xref: /linux/fs/dlm/lowcomms.c (revision a1ed04430f805364eeffe9afacf3ec0d152a8e83)
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 	int how;
164 
165 	void (*sockopts)(struct socket *sock);
166 	int (*bind)(struct socket *sock);
167 	int (*listen_validate)(void);
168 	void (*listen_sockopts)(struct socket *sock);
169 	int (*listen_bind)(struct socket *sock);
170 };
171 
172 static struct listen_sock_callbacks {
173 	void (*sk_error_report)(struct sock *);
174 	void (*sk_data_ready)(struct sock *);
175 	void (*sk_state_change)(struct sock *);
176 	void (*sk_write_space)(struct sock *);
177 } listen_sock;
178 
179 static struct listen_connection listen_con;
180 static struct sockaddr_storage dlm_local_addr[DLM_MAX_ADDR_COUNT];
181 static int dlm_local_count;
182 
183 /* Work queues */
184 static struct workqueue_struct *io_workqueue;
185 static struct workqueue_struct *process_workqueue;
186 
187 static struct hlist_head connection_hash[CONN_HASH_SIZE];
188 static DEFINE_SPINLOCK(connections_lock);
189 DEFINE_STATIC_SRCU(connections_srcu);
190 
191 static const struct dlm_proto_ops *dlm_proto_ops;
192 
193 #define DLM_IO_SUCCESS 0
194 #define DLM_IO_END 1
195 #define DLM_IO_EOF 2
196 #define DLM_IO_RESCHED 3
197 #define DLM_IO_FLUSH 4
198 
199 static void process_recv_sockets(struct work_struct *work);
200 static void process_send_sockets(struct work_struct *work);
201 static void process_dlm_messages(struct work_struct *work);
202 
203 static DECLARE_WORK(process_work, process_dlm_messages);
204 static DEFINE_SPINLOCK(processqueue_lock);
205 static bool process_dlm_messages_pending;
206 static DECLARE_WAIT_QUEUE_HEAD(processqueue_wq);
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(dlm_msg, 0);
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_srcu(con, &connection_hash[r], list,
275 			srcu_read_lock_held(&connections_srcu)) {
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_obj(*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_srcu(con, &connection_hash[i], list,
431 				srcu_read_lock_held(&connections_srcu)) {
432 			WARN_ON_ONCE(!con->addr_count);
433 
434 			spin_lock(&con->addrs_lock);
435 			for (addr_i = 0; addr_i < con->addr_count; addr_i++) {
436 				if (addr_compare(&con->addr[addr_i], addr)) {
437 					*nodeid = con->nodeid;
438 					*mark = con->mark;
439 					spin_unlock(&con->addrs_lock);
440 					srcu_read_unlock(&connections_srcu, idx);
441 					return 0;
442 				}
443 			}
444 			spin_unlock(&con->addrs_lock);
445 		}
446 	}
447 	srcu_read_unlock(&connections_srcu, idx);
448 
449 	return -ENOENT;
450 }
451 
452 static bool dlm_lowcomms_con_has_addr(const struct connection *con,
453 				      const struct sockaddr_storage *addr)
454 {
455 	int i;
456 
457 	for (i = 0; i < con->addr_count; i++) {
458 		if (addr_compare(&con->addr[i], addr))
459 			return true;
460 	}
461 
462 	return false;
463 }
464 
465 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr)
466 {
467 	struct connection *con;
468 	bool ret;
469 	int idx;
470 
471 	idx = srcu_read_lock(&connections_srcu);
472 	con = nodeid2con(nodeid, GFP_NOFS);
473 	if (!con) {
474 		srcu_read_unlock(&connections_srcu, idx);
475 		return -ENOMEM;
476 	}
477 
478 	spin_lock(&con->addrs_lock);
479 	if (!con->addr_count) {
480 		memcpy(&con->addr[0], addr, sizeof(*addr));
481 		con->addr_count = 1;
482 		con->mark = dlm_config.ci_mark;
483 		spin_unlock(&con->addrs_lock);
484 		srcu_read_unlock(&connections_srcu, idx);
485 		return 0;
486 	}
487 
488 	ret = dlm_lowcomms_con_has_addr(con, addr);
489 	if (ret) {
490 		spin_unlock(&con->addrs_lock);
491 		srcu_read_unlock(&connections_srcu, idx);
492 		return -EEXIST;
493 	}
494 
495 	if (con->addr_count >= DLM_MAX_ADDR_COUNT) {
496 		spin_unlock(&con->addrs_lock);
497 		srcu_read_unlock(&connections_srcu, idx);
498 		return -ENOSPC;
499 	}
500 
501 	memcpy(&con->addr[con->addr_count++], addr, sizeof(*addr));
502 	srcu_read_unlock(&connections_srcu, idx);
503 	spin_unlock(&con->addrs_lock);
504 	return 0;
505 }
506 
507 /* Data available on socket or listen socket received a connect */
508 static void lowcomms_data_ready(struct sock *sk)
509 {
510 	struct connection *con = sock2con(sk);
511 
512 	trace_sk_data_ready(sk);
513 
514 	set_bit(CF_RECV_INTR, &con->flags);
515 	lowcomms_queue_rwork(con);
516 }
517 
518 static void lowcomms_write_space(struct sock *sk)
519 {
520 	struct connection *con = sock2con(sk);
521 
522 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
523 
524 	spin_lock_bh(&con->writequeue_lock);
525 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
526 		con->sock->sk->sk_write_pending--;
527 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
528 	}
529 
530 	lowcomms_queue_swork(con);
531 	spin_unlock_bh(&con->writequeue_lock);
532 }
533 
534 static void lowcomms_state_change(struct sock *sk)
535 {
536 	/* SCTP layer is not calling sk_data_ready when the connection
537 	 * is done, so we catch the signal through here.
538 	 */
539 	if (sk->sk_shutdown & RCV_SHUTDOWN)
540 		lowcomms_data_ready(sk);
541 }
542 
543 static void lowcomms_listen_data_ready(struct sock *sk)
544 {
545 	trace_sk_data_ready(sk);
546 
547 	queue_work(io_workqueue, &listen_con.rwork);
548 }
549 
550 int dlm_lowcomms_connect_node(int nodeid)
551 {
552 	struct connection *con;
553 	int idx;
554 
555 	idx = srcu_read_lock(&connections_srcu);
556 	con = nodeid2con(nodeid, 0);
557 	if (WARN_ON_ONCE(!con)) {
558 		srcu_read_unlock(&connections_srcu, idx);
559 		return -ENOENT;
560 	}
561 
562 	down_read(&con->sock_lock);
563 	if (!con->sock) {
564 		spin_lock_bh(&con->writequeue_lock);
565 		lowcomms_queue_swork(con);
566 		spin_unlock_bh(&con->writequeue_lock);
567 	}
568 	up_read(&con->sock_lock);
569 	srcu_read_unlock(&connections_srcu, idx);
570 
571 	cond_resched();
572 	return 0;
573 }
574 
575 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
576 {
577 	struct connection *con;
578 	int idx;
579 
580 	idx = srcu_read_lock(&connections_srcu);
581 	con = nodeid2con(nodeid, 0);
582 	if (!con) {
583 		srcu_read_unlock(&connections_srcu, idx);
584 		return -ENOENT;
585 	}
586 
587 	spin_lock(&con->addrs_lock);
588 	con->mark = mark;
589 	spin_unlock(&con->addrs_lock);
590 	srcu_read_unlock(&connections_srcu, idx);
591 	return 0;
592 }
593 
594 static void lowcomms_error_report(struct sock *sk)
595 {
596 	struct connection *con = sock2con(sk);
597 	struct inet_sock *inet;
598 
599 	inet = inet_sk(sk);
600 	switch (sk->sk_family) {
601 	case AF_INET:
602 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
603 				   "sending to node %d at %pI4, dport %d, "
604 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
605 				   con->nodeid, &inet->inet_daddr,
606 				   ntohs(inet->inet_dport), sk->sk_err,
607 				   READ_ONCE(sk->sk_err_soft));
608 		break;
609 #if IS_ENABLED(CONFIG_IPV6)
610 	case AF_INET6:
611 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
612 				   "sending to node %d at %pI6c, "
613 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
614 				   con->nodeid, &sk->sk_v6_daddr,
615 				   ntohs(inet->inet_dport), sk->sk_err,
616 				   READ_ONCE(sk->sk_err_soft));
617 		break;
618 #endif
619 	default:
620 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
621 				   "invalid socket family %d set, "
622 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
623 				   sk->sk_family, sk->sk_err,
624 				   READ_ONCE(sk->sk_err_soft));
625 		break;
626 	}
627 
628 	dlm_midcomms_unack_msg_resend(con->nodeid);
629 
630 	listen_sock.sk_error_report(sk);
631 }
632 
633 static void restore_callbacks(struct sock *sk)
634 {
635 #ifdef CONFIG_LOCKDEP
636 	WARN_ON_ONCE(!lockdep_sock_is_held(sk));
637 #endif
638 
639 	sk->sk_user_data = NULL;
640 	sk->sk_data_ready = listen_sock.sk_data_ready;
641 	sk->sk_state_change = listen_sock.sk_state_change;
642 	sk->sk_write_space = listen_sock.sk_write_space;
643 	sk->sk_error_report = listen_sock.sk_error_report;
644 }
645 
646 /* Make a socket active */
647 static void add_sock(struct socket *sock, struct connection *con)
648 {
649 	struct sock *sk = sock->sk;
650 
651 	lock_sock(sk);
652 	con->sock = sock;
653 
654 	sk->sk_user_data = con;
655 	sk->sk_data_ready = lowcomms_data_ready;
656 	sk->sk_write_space = lowcomms_write_space;
657 	if (dlm_config.ci_protocol == DLM_PROTO_SCTP)
658 		sk->sk_state_change = lowcomms_state_change;
659 	sk->sk_allocation = GFP_NOFS;
660 	sk->sk_use_task_frag = false;
661 	sk->sk_error_report = lowcomms_error_report;
662 	release_sock(sk);
663 }
664 
665 /* Add the port number to an IPv6 or 4 sockaddr and return the address
666    length */
667 static void make_sockaddr(struct sockaddr_storage *saddr, __be16 port,
668 			  int *addr_len)
669 {
670 	saddr->ss_family =  dlm_local_addr[0].ss_family;
671 	if (saddr->ss_family == AF_INET) {
672 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
673 		in4_addr->sin_port = port;
674 		*addr_len = sizeof(struct sockaddr_in);
675 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
676 	} else {
677 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
678 		in6_addr->sin6_port = port;
679 		*addr_len = sizeof(struct sockaddr_in6);
680 	}
681 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
682 }
683 
684 static void dlm_page_release(struct kref *kref)
685 {
686 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
687 						  ref);
688 
689 	__free_page(e->page);
690 	dlm_free_writequeue(e);
691 }
692 
693 static void dlm_msg_release(struct kref *kref)
694 {
695 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
696 
697 	kref_put(&msg->entry->ref, dlm_page_release);
698 	dlm_free_msg(msg);
699 }
700 
701 static void free_entry(struct writequeue_entry *e)
702 {
703 	struct dlm_msg *msg, *tmp;
704 
705 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
706 		if (msg->orig_msg) {
707 			msg->orig_msg->retransmit = false;
708 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
709 		}
710 
711 		list_del(&msg->list);
712 		kref_put(&msg->ref, dlm_msg_release);
713 	}
714 
715 	list_del(&e->list);
716 	kref_put(&e->ref, dlm_page_release);
717 }
718 
719 static void dlm_close_sock(struct socket **sock)
720 {
721 	lock_sock((*sock)->sk);
722 	restore_callbacks((*sock)->sk);
723 	release_sock((*sock)->sk);
724 
725 	sock_release(*sock);
726 	*sock = NULL;
727 }
728 
729 static void allow_connection_io(struct connection *con)
730 {
731 	if (con->othercon)
732 		clear_bit(CF_IO_STOP, &con->othercon->flags);
733 	clear_bit(CF_IO_STOP, &con->flags);
734 }
735 
736 static void stop_connection_io(struct connection *con)
737 {
738 	if (con->othercon)
739 		stop_connection_io(con->othercon);
740 
741 	spin_lock_bh(&con->writequeue_lock);
742 	set_bit(CF_IO_STOP, &con->flags);
743 	spin_unlock_bh(&con->writequeue_lock);
744 
745 	down_write(&con->sock_lock);
746 	if (con->sock) {
747 		lock_sock(con->sock->sk);
748 		restore_callbacks(con->sock->sk);
749 		release_sock(con->sock->sk);
750 	}
751 	up_write(&con->sock_lock);
752 
753 	cancel_work_sync(&con->swork);
754 	cancel_work_sync(&con->rwork);
755 }
756 
757 /* Close a remote connection and tidy up */
758 static void close_connection(struct connection *con, bool and_other)
759 {
760 	struct writequeue_entry *e;
761 
762 	if (con->othercon && and_other)
763 		close_connection(con->othercon, false);
764 
765 	down_write(&con->sock_lock);
766 	if (!con->sock) {
767 		up_write(&con->sock_lock);
768 		return;
769 	}
770 
771 	dlm_close_sock(&con->sock);
772 
773 	/* if we send a writequeue entry only a half way, we drop the
774 	 * whole entry because reconnection and that we not start of the
775 	 * middle of a msg which will confuse the other end.
776 	 *
777 	 * we can always drop messages because retransmits, but what we
778 	 * cannot allow is to transmit half messages which may be processed
779 	 * at the other side.
780 	 *
781 	 * our policy is to start on a clean state when disconnects, we don't
782 	 * know what's send/received on transport layer in this case.
783 	 */
784 	spin_lock_bh(&con->writequeue_lock);
785 	if (!list_empty(&con->writequeue)) {
786 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
787 				     list);
788 		if (e->dirty)
789 			free_entry(e);
790 	}
791 	spin_unlock_bh(&con->writequeue_lock);
792 
793 	con->rx_leftover = 0;
794 	con->retries = 0;
795 	clear_bit(CF_APP_LIMITED, &con->flags);
796 	clear_bit(CF_RECV_PENDING, &con->flags);
797 	clear_bit(CF_SEND_PENDING, &con->flags);
798 	up_write(&con->sock_lock);
799 }
800 
801 static void shutdown_connection(struct connection *con, bool and_other)
802 {
803 	int ret;
804 
805 	if (con->othercon && and_other)
806 		shutdown_connection(con->othercon, false);
807 
808 	flush_workqueue(io_workqueue);
809 	down_read(&con->sock_lock);
810 	/* nothing to shutdown */
811 	if (!con->sock) {
812 		up_read(&con->sock_lock);
813 		return;
814 	}
815 
816 	ret = kernel_sock_shutdown(con->sock, dlm_proto_ops->how);
817 	up_read(&con->sock_lock);
818 	if (ret) {
819 		log_print("Connection %p failed to shutdown: %d will force close",
820 			  con, ret);
821 		goto force_close;
822 	} else {
823 		ret = wait_event_timeout(con->shutdown_wait, !con->sock,
824 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
825 		if (ret == 0) {
826 			log_print("Connection %p shutdown timed out, will force close",
827 				  con);
828 			goto force_close;
829 		}
830 	}
831 
832 	return;
833 
834 force_close:
835 	close_connection(con, false);
836 }
837 
838 static struct processqueue_entry *new_processqueue_entry(int nodeid,
839 							 int buflen)
840 {
841 	struct processqueue_entry *pentry;
842 
843 	pentry = kmalloc_obj(*pentry, GFP_NOFS);
844 	if (!pentry)
845 		return NULL;
846 
847 	pentry->buf = kmalloc(buflen, GFP_NOFS);
848 	if (!pentry->buf) {
849 		kfree(pentry);
850 		return NULL;
851 	}
852 
853 	pentry->nodeid = nodeid;
854 	return pentry;
855 }
856 
857 static void free_processqueue_entry(struct processqueue_entry *pentry)
858 {
859 	kfree(pentry->buf);
860 	kfree(pentry);
861 }
862 
863 static void process_dlm_messages(struct work_struct *work)
864 {
865 	struct processqueue_entry *pentry;
866 
867 	spin_lock_bh(&processqueue_lock);
868 	pentry = list_first_entry_or_null(&processqueue,
869 					  struct processqueue_entry, list);
870 	if (WARN_ON_ONCE(!pentry)) {
871 		process_dlm_messages_pending = false;
872 		spin_unlock_bh(&processqueue_lock);
873 		return;
874 	}
875 
876 	list_del(&pentry->list);
877 	if (atomic_dec_and_test(&processqueue_count))
878 		wake_up(&processqueue_wq);
879 	spin_unlock_bh(&processqueue_lock);
880 
881 	for (;;) {
882 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
883 					    pentry->buflen);
884 		free_processqueue_entry(pentry);
885 
886 		spin_lock_bh(&processqueue_lock);
887 		pentry = list_first_entry_or_null(&processqueue,
888 						  struct processqueue_entry, list);
889 		if (!pentry) {
890 			process_dlm_messages_pending = false;
891 			spin_unlock_bh(&processqueue_lock);
892 			break;
893 		}
894 
895 		list_del(&pentry->list);
896 		if (atomic_dec_and_test(&processqueue_count))
897 			wake_up(&processqueue_wq);
898 		spin_unlock_bh(&processqueue_lock);
899 	}
900 }
901 
902 /* Data received from remote end */
903 static int receive_from_sock(struct connection *con, int buflen)
904 {
905 	struct processqueue_entry *pentry;
906 	int ret, buflen_real;
907 	struct msghdr msg;
908 	struct kvec iov;
909 
910 	pentry = new_processqueue_entry(con->nodeid, buflen);
911 	if (!pentry)
912 		return DLM_IO_RESCHED;
913 
914 	memcpy(pentry->buf, con->rx_leftover_buf, con->rx_leftover);
915 
916 	/* calculate new buffer parameter regarding last receive and
917 	 * possible leftover bytes
918 	 */
919 	iov.iov_base = pentry->buf + con->rx_leftover;
920 	iov.iov_len = buflen - con->rx_leftover;
921 
922 	memset(&msg, 0, sizeof(msg));
923 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
924 	clear_bit(CF_RECV_INTR, &con->flags);
925 again:
926 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
927 			     msg.msg_flags);
928 	trace_dlm_recv(con->nodeid, ret);
929 	if (ret == -EAGAIN) {
930 		lock_sock(con->sock->sk);
931 		if (test_and_clear_bit(CF_RECV_INTR, &con->flags)) {
932 			release_sock(con->sock->sk);
933 			goto again;
934 		}
935 
936 		clear_bit(CF_RECV_PENDING, &con->flags);
937 		release_sock(con->sock->sk);
938 		free_processqueue_entry(pentry);
939 		return DLM_IO_END;
940 	} else if (ret == 0) {
941 		/* close will clear CF_RECV_PENDING */
942 		free_processqueue_entry(pentry);
943 		return DLM_IO_EOF;
944 	} else if (ret < 0) {
945 		free_processqueue_entry(pentry);
946 		return ret;
947 	}
948 
949 	/* new buflen according readed bytes and leftover from last receive */
950 	buflen_real = ret + con->rx_leftover;
951 	ret = dlm_validate_incoming_buffer(con->nodeid, pentry->buf,
952 					   buflen_real);
953 	if (ret < 0) {
954 		free_processqueue_entry(pentry);
955 		return ret;
956 	}
957 
958 	pentry->buflen = ret;
959 
960 	/* calculate leftover bytes from process and put it into begin of
961 	 * the receive buffer, so next receive we have the full message
962 	 * at the start address of the receive buffer.
963 	 */
964 	con->rx_leftover = buflen_real - ret;
965 	memmove(con->rx_leftover_buf, pentry->buf + ret,
966 		con->rx_leftover);
967 
968 	spin_lock_bh(&processqueue_lock);
969 	ret = atomic_inc_return(&processqueue_count);
970 	list_add_tail(&pentry->list, &processqueue);
971 	if (!process_dlm_messages_pending) {
972 		process_dlm_messages_pending = true;
973 		queue_work(process_workqueue, &process_work);
974 	}
975 	spin_unlock_bh(&processqueue_lock);
976 
977 	if (ret > DLM_MAX_PROCESS_BUFFERS)
978 		return DLM_IO_FLUSH;
979 
980 	return DLM_IO_SUCCESS;
981 }
982 
983 /* Listening socket is busy, accept a connection */
984 static int accept_from_sock(void)
985 {
986 	struct sockaddr_storage peeraddr;
987 	int len, idx, result, nodeid;
988 	struct connection *newcon;
989 	struct socket *newsock;
990 	unsigned int mark;
991 
992 	result = kernel_accept(listen_con.sock, &newsock, O_NONBLOCK);
993 	if (result == -EAGAIN)
994 		return DLM_IO_END;
995 	else if (result < 0)
996 		goto accept_err;
997 
998 	/* Get the connected socket's peer */
999 	memset(&peeraddr, 0, sizeof(peeraddr));
1000 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
1001 	if (len < 0) {
1002 		result = -ECONNABORTED;
1003 		goto accept_err;
1004 	}
1005 
1006 	/* Get the new node's NODEID */
1007 	make_sockaddr(&peeraddr, 0, &len);
1008 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1009 		switch (peeraddr.ss_family) {
1010 		case AF_INET: {
1011 			struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr;
1012 
1013 			log_print("connect from non cluster IPv4 node %pI4",
1014 				  &sin->sin_addr);
1015 			break;
1016 		}
1017 #if IS_ENABLED(CONFIG_IPV6)
1018 		case AF_INET6: {
1019 			struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr;
1020 
1021 			log_print("connect from non cluster IPv6 node %pI6c",
1022 				  &sin6->sin6_addr);
1023 			break;
1024 		}
1025 #endif
1026 		default:
1027 			log_print("invalid family from non cluster node");
1028 			break;
1029 		}
1030 
1031 		sock_release(newsock);
1032 		return -1;
1033 	}
1034 
1035 	log_print("got connection from %d", nodeid);
1036 
1037 	/*  Check to see if we already have a connection to this node. This
1038 	 *  could happen if the two nodes initiate a connection at roughly
1039 	 *  the same time and the connections cross on the wire.
1040 	 *  In this case we store the incoming one in "othercon"
1041 	 */
1042 	idx = srcu_read_lock(&connections_srcu);
1043 	newcon = nodeid2con(nodeid, 0);
1044 	if (WARN_ON_ONCE(!newcon)) {
1045 		srcu_read_unlock(&connections_srcu, idx);
1046 		result = -ENOENT;
1047 		goto accept_err;
1048 	}
1049 
1050 	sock_set_mark(newsock->sk, mark);
1051 
1052 	down_write(&newcon->sock_lock);
1053 	if (newcon->sock) {
1054 		struct connection *othercon = newcon->othercon;
1055 
1056 		if (!othercon) {
1057 			othercon = kzalloc_obj(*othercon, GFP_NOFS);
1058 			if (!othercon) {
1059 				log_print("failed to allocate incoming socket");
1060 				up_write(&newcon->sock_lock);
1061 				srcu_read_unlock(&connections_srcu, idx);
1062 				result = -ENOMEM;
1063 				goto accept_err;
1064 			}
1065 
1066 			dlm_con_init(othercon, nodeid);
1067 			lockdep_set_subclass(&othercon->sock_lock, 1);
1068 			newcon->othercon = othercon;
1069 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1070 		} else {
1071 			/* close other sock con if we have something new */
1072 			close_connection(othercon, false);
1073 		}
1074 
1075 		down_write(&othercon->sock_lock);
1076 		add_sock(newsock, othercon);
1077 
1078 		/* check if we receved something while adding */
1079 		lock_sock(othercon->sock->sk);
1080 		lowcomms_queue_rwork(othercon);
1081 		release_sock(othercon->sock->sk);
1082 		up_write(&othercon->sock_lock);
1083 	}
1084 	else {
1085 		/* accept copies the sk after we've saved the callbacks, so we
1086 		   don't want to save them a second time or comm errors will
1087 		   result in calling sk_error_report recursively. */
1088 		add_sock(newsock, newcon);
1089 
1090 		/* check if we receved something while adding */
1091 		lock_sock(newcon->sock->sk);
1092 		lowcomms_queue_rwork(newcon);
1093 		release_sock(newcon->sock->sk);
1094 	}
1095 	up_write(&newcon->sock_lock);
1096 	srcu_read_unlock(&connections_srcu, idx);
1097 
1098 	return DLM_IO_SUCCESS;
1099 
1100 accept_err:
1101 	if (newsock)
1102 		sock_release(newsock);
1103 
1104 	return result;
1105 }
1106 
1107 /*
1108  * writequeue_entry_complete - try to delete and free write queue entry
1109  * @e: write queue entry to try to delete
1110  * @completed: bytes completed
1111  *
1112  * writequeue_lock must be held.
1113  */
1114 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1115 {
1116 	e->offset += completed;
1117 	e->len -= completed;
1118 	/* signal that page was half way transmitted */
1119 	e->dirty = true;
1120 
1121 	if (e->len == 0 && e->users == 0)
1122 		free_entry(e);
1123 }
1124 
1125 /*
1126  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1127  */
1128 static int sctp_bind_addrs(struct socket *sock, __be16 port)
1129 {
1130 	struct sockaddr_storage localaddr;
1131 	struct sockaddr_unsized *addr = (struct sockaddr_unsized *)&localaddr;
1132 	int i, addr_len, result = 0;
1133 
1134 	for (i = 0; i < dlm_local_count; i++) {
1135 		memcpy(&localaddr, &dlm_local_addr[i], sizeof(localaddr));
1136 		make_sockaddr(&localaddr, port, &addr_len);
1137 
1138 		if (!i)
1139 			result = kernel_bind(sock, addr, addr_len);
1140 		else
1141 			result = sock_bind_add(sock->sk, addr, addr_len);
1142 
1143 		if (result < 0) {
1144 			log_print("Can't bind to %d addr number %d, %d.\n",
1145 				  port, i + 1, result);
1146 			break;
1147 		}
1148 	}
1149 	return result;
1150 }
1151 
1152 /* Get local addresses */
1153 static void init_local(void)
1154 {
1155 	struct sockaddr_storage sas;
1156 	int i;
1157 
1158 	dlm_local_count = 0;
1159 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1160 		if (dlm_our_addr(&sas, i))
1161 			break;
1162 
1163 		memcpy(&dlm_local_addr[dlm_local_count++], &sas, sizeof(sas));
1164 	}
1165 }
1166 
1167 static struct writequeue_entry *new_writequeue_entry(struct connection *con)
1168 {
1169 	struct writequeue_entry *entry;
1170 
1171 	entry = dlm_allocate_writequeue();
1172 	if (!entry)
1173 		return NULL;
1174 
1175 	entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
1176 	if (!entry->page) {
1177 		dlm_free_writequeue(entry);
1178 		return NULL;
1179 	}
1180 
1181 	entry->offset = 0;
1182 	entry->len = 0;
1183 	entry->end = 0;
1184 	entry->dirty = false;
1185 	entry->con = con;
1186 	entry->users = 1;
1187 	kref_init(&entry->ref);
1188 	return entry;
1189 }
1190 
1191 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1192 					     char **ppc, void (*cb)(void *data),
1193 					     void *data)
1194 {
1195 	struct writequeue_entry *e;
1196 
1197 	spin_lock_bh(&con->writequeue_lock);
1198 	if (!list_empty(&con->writequeue)) {
1199 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1200 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1201 			kref_get(&e->ref);
1202 
1203 			*ppc = page_address(e->page) + e->end;
1204 			if (cb)
1205 				cb(data);
1206 
1207 			e->end += len;
1208 			e->users++;
1209 			goto out;
1210 		}
1211 	}
1212 
1213 	e = new_writequeue_entry(con);
1214 	if (!e)
1215 		goto out;
1216 
1217 	kref_get(&e->ref);
1218 	*ppc = page_address(e->page);
1219 	e->end += len;
1220 	if (cb)
1221 		cb(data);
1222 
1223 	list_add_tail(&e->list, &con->writequeue);
1224 
1225 out:
1226 	spin_unlock_bh(&con->writequeue_lock);
1227 	return e;
1228 };
1229 
1230 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1231 						char **ppc, void (*cb)(void *data),
1232 						void *data)
1233 {
1234 	struct writequeue_entry *e;
1235 	struct dlm_msg *msg;
1236 
1237 	msg = dlm_allocate_msg();
1238 	if (!msg)
1239 		return NULL;
1240 
1241 	kref_init(&msg->ref);
1242 
1243 	e = new_wq_entry(con, len, ppc, cb, data);
1244 	if (!e) {
1245 		dlm_free_msg(msg);
1246 		return NULL;
1247 	}
1248 
1249 	msg->retransmit = false;
1250 	msg->orig_msg = NULL;
1251 	msg->ppc = *ppc;
1252 	msg->len = len;
1253 	msg->entry = e;
1254 
1255 	return msg;
1256 }
1257 
1258 /* avoid false positive for nodes_srcu, unlock happens in
1259  * dlm_lowcomms_commit_msg which is a must call if success
1260  */
1261 #ifndef __CHECKER__
1262 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, char **ppc,
1263 				     void (*cb)(void *data), void *data)
1264 {
1265 	struct connection *con;
1266 	struct dlm_msg *msg;
1267 	int idx;
1268 
1269 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1270 	    len < sizeof(struct dlm_header)) {
1271 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1272 		log_print("failed to allocate a buffer of size %d", len);
1273 		WARN_ON_ONCE(1);
1274 		return NULL;
1275 	}
1276 
1277 	idx = srcu_read_lock(&connections_srcu);
1278 	con = nodeid2con(nodeid, 0);
1279 	if (WARN_ON_ONCE(!con)) {
1280 		srcu_read_unlock(&connections_srcu, idx);
1281 		return NULL;
1282 	}
1283 
1284 	msg = dlm_lowcomms_new_msg_con(con, len, ppc, cb, data);
1285 	if (!msg) {
1286 		srcu_read_unlock(&connections_srcu, idx);
1287 		return NULL;
1288 	}
1289 
1290 	/* for dlm_lowcomms_commit_msg() */
1291 	kref_get(&msg->ref);
1292 	/* we assume if successful commit must called */
1293 	msg->idx = idx;
1294 	return msg;
1295 }
1296 #endif
1297 
1298 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1299 {
1300 	struct writequeue_entry *e = msg->entry;
1301 	struct connection *con = e->con;
1302 	int users;
1303 
1304 	spin_lock_bh(&con->writequeue_lock);
1305 	kref_get(&msg->ref);
1306 	list_add(&msg->list, &e->msgs);
1307 
1308 	users = --e->users;
1309 	if (users)
1310 		goto out;
1311 
1312 	e->len = DLM_WQ_LENGTH_BYTES(e);
1313 
1314 	lowcomms_queue_swork(con);
1315 
1316 out:
1317 	spin_unlock_bh(&con->writequeue_lock);
1318 	return;
1319 }
1320 
1321 /* avoid false positive for nodes_srcu, lock was happen in
1322  * dlm_lowcomms_new_msg
1323  */
1324 #ifndef __CHECKER__
1325 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1326 {
1327 	_dlm_lowcomms_commit_msg(msg);
1328 	srcu_read_unlock(&connections_srcu, msg->idx);
1329 	/* because dlm_lowcomms_new_msg() */
1330 	kref_put(&msg->ref, dlm_msg_release);
1331 }
1332 #endif
1333 
1334 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1335 {
1336 	kref_put(&msg->ref, dlm_msg_release);
1337 }
1338 
1339 /* does not held connections_srcu, usage lowcomms_error_report only */
1340 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1341 {
1342 	struct dlm_msg *msg_resend;
1343 	char *ppc;
1344 
1345 	if (msg->retransmit)
1346 		return 1;
1347 
1348 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len, &ppc,
1349 					      NULL, NULL);
1350 	if (!msg_resend)
1351 		return -ENOMEM;
1352 
1353 	msg->retransmit = true;
1354 	kref_get(&msg->ref);
1355 	msg_resend->orig_msg = msg;
1356 
1357 	memcpy(ppc, msg->ppc, msg->len);
1358 	_dlm_lowcomms_commit_msg(msg_resend);
1359 	dlm_lowcomms_put_msg(msg_resend);
1360 
1361 	return 0;
1362 }
1363 
1364 /* Send a message */
1365 static int send_to_sock(struct connection *con)
1366 {
1367 	struct writequeue_entry *e;
1368 	struct bio_vec bvec;
1369 	struct msghdr msg = {
1370 		.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT | MSG_NOSIGNAL,
1371 	};
1372 	int len, offset, ret;
1373 
1374 	spin_lock_bh(&con->writequeue_lock);
1375 	e = con_next_wq(con);
1376 	if (!e) {
1377 		clear_bit(CF_SEND_PENDING, &con->flags);
1378 		spin_unlock_bh(&con->writequeue_lock);
1379 		return DLM_IO_END;
1380 	}
1381 
1382 	len = e->len;
1383 	offset = e->offset;
1384 	WARN_ON_ONCE(len == 0 && e->users == 0);
1385 	spin_unlock_bh(&con->writequeue_lock);
1386 
1387 	bvec_set_page(&bvec, e->page, len, offset);
1388 	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, len);
1389 	ret = sock_sendmsg(con->sock, &msg);
1390 	trace_dlm_send(con->nodeid, ret);
1391 	if (ret == -EAGAIN || ret == 0) {
1392 		lock_sock(con->sock->sk);
1393 		spin_lock_bh(&con->writequeue_lock);
1394 		if (test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1395 		    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1396 			/* Notify TCP that we're limited by the
1397 			 * application window size.
1398 			 */
1399 			set_bit(SOCK_NOSPACE, &con->sock->sk->sk_socket->flags);
1400 			con->sock->sk->sk_write_pending++;
1401 
1402 			clear_bit(CF_SEND_PENDING, &con->flags);
1403 			spin_unlock_bh(&con->writequeue_lock);
1404 			release_sock(con->sock->sk);
1405 
1406 			/* wait for write_space() event */
1407 			return DLM_IO_END;
1408 		}
1409 		spin_unlock_bh(&con->writequeue_lock);
1410 		release_sock(con->sock->sk);
1411 
1412 		return DLM_IO_RESCHED;
1413 	} else if (ret < 0) {
1414 		return ret;
1415 	}
1416 
1417 	spin_lock_bh(&con->writequeue_lock);
1418 	writequeue_entry_complete(e, ret);
1419 	spin_unlock_bh(&con->writequeue_lock);
1420 
1421 	return DLM_IO_SUCCESS;
1422 }
1423 
1424 static void clean_one_writequeue(struct connection *con)
1425 {
1426 	struct writequeue_entry *e, *safe;
1427 
1428 	spin_lock_bh(&con->writequeue_lock);
1429 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1430 		free_entry(e);
1431 	}
1432 	spin_unlock_bh(&con->writequeue_lock);
1433 }
1434 
1435 static void connection_release(struct rcu_head *rcu)
1436 {
1437 	struct connection *con = container_of(rcu, struct connection, rcu);
1438 
1439 	WARN_ON_ONCE(!list_empty(&con->writequeue));
1440 	WARN_ON_ONCE(con->sock);
1441 	kfree(con);
1442 }
1443 
1444 /* Called from recovery when it knows that a node has
1445    left the cluster */
1446 int dlm_lowcomms_close(int nodeid)
1447 {
1448 	struct connection *con;
1449 	int idx;
1450 
1451 	log_print("closing connection to node %d", nodeid);
1452 
1453 	idx = srcu_read_lock(&connections_srcu);
1454 	con = nodeid2con(nodeid, 0);
1455 	if (WARN_ON_ONCE(!con)) {
1456 		srcu_read_unlock(&connections_srcu, idx);
1457 		return -ENOENT;
1458 	}
1459 
1460 	stop_connection_io(con);
1461 	log_print("io handling for node: %d stopped", nodeid);
1462 	close_connection(con, true);
1463 
1464 	spin_lock(&connections_lock);
1465 	hlist_del_rcu(&con->list);
1466 	spin_unlock(&connections_lock);
1467 
1468 	clean_one_writequeue(con);
1469 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1470 	if (con->othercon) {
1471 		clean_one_writequeue(con->othercon);
1472 		call_srcu(&connections_srcu, &con->othercon->rcu, connection_release);
1473 	}
1474 	srcu_read_unlock(&connections_srcu, idx);
1475 
1476 	/* for debugging we print when we are done to compare with other
1477 	 * messages in between. This function need to be correctly synchronized
1478 	 * with io handling
1479 	 */
1480 	log_print("closing connection to node %d done", nodeid);
1481 
1482 	return 0;
1483 }
1484 
1485 /* Receive worker function */
1486 static void process_recv_sockets(struct work_struct *work)
1487 {
1488 	struct connection *con = container_of(work, struct connection, rwork);
1489 	int ret, buflen;
1490 
1491 	down_read(&con->sock_lock);
1492 	if (!con->sock) {
1493 		up_read(&con->sock_lock);
1494 		return;
1495 	}
1496 
1497 	buflen = READ_ONCE(dlm_config.ci_buffer_size);
1498 	do {
1499 		ret = receive_from_sock(con, buflen);
1500 	} while (ret == DLM_IO_SUCCESS);
1501 	up_read(&con->sock_lock);
1502 
1503 	switch (ret) {
1504 	case DLM_IO_END:
1505 		/* CF_RECV_PENDING cleared */
1506 		break;
1507 	case DLM_IO_EOF:
1508 		close_connection(con, false);
1509 		wake_up(&con->shutdown_wait);
1510 		/* CF_RECV_PENDING cleared */
1511 		break;
1512 	case DLM_IO_FLUSH:
1513 		/* we can't flush the process_workqueue here because a
1514 		 * WQ_MEM_RECLAIM workequeue can occurr a deadlock for a non
1515 		 * WQ_MEM_RECLAIM workqueue such as process_workqueue. Instead
1516 		 * we have a waitqueue to wait until all messages are
1517 		 * processed.
1518 		 *
1519 		 * This handling is only necessary to backoff the sender and
1520 		 * not queue all messages from the socket layer into DLM
1521 		 * processqueue. When DLM is capable to parse multiple messages
1522 		 * on an e.g. per socket basis this handling can might be
1523 		 * removed. Especially in a message burst we are too slow to
1524 		 * process messages and the queue will fill up memory.
1525 		 */
1526 		wait_event(processqueue_wq, !atomic_read(&processqueue_count));
1527 		fallthrough;
1528 	case DLM_IO_RESCHED:
1529 		cond_resched();
1530 		queue_work(io_workqueue, &con->rwork);
1531 		/* CF_RECV_PENDING not cleared */
1532 		break;
1533 	default:
1534 		if (ret < 0) {
1535 			if (test_bit(CF_IS_OTHERCON, &con->flags)) {
1536 				close_connection(con, false);
1537 			} else {
1538 				spin_lock_bh(&con->writequeue_lock);
1539 				lowcomms_queue_swork(con);
1540 				spin_unlock_bh(&con->writequeue_lock);
1541 			}
1542 
1543 			/* CF_RECV_PENDING cleared for othercon
1544 			 * we trigger send queue if not already done
1545 			 * and process_send_sockets will handle it
1546 			 */
1547 			break;
1548 		}
1549 
1550 		WARN_ON_ONCE(1);
1551 		break;
1552 	}
1553 }
1554 
1555 static void process_listen_recv_socket(struct work_struct *work)
1556 {
1557 	int ret;
1558 
1559 	if (WARN_ON_ONCE(!listen_con.sock))
1560 		return;
1561 
1562 	do {
1563 		ret = accept_from_sock();
1564 	} while (ret == DLM_IO_SUCCESS);
1565 
1566 	if (ret < 0)
1567 		log_print("critical error accepting connection: %d", ret);
1568 }
1569 
1570 static int dlm_connect(struct connection *con)
1571 {
1572 	struct sockaddr_storage addr;
1573 	int result, addr_len;
1574 	struct socket *sock;
1575 	unsigned int mark;
1576 
1577 	memset(&addr, 0, sizeof(addr));
1578 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1579 				dlm_proto_ops->try_new_addr, &mark);
1580 	if (result < 0) {
1581 		log_print("no address for nodeid %d", con->nodeid);
1582 		return result;
1583 	}
1584 
1585 	/* Create a socket to communicate with */
1586 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1587 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1588 	if (result < 0)
1589 		return result;
1590 
1591 	sock_set_mark(sock->sk, mark);
1592 	dlm_proto_ops->sockopts(sock);
1593 
1594 	result = dlm_proto_ops->bind(sock);
1595 	if (result < 0) {
1596 		sock_release(sock);
1597 		return result;
1598 	}
1599 
1600 	add_sock(sock, con);
1601 
1602 	log_print_ratelimited("connecting to %d", con->nodeid);
1603 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1604 	result = kernel_connect(sock, (struct sockaddr_unsized *)&addr, addr_len, 0);
1605 	switch (result) {
1606 	case -EINPROGRESS:
1607 		/* not an error */
1608 		fallthrough;
1609 	case 0:
1610 		break;
1611 	default:
1612 		if (result < 0)
1613 			dlm_close_sock(&con->sock);
1614 
1615 		break;
1616 	}
1617 
1618 	return result;
1619 }
1620 
1621 /* Send worker function */
1622 static void process_send_sockets(struct work_struct *work)
1623 {
1624 	struct connection *con = container_of(work, struct connection, swork);
1625 	int ret;
1626 
1627 	WARN_ON_ONCE(test_bit(CF_IS_OTHERCON, &con->flags));
1628 
1629 	down_read(&con->sock_lock);
1630 	if (!con->sock) {
1631 		up_read(&con->sock_lock);
1632 		down_write(&con->sock_lock);
1633 		if (!con->sock) {
1634 			ret = dlm_connect(con);
1635 			switch (ret) {
1636 			case 0:
1637 				break;
1638 			default:
1639 				/* CF_SEND_PENDING not cleared */
1640 				up_write(&con->sock_lock);
1641 				log_print("connect to node %d try %d error %d",
1642 					  con->nodeid, con->retries++, ret);
1643 				msleep(1000);
1644 				/* For now we try forever to reconnect. In
1645 				 * future we should send a event to cluster
1646 				 * manager to fence itself after certain amount
1647 				 * of retries.
1648 				 */
1649 				queue_work(io_workqueue, &con->swork);
1650 				return;
1651 			}
1652 		}
1653 		downgrade_write(&con->sock_lock);
1654 	}
1655 
1656 	do {
1657 		ret = send_to_sock(con);
1658 	} while (ret == DLM_IO_SUCCESS);
1659 	up_read(&con->sock_lock);
1660 
1661 	switch (ret) {
1662 	case DLM_IO_END:
1663 		/* CF_SEND_PENDING cleared */
1664 		break;
1665 	case DLM_IO_RESCHED:
1666 		/* CF_SEND_PENDING not cleared */
1667 		cond_resched();
1668 		queue_work(io_workqueue, &con->swork);
1669 		break;
1670 	default:
1671 		if (ret < 0) {
1672 			close_connection(con, false);
1673 
1674 			/* CF_SEND_PENDING cleared */
1675 			spin_lock_bh(&con->writequeue_lock);
1676 			lowcomms_queue_swork(con);
1677 			spin_unlock_bh(&con->writequeue_lock);
1678 			break;
1679 		}
1680 
1681 		WARN_ON_ONCE(1);
1682 		break;
1683 	}
1684 }
1685 
1686 static void work_stop(void)
1687 {
1688 	if (io_workqueue) {
1689 		destroy_workqueue(io_workqueue);
1690 		io_workqueue = NULL;
1691 	}
1692 
1693 	if (process_workqueue) {
1694 		destroy_workqueue(process_workqueue);
1695 		process_workqueue = NULL;
1696 	}
1697 }
1698 
1699 static int work_start(void)
1700 {
1701 	io_workqueue = alloc_workqueue("dlm_io", WQ_HIGHPRI | WQ_MEM_RECLAIM |
1702 				       WQ_UNBOUND, 0);
1703 	if (!io_workqueue) {
1704 		log_print("can't start dlm_io");
1705 		return -ENOMEM;
1706 	}
1707 
1708 	process_workqueue = alloc_workqueue("dlm_process", WQ_HIGHPRI | WQ_BH | WQ_PERCPU, 0);
1709 	if (!process_workqueue) {
1710 		log_print("can't start dlm_process");
1711 		destroy_workqueue(io_workqueue);
1712 		io_workqueue = NULL;
1713 		return -ENOMEM;
1714 	}
1715 
1716 	return 0;
1717 }
1718 
1719 void dlm_lowcomms_shutdown(void)
1720 {
1721 	struct connection *con;
1722 	int i, idx;
1723 
1724 	/* stop lowcomms_listen_data_ready calls */
1725 	lock_sock(listen_con.sock->sk);
1726 	listen_con.sock->sk->sk_data_ready = listen_sock.sk_data_ready;
1727 	release_sock(listen_con.sock->sk);
1728 
1729 	cancel_work_sync(&listen_con.rwork);
1730 	dlm_close_sock(&listen_con.sock);
1731 
1732 	idx = srcu_read_lock(&connections_srcu);
1733 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1734 		hlist_for_each_entry_srcu(con, &connection_hash[i], list,
1735 				srcu_read_lock_held(&connections_srcu)) {
1736 			shutdown_connection(con, true);
1737 			stop_connection_io(con);
1738 			flush_workqueue(process_workqueue);
1739 			close_connection(con, true);
1740 
1741 			clean_one_writequeue(con);
1742 			if (con->othercon)
1743 				clean_one_writequeue(con->othercon);
1744 			allow_connection_io(con);
1745 		}
1746 	}
1747 	srcu_read_unlock(&connections_srcu, idx);
1748 }
1749 
1750 void dlm_lowcomms_stop(void)
1751 {
1752 	work_stop();
1753 	dlm_proto_ops = NULL;
1754 }
1755 
1756 static int dlm_listen_for_all(void)
1757 {
1758 	struct socket *sock;
1759 	int result;
1760 
1761 	log_print("Using %s for communications",
1762 		  dlm_proto_ops->name);
1763 
1764 	result = dlm_proto_ops->listen_validate();
1765 	if (result < 0)
1766 		return result;
1767 
1768 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1769 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1770 	if (result < 0) {
1771 		log_print("Can't create comms socket: %d", result);
1772 		return result;
1773 	}
1774 
1775 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1776 	dlm_proto_ops->listen_sockopts(sock);
1777 
1778 	result = dlm_proto_ops->listen_bind(sock);
1779 	if (result < 0)
1780 		goto out;
1781 
1782 	lock_sock(sock->sk);
1783 	listen_sock.sk_data_ready = sock->sk->sk_data_ready;
1784 	listen_sock.sk_write_space = sock->sk->sk_write_space;
1785 	listen_sock.sk_error_report = sock->sk->sk_error_report;
1786 	listen_sock.sk_state_change = sock->sk->sk_state_change;
1787 
1788 	listen_con.sock = sock;
1789 
1790 	sock->sk->sk_allocation = GFP_NOFS;
1791 	sock->sk->sk_use_task_frag = false;
1792 	sock->sk->sk_data_ready = lowcomms_listen_data_ready;
1793 	release_sock(sock->sk);
1794 
1795 	result = sock->ops->listen(sock, 128);
1796 	if (result < 0) {
1797 		dlm_close_sock(&listen_con.sock);
1798 		return result;
1799 	}
1800 
1801 	return 0;
1802 
1803 out:
1804 	sock_release(sock);
1805 	return result;
1806 }
1807 
1808 static int dlm_tcp_bind(struct socket *sock)
1809 {
1810 	struct sockaddr_storage src_addr;
1811 	int result, addr_len;
1812 
1813 	/* Bind to our cluster-known address connecting to avoid
1814 	 * routing problems.
1815 	 */
1816 	memcpy(&src_addr, &dlm_local_addr[0], sizeof(src_addr));
1817 	make_sockaddr(&src_addr, 0, &addr_len);
1818 
1819 	result = kernel_bind(sock, (struct sockaddr_unsized *)&src_addr,
1820 			     addr_len);
1821 	if (result < 0) {
1822 		/* This *may* not indicate a critical error */
1823 		log_print("could not bind for connect: %d", result);
1824 	}
1825 
1826 	return 0;
1827 }
1828 
1829 static int dlm_tcp_listen_validate(void)
1830 {
1831 	/* We don't support multi-homed hosts */
1832 	if (dlm_local_count > 1) {
1833 		log_print("Detect multi-homed hosts but use only the first IP address.");
1834 		log_print("Try SCTP, if you want to enable multi-link.");
1835 	}
1836 
1837 	return 0;
1838 }
1839 
1840 static void dlm_tcp_sockopts(struct socket *sock)
1841 {
1842 	/* Turn off Nagle's algorithm */
1843 	tcp_sock_set_nodelay(sock->sk);
1844 }
1845 
1846 static void dlm_tcp_listen_sockopts(struct socket *sock)
1847 {
1848 	dlm_tcp_sockopts(sock);
1849 	sock_set_reuseaddr(sock->sk);
1850 }
1851 
1852 static int dlm_tcp_listen_bind(struct socket *sock)
1853 {
1854 	int addr_len;
1855 
1856 	/* Bind to our port */
1857 	make_sockaddr(&dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1858 	return kernel_bind(sock, (struct sockaddr_unsized *)&dlm_local_addr[0],
1859 			   addr_len);
1860 }
1861 
1862 static const struct dlm_proto_ops dlm_tcp_ops = {
1863 	.name = "TCP",
1864 	.proto = IPPROTO_TCP,
1865 	.how = SHUT_WR,
1866 	.sockopts = dlm_tcp_sockopts,
1867 	.bind = dlm_tcp_bind,
1868 	.listen_validate = dlm_tcp_listen_validate,
1869 	.listen_sockopts = dlm_tcp_listen_sockopts,
1870 	.listen_bind = dlm_tcp_listen_bind,
1871 };
1872 
1873 static int dlm_sctp_bind(struct socket *sock)
1874 {
1875 	return sctp_bind_addrs(sock, 0);
1876 }
1877 
1878 static int dlm_sctp_listen_validate(void)
1879 {
1880 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1881 		log_print("SCTP is not enabled by this kernel");
1882 		return -EOPNOTSUPP;
1883 	}
1884 
1885 	request_module("sctp");
1886 	return 0;
1887 }
1888 
1889 static int dlm_sctp_bind_listen(struct socket *sock)
1890 {
1891 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1892 }
1893 
1894 static void dlm_sctp_sockopts(struct socket *sock)
1895 {
1896 	/* Turn off Nagle's algorithm */
1897 	sctp_sock_set_nodelay(sock->sk);
1898 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1899 }
1900 
1901 static const struct dlm_proto_ops dlm_sctp_ops = {
1902 	.name = "SCTP",
1903 	.proto = IPPROTO_SCTP,
1904 	.how = SHUT_RDWR,
1905 	.try_new_addr = true,
1906 	.sockopts = dlm_sctp_sockopts,
1907 	.bind = dlm_sctp_bind,
1908 	.listen_validate = dlm_sctp_listen_validate,
1909 	.listen_sockopts = dlm_sctp_sockopts,
1910 	.listen_bind = dlm_sctp_bind_listen,
1911 };
1912 
1913 int dlm_lowcomms_start(void)
1914 {
1915 	int error;
1916 
1917 	init_local();
1918 	if (!dlm_local_count) {
1919 		error = -ENOTCONN;
1920 		log_print("no local IP address has been set");
1921 		goto fail;
1922 	}
1923 
1924 	error = work_start();
1925 	if (error)
1926 		goto fail;
1927 
1928 	/* Start listening */
1929 	switch (dlm_config.ci_protocol) {
1930 	case DLM_PROTO_TCP:
1931 		dlm_proto_ops = &dlm_tcp_ops;
1932 		break;
1933 	case DLM_PROTO_SCTP:
1934 		dlm_proto_ops = &dlm_sctp_ops;
1935 		break;
1936 	default:
1937 		log_print("Invalid protocol identifier %d set",
1938 			  dlm_config.ci_protocol);
1939 		error = -EINVAL;
1940 		goto fail_proto_ops;
1941 	}
1942 
1943 	error = dlm_listen_for_all();
1944 	if (error)
1945 		goto fail_listen;
1946 
1947 	return 0;
1948 
1949 fail_listen:
1950 	dlm_proto_ops = NULL;
1951 fail_proto_ops:
1952 	work_stop();
1953 fail:
1954 	return error;
1955 }
1956 
1957 void dlm_lowcomms_init(void)
1958 {
1959 	int i;
1960 
1961 	for (i = 0; i < CONN_HASH_SIZE; i++)
1962 		INIT_HLIST_HEAD(&connection_hash[i]);
1963 
1964 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1965 }
1966 
1967 void dlm_lowcomms_exit(void)
1968 {
1969 	struct connection *con;
1970 	int i, idx;
1971 
1972 	idx = srcu_read_lock(&connections_srcu);
1973 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1974 		hlist_for_each_entry_srcu(con, &connection_hash[i], list,
1975 				srcu_read_lock_held(&connections_srcu)) {
1976 			spin_lock(&connections_lock);
1977 			hlist_del_rcu(&con->list);
1978 			spin_unlock(&connections_lock);
1979 
1980 			if (con->othercon)
1981 				call_srcu(&connections_srcu, &con->othercon->rcu,
1982 					  connection_release);
1983 			call_srcu(&connections_srcu, &con->rcu, connection_release);
1984 		}
1985 	}
1986 	srcu_read_unlock(&connections_srcu, idx);
1987 }
1988