *process_workqueue;
187 
188 static struct hlist_head connection_hash[CONN_HASH_SIZE];
189 static DEFINE_SPINLOCK(connections_lock);
190 DEFINE_STATIC_SRCU(connections_srcu);
191 
192 static const struct dlm_proto_ops *dlm_proto_ops;
193 
194 #define DLM_IO_SUCCESS 0
195 #define DLM_IO_END 1
196 #define DLM_IO_EOF 2
197 #define DLM_IO_RESCHED 3
198 #define DLM_IO_FLUSH 4
199 
200 static void process_recv_sockets(struct work_struct *work);
201 static void process_send_sockets(struct work_struct *work);
202 static void process_dlm_messages(struct work_struct *work);
203 
204 static DECLARE_WORK(process_work, process_dlm_messages);
205 static DEFINE_SPINLOCK(processqueue_lock);
206 static bool process_dlm_messages_pending;
207 static DECLARE_WAIT_QUEUE_HEAD(processqueue_wq);
208 static atomic_t processqueue_count;
209 static LIST_HEAD(processqueue);
210 
211 bool dlm_lowcomms_is_running(void)
212 {
213 	return !!listen_con.sock;
214 }
215 
216 static void lowcomms_queue_swork(struct connection *con)
217 {
218 	assert_spin_locked(&con->writequeue_lock);
219 
220 	if (!test_bit(CF_IO_STOP, &con->flags) &&
221 	    !test_bit(CF_APP_LIMITED, &con->flags) &&
222 	    !test_and_set_bit(CF_SEND_PENDING, &con->flags))
223 		queue_work(io_workqueue, &con->swork);
224 }
225 
226 static void lowcomms_queue_rwork(struct connection *con)
227 {
228 #ifdef CONFIG_LOCKDEP
229 	WARN_ON_ONCE(!lockdep_sock_is_held(con->sock->sk));
230 #endif
231 
232 	if (!test_bit(CF_IO_STOP, &con->flags) &&
233 	    !test_and_set_bit(CF_RECV_PENDING, &con->flags))
234 		queue_work(io_workqueue, &con->rwork);
235 }
236 
237 static void writequeue_entry_ctor(void *data)
238 {
239 	struct writequeue_entry *entry = data;
240 
241 	INIT_LIST_HEAD(&entry->msgs);
242 }
243 
244 struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void)
245 {
246 	return kmem_cache_create("dlm_writequeue", sizeof(struct writequeue_entry),
247 				 0, 0, writequeue_entry_ctor);
248 }
249 
250 struct kmem_cache *dlm_lowcomms_msg_cache_create(void)
251 {
252 	return KMEM_CACHE(dlm_msg, 0);
253 }
254 
255 /* need to held writequeue_lock */
256 static struct writequeue_entry *con_next_wq(struct connection *con)
257 {
258 	struct writequeue_entry *e;
259 
260 	e = list_first_entry_or_null(&con->writequeue, struct writequeue_entry,
261 				     list);
262 	/* if len is zero nothing is to send, if there are users filling
263 	 * buffers we wait until the users are done so we can send more.
264 	 */
265 	if (!e || e->users || e->len == 0)
266 		return NULL;
267 
268 	return e;
269 }
270 
271 static struct connection *__find_con(int nodeid, int r)
272 {
273 	struct connection *con;
274 
275 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
276 		if (con->nodeid == nodeid)
277 			return con;
278 	}
279 
280 	return NULL;
281 }
282 
283 static void dlm_con_init(struct connection *con, int nodeid)
284 {
285 	con->nodeid = nodeid;
286 	init_rwsem(&con->sock_lock);
287 	INIT_LIST_HEAD(&con->writequeue);
288 	spin_lock_init(&con->writequeue_lock);
289 	INIT_WORK(&con->swork, process_send_sockets);
290 	INIT_WORK(&con->rwork, process_recv_sockets);
291 	spin_lock_init(&con->addrs_lock);
292 	init_waitqueue_head(&con->shutdown_wait);
293 }
294 
295 /*
296  * If 'allocation' is zero then we don't attempt to create a new
297  * connection structure for this node.
298  */
299 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
300 {
301 	struct connection *con, *tmp;
302 	int r;
303 
304 	r = nodeid_hash(nodeid);
305 	con = __find_con(nodeid, r);
306 	if (con || !alloc)
307 		return con;
308 
309 	con = kzalloc(sizeof(*con), alloc);
310 	if (!con)
311 		return NULL;
312 
313 	dlm_con_init(con, nodeid);
314 
315 	spin_lock(&connections_lock);
316 	/* Because multiple workqueues/threads calls this function it can
317 	 * race on multiple cpu's. Instead of locking hot path __find_con()
318 	 * we just check in rare cases of recently added nodes again
319 	 * under protection of connections_lock. If this is the case we
320 	 * abort our connection creation and return the existing connection.
321 	 */
322 	tmp = __find_con(nodeid, r);
323 	if (tmp) {
324 		spin_unlock(&connections_lock);
325 		kfree(con);
326 		return tmp;
327 	}
328 
329 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
330 	spin_unlock(&connections_lock);
331 
332 	return con;
333 }
334 
335 static int addr_compare(const struct sockaddr_storage *x,
336 			const struct sockaddr_storage *y)
337 {
338 	switch (x->ss_family) {
339 	case AF_INET: {
340 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
341 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
342 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
343 			return 0;
344 		if (sinx->sin_port != siny->sin_port)
345 			return 0;
346 		break;
347 	}
348 	case AF_INET6: {
349 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
350 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
351 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
352 			return 0;
353 		if (sinx->sin6_port != siny->sin6_port)
354 			return 0;
355 		break;
356 	}
357 	default:
358 		return 0;
359 	}
360 	return 1;
361 }
362 
363 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
364 			  struct sockaddr *sa_out, bool try_new_addr,
365 			  unsigned int *mark)
366 {
367 	struct sockaddr_storage sas;
368 	struct connection *con;
369 	int idx;
370 
371 	if (!dlm_local_count)
372 		return -1;
373 
374 	idx = srcu_read_lock(&connections_srcu);
375 	con = nodeid2con(nodeid, 0);
376 	if (!con) {
377 		srcu_read_unlock(&connections_srcu, idx);
378 		return -ENOENT;
379 	}
380 
381 	spin_lock(&con->addrs_lock);
382 	if (!con->addr_count) {
383 		spin_unlock(&con->addrs_lock);
384 		srcu_read_unlock(&connections_srcu, idx);
385 		return -ENOENT;
386 	}
387 
388 	memcpy(&sas, &con->addr[con->curr_addr_index],
389 	       sizeof(struct sockaddr_storage));
390 
391 	if (try_new_addr) {
392 		con->curr_addr_index++;
393 		if (con->curr_addr_index == con->addr_count)
394 			con->curr_addr_index = 0;
395 	}
396 
397 	*mark = con->mark;
398 	spin_unlock(&con->addrs_lock);
399 
400 	if (sas_out)
401 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
402 
403 	if (!sa_out) {
404 		srcu_read_unlock(&connections_srcu, idx);
405 		return 0;
406 	}
407 
408 	if (dlm_local_addr[0].ss_family == AF_INET) {
409 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
410 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
411 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
412 	} else {
413 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
414 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
415 		ret6->sin6_addr = in6->sin6_addr;
416 	}
417 
418 	srcu_read_unlock(&connections_srcu, idx);
419 	return 0;
420 }
421 
422 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
423 			  unsigned int *mark)
424 {
425 	struct connection *con;
426 	int i, idx, addr_i;
427 
428 	idx = srcu_read_lock(&connections_srcu);
429 	for (i = 0; i < CONN_HASH_SIZE; i++) {
430 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
431 			WARN_ON_ONCE(!con->addr_count);
432 
433 			spin_lock(&con->addrs_lock);
434 			for (addr_i = 0; addr_i < con->addr_count; addr_i++) {
435 				if (addr_compare(&con->addr[addr_i], addr)) {
436 					*nodeid = con->nodeid;
437 					*mark = con->mark;
438 					spin_unlock(&con->addrs_lock);
439 					srcu_read_unlock(&connections_srcu, idx);
440 					return 0;
441 				}
442 			}
443 			spin_unlock(&con->addrs_lock);
444 		}
445 	}
446 	srcu_read_unlock(&connections_srcu, idx);
447 
448 	return -ENOENT;
449 }
450 
451 static bool dlm_lowcomms_con_has_addr(const struct connection *con,
452 				      const struct sockaddr_storage *addr)
453 {
454 	int i;
455 
456 	for (i = 0; i < con->addr_count; i++) {
457 		if (addr_compare(&con->addr[i], addr))
458 			return true;
459 	}
460 
461 	return false;
462 }
463 
464 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
465 {
466 	struct connection *con;
467 	bool ret, idx;
468 
469 	idx = srcu_read_lock(&connections_srcu);
470 	con = nodeid2con(nodeid, GFP_NOFS);
471 	if (!con) {
472 		srcu_read_unlock(&connections_srcu, idx);
473 		return -ENOMEM;
474 	}
475 
476 	spin_lock(&con->addrs_lock);
477 	if (!con->addr_count) {
478 		memcpy(&con->addr[0], addr, sizeof(*addr));
479 		con->addr_count = 1;
480 		con->mark = dlm_config.ci_mark;
481 		spin_unlock(&con->addrs_lock);
482 		srcu_read_unlock(&connections_srcu, idx);
483 		return 0;
484 	}
485 
486 	ret = dlm_lowcomms_con_has_addr(con, addr);
487 	if (ret) {
488 		spin_unlock(&con->addrs_lock);
489 		srcu_read_unlock(&connections_srcu, idx);
490 		return -EEXIST;
491 	}
492 
493 	if (con->addr_count >= DLM_MAX_ADDR_COUNT) {
494 		spin_unlock(&con->addrs_lock);
495 		srcu_read_unlock(&connections_srcu, idx);
496 		return -ENOSPC;
497 	}
498 
499 	memcpy(&con->addr[con->addr_count++], addr, sizeof(*addr));
500 	srcu_read_unlock(&connections_srcu, idx);
501 	spin_unlock(&con->addrs_lock);
502 	return 0;
503 }
504 
505 /* Data available on socket or listen socket received a connect */
506 static void lowcomms_data_ready(struct sock *sk)
507 {
508 	struct connection *con = sock2con(sk);
509 
510 	trace_sk_data_ready(sk);
511 
512 	set_bit(CF_RECV_INTR, &con->flags);
513 	lowcomms_queue_rwork(con);
514 }
515 
516 static void lowcomms_write_space(struct sock *sk)
517 {
518 	struct connection *con = sock2con(sk);
519 
520 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
521 
522 	spin_lock_bh(&con->writequeue_lock);
523 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
524 		con->sock->sk->sk_write_pending--;
525 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
526 	}
527 
528 	lowcomms_queue_swork(con);
529 	spin_unlock_bh(&con->writequeue_lock);
530 }
531 
532 static void lowcomms_state_change(struct sock *sk)
533 {
534 	/* SCTP layer is not calling sk_data_ready when the connection
535 	 * is done, so we catch the signal through here.
536 	 */
537 	if (sk->sk_shutdown == RCV_SHUTDOWN)
538 		lowcomms_data_ready(sk);
539 }
540 
541 static void lowcomms_listen_data_ready(struct sock *sk)
542 {
543 	trace_sk_data_ready(sk);
544 
545 	queue_work(io_workqueue, &listen_con.rwork);
546 }
547 
548 int dlm_lowcomms_connect_node(int nodeid)
549 {
550 	struct connection *con;
551 	int idx;
552 
553 	idx = srcu_read_lock(&connections_srcu);
554 	con = nodeid2con(nodeid, 0);
555 	if (WARN_ON_ONCE(!con)) {
556 		srcu_read_unlock(&connections_srcu, idx);
557 		return -ENOENT;
558 	}
559 
560 	down_read(&con->sock_lock);
561 	if (!con->sock) {
562 		spin_lock_bh(&con->writequeue_lock);
563 		lowcomms_queue_swork(con);
564 		spin_unlock_bh(&con->writequeue_lock);
565 	}
566 	up_read(&con->sock_lock);
567 	srcu_read_unlock(&connections_srcu, idx);
568 
569 	cond_resched();
570 	return 0;
571 }
572 
573 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
574 {
575 	struct connection *con;
576 	int idx;
577 
578 	idx = srcu_read_lock(&connections_srcu);
579 	con = nodeid2con(nodeid, 0);
580 	if (!con) {
581 		srcu_read_unlock(&connections_srcu, idx);
582 		return -ENOENT;
583 	}
584 
585 	spin_lock(&con->addrs_lock);
586 	con->mark = mark;
587 	spin_unlock(&con->addrs_lock);
588 	srcu_read_unlock(&connections_srcu, idx);
589 	return 0;
590 }
591 
592 static void lowcomms_error_report(struct sock *sk)
593 {
594 	struct connection *con = sock2con(sk);
595 	struct inet_sock *inet;
596 
597 	inet = inet_sk(sk);
598 	switch (sk->sk_family) {
599 	case AF_INET:
600 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
601 				   "sending to node %d at %pI4, dport %d, "
602 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
603 				   con->nodeid, &inet->inet_daddr,
604 				   ntohs(inet->inet_dport), sk->sk_err,
605 				   READ_ONCE(sk->sk_err_soft));
606 		break;
607 #if IS_ENABLED(CONFIG_IPV6)
608 	case AF_INET6:
609 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
610 				   "sending to node %d at %pI6c, "
611 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
612 				   con->nodeid, &sk->sk_v6_daddr,
613 				   ntohs(inet->inet_dport), sk->sk_err,
614 				   READ_ONCE(sk->sk_err_soft));
615 		break;
616 #endif
617 	default:
618 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
619 				   "invalid socket family %d set, "
620 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
621 				   sk->sk_family, sk->sk_err,
622 				   READ_ONCE(sk->sk_err_soft));
623 		break;
624 	}
625 
626 	dlm_midcomms_unack_msg_resend(con->nodeid);
627 
628 	listen_sock.sk_error_report(sk);
629 }
630 
631 static void restore_callbacks(struct sock *sk)
632 {
633 #ifdef CONFIG_LOCKDEP
634 	WARN_ON_ONCE(!lockdep_sock_is_held(sk));
635 #endif
636 
637 	sk->sk_user_data = NULL;
638 	sk->sk_data_ready = listen_sock.sk_data_ready;
639 	sk->sk_state_change = listen_sock.sk_state_change;
640 	sk->sk_write_space = listen_sock.sk_write_space;
641 	sk->sk_error_report = listen_sock.sk_error_report;
642 }
643 
644 /* Make a socket active */
645 static void add_sock(struct socket *sock, struct connection *con)
646 {
647 	struct sock *sk = sock->sk;
648 
649 	lock_sock(sk);
650 	con->sock = sock;
651 
652 	sk->sk_user_data = con;
653 	sk->sk_data_ready = lowcomms_data_ready;
654 	sk->sk_write_space = lowcomms_write_space;
655 	if (dlm_config.ci_protocol == DLM_PROTO_SCTP)
656 		sk->sk_state_change = lowcomms_state_change;
657 	sk->sk_allocation = GFP_NOFS;
658 	sk->sk_use_task_frag = false;
659 	sk->sk_error_report = lowcomms_error_report;
660 	release_sock(sk);
661 }
662 
663 /* Add the port number to an IPv6 or 4 sockaddr and return the address
664    length */
665 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
666 			  int *addr_len)
667 {
668 	saddr->ss_family =  dlm_local_addr[0].ss_family;
669 	if (saddr->ss_family == AF_INET) {
670 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
671 		in4_addr->sin_port = cpu_to_be16(port);
672 		*addr_len = sizeof(struct sockaddr_in);
673 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
674 	} else {
675 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
676 		in6_addr->sin6_port = cpu_to_be16(port);
677 		*addr_len = sizeof(struct sockaddr_in6);
678 	}
679 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
680 }
681 
682 static void dlm_page_release(struct kref *kref)
683 {
684 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
685 						  ref);
686 
687 	__free_page(e->page);
688 	dlm_free_writequeue(e);
689 }
690 
691 static void dlm_msg_release(struct kref *kref)
692 {
693 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
694 
695 	kref_put(&msg->entry->ref, dlm_page_release);
696 	dlm_free_msg(msg);
697 }
698 
699 static void free_entry(struct writequeue_entry *e)
700 {
701 	struct dlm_msg *msg, *tmp;
702 
703 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
704 		if (msg->orig_msg) {
705 			msg->orig_msg->retransmit = false;
706 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
707 		}
708 
709 		list_del(&msg->list);
710 		kref_put(&msg->ref, dlm_msg_release);
711 	}
712 
713 	list_del(&e->list);
714 	kref_put(&e->ref, dlm_page_release);
715 }
716 
717 static void dlm_close_sock(struct socket **sock)
718 {
719 	lock_sock((*sock)->sk);
720 	restore_callbacks((*sock)->sk);
721 	release_sock((*sock)->sk);
722 
723 	sock_release(*sock);
724 	*sock = NULL;
725 }
726 
727 static void allow_connection_io(struct connection *con)
728 {
729 	if (con->othercon)
730 		clear_bit(CF_IO_STOP, &con->othercon->flags);
731 	clear_bit(CF_IO_STOP, &con->flags);
732 }
733 
734 static void stop_connection_io(struct connection *con)
735 {
736 	if (con->othercon)
737 		stop_connection_io(con->othercon);
738 
739 	spin_lock_bh(&con->writequeue_lock);
740 	set_bit(CF_IO_STOP, &con->flags);
741 	spin_unlock_bh(&con->writequeue_lock);
742 
743 	down_write(&con->sock_lock);
744 	if (con->sock) {
745 		lock_sock(con->sock->sk);
746 		restore_callbacks(con->sock->sk);
747 		release_sock(con->sock->sk);
748 	}
749 	up_write(&con->sock_lock);
750 
751 	cancel_work_sync(&con->swork);
752 	cancel_work_sync(&con->rwork);
753 }
754 
755 /* Close a remote connection and tidy up */
756 static void close_connection(struct connection *con, bool and_other)
757 {
758 	struct writequeue_entry *e;
759 
760 	if (con->othercon && and_other)
761 		close_connection(con->othercon, false);
762 
763 	down_write(&con->sock_lock);
764 	if (!con->sock) {
765 		up_write(&con->sock_lock);
766 		return;
767 	}
768 
769 	dlm_close_sock(&con->sock);
770 
771 	/* if we send a writequeue entry only a half way, we drop the
772 	 * whole entry because reconnection and that we not start of the
773 	 * middle of a msg which will confuse the other end.
774 	 *
775 	 * we can always drop messages because retransmits, but what we
776 	 * cannot allow is to transmit half messages which may be processed
777 	 * at the other side.
778 	 *
779 	 * our policy is to start on a clean state when disconnects, we don't
780 	 * know what's send/received on transport layer in this case.
781 	 */
782 	spin_lock_bh(&con->writequeue_lock);
783 	if (!list_empty(&con->writequeue)) {
784 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
785 				     list);
786 		if (e->dirty)
787 			free_entry(e);
788 	}
789 	spin_unlock_bh(&con->writequeue_lock);
790 
791 	con->rx_leftover = 0;
792 	con->retries = 0;
793 	clear_bit(CF_APP_LIMITED, &con->flags);
794 	clear_bit(CF_RECV_PENDING, &con->flags);
795 	clear_bit(CF_SEND_PENDING, &con->flags);
796 	up_write(&con->sock_lock);
797 }
798 
799 static void shutdown_connection(struct connection *con, bool and_other)
800 {
801 	int ret;
802 
803 	if (con->othercon && and_other)
804 		shutdown_connection(con->othercon, false);
805 
806 	flush_workqueue(io_workqueue);
807 	down_read(&con->sock_lock);
808 	/* nothing to shutdown */
809 	if (!con->sock) {
810 		up_read(&con->sock_lock);
811 		return;
812 	}
813 
814 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
815 	up_read(&con->sock_lock);
816 	if (ret) {
817 		log_print("Connection %p failed to shutdown: %d will force close",
818 			  con, ret);
819 		goto force_close;
820 	} else {
821 		ret = wait_event_timeout(con->shutdown_wait, !con->sock,
822 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
823 		if (ret == 0) {
824 			log_print("Connection %p shutdown timed out, will force close",
825 				  con);
826 			goto force_close;
827 		}
828 	}
829 
830 	return;
831 
832 force_close:
833 	close_connection(con, false);
834 }
835 
836 static struct processqueue_entry *new_processqueue_entry(int nodeid,
837 							 int buflen)
838 {
839 	struct processqueue_entry *pentry;
840 
841 	pentry = kmalloc(sizeof(*pentry), GFP_NOFS);
842 	if (!pentry)
843 		return NULL;
844 
845 	pentry->buf = kmalloc(buflen, GFP_NOFS);
846 	if (!pentry->buf) {
847 		kfree(pentry);
848 		return NULL;
849 	}
850 
851 	pentry->nodeid = nodeid;
852 	return pentry;
853 }
854 
855 static void free_processqueue_entry(struct processqueue_entry *pentry)
856 {
857 	kfree(pentry->buf);
858 	kfree(pentry);
859 }
860 
861 struct dlm_processed_nodes {
862 	int nodeid;
863 
864 	struct list_head list;
865 };
866 
867 static void process_dlm_messages(struct work_struct *work)
868 {
869 	struct processqueue_entry *pentry;
870 
871 	spin_lock_bh(&processqueue_lock);
872 	pentry = list_first_entry_or_null(&processqueue,
873 					  struct processqueue_entry, list);
874 	if (WARN_ON_ONCE(!pentry)) {
875 		process_dlm_messages_pending = false;
876 		spin_unlock_bh(&processqueue_lock);
877 		return;
878 	}
879 
880 	list_del(&pentry->list);
881 	if (atomic_dec_and_test(&processqueue_count))
882 		wake_up(&processqueue_wq);
883 	spin_unlock_bh(&processqueue_lock);
884 
885 	for (;;) {
886 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
887 					    pentry->buflen);
888 		free_processqueue_entry(pentry);
889 
890 		spin_lock_bh(&processqueue_lock);
891 		pentry = list_first_entry_or_null(&processqueue,
892 						  struct processqueue_entry, list);
893 		if (!pentry) {
894 			process_dlm_messages_pending = false;
895 			spin_unlock_bh(&processqueue_lock);
896 			break;
897 		}
898 
899 		list_del(&pentry->list);
900 		if (atomic_dec_and_test(&processqueue_count))
901 			wake_up(&processqueue_wq);
902 		spin_unlock_bh(&processqueue_lock);
903 	}
904 }
905 
906 /* Data received from remote end */
907 static int receive_from_sock(struct connection *con, int buflen)
908 {
909 	struct processqueue_entry *pentry;
910 	int ret, buflen_real;
911 	struct msghdr msg;
912 	struct kvec iov;
913 
914 	pentry = new_processqueue_entry(con->nodeid, buflen);
915 	if (!pentry)
916 		return DLM_IO_RESCHED;
917 
918 	memcpy(pentry->buf, con->rx_leftover_buf, con->rx_leftover);
919 
920 	/* calculate new buffer parameter regarding last receive and
921 	 * possible leftover bytes
922 	 */
923 	iov.iov_base = pentry->buf + con->rx_leftover;
924 	iov.iov_len = buflen - con->rx_leftover;
925 
926 	memset(&msg, 0, sizeof(msg));
927 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
928 	clear_bit(CF_RECV_INTR, &con->flags);
929 again:
930 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
931 			     msg.msg_flags);
932 	trace_dlm_recv(con->nodeid, ret);
933 	if (ret == -EAGAIN) {
934 		lock_sock(con->sock->sk);
935 		if (test_and_clear_bit(CF_RECV_INTR, &con->flags)) {
936 			release_sock(con->sock->sk);
937 			goto again;
938 		}
939 
940 		clear_bit(CF_RECV_PENDING, &con->flags);
941 		release_sock(con->sock->sk);
942 		free_processqueue_entry(pentry);
943 		return DLM_IO_END;
944 	} else if (ret == 0) {
945 		/* close will clear CF_RECV_PENDING */
946 		free_processqueue_entry(pentry);
947 		return DLM_IO_EOF;
948 	} else if (ret < 0) {
949 		free_processqueue_entry(pentry);
950 		return ret;
951 	}
952 
953 	/* new buflen according readed bytes and leftover from last receive */
954 	buflen_real = ret + con->rx_leftover;
955 	ret = dlm_validate_incoming_buffer(con->nodeid, pentry->buf,
956 					   buflen_real);
957 	if (ret < 0) {
958 		free_processqueue_entry(pentry);
959 		return ret;
960 	}
961 
962 	pentry->buflen = ret;
963 
964 	/* calculate leftover bytes from process and put it into begin of
965 	 * the receive buffer, so next receive we have the full message
966 	 * at the start address of the receive buffer.
967 	 */
968 	con->rx_leftover = buflen_real - ret;
969 	memmove(con->rx_leftover_buf, pentry->buf + ret,
970 		con->rx_leftover);
971 
972 	spin_lock_bh(&processqueue_lock);
973 	ret = atomic_inc_return(&processqueue_count);
974 	list_add_tail(&pentry->list, &processqueue);
975 	if (!process_dlm_messages_pending) {
976 		process_dlm_messages_pending = true;
977 		queue_work(process_workqueue, &process_work);
978 	}
979 	spin_unlock_bh(&processqueue_lock);
980 
981 	if (ret > DLM_MAX_PROCESS_BUFFERS)
982 		return DLM_IO_FLUSH;
983 
984 	return DLM_IO_SUCCESS;
985 }
986 
987 /* Listening socket is busy, accept a connection */
988 static int accept_from_sock(void)
989 {
990 	struct sockaddr_storage peeraddr;
991 	int len, idx, result, nodeid;
992 	struct connection *newcon;
993 	struct socket *newsock;
994 	unsigned int mark;
995 
996 	result = kernel_accept(listen_con.sock, &newsock, O_NONBLOCK);
997 	if (result == -EAGAIN)
998 		return DLM_IO_END;
999 	else if (result < 0)
1000 		goto accept_err;
1001 
1002 	/* Get the connected socket's peer */
1003 	memset(&peeraddr, 0, sizeof(peeraddr));
1004 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1005 	if (len < 0) {
1006 		result = -ECONNABORTED;
1007 		goto accept_err;
1008 	}
1009 
1010 	/* Get the new node's NODEID */
1011 	make_sockaddr(&peeraddr, 0, &len);
1012 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1013 		switch (peeraddr.ss_family) {
1014 		case AF_INET: {
1015 			struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr;
1016 
1017 			log_print("connect from non cluster IPv4 node %pI4",
1018 				  &sin->sin_addr);
1019 			break;
1020 		}
1021 #if IS_ENABLED(CONFIG_IPV6)
1022 		case AF_INET6: {
1023 			struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr;
1024 
1025 			log_print("connect from non cluster IPv6 node %pI6c",
1026 				  &sin6->sin6_addr);
1027 			break;
1028 		}
1029 #endif
1030 		default:
1031 			log_print("invalid family from non cluster node");
1032 			break;
1033 		}
1034 
1035 		sock_release(newsock);
1036 		return -1;
1037 	}
1038 
1039 	log_print("got connection from %d", nodeid);
1040 
1041 	/*  Check to see if we already have a connection to this node. This
1042 	 *  could happen if the two nodes initiate a connection at roughly
1043 	 *  the same time and the connections cross on the wire.
1044 	 *  In this case we store the incoming one in "othercon"
1045 	 */
1046 	idx = srcu_read_lock(&connections_srcu);
1047 	newcon = nodeid2con(nodeid, 0);
1048 	if (WARN_ON_ONCE(!newcon)) {
1049 		srcu_read_unlock(&connections_srcu, idx);
1050 		result = -ENOENT;
1051 		goto accept_err;
1052 	}
1053 
1054 	sock_set_mark(newsock->sk, mark);
1055 
1056 	down_write(&newcon->sock_lock);
1057 	if (newcon->sock) {
1058 		struct connection *othercon = newcon->othercon;
1059 
1060 		if (!othercon) {
1061 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1062 			if (!othercon) {
1063 				log_print("failed to allocate incoming socket");
1064 				up_write(&newcon->sock_lock);
1065 				srcu_read_unlock(&connections_srcu, idx);
1066 				result = -ENOMEM;
1067 				goto accept_err;
1068 			}
1069 
1070 			dlm_con_init(othercon, nodeid);
1071 			lockdep_set_subclass(&othercon->sock_lock, 1);
1072 			newcon->othercon = othercon;
1073 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1074 		} else {
1075 			/* close other sock con if we have something new */
1076 			close_connection(othercon, false);
1077 		}
1078 
1079 		down_write(&othercon->sock_lock);
1080 		add_sock(newsock, othercon);
1081 
1082 		/* check if we receved something while adding */
1083 		lock_sock(othercon->sock->sk);
1084 		lowcomms_queue_rwork(othercon);
1085 		release_sock(othercon->sock->sk);
1086 		up_write(&othercon->sock_lock);
1087 	}
1088 	else {
1089 		/* accept copies the sk after we've saved the callbacks, so we
1090 		   don't want to save them a second time or comm errors will
1091 		   result in calling sk_error_report recursively. */
1092 		add_sock(newsock, newcon);
1093 
1094 		/* check if we receved something while adding */
1095 		lock_sock(newcon->sock->sk);
1096 		lowcomms_queue_rwork(newcon);
1097 		release_sock(newcon->sock->sk);
1098 	}
1099 	up_write(&newcon->sock_lock);
1100 	srcu_read_unlock(&connections_srcu, idx);
1101 
1102 	return DLM_IO_SUCCESS;
1103 
1104 accept_err:
1105 	if (newsock)
1106 		sock_release(newsock);
1107 
1108 	return result;
1109 }
1110 
1111 /*
1112  * writequeue_entry_complete - try to delete and free write queue entry
1113  * @e: write queue entry to try to delete
1114  * @completed: bytes completed
1115  *
1116  * writequeue_lock must be held.
1117  */
1118 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1119 {
1120 	e->offset += completed;
1121 	e->len -= completed;
1122 	/* signal that page was half way transmitted */
1123 	e->dirty = true;
1124 
1125 	if (e->len == 0 && e->users == 0)
1126 		free_entry(e);
1127 }
1128 
1129 /*
1130  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1131  */
1132 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1133 {
1134 	struct sockaddr_storage localaddr;
1135 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1136 	int i, addr_len, result = 0;
1137 
1138 	for (i = 0; i < dlm_local_count; i++) {
1139 		memcpy(&localaddr, &dlm_local_addr[i], sizeof(localaddr));
1140 		make_sockaddr(&localaddr, port, &addr_len);
1141 
1142 		if (!i)
1143 			result = kernel_bind(sock, addr, addr_len);
1144 		else
1145 			result = sock_bind_add(sock->sk, addr, addr_len);
1146 
1147 		if (result < 0) {
1148 			log_print("Can't bind to %d addr number %d, %d.\n",
1149 				  port, i + 1, result);
1150 			break;
1151 		}
1152 	}
1153 	return result;
1154 }
1155 
1156 /* Get local addresses */
1157 static void init_local(void)
1158 {
1159 	struct sockaddr_storage sas;
1160 	int i;
1161 
1162 	dlm_local_count = 0;
1163 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1164 		if (dlm_our_addr(&sas, i))
1165 			break;
1166 
1167 		memcpy(&dlm_local_addr[dlm_local_count++], &sas, sizeof(sas));
1168 	}
1169 }
1170 
1171 static struct writequeue_entry *new_writequeue_entry(struct connection *con)
1172 {
1173 	struct writequeue_entry *entry;
1174 
1175 	entry = dlm_allocate_writequeue();
1176 	if (!entry)
1177 		return NULL;
1178 
1179 	entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
1180 	if (!entry->page) {
1181 		dlm_free_writequeue(entry);
1182 		return NULL;
1183 	}
1184 
1185 	entry->offset = 0;
1186 	entry->len = 0;
1187 	entry->end = 0;
1188 	entry->dirty = false;
1189 	entry->con = con;
1190 	entry->users = 1;
1191 	kref_init(&entry->ref);
1192 	return entry;
1193 }
1194 
1195 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1196 					     char **ppc, void (*cb)(void *data),
1197 					     void *data)
1198 {
1199 	struct writequeue_entry *e;
1200 
1201 	spin_lock_bh(&con->writequeue_lock);
1202 	if (!list_empty(&con->writequeue)) {
1203 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1204 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1205 			kref_get(&e->ref);
1206 
1207 			*ppc = page_address(e->page) + e->end;
1208 			if (cb)
1209 				cb(data);
1210 
1211 			e->end += len;
1212 			e->users++;
1213 			goto out;
1214 		}
1215 	}
1216 
1217 	e = new_writequeue_entry(con);
1218 	if (!e)
1219 		goto out;
1220 
1221 	kref_get(&e->ref);
1222 	*ppc = page_address(e->page);
1223 	e->end += len;
1224 	if (cb)
1225 		cb(data);
1226 
1227 	list_add_tail(&e->list, &con->writequeue);
1228 
1229 out:
1230 	spin_unlock_bh(&con->writequeue_lock);
1231 	return e;
1232 };
1233 
1234 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1235 						char **ppc, void (*cb)(void *data),
1236 						void *data)
1237 {
1238 	struct writequeue_entry *e;
1239 	struct dlm_msg *msg;
1240 
1241 	msg = dlm_allocate_msg();
1242 	if (!msg)
1243 		return NULL;
1244 
1245 	kref_init(&msg->ref);
1246 
1247 	e = new_wq_entry(con, len, ppc, cb, data);
1248 	if (!e) {
1249 		dlm_free_msg(msg);
1250 		return NULL;
1251 	}
1252 
1253 	msg->retransmit = false;
1254 	msg->orig_msg = NULL;
1255 	msg->ppc = *ppc;
1256 	msg->len = len;
1257 	msg->entry = e;
1258 
1259 	return msg;
1260 }
1261 
1262 /* avoid false positive for nodes_srcu, unlock happens in
1263  * dlm_lowcomms_commit_msg which is a must call if success
1264  */
1265 #ifndef __CHECKER__
1266 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
1267 				     void (*cb)(void *data), void *data)
1268 {
1269 	struct connection *con;
1270 	struct dlm_msg *msg;
1271 	int idx;
1272 
1273 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1274 	    len < sizeof(struct dlm_header)) {
1275 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1276 		log_print("failed to allocate a buffer of size %d", len);
1277 		WARN_ON_ONCE(1);
1278 		return NULL;
1279 	}
1280 
1281 	idx = srcu_read_lock(&connections_srcu);
1282 	con = nodeid2con(nodeid, 0);
1283 	if (WARN_ON_ONCE(!con)) {
1284 		srcu_read_unlock(&connections_srcu, idx);
1285 		return NULL;
1286 	}
1287 
1288 	msg = dlm_lowcomms_new_msg_con(con, len, ppc, cb, data);
1289 	if (!msg) {
1290 		srcu_read_unlock(&connections_srcu, idx);
1291 		return NULL;
1292 	}
1293 
1294 	/* for dlm_lowcomms_commit_msg() */
1295 	kref_get(&msg->ref);
1296 	/* we assume if successful commit must called */
1297 	msg->idx = idx;
1298 	return msg;
1299 }
1300 #endif
1301 
1302 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1303 {
1304 	struct writequeue_entry *e = msg->entry;
1305 	struct connection *con = e->con;
1306 	int users;
1307 
1308 	spin_lock_bh(&con->writequeue_lock);
1309 	kref_get(&msg->ref);
1310 	list_add(&msg->list, &e->msgs);
1311 
1312 	users = --e->users;
1313 	if (users)
1314 		goto out;
1315 
1316 	e->len = DLM_WQ_LENGTH_BYTES(e);
1317 
1318 	lowcomms_queue_swork(con);
1319 
1320 out:
1321 	spin_unlock_bh(&con->writequeue_lock);
1322 	return;
1323 }
1324 
1325 /* avoid false positive for nodes_srcu, lock was happen in
1326  * dlm_lowcomms_new_msg
1327  */
1328 #ifndef __CHECKER__
1329 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1330 {
1331 	_dlm_lowcomms_commit_msg(msg);
1332 	srcu_read_unlock(&connections_srcu, msg->idx);
1333 	/* because dlm_lowcomms_new_msg() */
1334 	kref_put(&msg->ref, dlm_msg_release);
1335 }
1336 #endif
1337 
1338 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1339 {
1340 	kref_put(&msg->ref, dlm_msg_release);
1341 }
1342 
1343 /* does not held connections_srcu, usage lowcomms_error_report only */
1344 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1345 {
1346 	struct dlm_msg *msg_resend;
1347 	char *ppc;
1348 
1349 	if (msg->retransmit)
1350 		return 1;
1351 
1352 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, &ppc,
1353 					      NULL, NULL);
1354 	if (!msg_resend)
1355 		return -ENOMEM;
1356 
1357 	msg->retransmit = true;
1358 	kref_get(&msg->ref);
1359 	msg_resend->orig_msg = msg;
1360 
1361 	memcpy(ppc, msg->ppc, msg->len);
1362 	_dlm_lowcomms_commit_msg(msg_resend);
1363 	dlm_lowcomms_put_msg(msg_resend);
1364 
1365 	return 0;
1366 }
1367 
1368 /* Send a message */
1369 static int send_to_sock(struct connection *con)
1370 {
1371 	struct writequeue_entry *e;
1372 	struct bio_vec bvec;
1373 	struct msghdr msg = {
1374 		.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT | MSG_NOSIGNAL,
1375 	};
1376 	int len, offset, ret;
1377 
1378 	spin_lock_bh(&con->writequeue_lock);
1379 	e = con_next_wq(con);
1380 	if (!e) {
1381 		clear_bit(CF_SEND_PENDING, &con->flags);
1382 		spin_unlock_bh(&con->writequeue_lock);
1383 		return DLM_IO_END;
1384 	}
1385 
1386 	len = e->len;
1387 	offset = e->offset;
1388 	WARN_ON_ONCE(len == 0 && e->users == 0);
1389 	spin_unlock_bh(&con->writequeue_lock);
1390 
1391 	bvec_set_page(&bvec, e->page, len, offset);
1392 	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, len);
1393 	ret = sock_sendmsg(con->sock, &msg);
1394 	trace_dlm_send(con->nodeid, ret);
1395 	if (ret == -EAGAIN || ret == 0) {
1396 		lock_sock(con->sock->sk);
1397 		spin_lock_bh(&con->writequeue_lock);
1398 		if (test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1399 		    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1400 			/* Notify TCP that we're limited by the
1401 			 * application window size.
1402 			 */
1403 			set_bit(SOCK_NOSPACE, &con->sock->sk->sk_socket->flags);
1404 			con->sock->sk->sk_write_pending++;
1405 
1406 			clear_bit(CF_SEND_PENDING, &con->flags);
1407 			spin_unlock_bh(&con->writequeue_lock);
1408 			release_sock(con->sock->sk);
1409 
1410 			/* wait for write_space() event */
1411 			return DLM_IO_END;
1412 		}
1413 		spin_unlock_bh(&con->writequeue_lock);
1414 		release_sock(con->sock->sk);
1415 
1416 		return DLM_IO_RESCHED;
1417 	} else if (ret < 0) {
1418 		return ret;
1419 	}
1420 
1421 	spin_lock_bh(&con->writequeue_lock);
1422 	writequeue_entry_complete(e, ret);
1423 	spin_unlock_bh(&con->writequeue_lock);
1424 
1425 	return DLM_IO_SUCCESS;
1426 }
1427 
1428 static void clean_one_writequeue(struct connection *con)
1429 {
1430 	struct writequeue_entry *e, *safe;
1431 
1432 	spin_lock_bh(&con->writequeue_lock);
1433 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1434 		free_entry(e);
1435 	}
1436 	spin_unlock_bh(&con->writequeue_lock);
1437 }
1438 
1439 static void connection_release(struct rcu_head *rcu)
1440 {
1441 	struct connection *con = container_of(rcu, struct connection, rcu);
1442 
1443 	WARN_ON_ONCE(!list_empty(&con->writequeue));
1444 	WARN_ON_ONCE(con->sock);
1445 	kfree(con);
1446 }
1447 
1448 /* Called from recovery when it knows that a node has
1449    left the cluster */
1450 int dlm_lowcomms_close(int nodeid)
1451 {
1452 	struct connection *con;
1453 	int idx;
1454 
1455 	log_print("closing connection to node %d", nodeid);
1456 
1457 	idx = srcu_read_lock(&connections_srcu);
1458 	con = nodeid2con(nodeid, 0);
1459 	if (WARN_ON_ONCE(!con)) {
1460 		srcu_read_unlock(&connections_srcu, idx);
1461 		return -ENOENT;
1462 	}
1463 
1464 	stop_connection_io(con);
1465 	log_print("io handling for node: %d stopped", nodeid);
1466 	close_connection(con, true);
1467 
1468 	spin_lock(&connections_lock);
1469 	hlist_del_rcu(&con->list);
1470 	spin_unlock(&connections_lock);
1471 
1472 	clean_one_writequeue(con);
1473 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1474 	if (con->othercon) {
1475 		clean_one_writequeue(con->othercon);
1476 		call_srcu(&connections_srcu, &con->othercon->rcu, connection_release);
1477 	}
1478 	srcu_read_unlock(&connections_srcu, idx);
1479 
1480 	/* for debugging we print when we are done to compare with other
1481 	 * messages in between. This function need to be correctly synchronized
1482 	 * with io handling
1483 	 */
1484 	log_print("closing connection to node %d done", nodeid);
1485 
1486 	return 0;
1487 }
1488 
1489 /* Receive worker function */
1490 static void process_recv_sockets(struct work_struct *work)
1491 {
1492 	struct connection *con = container_of(work, struct connection, rwork);
1493 	int ret, buflen;
1494 
1495 	down_read(&con->sock_lock);
1496 	if (!con->sock) {
1497 		up_read(&con->sock_lock);
1498 		return;
1499 	}
1500 
1501 	buflen = READ_ONCE(dlm_config.ci_buffer_size);
1502 	do {
1503 		ret = receive_from_sock(con, buflen);
1504 	} while (ret == DLM_IO_SUCCESS);
1505 	up_read(&con->sock_lock);
1506 
1507 	switch (ret) {
1508 	case DLM_IO_END:
1509 		/* CF_RECV_PENDING cleared */
1510 		break;
1511 	case DLM_IO_EOF:
1512 		close_connection(con, false);
1513 		wake_up(&con->shutdown_wait);
1514 		/* CF_RECV_PENDING cleared */
1515 		break;
1516 	case DLM_IO_FLUSH:
1517 		/* we can't flush the process_workqueue here because a
1518 		 * WQ_MEM_RECLAIM workequeue can occurr a deadlock for a non
1519 		 * WQ_MEM_RECLAIM workqueue such as process_workqueue. Instead
1520 		 * we have a waitqueue to wait until all messages are
1521 		 * processed.
1522 		 *
1523 		 * This handling is only necessary to backoff the sender and
1524 		 * not queue all messages from the socket layer into DLM
1525 		 * processqueue. When DLM is capable to parse multiple messages
1526 		 * on an e.g. per socket basis this handling can might be
1527 		 * removed. Especially in a message burst we are too slow to
1528 		 * process messages and the queue will fill up memory.
1529 		 */
1530 		wait_event(processqueue_wq, !atomic_read(&processqueue_count));
1531 		fallthrough;
1532 	case DLM_IO_RESCHED:
1533 		cond_resched();
1534 		queue_work(io_workqueue, &con->rwork);
1535 		/* CF_RECV_PENDING not cleared */
1536 		break;
1537 	default:
1538 		if (ret < 0) {
1539 			if (test_bit(CF_IS_OTHERCON, &con->flags)) {
1540 				close_connection(con, false);
1541 			} else {
1542 				spin_lock_bh(&con->writequeue_lock);
1543 				lowcomms_queue_swork(con);
1544 				spin_unlock_bh(&con->writequeue_lock);
1545 			}
1546 
1547 			/* CF_RECV_PENDING cleared for othercon
1548 			 * we trigger send queue if not already done
1549 			 * and process_send_sockets will handle it
1550 			 */
1551 			break;
1552 		}
1553 
1554 		WARN_ON_ONCE(1);
1555 		break;
1556 	}
1557 }
1558 
1559 static void process_listen_recv_socket(struct work_struct *work)
1560 {
1561 	int ret;
1562 
1563 	if (WARN_ON_ONCE(!listen_con.sock))
1564 		return;
1565 
1566 	do {
1567 		ret = accept_from_sock();
1568 	} while (ret == DLM_IO_SUCCESS);
1569 
1570 	if (ret < 0)
1571 		log_print("critical error accepting connection: %d", ret);
1572 }
1573 
1574 static int dlm_connect(struct connection *con)
1575 {
1576 	struct sockaddr_storage addr;
1577 	int result, addr_len;
1578 	struct socket *sock;
1579 	unsigned int mark;
1580 
1581 	memset(&addr, 0, sizeof(addr));
1582 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1583 				dlm_proto_ops->try_new_addr, &mark);
1584 	if (result < 0) {
1585 		log_print("no address for nodeid %d", con->nodeid);
1586 		return result;
1587 	}
1588 
1589 	/* Create a socket to communicate with */
1590 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1591 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1592 	if (result < 0)
1593 		return result;
1594 
1595 	sock_set_mark(sock->sk, mark);
1596 	dlm_proto_ops->sockopts(sock);
1597 
1598 	result = dlm_proto_ops->bind(sock);
1599 	if (result < 0) {
1600 		sock_release(sock);
1601 		return result;
1602 	}
1603 
1604 	add_sock(sock, con);
1605 
1606 	log_print_ratelimited("connecting to %d", con->nodeid);
1607 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1608 	result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1609 					addr_len);
1610 	switch (result) {
1611 	case -EINPROGRESS:
1612 		/* not an error */
1613 		fallthrough;
1614 	case 0:
1615 		break;
1616 	default:
1617 		if (result < 0)
1618 			dlm_close_sock(&con->sock);
1619 
1620 		break;
1621 	}
1622 
1623 	return result;
1624 }
1625 
1626 /* Send worker function */
1627 static void process_send_sockets(struct work_struct *work)
1628 {
1629 	struct connection *con = container_of(work, struct connection, swork);
1630 	int ret;
1631 
1632 	WARN_ON_ONCE(test_bit(CF_IS_OTHERCON, &con->flags));
1633 
1634 	down_read(&con->sock_lock);
1635 	if (!con->sock) {
1636 		up_read(&con->sock_lock);
1637 		down_write(&con->sock_lock);
1638 		if (!con->sock) {
1639 			ret = dlm_connect(con);
1640 			switch (ret) {
1641 			case 0:
1642 				break;
1643 			case -EINPROGRESS:
1644 				/* avoid spamming resched on connection
1645 				 * we might can switch to a state_change
1646 				 * event based mechanism if established
1647 				 */
1648 				msleep(100);
1649 				break;
1650 			default:
1651 				/* CF_SEND_PENDING not cleared */
1652 				up_write(&con->sock_lock);
1653 				log_print("connect to node %d try %d error %d",
1654 					  con->nodeid, con->retries++, ret);
1655 				msleep(1000);
1656 				/* For now we try forever to reconnect. In
1657 				 * future we should send a event to cluster
1658 				 * manager to fence itself after certain amount
1659 				 * of retries.
1660 				 */
1661 				queue_work(io_workqueue, &con->swork);
1662 				return;
1663 			}
1664 		}
1665 		downgrade_write(&con->sock_lock);
1666 	}
1667 
1668 	do {
1669 		ret = send_to_sock(con);
1670 	} while (ret == DLM_IO_SUCCESS);
1671 	up_read(&con->sock_lock);
1672 
1673 	switch (ret) {
1674 	case DLM_IO_END:
1675 		/* CF_SEND_PENDING cleared */
1676 		break;
1677 	case DLM_IO_RESCHED:
1678 		/* CF_SEND_PENDING not cleared */
1679 		cond_resched();
1680 		queue_work(io_workqueue, &con->swork);
1681 		break;
1682 	default:
1683 		if (ret < 0) {
1684 			close_connection(con, false);
1685 
1686 			/* CF_SEND_PENDING cleared */
1687 			spin_lock_bh(&con->writequeue_lock);
1688 			lowcomms_queue_swork(con);
1689 			spin_unlock_bh(&con->writequeue_lock);
1690 			break;
1691 		}
1692 
1693 		WARN_ON_ONCE(1);
1694 		break;
1695 	}
1696 }
1697 
1698 static void work_stop(void)
1699 {
1700 	if (io_workqueue) {
1701 		destroy_workqueue(io_workqueue);
1702 		io_workqueue = NULL;
1703 	}
1704 
1705 	if (process_workqueue) {
1706 		destroy_workqueue(process_workqueue);
1707 		process_workqueue = NULL;
1708 	}
1709 }
1710 
1711 static int work_start(void)
1712 {
1713 	io_workqueue = alloc_workqueue("dlm_io", WQ_HIGHPRI | WQ_MEM_RECLAIM |
1714 				       WQ_UNBOUND, 0);
1715 	if (!io_workqueue) {
1716 		log_print("can't start dlm_io");
1717 		return -ENOMEM;
1718 	}
1719 
1720 	process_workqueue = alloc_workqueue("dlm_process", WQ_HIGHPRI | WQ_BH, 0);
1721 	if (!process_workqueue) {
1722 		log_print("can't start dlm_process");
1723 		destroy_workqueue(io_workqueue);
1724 		io_workqueue = NULL;
1725 		return -ENOMEM;
1726 	}
1727 
1728 	return 0;
1729 }
1730 
1731 void dlm_lowcomms_shutdown(void)
1732 {
1733 	struct connection *con;
1734 	int i, idx;
1735 
1736 	/* stop lowcomms_listen_data_ready calls */
1737 	lock_sock(listen_con.sock->sk);
1738 	listen_con.sock->sk->sk_data_ready = listen_sock.sk_data_ready;
1739 	release_sock(listen_con.sock->sk);
1740 
1741 	cancel_work_sync(&listen_con.rwork);
1742 	dlm_close_sock(&listen_con.sock);
1743 
1744 	idx = srcu_read_lock(&connections_srcu);
1745 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1746 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
1747 			shutdown_connection(con, true);
1748 			stop_connection_io(con);
1749 			flush_workqueue(process_workqueue);
1750 			close_connection(con, true);
1751 
1752 			clean_one_writequeue(con);
1753 			if (con->othercon)
1754 				clean_one_writequeue(con->othercon);
1755 			allow_connection_io(con);
1756 		}
1757 	}
1758 	srcu_read_unlock(&connections_srcu, idx);
1759 }
1760 
1761 void dlm_lowcomms_stop(void)
1762 {
1763 	work_stop();
1764 	dlm_proto_ops = NULL;
1765 }
1766 
1767 static int dlm_listen_for_all(void)
1768 {
1769 	struct socket *sock;
1770 	int result;
1771 
1772 	log_print("Using %s for communications",
1773 		  dlm_proto_ops->name);
1774 
1775 	result = dlm_proto_ops->listen_validate();
1776 	if (result < 0)
1777 		return result;
1778 
1779 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1780 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1781 	if (result < 0) {
1782 		log_print("Can't create comms socket: %d", result);
1783 		return result;
1784 	}
1785 
1786 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1787 	dlm_proto_ops->listen_sockopts(sock);
1788 
1789 	result = dlm_proto_ops->listen_bind(sock);
1790 	if (result < 0)
1791 		goto out;
1792 
1793 	lock_sock(sock->sk);
1794 	listen_sock.sk_data_ready = sock->sk->sk_data_ready;
1795 	listen_sock.sk_write_space = sock->sk->sk_write_space;
1796 	listen_sock.sk_error_report = sock->sk->sk_error_report;
1797 	listen_sock.sk_state_change = sock->sk->sk_state_change;
1798 
1799 	listen_con.sock = sock;
1800 
1801 	sock->sk->sk_allocation = GFP_NOFS;
1802 	sock->sk->sk_use_task_frag = false;
1803 	sock->sk->sk_data_ready = lowcomms_listen_data_ready;
1804 	release_sock(sock->sk);
1805 
1806 	result = sock->ops->listen(sock, 128);
1807 	if (result < 0) {
1808 		dlm_close_sock(&listen_con.sock);
1809 		return result;
1810 	}
1811 
1812 	return 0;
1813 
1814 out:
1815 	sock_release(sock);
1816 	return result;
1817 }
1818 
1819 static int dlm_tcp_bind(struct socket *sock)
1820 {
1821 	struct sockaddr_storage src_addr;
1822 	int result, addr_len;
1823 
1824 	/* Bind to our cluster-known address connecting to avoid
1825 	 * routing problems.
1826 	 */
1827 	memcpy(&src_addr, &dlm_local_addr[0], sizeof(src_addr));
1828 	make_sockaddr(&src_addr, 0, &addr_len);
1829 
1830 	result = kernel_bind(sock, (struct sockaddr *)&src_addr,
1831 			     addr_len);
1832 	if (result < 0) {
1833 		/* This *may* not indicate a critical error */
1834 		log_print("could not bind for connect: %d", result);
1835 	}
1836 
1837 	return 0;
1838 }
1839 
1840 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1841 			   struct sockaddr *addr, int addr_len)
1842 {
1843 	return kernel_connect(sock, addr, addr_len, O_NONBLOCK);
1844 }
1845 
1846 static int dlm_tcp_listen_validate(void)
1847 {
1848 	/* We don't support multi-homed hosts */
1849 	if (dlm_local_count > 1) {
1850 		log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1851 		return -EINVAL;
1852 	}
1853 
1854 	return 0;
1855 }
1856 
1857 static void dlm_tcp_sockopts(struct socket *sock)
1858 {
1859 	/* Turn off Nagle's algorithm */
1860 	tcp_sock_set_nodelay(sock->sk);
1861 }
1862 
1863 static void dlm_tcp_listen_sockopts(struct socket *sock)
1864 {
1865 	dlm_tcp_sockopts(sock);
1866 	sock_set_reuseaddr(sock->sk);
1867 }
1868 
1869 static int dlm_tcp_listen_bind(struct socket *sock)
1870 {
1871 	int addr_len;
1872 
1873 	/* Bind to our port */
1874 	make_sockaddr(&dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1875 	return kernel_bind(sock, (struct sockaddr *)&dlm_local_addr[0],
1876 			   addr_len);
1877 }
1878 
1879 static const struct dlm_proto_ops dlm_tcp_ops = {
1880 	.name = "TCP",
1881 	.proto = IPPROTO_TCP,
1882 	.connect = dlm_tcp_connect,
1883 	.sockopts = dlm_tcp_sockopts,
1884 	.bind = dlm_tcp_bind,
1885 	.listen_validate = dlm_tcp_listen_validate,
1886 	.listen_sockopts = dlm_tcp_listen_sockopts,
1887 	.listen_bind = dlm_tcp_listen_bind,
1888 };
1889 
1890 static int dlm_sctp_bind(struct socket *sock)
1891 {
1892 	return sctp_bind_addrs(sock, 0);
1893 }
1894 
1895 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1896 			    struct sockaddr *addr, int addr_len)
1897 {
1898 	int ret;
1899 
1900 	/*
1901 	 * Make kernel_connect() function return in specified time,
1902 	 * since O_NONBLOCK argument in connect() function does not work here,
1903 	 * then, we should restore the default value of this attribute.
1904 	 */
1905 	sock_set_sndtimeo(sock->sk, 5);
1906 	ret = kernel_connect(sock, addr, addr_len, 0);
1907 	sock_set_sndtimeo(sock->sk, 0);
1908 	return ret;
1909 }
1910 
1911 static int dlm_sctp_listen_validate(void)
1912 {
1913 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1914 		log_print("SCTP is not enabled by this kernel");
1915 		return -EOPNOTSUPP;
1916 	}
1917 
1918 	request_module("sctp");
1919 	return 0;
1920 }
1921 
1922 static int dlm_sctp_bind_listen(struct socket *sock)
1923 {
1924 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1925 }
1926 
1927 static void dlm_sctp_sockopts(struct socket *sock)
1928 {
1929 	/* Turn off Nagle's algorithm */
1930 	sctp_sock_set_nodelay(sock->sk);
1931 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1932 }
1933 
1934 static const struct dlm_proto_ops dlm_sctp_ops = {
1935 	.name = "SCTP",
1936 	.proto = IPPROTO_SCTP,
1937 	.try_new_addr = true,
1938 	.connect = dlm_sctp_connect,
1939 	.sockopts = dlm_sctp_sockopts,
1940 	.bind = dlm_sctp_bind,
1941 	.listen_validate = dlm_sctp_listen_validate,
1942 	.listen_sockopts = dlm_sctp_sockopts,
1943 	.listen_bind = dlm_sctp_bind_listen,
1944 };
1945 
1946 int dlm_lowcomms_start(void)
1947 {
1948 	int error;
1949 
1950 	init_local();
1951 	if (!dlm_local_count) {
1952 		error = -ENOTCONN;
1953 		log_print("no local IP address has been set");
1954 		goto fail;
1955 	}
1956 
1957 	error = work_start();
1958 	if (error)
1959 		goto fail;
1960 
1961 	/* Start listening */
1962 	switch (dlm_config.ci_protocol) {
1963 	case DLM_PROTO_TCP:
1964 		dlm_proto_ops = &dlm_tcp_ops;
1965 		break;
1966 	case DLM_PROTO_SCTP:
1967 		dlm_proto_ops = &dlm_sctp_ops;
1968 		break;
1969 	default:
1970 		log_print("Invalid protocol identifier %d set",
1971 			  dlm_config.ci_protocol);
1972 		error = -EINVAL;
1973 		goto fail_proto_ops;
1974 	}
1975 
1976 	error = dlm_listen_for_all();
1977 	if (error)
1978 		goto fail_listen;
1979 
1980 	return 0;
1981 
1982 fail_listen:
1983 	dlm_proto_ops = NULL;
1984 fail_proto_ops:
1985 	work_stop();
1986 fail:
1987 	return error;
1988 }
1989 
1990 void dlm_lowcomms_init(void)
1991 {
1992 	int i;
1993 
1994 	for (i = 0; i < CONN_HASH_SIZE; i++)
1995 		INIT_HLIST_HEAD(&connection_hash[i]);
1996 
1997 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1998 }
1999 
2000 void dlm_lowcomms_exit(void)
2001 {
2002 	struct connection *con;
2003 	int i, idx;
2004 
2005 	idx = srcu_read_lock(&connections_srcu);
2006 	for (i = 0; i < CONN_HASH_SIZE; i++) {
2007 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
2008 			spin_lock(&connections_lock);
2009 			hlist_del_rcu(&con->list);
2010 			spin_unlock(&connections_lock);
2011 
2012 			if (con->othercon)
2013 				call_srcu(&connections_srcu, &con->othercon->rcu,
2014 					  connection_release);
2015 			call_srcu(&connections_srcu, &con->rcu, connection_release);
2016 		}
2017 	}
2018 	srcu_read_unlock(&connections_srcu, idx);
2019 }
2020