xref: /linux/fs/dlm/lowcomms.c (revision b58b13f156c00c2457035b7071eaaac105fe6836)
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 atomic_t processqueue_count;
208 static LIST_HEAD(processqueue);
209 
210 bool dlm_lowcomms_is_running(void)
211 {
212 	return !!listen_con.sock;
213 }
214 
215 static void lowcomms_queue_swork(struct connection *con)
216 {
217 	assert_spin_locked(&con->writequeue_lock);
218 
219 	if (!test_bit(CF_IO_STOP, &con->flags) &&
220 	    !test_bit(CF_APP_LIMITED, &con->flags) &&
221 	    !test_and_set_bit(CF_SEND_PENDING, &con->flags))
222 		queue_work(io_workqueue, &con->swork);
223 }
224 
225 static void lowcomms_queue_rwork(struct connection *con)
226 {
227 #ifdef CONFIG_LOCKDEP
228 	WARN_ON_ONCE(!lockdep_sock_is_held(con->sock->sk));
229 #endif
230 
231 	if (!test_bit(CF_IO_STOP, &con->flags) &&
232 	    !test_and_set_bit(CF_RECV_PENDING, &con->flags))
233 		queue_work(io_workqueue, &con->rwork);
234 }
235 
236 static void writequeue_entry_ctor(void *data)
237 {
238 	struct writequeue_entry *entry = data;
239 
240 	INIT_LIST_HEAD(&entry->msgs);
241 }
242 
243 struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void)
244 {
245 	return kmem_cache_create("dlm_writequeue", sizeof(struct writequeue_entry),
246 				 0, 0, writequeue_entry_ctor);
247 }
248 
249 struct kmem_cache *dlm_lowcomms_msg_cache_create(void)
250 {
251 	return kmem_cache_create("dlm_msg", sizeof(struct dlm_msg), 0, 0, NULL);
252 }
253 
254 /* need to held writequeue_lock */
255 static struct writequeue_entry *con_next_wq(struct connection *con)
256 {
257 	struct writequeue_entry *e;
258 
259 	e = list_first_entry_or_null(&con->writequeue, struct writequeue_entry,
260 				     list);
261 	/* if len is zero nothing is to send, if there are users filling
262 	 * buffers we wait until the users are done so we can send more.
263 	 */
264 	if (!e || e->users || e->len == 0)
265 		return NULL;
266 
267 	return e;
268 }
269 
270 static struct connection *__find_con(int nodeid, int r)
271 {
272 	struct connection *con;
273 
274 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
275 		if (con->nodeid == nodeid)
276 			return con;
277 	}
278 
279 	return NULL;
280 }
281 
282 static void dlm_con_init(struct connection *con, int nodeid)
283 {
284 	con->nodeid = nodeid;
285 	init_rwsem(&con->sock_lock);
286 	INIT_LIST_HEAD(&con->writequeue);
287 	spin_lock_init(&con->writequeue_lock);
288 	INIT_WORK(&con->swork, process_send_sockets);
289 	INIT_WORK(&con->rwork, process_recv_sockets);
290 	spin_lock_init(&con->addrs_lock);
291 	init_waitqueue_head(&con->shutdown_wait);
292 }
293 
294 /*
295  * If 'allocation' is zero then we don't attempt to create a new
296  * connection structure for this node.
297  */
298 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
299 {
300 	struct connection *con, *tmp;
301 	int r;
302 
303 	r = nodeid_hash(nodeid);
304 	con = __find_con(nodeid, r);
305 	if (con || !alloc)
306 		return con;
307 
308 	con = kzalloc(sizeof(*con), alloc);
309 	if (!con)
310 		return NULL;
311 
312 	dlm_con_init(con, nodeid);
313 
314 	spin_lock(&connections_lock);
315 	/* Because multiple workqueues/threads calls this function it can
316 	 * race on multiple cpu's. Instead of locking hot path __find_con()
317 	 * we just check in rare cases of recently added nodes again
318 	 * under protection of connections_lock. If this is the case we
319 	 * abort our connection creation and return the existing connection.
320 	 */
321 	tmp = __find_con(nodeid, r);
322 	if (tmp) {
323 		spin_unlock(&connections_lock);
324 		kfree(con);
325 		return tmp;
326 	}
327 
328 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
329 	spin_unlock(&connections_lock);
330 
331 	return con;
332 }
333 
334 static int addr_compare(const struct sockaddr_storage *x,
335 			const struct sockaddr_storage *y)
336 {
337 	switch (x->ss_family) {
338 	case AF_INET: {
339 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
340 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
341 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
342 			return 0;
343 		if (sinx->sin_port != siny->sin_port)
344 			return 0;
345 		break;
346 	}
347 	case AF_INET6: {
348 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
349 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
350 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
351 			return 0;
352 		if (sinx->sin6_port != siny->sin6_port)
353 			return 0;
354 		break;
355 	}
356 	default:
357 		return 0;
358 	}
359 	return 1;
360 }
361 
362 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
363 			  struct sockaddr *sa_out, bool try_new_addr,
364 			  unsigned int *mark)
365 {
366 	struct sockaddr_storage sas;
367 	struct connection *con;
368 	int idx;
369 
370 	if (!dlm_local_count)
371 		return -1;
372 
373 	idx = srcu_read_lock(&connections_srcu);
374 	con = nodeid2con(nodeid, 0);
375 	if (!con) {
376 		srcu_read_unlock(&connections_srcu, idx);
377 		return -ENOENT;
378 	}
379 
380 	spin_lock(&con->addrs_lock);
381 	if (!con->addr_count) {
382 		spin_unlock(&con->addrs_lock);
383 		srcu_read_unlock(&connections_srcu, idx);
384 		return -ENOENT;
385 	}
386 
387 	memcpy(&sas, &con->addr[con->curr_addr_index],
388 	       sizeof(struct sockaddr_storage));
389 
390 	if (try_new_addr) {
391 		con->curr_addr_index++;
392 		if (con->curr_addr_index == con->addr_count)
393 			con->curr_addr_index = 0;
394 	}
395 
396 	*mark = con->mark;
397 	spin_unlock(&con->addrs_lock);
398 
399 	if (sas_out)
400 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
401 
402 	if (!sa_out) {
403 		srcu_read_unlock(&connections_srcu, idx);
404 		return 0;
405 	}
406 
407 	if (dlm_local_addr[0].ss_family == AF_INET) {
408 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
409 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
410 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
411 	} else {
412 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
413 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
414 		ret6->sin6_addr = in6->sin6_addr;
415 	}
416 
417 	srcu_read_unlock(&connections_srcu, idx);
418 	return 0;
419 }
420 
421 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
422 			  unsigned int *mark)
423 {
424 	struct connection *con;
425 	int i, idx, addr_i;
426 
427 	idx = srcu_read_lock(&connections_srcu);
428 	for (i = 0; i < CONN_HASH_SIZE; i++) {
429 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
430 			WARN_ON_ONCE(!con->addr_count);
431 
432 			spin_lock(&con->addrs_lock);
433 			for (addr_i = 0; addr_i < con->addr_count; addr_i++) {
434 				if (addr_compare(&con->addr[addr_i], addr)) {
435 					*nodeid = con->nodeid;
436 					*mark = con->mark;
437 					spin_unlock(&con->addrs_lock);
438 					srcu_read_unlock(&connections_srcu, idx);
439 					return 0;
440 				}
441 			}
442 			spin_unlock(&con->addrs_lock);
443 		}
444 	}
445 	srcu_read_unlock(&connections_srcu, idx);
446 
447 	return -ENOENT;
448 }
449 
450 static bool dlm_lowcomms_con_has_addr(const struct connection *con,
451 				      const struct sockaddr_storage *addr)
452 {
453 	int i;
454 
455 	for (i = 0; i < con->addr_count; i++) {
456 		if (addr_compare(&con->addr[i], addr))
457 			return true;
458 	}
459 
460 	return false;
461 }
462 
463 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
464 {
465 	struct connection *con;
466 	bool ret, idx;
467 
468 	idx = srcu_read_lock(&connections_srcu);
469 	con = nodeid2con(nodeid, GFP_NOFS);
470 	if (!con) {
471 		srcu_read_unlock(&connections_srcu, idx);
472 		return -ENOMEM;
473 	}
474 
475 	spin_lock(&con->addrs_lock);
476 	if (!con->addr_count) {
477 		memcpy(&con->addr[0], addr, sizeof(*addr));
478 		con->addr_count = 1;
479 		con->mark = dlm_config.ci_mark;
480 		spin_unlock(&con->addrs_lock);
481 		srcu_read_unlock(&connections_srcu, idx);
482 		return 0;
483 	}
484 
485 	ret = dlm_lowcomms_con_has_addr(con, addr);
486 	if (ret) {
487 		spin_unlock(&con->addrs_lock);
488 		srcu_read_unlock(&connections_srcu, idx);
489 		return -EEXIST;
490 	}
491 
492 	if (con->addr_count >= DLM_MAX_ADDR_COUNT) {
493 		spin_unlock(&con->addrs_lock);
494 		srcu_read_unlock(&connections_srcu, idx);
495 		return -ENOSPC;
496 	}
497 
498 	memcpy(&con->addr[con->addr_count++], addr, sizeof(*addr));
499 	srcu_read_unlock(&connections_srcu, idx);
500 	spin_unlock(&con->addrs_lock);
501 	return 0;
502 }
503 
504 /* Data available on socket or listen socket received a connect */
505 static void lowcomms_data_ready(struct sock *sk)
506 {
507 	struct connection *con = sock2con(sk);
508 
509 	trace_sk_data_ready(sk);
510 
511 	set_bit(CF_RECV_INTR, &con->flags);
512 	lowcomms_queue_rwork(con);
513 }
514 
515 static void lowcomms_write_space(struct sock *sk)
516 {
517 	struct connection *con = sock2con(sk);
518 
519 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
520 
521 	spin_lock_bh(&con->writequeue_lock);
522 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
523 		con->sock->sk->sk_write_pending--;
524 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
525 	}
526 
527 	lowcomms_queue_swork(con);
528 	spin_unlock_bh(&con->writequeue_lock);
529 }
530 
531 static void lowcomms_state_change(struct sock *sk)
532 {
533 	/* SCTP layer is not calling sk_data_ready when the connection
534 	 * is done, so we catch the signal through here.
535 	 */
536 	if (sk->sk_shutdown == RCV_SHUTDOWN)
537 		lowcomms_data_ready(sk);
538 }
539 
540 static void lowcomms_listen_data_ready(struct sock *sk)
541 {
542 	trace_sk_data_ready(sk);
543 
544 	queue_work(io_workqueue, &listen_con.rwork);
545 }
546 
547 int dlm_lowcomms_connect_node(int nodeid)
548 {
549 	struct connection *con;
550 	int idx;
551 
552 	idx = srcu_read_lock(&connections_srcu);
553 	con = nodeid2con(nodeid, 0);
554 	if (WARN_ON_ONCE(!con)) {
555 		srcu_read_unlock(&connections_srcu, idx);
556 		return -ENOENT;
557 	}
558 
559 	down_read(&con->sock_lock);
560 	if (!con->sock) {
561 		spin_lock_bh(&con->writequeue_lock);
562 		lowcomms_queue_swork(con);
563 		spin_unlock_bh(&con->writequeue_lock);
564 	}
565 	up_read(&con->sock_lock);
566 	srcu_read_unlock(&connections_srcu, idx);
567 
568 	cond_resched();
569 	return 0;
570 }
571 
572 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
573 {
574 	struct connection *con;
575 	int idx;
576 
577 	idx = srcu_read_lock(&connections_srcu);
578 	con = nodeid2con(nodeid, 0);
579 	if (!con) {
580 		srcu_read_unlock(&connections_srcu, idx);
581 		return -ENOENT;
582 	}
583 
584 	spin_lock(&con->addrs_lock);
585 	con->mark = mark;
586 	spin_unlock(&con->addrs_lock);
587 	srcu_read_unlock(&connections_srcu, idx);
588 	return 0;
589 }
590 
591 static void lowcomms_error_report(struct sock *sk)
592 {
593 	struct connection *con = sock2con(sk);
594 	struct inet_sock *inet;
595 
596 	inet = inet_sk(sk);
597 	switch (sk->sk_family) {
598 	case AF_INET:
599 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
600 				   "sending to node %d at %pI4, dport %d, "
601 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
602 				   con->nodeid, &inet->inet_daddr,
603 				   ntohs(inet->inet_dport), sk->sk_err,
604 				   READ_ONCE(sk->sk_err_soft));
605 		break;
606 #if IS_ENABLED(CONFIG_IPV6)
607 	case AF_INET6:
608 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
609 				   "sending to node %d at %pI6c, "
610 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
611 				   con->nodeid, &sk->sk_v6_daddr,
612 				   ntohs(inet->inet_dport), sk->sk_err,
613 				   READ_ONCE(sk->sk_err_soft));
614 		break;
615 #endif
616 	default:
617 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
618 				   "invalid socket family %d set, "
619 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
620 				   sk->sk_family, sk->sk_err,
621 				   READ_ONCE(sk->sk_err_soft));
622 		break;
623 	}
624 
625 	dlm_midcomms_unack_msg_resend(con->nodeid);
626 
627 	listen_sock.sk_error_report(sk);
628 }
629 
630 static void restore_callbacks(struct sock *sk)
631 {
632 #ifdef CONFIG_LOCKDEP
633 	WARN_ON_ONCE(!lockdep_sock_is_held(sk));
634 #endif
635 
636 	sk->sk_user_data = NULL;
637 	sk->sk_data_ready = listen_sock.sk_data_ready;
638 	sk->sk_state_change = listen_sock.sk_state_change;
639 	sk->sk_write_space = listen_sock.sk_write_space;
640 	sk->sk_error_report = listen_sock.sk_error_report;
641 }
642 
643 /* Make a socket active */
644 static void add_sock(struct socket *sock, struct connection *con)
645 {
646 	struct sock *sk = sock->sk;
647 
648 	lock_sock(sk);
649 	con->sock = sock;
650 
651 	sk->sk_user_data = con;
652 	sk->sk_data_ready = lowcomms_data_ready;
653 	sk->sk_write_space = lowcomms_write_space;
654 	if (dlm_config.ci_protocol == DLM_PROTO_SCTP)
655 		sk->sk_state_change = lowcomms_state_change;
656 	sk->sk_allocation = GFP_NOFS;
657 	sk->sk_use_task_frag = false;
658 	sk->sk_error_report = lowcomms_error_report;
659 	release_sock(sk);
660 }
661 
662 /* Add the port number to an IPv6 or 4 sockaddr and return the address
663    length */
664 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
665 			  int *addr_len)
666 {
667 	saddr->ss_family =  dlm_local_addr[0].ss_family;
668 	if (saddr->ss_family == AF_INET) {
669 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
670 		in4_addr->sin_port = cpu_to_be16(port);
671 		*addr_len = sizeof(struct sockaddr_in);
672 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
673 	} else {
674 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
675 		in6_addr->sin6_port = cpu_to_be16(port);
676 		*addr_len = sizeof(struct sockaddr_in6);
677 	}
678 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
679 }
680 
681 static void dlm_page_release(struct kref *kref)
682 {
683 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
684 						  ref);
685 
686 	__free_page(e->page);
687 	dlm_free_writequeue(e);
688 }
689 
690 static void dlm_msg_release(struct kref *kref)
691 {
692 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
693 
694 	kref_put(&msg->entry->ref, dlm_page_release);
695 	dlm_free_msg(msg);
696 }
697 
698 static void free_entry(struct writequeue_entry *e)
699 {
700 	struct dlm_msg *msg, *tmp;
701 
702 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
703 		if (msg->orig_msg) {
704 			msg->orig_msg->retransmit = false;
705 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
706 		}
707 
708 		list_del(&msg->list);
709 		kref_put(&msg->ref, dlm_msg_release);
710 	}
711 
712 	list_del(&e->list);
713 	kref_put(&e->ref, dlm_page_release);
714 }
715 
716 static void dlm_close_sock(struct socket **sock)
717 {
718 	lock_sock((*sock)->sk);
719 	restore_callbacks((*sock)->sk);
720 	release_sock((*sock)->sk);
721 
722 	sock_release(*sock);
723 	*sock = NULL;
724 }
725 
726 static void allow_connection_io(struct connection *con)
727 {
728 	if (con->othercon)
729 		clear_bit(CF_IO_STOP, &con->othercon->flags);
730 	clear_bit(CF_IO_STOP, &con->flags);
731 }
732 
733 static void stop_connection_io(struct connection *con)
734 {
735 	if (con->othercon)
736 		stop_connection_io(con->othercon);
737 
738 	spin_lock_bh(&con->writequeue_lock);
739 	set_bit(CF_IO_STOP, &con->flags);
740 	spin_unlock_bh(&con->writequeue_lock);
741 
742 	down_write(&con->sock_lock);
743 	if (con->sock) {
744 		lock_sock(con->sock->sk);
745 		restore_callbacks(con->sock->sk);
746 		release_sock(con->sock->sk);
747 	}
748 	up_write(&con->sock_lock);
749 
750 	cancel_work_sync(&con->swork);
751 	cancel_work_sync(&con->rwork);
752 }
753 
754 /* Close a remote connection and tidy up */
755 static void close_connection(struct connection *con, bool and_other)
756 {
757 	struct writequeue_entry *e;
758 
759 	if (con->othercon && and_other)
760 		close_connection(con->othercon, false);
761 
762 	down_write(&con->sock_lock);
763 	if (!con->sock) {
764 		up_write(&con->sock_lock);
765 		return;
766 	}
767 
768 	dlm_close_sock(&con->sock);
769 
770 	/* if we send a writequeue entry only a half way, we drop the
771 	 * whole entry because reconnection and that we not start of the
772 	 * middle of a msg which will confuse the other end.
773 	 *
774 	 * we can always drop messages because retransmits, but what we
775 	 * cannot allow is to transmit half messages which may be processed
776 	 * at the other side.
777 	 *
778 	 * our policy is to start on a clean state when disconnects, we don't
779 	 * know what's send/received on transport layer in this case.
780 	 */
781 	spin_lock_bh(&con->writequeue_lock);
782 	if (!list_empty(&con->writequeue)) {
783 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
784 				     list);
785 		if (e->dirty)
786 			free_entry(e);
787 	}
788 	spin_unlock_bh(&con->writequeue_lock);
789 
790 	con->rx_leftover = 0;
791 	con->retries = 0;
792 	clear_bit(CF_APP_LIMITED, &con->flags);
793 	clear_bit(CF_RECV_PENDING, &con->flags);
794 	clear_bit(CF_SEND_PENDING, &con->flags);
795 	up_write(&con->sock_lock);
796 }
797 
798 static void shutdown_connection(struct connection *con, bool and_other)
799 {
800 	int ret;
801 
802 	if (con->othercon && and_other)
803 		shutdown_connection(con->othercon, false);
804 
805 	flush_workqueue(io_workqueue);
806 	down_read(&con->sock_lock);
807 	/* nothing to shutdown */
808 	if (!con->sock) {
809 		up_read(&con->sock_lock);
810 		return;
811 	}
812 
813 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
814 	up_read(&con->sock_lock);
815 	if (ret) {
816 		log_print("Connection %p failed to shutdown: %d will force close",
817 			  con, ret);
818 		goto force_close;
819 	} else {
820 		ret = wait_event_timeout(con->shutdown_wait, !con->sock,
821 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
822 		if (ret == 0) {
823 			log_print("Connection %p shutdown timed out, will force close",
824 				  con);
825 			goto force_close;
826 		}
827 	}
828 
829 	return;
830 
831 force_close:
832 	close_connection(con, false);
833 }
834 
835 static struct processqueue_entry *new_processqueue_entry(int nodeid,
836 							 int buflen)
837 {
838 	struct processqueue_entry *pentry;
839 
840 	pentry = kmalloc(sizeof(*pentry), GFP_NOFS);
841 	if (!pentry)
842 		return NULL;
843 
844 	pentry->buf = kmalloc(buflen, GFP_NOFS);
845 	if (!pentry->buf) {
846 		kfree(pentry);
847 		return NULL;
848 	}
849 
850 	pentry->nodeid = nodeid;
851 	return pentry;
852 }
853 
854 static void free_processqueue_entry(struct processqueue_entry *pentry)
855 {
856 	kfree(pentry->buf);
857 	kfree(pentry);
858 }
859 
860 struct dlm_processed_nodes {
861 	int nodeid;
862 
863 	struct list_head list;
864 };
865 
866 static void process_dlm_messages(struct work_struct *work)
867 {
868 	struct processqueue_entry *pentry;
869 
870 	spin_lock(&processqueue_lock);
871 	pentry = list_first_entry_or_null(&processqueue,
872 					  struct processqueue_entry, list);
873 	if (WARN_ON_ONCE(!pentry)) {
874 		process_dlm_messages_pending = false;
875 		spin_unlock(&processqueue_lock);
876 		return;
877 	}
878 
879 	list_del(&pentry->list);
880 	atomic_dec(&processqueue_count);
881 	spin_unlock(&processqueue_lock);
882 
883 	for (;;) {
884 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
885 					    pentry->buflen);
886 		free_processqueue_entry(pentry);
887 
888 		spin_lock(&processqueue_lock);
889 		pentry = list_first_entry_or_null(&processqueue,
890 						  struct processqueue_entry, list);
891 		if (!pentry) {
892 			process_dlm_messages_pending = false;
893 			spin_unlock(&processqueue_lock);
894 			break;
895 		}
896 
897 		list_del(&pentry->list);
898 		atomic_dec(&processqueue_count);
899 		spin_unlock(&processqueue_lock);
900 	}
901 }
902 
903 /* Data received from remote end */
904 static int receive_from_sock(struct connection *con, int buflen)
905 {
906 	struct processqueue_entry *pentry;
907 	int ret, buflen_real;
908 	struct msghdr msg;
909 	struct kvec iov;
910 
911 	pentry = new_processqueue_entry(con->nodeid, buflen);
912 	if (!pentry)
913 		return DLM_IO_RESCHED;
914 
915 	memcpy(pentry->buf, con->rx_leftover_buf, con->rx_leftover);
916 
917 	/* calculate new buffer parameter regarding last receive and
918 	 * possible leftover bytes
919 	 */
920 	iov.iov_base = pentry->buf + con->rx_leftover;
921 	iov.iov_len = buflen - con->rx_leftover;
922 
923 	memset(&msg, 0, sizeof(msg));
924 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
925 	clear_bit(CF_RECV_INTR, &con->flags);
926 again:
927 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
928 			     msg.msg_flags);
929 	trace_dlm_recv(con->nodeid, ret);
930 	if (ret == -EAGAIN) {
931 		lock_sock(con->sock->sk);
932 		if (test_and_clear_bit(CF_RECV_INTR, &con->flags)) {
933 			release_sock(con->sock->sk);
934 			goto again;
935 		}
936 
937 		clear_bit(CF_RECV_PENDING, &con->flags);
938 		release_sock(con->sock->sk);
939 		free_processqueue_entry(pentry);
940 		return DLM_IO_END;
941 	} else if (ret == 0) {
942 		/* close will clear CF_RECV_PENDING */
943 		free_processqueue_entry(pentry);
944 		return DLM_IO_EOF;
945 	} else if (ret < 0) {
946 		free_processqueue_entry(pentry);
947 		return ret;
948 	}
949 
950 	/* new buflen according readed bytes and leftover from last receive */
951 	buflen_real = ret + con->rx_leftover;
952 	ret = dlm_validate_incoming_buffer(con->nodeid, pentry->buf,
953 					   buflen_real);
954 	if (ret < 0) {
955 		free_processqueue_entry(pentry);
956 		return ret;
957 	}
958 
959 	pentry->buflen = ret;
960 
961 	/* calculate leftover bytes from process and put it into begin of
962 	 * the receive buffer, so next receive we have the full message
963 	 * at the start address of the receive buffer.
964 	 */
965 	con->rx_leftover = buflen_real - ret;
966 	memmove(con->rx_leftover_buf, pentry->buf + ret,
967 		con->rx_leftover);
968 
969 	spin_lock(&processqueue_lock);
970 	ret = atomic_inc_return(&processqueue_count);
971 	list_add_tail(&pentry->list, &processqueue);
972 	if (!process_dlm_messages_pending) {
973 		process_dlm_messages_pending = true;
974 		queue_work(process_workqueue, &process_work);
975 	}
976 	spin_unlock(&processqueue_lock);
977 
978 	if (ret > DLM_MAX_PROCESS_BUFFERS)
979 		return DLM_IO_FLUSH;
980 
981 	return DLM_IO_SUCCESS;
982 }
983 
984 /* Listening socket is busy, accept a connection */
985 static int accept_from_sock(void)
986 {
987 	struct sockaddr_storage peeraddr;
988 	int len, idx, result, nodeid;
989 	struct connection *newcon;
990 	struct socket *newsock;
991 	unsigned int mark;
992 
993 	result = kernel_accept(listen_con.sock, &newsock, O_NONBLOCK);
994 	if (result == -EAGAIN)
995 		return DLM_IO_END;
996 	else if (result < 0)
997 		goto accept_err;
998 
999 	/* Get the connected socket's peer */
1000 	memset(&peeraddr, 0, sizeof(peeraddr));
1001 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1002 	if (len < 0) {
1003 		result = -ECONNABORTED;
1004 		goto accept_err;
1005 	}
1006 
1007 	/* Get the new node's NODEID */
1008 	make_sockaddr(&peeraddr, 0, &len);
1009 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1010 		switch (peeraddr.ss_family) {
1011 		case AF_INET: {
1012 			struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr;
1013 
1014 			log_print("connect from non cluster IPv4 node %pI4",
1015 				  &sin->sin_addr);
1016 			break;
1017 		}
1018 #if IS_ENABLED(CONFIG_IPV6)
1019 		case AF_INET6: {
1020 			struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr;
1021 
1022 			log_print("connect from non cluster IPv6 node %pI6c",
1023 				  &sin6->sin6_addr);
1024 			break;
1025 		}
1026 #endif
1027 		default:
1028 			log_print("invalid family from non cluster node");
1029 			break;
1030 		}
1031 
1032 		sock_release(newsock);
1033 		return -1;
1034 	}
1035 
1036 	log_print("got connection from %d", nodeid);
1037 
1038 	/*  Check to see if we already have a connection to this node. This
1039 	 *  could happen if the two nodes initiate a connection at roughly
1040 	 *  the same time and the connections cross on the wire.
1041 	 *  In this case we store the incoming one in "othercon"
1042 	 */
1043 	idx = srcu_read_lock(&connections_srcu);
1044 	newcon = nodeid2con(nodeid, 0);
1045 	if (WARN_ON_ONCE(!newcon)) {
1046 		srcu_read_unlock(&connections_srcu, idx);
1047 		result = -ENOENT;
1048 		goto accept_err;
1049 	}
1050 
1051 	sock_set_mark(newsock->sk, mark);
1052 
1053 	down_write(&newcon->sock_lock);
1054 	if (newcon->sock) {
1055 		struct connection *othercon = newcon->othercon;
1056 
1057 		if (!othercon) {
1058 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1059 			if (!othercon) {
1060 				log_print("failed to allocate incoming socket");
1061 				up_write(&newcon->sock_lock);
1062 				srcu_read_unlock(&connections_srcu, idx);
1063 				result = -ENOMEM;
1064 				goto accept_err;
1065 			}
1066 
1067 			dlm_con_init(othercon, nodeid);
1068 			lockdep_set_subclass(&othercon->sock_lock, 1);
1069 			newcon->othercon = othercon;
1070 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1071 		} else {
1072 			/* close other sock con if we have something new */
1073 			close_connection(othercon, false);
1074 		}
1075 
1076 		down_write(&othercon->sock_lock);
1077 		add_sock(newsock, othercon);
1078 
1079 		/* check if we receved something while adding */
1080 		lock_sock(othercon->sock->sk);
1081 		lowcomms_queue_rwork(othercon);
1082 		release_sock(othercon->sock->sk);
1083 		up_write(&othercon->sock_lock);
1084 	}
1085 	else {
1086 		/* accept copies the sk after we've saved the callbacks, so we
1087 		   don't want to save them a second time or comm errors will
1088 		   result in calling sk_error_report recursively. */
1089 		add_sock(newsock, newcon);
1090 
1091 		/* check if we receved something while adding */
1092 		lock_sock(newcon->sock->sk);
1093 		lowcomms_queue_rwork(newcon);
1094 		release_sock(newcon->sock->sk);
1095 	}
1096 	up_write(&newcon->sock_lock);
1097 	srcu_read_unlock(&connections_srcu, idx);
1098 
1099 	return DLM_IO_SUCCESS;
1100 
1101 accept_err:
1102 	if (newsock)
1103 		sock_release(newsock);
1104 
1105 	return result;
1106 }
1107 
1108 /*
1109  * writequeue_entry_complete - try to delete and free write queue entry
1110  * @e: write queue entry to try to delete
1111  * @completed: bytes completed
1112  *
1113  * writequeue_lock must be held.
1114  */
1115 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1116 {
1117 	e->offset += completed;
1118 	e->len -= completed;
1119 	/* signal that page was half way transmitted */
1120 	e->dirty = true;
1121 
1122 	if (e->len == 0 && e->users == 0)
1123 		free_entry(e);
1124 }
1125 
1126 /*
1127  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1128  */
1129 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1130 {
1131 	struct sockaddr_storage localaddr;
1132 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1133 	int i, addr_len, result = 0;
1134 
1135 	for (i = 0; i < dlm_local_count; i++) {
1136 		memcpy(&localaddr, &dlm_local_addr[i], sizeof(localaddr));
1137 		make_sockaddr(&localaddr, port, &addr_len);
1138 
1139 		if (!i)
1140 			result = kernel_bind(sock, addr, addr_len);
1141 		else
1142 			result = sock_bind_add(sock->sk, addr, addr_len);
1143 
1144 		if (result < 0) {
1145 			log_print("Can't bind to %d addr number %d, %d.\n",
1146 				  port, i + 1, result);
1147 			break;
1148 		}
1149 	}
1150 	return result;
1151 }
1152 
1153 /* Get local addresses */
1154 static void init_local(void)
1155 {
1156 	struct sockaddr_storage sas;
1157 	int i;
1158 
1159 	dlm_local_count = 0;
1160 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1161 		if (dlm_our_addr(&sas, i))
1162 			break;
1163 
1164 		memcpy(&dlm_local_addr[dlm_local_count++], &sas, sizeof(sas));
1165 	}
1166 }
1167 
1168 static struct writequeue_entry *new_writequeue_entry(struct connection *con)
1169 {
1170 	struct writequeue_entry *entry;
1171 
1172 	entry = dlm_allocate_writequeue();
1173 	if (!entry)
1174 		return NULL;
1175 
1176 	entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
1177 	if (!entry->page) {
1178 		dlm_free_writequeue(entry);
1179 		return NULL;
1180 	}
1181 
1182 	entry->offset = 0;
1183 	entry->len = 0;
1184 	entry->end = 0;
1185 	entry->dirty = false;
1186 	entry->con = con;
1187 	entry->users = 1;
1188 	kref_init(&entry->ref);
1189 	return entry;
1190 }
1191 
1192 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1193 					     char **ppc, void (*cb)(void *data),
1194 					     void *data)
1195 {
1196 	struct writequeue_entry *e;
1197 
1198 	spin_lock_bh(&con->writequeue_lock);
1199 	if (!list_empty(&con->writequeue)) {
1200 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1201 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1202 			kref_get(&e->ref);
1203 
1204 			*ppc = page_address(e->page) + e->end;
1205 			if (cb)
1206 				cb(data);
1207 
1208 			e->end += len;
1209 			e->users++;
1210 			goto out;
1211 		}
1212 	}
1213 
1214 	e = new_writequeue_entry(con);
1215 	if (!e)
1216 		goto out;
1217 
1218 	kref_get(&e->ref);
1219 	*ppc = page_address(e->page);
1220 	e->end += len;
1221 	if (cb)
1222 		cb(data);
1223 
1224 	list_add_tail(&e->list, &con->writequeue);
1225 
1226 out:
1227 	spin_unlock_bh(&con->writequeue_lock);
1228 	return e;
1229 };
1230 
1231 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1232 						gfp_t allocation, char **ppc,
1233 						void (*cb)(void *data),
1234 						void *data)
1235 {
1236 	struct writequeue_entry *e;
1237 	struct dlm_msg *msg;
1238 
1239 	msg = dlm_allocate_msg(allocation);
1240 	if (!msg)
1241 		return NULL;
1242 
1243 	kref_init(&msg->ref);
1244 
1245 	e = new_wq_entry(con, len, ppc, cb, data);
1246 	if (!e) {
1247 		dlm_free_msg(msg);
1248 		return NULL;
1249 	}
1250 
1251 	msg->retransmit = false;
1252 	msg->orig_msg = NULL;
1253 	msg->ppc = *ppc;
1254 	msg->len = len;
1255 	msg->entry = e;
1256 
1257 	return msg;
1258 }
1259 
1260 /* avoid false positive for nodes_srcu, unlock happens in
1261  * dlm_lowcomms_commit_msg which is a must call if success
1262  */
1263 #ifndef __CHECKER__
1264 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1265 				     char **ppc, void (*cb)(void *data),
1266 				     void *data)
1267 {
1268 	struct connection *con;
1269 	struct dlm_msg *msg;
1270 	int idx;
1271 
1272 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1273 	    len < sizeof(struct dlm_header)) {
1274 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1275 		log_print("failed to allocate a buffer of size %d", len);
1276 		WARN_ON_ONCE(1);
1277 		return NULL;
1278 	}
1279 
1280 	idx = srcu_read_lock(&connections_srcu);
1281 	con = nodeid2con(nodeid, 0);
1282 	if (WARN_ON_ONCE(!con)) {
1283 		srcu_read_unlock(&connections_srcu, idx);
1284 		return NULL;
1285 	}
1286 
1287 	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data);
1288 	if (!msg) {
1289 		srcu_read_unlock(&connections_srcu, idx);
1290 		return NULL;
1291 	}
1292 
1293 	/* for dlm_lowcomms_commit_msg() */
1294 	kref_get(&msg->ref);
1295 	/* we assume if successful commit must called */
1296 	msg->idx = idx;
1297 	return msg;
1298 }
1299 #endif
1300 
1301 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1302 {
1303 	struct writequeue_entry *e = msg->entry;
1304 	struct connection *con = e->con;
1305 	int users;
1306 
1307 	spin_lock_bh(&con->writequeue_lock);
1308 	kref_get(&msg->ref);
1309 	list_add(&msg->list, &e->msgs);
1310 
1311 	users = --e->users;
1312 	if (users)
1313 		goto out;
1314 
1315 	e->len = DLM_WQ_LENGTH_BYTES(e);
1316 
1317 	lowcomms_queue_swork(con);
1318 
1319 out:
1320 	spin_unlock_bh(&con->writequeue_lock);
1321 	return;
1322 }
1323 
1324 /* avoid false positive for nodes_srcu, lock was happen in
1325  * dlm_lowcomms_new_msg
1326  */
1327 #ifndef __CHECKER__
1328 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1329 {
1330 	_dlm_lowcomms_commit_msg(msg);
1331 	srcu_read_unlock(&connections_srcu, msg->idx);
1332 	/* because dlm_lowcomms_new_msg() */
1333 	kref_put(&msg->ref, dlm_msg_release);
1334 }
1335 #endif
1336 
1337 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1338 {
1339 	kref_put(&msg->ref, dlm_msg_release);
1340 }
1341 
1342 /* does not held connections_srcu, usage lowcomms_error_report only */
1343 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1344 {
1345 	struct dlm_msg *msg_resend;
1346 	char *ppc;
1347 
1348 	if (msg->retransmit)
1349 		return 1;
1350 
1351 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1352 					      GFP_ATOMIC, &ppc, NULL, NULL);
1353 	if (!msg_resend)
1354 		return -ENOMEM;
1355 
1356 	msg->retransmit = true;
1357 	kref_get(&msg->ref);
1358 	msg_resend->orig_msg = msg;
1359 
1360 	memcpy(ppc, msg->ppc, msg->len);
1361 	_dlm_lowcomms_commit_msg(msg_resend);
1362 	dlm_lowcomms_put_msg(msg_resend);
1363 
1364 	return 0;
1365 }
1366 
1367 /* Send a message */
1368 static int send_to_sock(struct connection *con)
1369 {
1370 	struct writequeue_entry *e;
1371 	struct bio_vec bvec;
1372 	struct msghdr msg = {
1373 		.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT | MSG_NOSIGNAL,
1374 	};
1375 	int len, offset, ret;
1376 
1377 	spin_lock_bh(&con->writequeue_lock);
1378 	e = con_next_wq(con);
1379 	if (!e) {
1380 		clear_bit(CF_SEND_PENDING, &con->flags);
1381 		spin_unlock_bh(&con->writequeue_lock);
1382 		return DLM_IO_END;
1383 	}
1384 
1385 	len = e->len;
1386 	offset = e->offset;
1387 	WARN_ON_ONCE(len == 0 && e->users == 0);
1388 	spin_unlock_bh(&con->writequeue_lock);
1389 
1390 	bvec_set_page(&bvec, e->page, len, offset);
1391 	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, len);
1392 	ret = sock_sendmsg(con->sock, &msg);
1393 	trace_dlm_send(con->nodeid, ret);
1394 	if (ret == -EAGAIN || ret == 0) {
1395 		lock_sock(con->sock->sk);
1396 		spin_lock_bh(&con->writequeue_lock);
1397 		if (test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1398 		    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1399 			/* Notify TCP that we're limited by the
1400 			 * application window size.
1401 			 */
1402 			set_bit(SOCK_NOSPACE, &con->sock->sk->sk_socket->flags);
1403 			con->sock->sk->sk_write_pending++;
1404 
1405 			clear_bit(CF_SEND_PENDING, &con->flags);
1406 			spin_unlock_bh(&con->writequeue_lock);
1407 			release_sock(con->sock->sk);
1408 
1409 			/* wait for write_space() event */
1410 			return DLM_IO_END;
1411 		}
1412 		spin_unlock_bh(&con->writequeue_lock);
1413 		release_sock(con->sock->sk);
1414 
1415 		return DLM_IO_RESCHED;
1416 	} else if (ret < 0) {
1417 		return ret;
1418 	}
1419 
1420 	spin_lock_bh(&con->writequeue_lock);
1421 	writequeue_entry_complete(e, ret);
1422 	spin_unlock_bh(&con->writequeue_lock);
1423 
1424 	return DLM_IO_SUCCESS;
1425 }
1426 
1427 static void clean_one_writequeue(struct connection *con)
1428 {
1429 	struct writequeue_entry *e, *safe;
1430 
1431 	spin_lock_bh(&con->writequeue_lock);
1432 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1433 		free_entry(e);
1434 	}
1435 	spin_unlock_bh(&con->writequeue_lock);
1436 }
1437 
1438 static void connection_release(struct rcu_head *rcu)
1439 {
1440 	struct connection *con = container_of(rcu, struct connection, rcu);
1441 
1442 	WARN_ON_ONCE(!list_empty(&con->writequeue));
1443 	WARN_ON_ONCE(con->sock);
1444 	kfree(con);
1445 }
1446 
1447 /* Called from recovery when it knows that a node has
1448    left the cluster */
1449 int dlm_lowcomms_close(int nodeid)
1450 {
1451 	struct connection *con;
1452 	int idx;
1453 
1454 	log_print("closing connection to node %d", nodeid);
1455 
1456 	idx = srcu_read_lock(&connections_srcu);
1457 	con = nodeid2con(nodeid, 0);
1458 	if (WARN_ON_ONCE(!con)) {
1459 		srcu_read_unlock(&connections_srcu, idx);
1460 		return -ENOENT;
1461 	}
1462 
1463 	stop_connection_io(con);
1464 	log_print("io handling for node: %d stopped", nodeid);
1465 	close_connection(con, true);
1466 
1467 	spin_lock(&connections_lock);
1468 	hlist_del_rcu(&con->list);
1469 	spin_unlock(&connections_lock);
1470 
1471 	clean_one_writequeue(con);
1472 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1473 	if (con->othercon) {
1474 		clean_one_writequeue(con->othercon);
1475 		call_srcu(&connections_srcu, &con->othercon->rcu, connection_release);
1476 	}
1477 	srcu_read_unlock(&connections_srcu, idx);
1478 
1479 	/* for debugging we print when we are done to compare with other
1480 	 * messages in between. This function need to be correctly synchronized
1481 	 * with io handling
1482 	 */
1483 	log_print("closing connection to node %d done", nodeid);
1484 
1485 	return 0;
1486 }
1487 
1488 /* Receive worker function */
1489 static void process_recv_sockets(struct work_struct *work)
1490 {
1491 	struct connection *con = container_of(work, struct connection, rwork);
1492 	int ret, buflen;
1493 
1494 	down_read(&con->sock_lock);
1495 	if (!con->sock) {
1496 		up_read(&con->sock_lock);
1497 		return;
1498 	}
1499 
1500 	buflen = READ_ONCE(dlm_config.ci_buffer_size);
1501 	do {
1502 		ret = receive_from_sock(con, buflen);
1503 	} while (ret == DLM_IO_SUCCESS);
1504 	up_read(&con->sock_lock);
1505 
1506 	switch (ret) {
1507 	case DLM_IO_END:
1508 		/* CF_RECV_PENDING cleared */
1509 		break;
1510 	case DLM_IO_EOF:
1511 		close_connection(con, false);
1512 		wake_up(&con->shutdown_wait);
1513 		/* CF_RECV_PENDING cleared */
1514 		break;
1515 	case DLM_IO_FLUSH:
1516 		flush_workqueue(process_workqueue);
1517 		fallthrough;
1518 	case DLM_IO_RESCHED:
1519 		cond_resched();
1520 		queue_work(io_workqueue, &con->rwork);
1521 		/* CF_RECV_PENDING not cleared */
1522 		break;
1523 	default:
1524 		if (ret < 0) {
1525 			if (test_bit(CF_IS_OTHERCON, &con->flags)) {
1526 				close_connection(con, false);
1527 			} else {
1528 				spin_lock_bh(&con->writequeue_lock);
1529 				lowcomms_queue_swork(con);
1530 				spin_unlock_bh(&con->writequeue_lock);
1531 			}
1532 
1533 			/* CF_RECV_PENDING cleared for othercon
1534 			 * we trigger send queue if not already done
1535 			 * and process_send_sockets will handle it
1536 			 */
1537 			break;
1538 		}
1539 
1540 		WARN_ON_ONCE(1);
1541 		break;
1542 	}
1543 }
1544 
1545 static void process_listen_recv_socket(struct work_struct *work)
1546 {
1547 	int ret;
1548 
1549 	if (WARN_ON_ONCE(!listen_con.sock))
1550 		return;
1551 
1552 	do {
1553 		ret = accept_from_sock();
1554 	} while (ret == DLM_IO_SUCCESS);
1555 
1556 	if (ret < 0)
1557 		log_print("critical error accepting connection: %d", ret);
1558 }
1559 
1560 static int dlm_connect(struct connection *con)
1561 {
1562 	struct sockaddr_storage addr;
1563 	int result, addr_len;
1564 	struct socket *sock;
1565 	unsigned int mark;
1566 
1567 	memset(&addr, 0, sizeof(addr));
1568 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1569 				dlm_proto_ops->try_new_addr, &mark);
1570 	if (result < 0) {
1571 		log_print("no address for nodeid %d", con->nodeid);
1572 		return result;
1573 	}
1574 
1575 	/* Create a socket to communicate with */
1576 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1577 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1578 	if (result < 0)
1579 		return result;
1580 
1581 	sock_set_mark(sock->sk, mark);
1582 	dlm_proto_ops->sockopts(sock);
1583 
1584 	result = dlm_proto_ops->bind(sock);
1585 	if (result < 0) {
1586 		sock_release(sock);
1587 		return result;
1588 	}
1589 
1590 	add_sock(sock, con);
1591 
1592 	log_print_ratelimited("connecting to %d", con->nodeid);
1593 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1594 	result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1595 					addr_len);
1596 	switch (result) {
1597 	case -EINPROGRESS:
1598 		/* not an error */
1599 		fallthrough;
1600 	case 0:
1601 		break;
1602 	default:
1603 		if (result < 0)
1604 			dlm_close_sock(&con->sock);
1605 
1606 		break;
1607 	}
1608 
1609 	return result;
1610 }
1611 
1612 /* Send worker function */
1613 static void process_send_sockets(struct work_struct *work)
1614 {
1615 	struct connection *con = container_of(work, struct connection, swork);
1616 	int ret;
1617 
1618 	WARN_ON_ONCE(test_bit(CF_IS_OTHERCON, &con->flags));
1619 
1620 	down_read(&con->sock_lock);
1621 	if (!con->sock) {
1622 		up_read(&con->sock_lock);
1623 		down_write(&con->sock_lock);
1624 		if (!con->sock) {
1625 			ret = dlm_connect(con);
1626 			switch (ret) {
1627 			case 0:
1628 				break;
1629 			case -EINPROGRESS:
1630 				/* avoid spamming resched on connection
1631 				 * we might can switch to a state_change
1632 				 * event based mechanism if established
1633 				 */
1634 				msleep(100);
1635 				break;
1636 			default:
1637 				/* CF_SEND_PENDING not cleared */
1638 				up_write(&con->sock_lock);
1639 				log_print("connect to node %d try %d error %d",
1640 					  con->nodeid, con->retries++, ret);
1641 				msleep(1000);
1642 				/* For now we try forever to reconnect. In
1643 				 * future we should send a event to cluster
1644 				 * manager to fence itself after certain amount
1645 				 * of retries.
1646 				 */
1647 				queue_work(io_workqueue, &con->swork);
1648 				return;
1649 			}
1650 		}
1651 		downgrade_write(&con->sock_lock);
1652 	}
1653 
1654 	do {
1655 		ret = send_to_sock(con);
1656 	} while (ret == DLM_IO_SUCCESS);
1657 	up_read(&con->sock_lock);
1658 
1659 	switch (ret) {
1660 	case DLM_IO_END:
1661 		/* CF_SEND_PENDING cleared */
1662 		break;
1663 	case DLM_IO_RESCHED:
1664 		/* CF_SEND_PENDING not cleared */
1665 		cond_resched();
1666 		queue_work(io_workqueue, &con->swork);
1667 		break;
1668 	default:
1669 		if (ret < 0) {
1670 			close_connection(con, false);
1671 
1672 			/* CF_SEND_PENDING cleared */
1673 			spin_lock_bh(&con->writequeue_lock);
1674 			lowcomms_queue_swork(con);
1675 			spin_unlock_bh(&con->writequeue_lock);
1676 			break;
1677 		}
1678 
1679 		WARN_ON_ONCE(1);
1680 		break;
1681 	}
1682 }
1683 
1684 static void work_stop(void)
1685 {
1686 	if (io_workqueue) {
1687 		destroy_workqueue(io_workqueue);
1688 		io_workqueue = NULL;
1689 	}
1690 
1691 	if (process_workqueue) {
1692 		destroy_workqueue(process_workqueue);
1693 		process_workqueue = NULL;
1694 	}
1695 }
1696 
1697 static int work_start(void)
1698 {
1699 	io_workqueue = alloc_workqueue("dlm_io", WQ_HIGHPRI | WQ_MEM_RECLAIM |
1700 				       WQ_UNBOUND, 0);
1701 	if (!io_workqueue) {
1702 		log_print("can't start dlm_io");
1703 		return -ENOMEM;
1704 	}
1705 
1706 	/* ordered dlm message process queue,
1707 	 * should be converted to a tasklet
1708 	 */
1709 	process_workqueue = alloc_ordered_workqueue("dlm_process",
1710 						    WQ_HIGHPRI | WQ_MEM_RECLAIM);
1711 	if (!process_workqueue) {
1712 		log_print("can't start dlm_process");
1713 		destroy_workqueue(io_workqueue);
1714 		io_workqueue = NULL;
1715 		return -ENOMEM;
1716 	}
1717 
1718 	return 0;
1719 }
1720 
1721 void dlm_lowcomms_shutdown(void)
1722 {
1723 	struct connection *con;
1724 	int i, idx;
1725 
1726 	/* stop lowcomms_listen_data_ready calls */
1727 	lock_sock(listen_con.sock->sk);
1728 	listen_con.sock->sk->sk_data_ready = listen_sock.sk_data_ready;
1729 	release_sock(listen_con.sock->sk);
1730 
1731 	cancel_work_sync(&listen_con.rwork);
1732 	dlm_close_sock(&listen_con.sock);
1733 
1734 	idx = srcu_read_lock(&connections_srcu);
1735 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1736 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
1737 			shutdown_connection(con, true);
1738 			stop_connection_io(con);
1739 			flush_workqueue(process_workqueue);
1740 			close_connection(con, true);
1741 
1742 			clean_one_writequeue(con);
1743 			if (con->othercon)
1744 				clean_one_writequeue(con->othercon);
1745 			allow_connection_io(con);
1746 		}
1747 	}
1748 	srcu_read_unlock(&connections_srcu, idx);
1749 }
1750 
1751 void dlm_lowcomms_stop(void)
1752 {
1753 	work_stop();
1754 	dlm_proto_ops = NULL;
1755 }
1756 
1757 static int dlm_listen_for_all(void)
1758 {
1759 	struct socket *sock;
1760 	int result;
1761 
1762 	log_print("Using %s for communications",
1763 		  dlm_proto_ops->name);
1764 
1765 	result = dlm_proto_ops->listen_validate();
1766 	if (result < 0)
1767 		return result;
1768 
1769 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1770 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1771 	if (result < 0) {
1772 		log_print("Can't create comms socket: %d", result);
1773 		return result;
1774 	}
1775 
1776 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1777 	dlm_proto_ops->listen_sockopts(sock);
1778 
1779 	result = dlm_proto_ops->listen_bind(sock);
1780 	if (result < 0)
1781 		goto out;
1782 
1783 	lock_sock(sock->sk);
1784 	listen_sock.sk_data_ready = sock->sk->sk_data_ready;
1785 	listen_sock.sk_write_space = sock->sk->sk_write_space;
1786 	listen_sock.sk_error_report = sock->sk->sk_error_report;
1787 	listen_sock.sk_state_change = sock->sk->sk_state_change;
1788 
1789 	listen_con.sock = sock;
1790 
1791 	sock->sk->sk_allocation = GFP_NOFS;
1792 	sock->sk->sk_use_task_frag = false;
1793 	sock->sk->sk_data_ready = lowcomms_listen_data_ready;
1794 	release_sock(sock->sk);
1795 
1796 	result = sock->ops->listen(sock, 128);
1797 	if (result < 0) {
1798 		dlm_close_sock(&listen_con.sock);
1799 		return result;
1800 	}
1801 
1802 	return 0;
1803 
1804 out:
1805 	sock_release(sock);
1806 	return result;
1807 }
1808 
1809 static int dlm_tcp_bind(struct socket *sock)
1810 {
1811 	struct sockaddr_storage src_addr;
1812 	int result, addr_len;
1813 
1814 	/* Bind to our cluster-known address connecting to avoid
1815 	 * routing problems.
1816 	 */
1817 	memcpy(&src_addr, &dlm_local_addr[0], sizeof(src_addr));
1818 	make_sockaddr(&src_addr, 0, &addr_len);
1819 
1820 	result = kernel_bind(sock, (struct sockaddr *)&src_addr,
1821 			     addr_len);
1822 	if (result < 0) {
1823 		/* This *may* not indicate a critical error */
1824 		log_print("could not bind for connect: %d", result);
1825 	}
1826 
1827 	return 0;
1828 }
1829 
1830 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1831 			   struct sockaddr *addr, int addr_len)
1832 {
1833 	return kernel_connect(sock, addr, addr_len, O_NONBLOCK);
1834 }
1835 
1836 static int dlm_tcp_listen_validate(void)
1837 {
1838 	/* We don't support multi-homed hosts */
1839 	if (dlm_local_count > 1) {
1840 		log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1841 		return -EINVAL;
1842 	}
1843 
1844 	return 0;
1845 }
1846 
1847 static void dlm_tcp_sockopts(struct socket *sock)
1848 {
1849 	/* Turn off Nagle's algorithm */
1850 	tcp_sock_set_nodelay(sock->sk);
1851 }
1852 
1853 static void dlm_tcp_listen_sockopts(struct socket *sock)
1854 {
1855 	dlm_tcp_sockopts(sock);
1856 	sock_set_reuseaddr(sock->sk);
1857 }
1858 
1859 static int dlm_tcp_listen_bind(struct socket *sock)
1860 {
1861 	int addr_len;
1862 
1863 	/* Bind to our port */
1864 	make_sockaddr(&dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1865 	return kernel_bind(sock, (struct sockaddr *)&dlm_local_addr[0],
1866 			   addr_len);
1867 }
1868 
1869 static const struct dlm_proto_ops dlm_tcp_ops = {
1870 	.name = "TCP",
1871 	.proto = IPPROTO_TCP,
1872 	.connect = dlm_tcp_connect,
1873 	.sockopts = dlm_tcp_sockopts,
1874 	.bind = dlm_tcp_bind,
1875 	.listen_validate = dlm_tcp_listen_validate,
1876 	.listen_sockopts = dlm_tcp_listen_sockopts,
1877 	.listen_bind = dlm_tcp_listen_bind,
1878 };
1879 
1880 static int dlm_sctp_bind(struct socket *sock)
1881 {
1882 	return sctp_bind_addrs(sock, 0);
1883 }
1884 
1885 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1886 			    struct sockaddr *addr, int addr_len)
1887 {
1888 	int ret;
1889 
1890 	/*
1891 	 * Make kernel_connect() function return in specified time,
1892 	 * since O_NONBLOCK argument in connect() function does not work here,
1893 	 * then, we should restore the default value of this attribute.
1894 	 */
1895 	sock_set_sndtimeo(sock->sk, 5);
1896 	ret = kernel_connect(sock, addr, addr_len, 0);
1897 	sock_set_sndtimeo(sock->sk, 0);
1898 	return ret;
1899 }
1900 
1901 static int dlm_sctp_listen_validate(void)
1902 {
1903 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1904 		log_print("SCTP is not enabled by this kernel");
1905 		return -EOPNOTSUPP;
1906 	}
1907 
1908 	request_module("sctp");
1909 	return 0;
1910 }
1911 
1912 static int dlm_sctp_bind_listen(struct socket *sock)
1913 {
1914 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1915 }
1916 
1917 static void dlm_sctp_sockopts(struct socket *sock)
1918 {
1919 	/* Turn off Nagle's algorithm */
1920 	sctp_sock_set_nodelay(sock->sk);
1921 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1922 }
1923 
1924 static const struct dlm_proto_ops dlm_sctp_ops = {
1925 	.name = "SCTP",
1926 	.proto = IPPROTO_SCTP,
1927 	.try_new_addr = true,
1928 	.connect = dlm_sctp_connect,
1929 	.sockopts = dlm_sctp_sockopts,
1930 	.bind = dlm_sctp_bind,
1931 	.listen_validate = dlm_sctp_listen_validate,
1932 	.listen_sockopts = dlm_sctp_sockopts,
1933 	.listen_bind = dlm_sctp_bind_listen,
1934 };
1935 
1936 int dlm_lowcomms_start(void)
1937 {
1938 	int error;
1939 
1940 	init_local();
1941 	if (!dlm_local_count) {
1942 		error = -ENOTCONN;
1943 		log_print("no local IP address has been set");
1944 		goto fail;
1945 	}
1946 
1947 	error = work_start();
1948 	if (error)
1949 		goto fail;
1950 
1951 	/* Start listening */
1952 	switch (dlm_config.ci_protocol) {
1953 	case DLM_PROTO_TCP:
1954 		dlm_proto_ops = &dlm_tcp_ops;
1955 		break;
1956 	case DLM_PROTO_SCTP:
1957 		dlm_proto_ops = &dlm_sctp_ops;
1958 		break;
1959 	default:
1960 		log_print("Invalid protocol identifier %d set",
1961 			  dlm_config.ci_protocol);
1962 		error = -EINVAL;
1963 		goto fail_proto_ops;
1964 	}
1965 
1966 	error = dlm_listen_for_all();
1967 	if (error)
1968 		goto fail_listen;
1969 
1970 	return 0;
1971 
1972 fail_listen:
1973 	dlm_proto_ops = NULL;
1974 fail_proto_ops:
1975 	work_stop();
1976 fail:
1977 	return error;
1978 }
1979 
1980 void dlm_lowcomms_init(void)
1981 {
1982 	int i;
1983 
1984 	for (i = 0; i < CONN_HASH_SIZE; i++)
1985 		INIT_HLIST_HEAD(&connection_hash[i]);
1986 
1987 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1988 }
1989 
1990 void dlm_lowcomms_exit(void)
1991 {
1992 	struct connection *con;
1993 	int i, idx;
1994 
1995 	idx = srcu_read_lock(&connections_srcu);
1996 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1997 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
1998 			spin_lock(&connections_lock);
1999 			hlist_del_rcu(&con->list);
2000 			spin_unlock(&connections_lock);
2001 
2002 			if (con->othercon)
2003 				call_srcu(&connections_srcu, &con->othercon->rcu,
2004 					  connection_release);
2005 			call_srcu(&connections_srcu, &con->rcu, connection_release);
2006 		}
2007 	}
2008 	srcu_read_unlock(&connections_srcu, idx);
2009 }
2010