xref: /openbmc/linux/fs/dlm/lowcomms.c (revision c018ab3e)
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 NEEDED_RMEM (4*1024*1024)
67 
68 struct connection {
69 	struct socket *sock;	/* NULL if not connected */
70 	uint32_t nodeid;	/* So we know who we are in the list */
71 	/* this semaphore is used to allow parallel recv/send in read
72 	 * lock mode. When we release a sock we need to held the write lock.
73 	 *
74 	 * However this is locking code and not nice. When we remove the
75 	 * othercon handling we can look into other mechanism to synchronize
76 	 * io handling to call sock_release() at the right time.
77 	 */
78 	struct rw_semaphore sock_lock;
79 	unsigned long flags;
80 #define CF_APP_LIMITED 0
81 #define CF_RECV_PENDING 1
82 #define CF_SEND_PENDING 2
83 #define CF_RECV_INTR 3
84 #define CF_IO_STOP 4
85 #define CF_IS_OTHERCON 5
86 	struct list_head writequeue;  /* List of outgoing writequeue_entries */
87 	spinlock_t writequeue_lock;
88 	int retries;
89 	struct hlist_node list;
90 	/* due some connect()/accept() races we currently have this cross over
91 	 * connection attempt second connection for one node.
92 	 *
93 	 * There is a solution to avoid the race by introducing a connect
94 	 * rule as e.g. our_nodeid > nodeid_to_connect who is allowed to
95 	 * connect. Otherside can connect but will only be considered that
96 	 * the other side wants to have a reconnect.
97 	 *
98 	 * However changing to this behaviour will break backwards compatible.
99 	 * In a DLM protocol major version upgrade we should remove this!
100 	 */
101 	struct connection *othercon;
102 	struct work_struct rwork; /* receive worker */
103 	struct work_struct swork; /* send worker */
104 	wait_queue_head_t shutdown_wait;
105 	unsigned char rx_leftover_buf[DLM_MAX_SOCKET_BUFSIZE];
106 	int rx_leftover;
107 	int mark;
108 	int addr_count;
109 	int curr_addr_index;
110 	struct sockaddr_storage addr[DLM_MAX_ADDR_COUNT];
111 	spinlock_t addrs_lock;
112 	struct rcu_head rcu;
113 };
114 #define sock2con(x) ((struct connection *)(x)->sk_user_data)
115 
116 struct listen_connection {
117 	struct socket *sock;
118 	struct work_struct rwork;
119 };
120 
121 #define DLM_WQ_REMAIN_BYTES(e) (PAGE_SIZE - e->end)
122 #define DLM_WQ_LENGTH_BYTES(e) (e->end - e->offset)
123 
124 /* An entry waiting to be sent */
125 struct writequeue_entry {
126 	struct list_head list;
127 	struct page *page;
128 	int offset;
129 	int len;
130 	int end;
131 	int users;
132 	bool dirty;
133 	struct connection *con;
134 	struct list_head msgs;
135 	struct kref ref;
136 };
137 
138 struct dlm_msg {
139 	struct writequeue_entry *entry;
140 	struct dlm_msg *orig_msg;
141 	bool retransmit;
142 	void *ppc;
143 	int len;
144 	int idx; /* new()/commit() idx exchange */
145 
146 	struct list_head list;
147 	struct kref ref;
148 };
149 
150 struct processqueue_entry {
151 	unsigned char *buf;
152 	int nodeid;
153 	int buflen;
154 
155 	struct list_head list;
156 };
157 
158 struct dlm_proto_ops {
159 	bool try_new_addr;
160 	const char *name;
161 	int proto;
162 
163 	int (*connect)(struct connection *con, struct socket *sock,
164 		       struct sockaddr *addr, int addr_len);
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 
198 static void process_recv_sockets(struct work_struct *work);
199 static void process_send_sockets(struct work_struct *work);
200 static void process_dlm_messages(struct work_struct *work);
201 
202 static DECLARE_WORK(process_work, process_dlm_messages);
203 static DEFINE_SPINLOCK(processqueue_lock);
204 static bool process_dlm_messages_pending;
205 static LIST_HEAD(processqueue);
206 
dlm_lowcomms_is_running(void)207 bool dlm_lowcomms_is_running(void)
208 {
209 	return !!listen_con.sock;
210 }
211 
lowcomms_queue_swork(struct connection * con)212 static void lowcomms_queue_swork(struct connection *con)
213 {
214 	assert_spin_locked(&con->writequeue_lock);
215 
216 	if (!test_bit(CF_IO_STOP, &con->flags) &&
217 	    !test_bit(CF_APP_LIMITED, &con->flags) &&
218 	    !test_and_set_bit(CF_SEND_PENDING, &con->flags))
219 		queue_work(io_workqueue, &con->swork);
220 }
221 
lowcomms_queue_rwork(struct connection * con)222 static void lowcomms_queue_rwork(struct connection *con)
223 {
224 #ifdef CONFIG_LOCKDEP
225 	WARN_ON_ONCE(!lockdep_sock_is_held(con->sock->sk));
226 #endif
227 
228 	if (!test_bit(CF_IO_STOP, &con->flags) &&
229 	    !test_and_set_bit(CF_RECV_PENDING, &con->flags))
230 		queue_work(io_workqueue, &con->rwork);
231 }
232 
writequeue_entry_ctor(void * data)233 static void writequeue_entry_ctor(void *data)
234 {
235 	struct writequeue_entry *entry = data;
236 
237 	INIT_LIST_HEAD(&entry->msgs);
238 }
239 
dlm_lowcomms_writequeue_cache_create(void)240 struct kmem_cache *dlm_lowcomms_writequeue_cache_create(void)
241 {
242 	return kmem_cache_create("dlm_writequeue", sizeof(struct writequeue_entry),
243 				 0, 0, writequeue_entry_ctor);
244 }
245 
dlm_lowcomms_msg_cache_create(void)246 struct kmem_cache *dlm_lowcomms_msg_cache_create(void)
247 {
248 	return kmem_cache_create("dlm_msg", sizeof(struct dlm_msg), 0, 0, NULL);
249 }
250 
251 /* need to held writequeue_lock */
con_next_wq(struct connection * con)252 static struct writequeue_entry *con_next_wq(struct connection *con)
253 {
254 	struct writequeue_entry *e;
255 
256 	e = list_first_entry_or_null(&con->writequeue, struct writequeue_entry,
257 				     list);
258 	/* if len is zero nothing is to send, if there are users filling
259 	 * buffers we wait until the users are done so we can send more.
260 	 */
261 	if (!e || e->users || e->len == 0)
262 		return NULL;
263 
264 	return e;
265 }
266 
__find_con(int nodeid,int r)267 static struct connection *__find_con(int nodeid, int r)
268 {
269 	struct connection *con;
270 
271 	hlist_for_each_entry_rcu(con, &connection_hash[r], list) {
272 		if (con->nodeid == nodeid)
273 			return con;
274 	}
275 
276 	return NULL;
277 }
278 
dlm_con_init(struct connection * con,int nodeid)279 static void dlm_con_init(struct connection *con, int nodeid)
280 {
281 	con->nodeid = nodeid;
282 	init_rwsem(&con->sock_lock);
283 	INIT_LIST_HEAD(&con->writequeue);
284 	spin_lock_init(&con->writequeue_lock);
285 	INIT_WORK(&con->swork, process_send_sockets);
286 	INIT_WORK(&con->rwork, process_recv_sockets);
287 	spin_lock_init(&con->addrs_lock);
288 	init_waitqueue_head(&con->shutdown_wait);
289 }
290 
291 /*
292  * If 'allocation' is zero then we don't attempt to create a new
293  * connection structure for this node.
294  */
nodeid2con(int nodeid,gfp_t alloc)295 static struct connection *nodeid2con(int nodeid, gfp_t alloc)
296 {
297 	struct connection *con, *tmp;
298 	int r;
299 
300 	r = nodeid_hash(nodeid);
301 	con = __find_con(nodeid, r);
302 	if (con || !alloc)
303 		return con;
304 
305 	con = kzalloc(sizeof(*con), alloc);
306 	if (!con)
307 		return NULL;
308 
309 	dlm_con_init(con, nodeid);
310 
311 	spin_lock(&connections_lock);
312 	/* Because multiple workqueues/threads calls this function it can
313 	 * race on multiple cpu's. Instead of locking hot path __find_con()
314 	 * we just check in rare cases of recently added nodes again
315 	 * under protection of connections_lock. If this is the case we
316 	 * abort our connection creation and return the existing connection.
317 	 */
318 	tmp = __find_con(nodeid, r);
319 	if (tmp) {
320 		spin_unlock(&connections_lock);
321 		kfree(con);
322 		return tmp;
323 	}
324 
325 	hlist_add_head_rcu(&con->list, &connection_hash[r]);
326 	spin_unlock(&connections_lock);
327 
328 	return con;
329 }
330 
addr_compare(const struct sockaddr_storage * x,const struct sockaddr_storage * y)331 static int addr_compare(const struct sockaddr_storage *x,
332 			const struct sockaddr_storage *y)
333 {
334 	switch (x->ss_family) {
335 	case AF_INET: {
336 		struct sockaddr_in *sinx = (struct sockaddr_in *)x;
337 		struct sockaddr_in *siny = (struct sockaddr_in *)y;
338 		if (sinx->sin_addr.s_addr != siny->sin_addr.s_addr)
339 			return 0;
340 		if (sinx->sin_port != siny->sin_port)
341 			return 0;
342 		break;
343 	}
344 	case AF_INET6: {
345 		struct sockaddr_in6 *sinx = (struct sockaddr_in6 *)x;
346 		struct sockaddr_in6 *siny = (struct sockaddr_in6 *)y;
347 		if (!ipv6_addr_equal(&sinx->sin6_addr, &siny->sin6_addr))
348 			return 0;
349 		if (sinx->sin6_port != siny->sin6_port)
350 			return 0;
351 		break;
352 	}
353 	default:
354 		return 0;
355 	}
356 	return 1;
357 }
358 
nodeid_to_addr(int nodeid,struct sockaddr_storage * sas_out,struct sockaddr * sa_out,bool try_new_addr,unsigned int * mark)359 static int nodeid_to_addr(int nodeid, struct sockaddr_storage *sas_out,
360 			  struct sockaddr *sa_out, bool try_new_addr,
361 			  unsigned int *mark)
362 {
363 	struct sockaddr_storage sas;
364 	struct connection *con;
365 	int idx;
366 
367 	if (!dlm_local_count)
368 		return -1;
369 
370 	idx = srcu_read_lock(&connections_srcu);
371 	con = nodeid2con(nodeid, 0);
372 	if (!con) {
373 		srcu_read_unlock(&connections_srcu, idx);
374 		return -ENOENT;
375 	}
376 
377 	spin_lock(&con->addrs_lock);
378 	if (!con->addr_count) {
379 		spin_unlock(&con->addrs_lock);
380 		srcu_read_unlock(&connections_srcu, idx);
381 		return -ENOENT;
382 	}
383 
384 	memcpy(&sas, &con->addr[con->curr_addr_index],
385 	       sizeof(struct sockaddr_storage));
386 
387 	if (try_new_addr) {
388 		con->curr_addr_index++;
389 		if (con->curr_addr_index == con->addr_count)
390 			con->curr_addr_index = 0;
391 	}
392 
393 	*mark = con->mark;
394 	spin_unlock(&con->addrs_lock);
395 
396 	if (sas_out)
397 		memcpy(sas_out, &sas, sizeof(struct sockaddr_storage));
398 
399 	if (!sa_out) {
400 		srcu_read_unlock(&connections_srcu, idx);
401 		return 0;
402 	}
403 
404 	if (dlm_local_addr[0].ss_family == AF_INET) {
405 		struct sockaddr_in *in4  = (struct sockaddr_in *) &sas;
406 		struct sockaddr_in *ret4 = (struct sockaddr_in *) sa_out;
407 		ret4->sin_addr.s_addr = in4->sin_addr.s_addr;
408 	} else {
409 		struct sockaddr_in6 *in6  = (struct sockaddr_in6 *) &sas;
410 		struct sockaddr_in6 *ret6 = (struct sockaddr_in6 *) sa_out;
411 		ret6->sin6_addr = in6->sin6_addr;
412 	}
413 
414 	srcu_read_unlock(&connections_srcu, idx);
415 	return 0;
416 }
417 
addr_to_nodeid(struct sockaddr_storage * addr,int * nodeid,unsigned int * mark)418 static int addr_to_nodeid(struct sockaddr_storage *addr, int *nodeid,
419 			  unsigned int *mark)
420 {
421 	struct connection *con;
422 	int i, idx, addr_i;
423 
424 	idx = srcu_read_lock(&connections_srcu);
425 	for (i = 0; i < CONN_HASH_SIZE; i++) {
426 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
427 			WARN_ON_ONCE(!con->addr_count);
428 
429 			spin_lock(&con->addrs_lock);
430 			for (addr_i = 0; addr_i < con->addr_count; addr_i++) {
431 				if (addr_compare(&con->addr[addr_i], addr)) {
432 					*nodeid = con->nodeid;
433 					*mark = con->mark;
434 					spin_unlock(&con->addrs_lock);
435 					srcu_read_unlock(&connections_srcu, idx);
436 					return 0;
437 				}
438 			}
439 			spin_unlock(&con->addrs_lock);
440 		}
441 	}
442 	srcu_read_unlock(&connections_srcu, idx);
443 
444 	return -ENOENT;
445 }
446 
dlm_lowcomms_con_has_addr(const struct connection * con,const struct sockaddr_storage * addr)447 static bool dlm_lowcomms_con_has_addr(const struct connection *con,
448 				      const struct sockaddr_storage *addr)
449 {
450 	int i;
451 
452 	for (i = 0; i < con->addr_count; i++) {
453 		if (addr_compare(&con->addr[i], addr))
454 			return true;
455 	}
456 
457 	return false;
458 }
459 
dlm_lowcomms_addr(int nodeid,struct sockaddr_storage * addr,int len)460 int dlm_lowcomms_addr(int nodeid, struct sockaddr_storage *addr, int len)
461 {
462 	struct connection *con;
463 	bool ret, idx;
464 
465 	idx = srcu_read_lock(&connections_srcu);
466 	con = nodeid2con(nodeid, GFP_NOFS);
467 	if (!con) {
468 		srcu_read_unlock(&connections_srcu, idx);
469 		return -ENOMEM;
470 	}
471 
472 	spin_lock(&con->addrs_lock);
473 	if (!con->addr_count) {
474 		memcpy(&con->addr[0], addr, sizeof(*addr));
475 		con->addr_count = 1;
476 		con->mark = dlm_config.ci_mark;
477 		spin_unlock(&con->addrs_lock);
478 		srcu_read_unlock(&connections_srcu, idx);
479 		return 0;
480 	}
481 
482 	ret = dlm_lowcomms_con_has_addr(con, addr);
483 	if (ret) {
484 		spin_unlock(&con->addrs_lock);
485 		srcu_read_unlock(&connections_srcu, idx);
486 		return -EEXIST;
487 	}
488 
489 	if (con->addr_count >= DLM_MAX_ADDR_COUNT) {
490 		spin_unlock(&con->addrs_lock);
491 		srcu_read_unlock(&connections_srcu, idx);
492 		return -ENOSPC;
493 	}
494 
495 	memcpy(&con->addr[con->addr_count++], addr, sizeof(*addr));
496 	srcu_read_unlock(&connections_srcu, idx);
497 	spin_unlock(&con->addrs_lock);
498 	return 0;
499 }
500 
501 /* Data available on socket or listen socket received a connect */
lowcomms_data_ready(struct sock * sk)502 static void lowcomms_data_ready(struct sock *sk)
503 {
504 	struct connection *con = sock2con(sk);
505 
506 	trace_sk_data_ready(sk);
507 
508 	set_bit(CF_RECV_INTR, &con->flags);
509 	lowcomms_queue_rwork(con);
510 }
511 
lowcomms_write_space(struct sock * sk)512 static void lowcomms_write_space(struct sock *sk)
513 {
514 	struct connection *con = sock2con(sk);
515 
516 	clear_bit(SOCK_NOSPACE, &con->sock->flags);
517 
518 	spin_lock_bh(&con->writequeue_lock);
519 	if (test_and_clear_bit(CF_APP_LIMITED, &con->flags)) {
520 		con->sock->sk->sk_write_pending--;
521 		clear_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags);
522 	}
523 
524 	lowcomms_queue_swork(con);
525 	spin_unlock_bh(&con->writequeue_lock);
526 }
527 
lowcomms_state_change(struct sock * sk)528 static void lowcomms_state_change(struct sock *sk)
529 {
530 	/* SCTP layer is not calling sk_data_ready when the connection
531 	 * is done, so we catch the signal through here.
532 	 */
533 	if (sk->sk_shutdown == RCV_SHUTDOWN)
534 		lowcomms_data_ready(sk);
535 }
536 
lowcomms_listen_data_ready(struct sock * sk)537 static void lowcomms_listen_data_ready(struct sock *sk)
538 {
539 	trace_sk_data_ready(sk);
540 
541 	queue_work(io_workqueue, &listen_con.rwork);
542 }
543 
dlm_lowcomms_connect_node(int nodeid)544 int dlm_lowcomms_connect_node(int nodeid)
545 {
546 	struct connection *con;
547 	int idx;
548 
549 	idx = srcu_read_lock(&connections_srcu);
550 	con = nodeid2con(nodeid, 0);
551 	if (WARN_ON_ONCE(!con)) {
552 		srcu_read_unlock(&connections_srcu, idx);
553 		return -ENOENT;
554 	}
555 
556 	down_read(&con->sock_lock);
557 	if (!con->sock) {
558 		spin_lock_bh(&con->writequeue_lock);
559 		lowcomms_queue_swork(con);
560 		spin_unlock_bh(&con->writequeue_lock);
561 	}
562 	up_read(&con->sock_lock);
563 	srcu_read_unlock(&connections_srcu, idx);
564 
565 	cond_resched();
566 	return 0;
567 }
568 
dlm_lowcomms_nodes_set_mark(int nodeid,unsigned int mark)569 int dlm_lowcomms_nodes_set_mark(int nodeid, unsigned int mark)
570 {
571 	struct connection *con;
572 	int idx;
573 
574 	idx = srcu_read_lock(&connections_srcu);
575 	con = nodeid2con(nodeid, 0);
576 	if (!con) {
577 		srcu_read_unlock(&connections_srcu, idx);
578 		return -ENOENT;
579 	}
580 
581 	spin_lock(&con->addrs_lock);
582 	con->mark = mark;
583 	spin_unlock(&con->addrs_lock);
584 	srcu_read_unlock(&connections_srcu, idx);
585 	return 0;
586 }
587 
lowcomms_error_report(struct sock * sk)588 static void lowcomms_error_report(struct sock *sk)
589 {
590 	struct connection *con = sock2con(sk);
591 	struct inet_sock *inet;
592 
593 	inet = inet_sk(sk);
594 	switch (sk->sk_family) {
595 	case AF_INET:
596 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
597 				   "sending to node %d at %pI4, dport %d, "
598 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
599 				   con->nodeid, &inet->inet_daddr,
600 				   ntohs(inet->inet_dport), sk->sk_err,
601 				   READ_ONCE(sk->sk_err_soft));
602 		break;
603 #if IS_ENABLED(CONFIG_IPV6)
604 	case AF_INET6:
605 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
606 				   "sending to node %d at %pI6c, "
607 				   "dport %d, sk_err=%d/%d\n", dlm_our_nodeid(),
608 				   con->nodeid, &sk->sk_v6_daddr,
609 				   ntohs(inet->inet_dport), sk->sk_err,
610 				   READ_ONCE(sk->sk_err_soft));
611 		break;
612 #endif
613 	default:
614 		printk_ratelimited(KERN_ERR "dlm: node %d: socket error "
615 				   "invalid socket family %d set, "
616 				   "sk_err=%d/%d\n", dlm_our_nodeid(),
617 				   sk->sk_family, sk->sk_err,
618 				   READ_ONCE(sk->sk_err_soft));
619 		break;
620 	}
621 
622 	dlm_midcomms_unack_msg_resend(con->nodeid);
623 
624 	listen_sock.sk_error_report(sk);
625 }
626 
restore_callbacks(struct sock * sk)627 static void restore_callbacks(struct sock *sk)
628 {
629 #ifdef CONFIG_LOCKDEP
630 	WARN_ON_ONCE(!lockdep_sock_is_held(sk));
631 #endif
632 
633 	sk->sk_user_data = NULL;
634 	sk->sk_data_ready = listen_sock.sk_data_ready;
635 	sk->sk_state_change = listen_sock.sk_state_change;
636 	sk->sk_write_space = listen_sock.sk_write_space;
637 	sk->sk_error_report = listen_sock.sk_error_report;
638 }
639 
640 /* Make a socket active */
add_sock(struct socket * sock,struct connection * con)641 static void add_sock(struct socket *sock, struct connection *con)
642 {
643 	struct sock *sk = sock->sk;
644 
645 	lock_sock(sk);
646 	con->sock = sock;
647 
648 	sk->sk_user_data = con;
649 	sk->sk_data_ready = lowcomms_data_ready;
650 	sk->sk_write_space = lowcomms_write_space;
651 	if (dlm_config.ci_protocol == DLM_PROTO_SCTP)
652 		sk->sk_state_change = lowcomms_state_change;
653 	sk->sk_allocation = GFP_NOFS;
654 	sk->sk_use_task_frag = false;
655 	sk->sk_error_report = lowcomms_error_report;
656 	release_sock(sk);
657 }
658 
659 /* Add the port number to an IPv6 or 4 sockaddr and return the address
660    length */
make_sockaddr(struct sockaddr_storage * saddr,uint16_t port,int * addr_len)661 static void make_sockaddr(struct sockaddr_storage *saddr, uint16_t port,
662 			  int *addr_len)
663 {
664 	saddr->ss_family =  dlm_local_addr[0].ss_family;
665 	if (saddr->ss_family == AF_INET) {
666 		struct sockaddr_in *in4_addr = (struct sockaddr_in *)saddr;
667 		in4_addr->sin_port = cpu_to_be16(port);
668 		*addr_len = sizeof(struct sockaddr_in);
669 		memset(&in4_addr->sin_zero, 0, sizeof(in4_addr->sin_zero));
670 	} else {
671 		struct sockaddr_in6 *in6_addr = (struct sockaddr_in6 *)saddr;
672 		in6_addr->sin6_port = cpu_to_be16(port);
673 		*addr_len = sizeof(struct sockaddr_in6);
674 	}
675 	memset((char *)saddr + *addr_len, 0, sizeof(struct sockaddr_storage) - *addr_len);
676 }
677 
dlm_page_release(struct kref * kref)678 static void dlm_page_release(struct kref *kref)
679 {
680 	struct writequeue_entry *e = container_of(kref, struct writequeue_entry,
681 						  ref);
682 
683 	__free_page(e->page);
684 	dlm_free_writequeue(e);
685 }
686 
dlm_msg_release(struct kref * kref)687 static void dlm_msg_release(struct kref *kref)
688 {
689 	struct dlm_msg *msg = container_of(kref, struct dlm_msg, ref);
690 
691 	kref_put(&msg->entry->ref, dlm_page_release);
692 	dlm_free_msg(msg);
693 }
694 
free_entry(struct writequeue_entry * e)695 static void free_entry(struct writequeue_entry *e)
696 {
697 	struct dlm_msg *msg, *tmp;
698 
699 	list_for_each_entry_safe(msg, tmp, &e->msgs, list) {
700 		if (msg->orig_msg) {
701 			msg->orig_msg->retransmit = false;
702 			kref_put(&msg->orig_msg->ref, dlm_msg_release);
703 		}
704 
705 		list_del(&msg->list);
706 		kref_put(&msg->ref, dlm_msg_release);
707 	}
708 
709 	list_del(&e->list);
710 	kref_put(&e->ref, dlm_page_release);
711 }
712 
dlm_close_sock(struct socket ** sock)713 static void dlm_close_sock(struct socket **sock)
714 {
715 	lock_sock((*sock)->sk);
716 	restore_callbacks((*sock)->sk);
717 	release_sock((*sock)->sk);
718 
719 	sock_release(*sock);
720 	*sock = NULL;
721 }
722 
allow_connection_io(struct connection * con)723 static void allow_connection_io(struct connection *con)
724 {
725 	if (con->othercon)
726 		clear_bit(CF_IO_STOP, &con->othercon->flags);
727 	clear_bit(CF_IO_STOP, &con->flags);
728 }
729 
stop_connection_io(struct connection * con)730 static void stop_connection_io(struct connection *con)
731 {
732 	if (con->othercon)
733 		stop_connection_io(con->othercon);
734 
735 	spin_lock_bh(&con->writequeue_lock);
736 	set_bit(CF_IO_STOP, &con->flags);
737 	spin_unlock_bh(&con->writequeue_lock);
738 
739 	down_write(&con->sock_lock);
740 	if (con->sock) {
741 		lock_sock(con->sock->sk);
742 		restore_callbacks(con->sock->sk);
743 		release_sock(con->sock->sk);
744 	}
745 	up_write(&con->sock_lock);
746 
747 	cancel_work_sync(&con->swork);
748 	cancel_work_sync(&con->rwork);
749 }
750 
751 /* Close a remote connection and tidy up */
close_connection(struct connection * con,bool and_other)752 static void close_connection(struct connection *con, bool and_other)
753 {
754 	struct writequeue_entry *e;
755 
756 	if (con->othercon && and_other)
757 		close_connection(con->othercon, false);
758 
759 	down_write(&con->sock_lock);
760 	if (!con->sock) {
761 		up_write(&con->sock_lock);
762 		return;
763 	}
764 
765 	dlm_close_sock(&con->sock);
766 
767 	/* if we send a writequeue entry only a half way, we drop the
768 	 * whole entry because reconnection and that we not start of the
769 	 * middle of a msg which will confuse the other end.
770 	 *
771 	 * we can always drop messages because retransmits, but what we
772 	 * cannot allow is to transmit half messages which may be processed
773 	 * at the other side.
774 	 *
775 	 * our policy is to start on a clean state when disconnects, we don't
776 	 * know what's send/received on transport layer in this case.
777 	 */
778 	spin_lock_bh(&con->writequeue_lock);
779 	if (!list_empty(&con->writequeue)) {
780 		e = list_first_entry(&con->writequeue, struct writequeue_entry,
781 				     list);
782 		if (e->dirty)
783 			free_entry(e);
784 	}
785 	spin_unlock_bh(&con->writequeue_lock);
786 
787 	con->rx_leftover = 0;
788 	con->retries = 0;
789 	clear_bit(CF_APP_LIMITED, &con->flags);
790 	clear_bit(CF_RECV_PENDING, &con->flags);
791 	clear_bit(CF_SEND_PENDING, &con->flags);
792 	up_write(&con->sock_lock);
793 }
794 
shutdown_connection(struct connection * con,bool and_other)795 static void shutdown_connection(struct connection *con, bool and_other)
796 {
797 	int ret;
798 
799 	if (con->othercon && and_other)
800 		shutdown_connection(con->othercon, false);
801 
802 	flush_workqueue(io_workqueue);
803 	down_read(&con->sock_lock);
804 	/* nothing to shutdown */
805 	if (!con->sock) {
806 		up_read(&con->sock_lock);
807 		return;
808 	}
809 
810 	ret = kernel_sock_shutdown(con->sock, SHUT_WR);
811 	up_read(&con->sock_lock);
812 	if (ret) {
813 		log_print("Connection %p failed to shutdown: %d will force close",
814 			  con, ret);
815 		goto force_close;
816 	} else {
817 		ret = wait_event_timeout(con->shutdown_wait, !con->sock,
818 					 DLM_SHUTDOWN_WAIT_TIMEOUT);
819 		if (ret == 0) {
820 			log_print("Connection %p shutdown timed out, will force close",
821 				  con);
822 			goto force_close;
823 		}
824 	}
825 
826 	return;
827 
828 force_close:
829 	close_connection(con, false);
830 }
831 
new_processqueue_entry(int nodeid,int buflen)832 static struct processqueue_entry *new_processqueue_entry(int nodeid,
833 							 int buflen)
834 {
835 	struct processqueue_entry *pentry;
836 
837 	pentry = kmalloc(sizeof(*pentry), GFP_NOFS);
838 	if (!pentry)
839 		return NULL;
840 
841 	pentry->buf = kmalloc(buflen, GFP_NOFS);
842 	if (!pentry->buf) {
843 		kfree(pentry);
844 		return NULL;
845 	}
846 
847 	pentry->nodeid = nodeid;
848 	return pentry;
849 }
850 
free_processqueue_entry(struct processqueue_entry * pentry)851 static void free_processqueue_entry(struct processqueue_entry *pentry)
852 {
853 	kfree(pentry->buf);
854 	kfree(pentry);
855 }
856 
857 struct dlm_processed_nodes {
858 	int nodeid;
859 
860 	struct list_head list;
861 };
862 
process_dlm_messages(struct work_struct * work)863 static void process_dlm_messages(struct work_struct *work)
864 {
865 	struct processqueue_entry *pentry;
866 
867 	spin_lock(&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(&processqueue_lock);
873 		return;
874 	}
875 
876 	list_del(&pentry->list);
877 	spin_unlock(&processqueue_lock);
878 
879 	for (;;) {
880 		dlm_process_incoming_buffer(pentry->nodeid, pentry->buf,
881 					    pentry->buflen);
882 		free_processqueue_entry(pentry);
883 
884 		spin_lock(&processqueue_lock);
885 		pentry = list_first_entry_or_null(&processqueue,
886 						  struct processqueue_entry, list);
887 		if (!pentry) {
888 			process_dlm_messages_pending = false;
889 			spin_unlock(&processqueue_lock);
890 			break;
891 		}
892 
893 		list_del(&pentry->list);
894 		spin_unlock(&processqueue_lock);
895 	}
896 }
897 
898 /* Data received from remote end */
receive_from_sock(struct connection * con,int buflen)899 static int receive_from_sock(struct connection *con, int buflen)
900 {
901 	struct processqueue_entry *pentry;
902 	int ret, buflen_real;
903 	struct msghdr msg;
904 	struct kvec iov;
905 
906 	pentry = new_processqueue_entry(con->nodeid, buflen);
907 	if (!pentry)
908 		return DLM_IO_RESCHED;
909 
910 	memcpy(pentry->buf, con->rx_leftover_buf, con->rx_leftover);
911 
912 	/* calculate new buffer parameter regarding last receive and
913 	 * possible leftover bytes
914 	 */
915 	iov.iov_base = pentry->buf + con->rx_leftover;
916 	iov.iov_len = buflen - con->rx_leftover;
917 
918 	memset(&msg, 0, sizeof(msg));
919 	msg.msg_flags = MSG_DONTWAIT | MSG_NOSIGNAL;
920 	clear_bit(CF_RECV_INTR, &con->flags);
921 again:
922 	ret = kernel_recvmsg(con->sock, &msg, &iov, 1, iov.iov_len,
923 			     msg.msg_flags);
924 	trace_dlm_recv(con->nodeid, ret);
925 	if (ret == -EAGAIN) {
926 		lock_sock(con->sock->sk);
927 		if (test_and_clear_bit(CF_RECV_INTR, &con->flags)) {
928 			release_sock(con->sock->sk);
929 			goto again;
930 		}
931 
932 		clear_bit(CF_RECV_PENDING, &con->flags);
933 		release_sock(con->sock->sk);
934 		free_processqueue_entry(pentry);
935 		return DLM_IO_END;
936 	} else if (ret == 0) {
937 		/* close will clear CF_RECV_PENDING */
938 		free_processqueue_entry(pentry);
939 		return DLM_IO_EOF;
940 	} else if (ret < 0) {
941 		free_processqueue_entry(pentry);
942 		return ret;
943 	}
944 
945 	/* new buflen according readed bytes and leftover from last receive */
946 	buflen_real = ret + con->rx_leftover;
947 	ret = dlm_validate_incoming_buffer(con->nodeid, pentry->buf,
948 					   buflen_real);
949 	if (ret < 0) {
950 		free_processqueue_entry(pentry);
951 		return ret;
952 	}
953 
954 	pentry->buflen = ret;
955 
956 	/* calculate leftover bytes from process and put it into begin of
957 	 * the receive buffer, so next receive we have the full message
958 	 * at the start address of the receive buffer.
959 	 */
960 	con->rx_leftover = buflen_real - ret;
961 	memmove(con->rx_leftover_buf, pentry->buf + ret,
962 		con->rx_leftover);
963 
964 	spin_lock(&processqueue_lock);
965 	list_add_tail(&pentry->list, &processqueue);
966 	if (!process_dlm_messages_pending) {
967 		process_dlm_messages_pending = true;
968 		queue_work(process_workqueue, &process_work);
969 	}
970 	spin_unlock(&processqueue_lock);
971 
972 	return DLM_IO_SUCCESS;
973 }
974 
975 /* Listening socket is busy, accept a connection */
accept_from_sock(void)976 static int accept_from_sock(void)
977 {
978 	struct sockaddr_storage peeraddr;
979 	int len, idx, result, nodeid;
980 	struct connection *newcon;
981 	struct socket *newsock;
982 	unsigned int mark;
983 
984 	result = kernel_accept(listen_con.sock, &newsock, O_NONBLOCK);
985 	if (result == -EAGAIN)
986 		return DLM_IO_END;
987 	else if (result < 0)
988 		goto accept_err;
989 
990 	/* Get the connected socket's peer */
991 	memset(&peeraddr, 0, sizeof(peeraddr));
992 	len = newsock->ops->getname(newsock, (struct sockaddr *)&peeraddr, 2);
993 	if (len < 0) {
994 		result = -ECONNABORTED;
995 		goto accept_err;
996 	}
997 
998 	/* Get the new node's NODEID */
999 	make_sockaddr(&peeraddr, 0, &len);
1000 	if (addr_to_nodeid(&peeraddr, &nodeid, &mark)) {
1001 		switch (peeraddr.ss_family) {
1002 		case AF_INET: {
1003 			struct sockaddr_in *sin = (struct sockaddr_in *)&peeraddr;
1004 
1005 			log_print("connect from non cluster IPv4 node %pI4",
1006 				  &sin->sin_addr);
1007 			break;
1008 		}
1009 #if IS_ENABLED(CONFIG_IPV6)
1010 		case AF_INET6: {
1011 			struct sockaddr_in6 *sin6 = (struct sockaddr_in6 *)&peeraddr;
1012 
1013 			log_print("connect from non cluster IPv6 node %pI6c",
1014 				  &sin6->sin6_addr);
1015 			break;
1016 		}
1017 #endif
1018 		default:
1019 			log_print("invalid family from non cluster node");
1020 			break;
1021 		}
1022 
1023 		sock_release(newsock);
1024 		return -1;
1025 	}
1026 
1027 	log_print("got connection from %d", nodeid);
1028 
1029 	/*  Check to see if we already have a connection to this node. This
1030 	 *  could happen if the two nodes initiate a connection at roughly
1031 	 *  the same time and the connections cross on the wire.
1032 	 *  In this case we store the incoming one in "othercon"
1033 	 */
1034 	idx = srcu_read_lock(&connections_srcu);
1035 	newcon = nodeid2con(nodeid, 0);
1036 	if (WARN_ON_ONCE(!newcon)) {
1037 		srcu_read_unlock(&connections_srcu, idx);
1038 		result = -ENOENT;
1039 		goto accept_err;
1040 	}
1041 
1042 	sock_set_mark(newsock->sk, mark);
1043 
1044 	down_write(&newcon->sock_lock);
1045 	if (newcon->sock) {
1046 		struct connection *othercon = newcon->othercon;
1047 
1048 		if (!othercon) {
1049 			othercon = kzalloc(sizeof(*othercon), GFP_NOFS);
1050 			if (!othercon) {
1051 				log_print("failed to allocate incoming socket");
1052 				up_write(&newcon->sock_lock);
1053 				srcu_read_unlock(&connections_srcu, idx);
1054 				result = -ENOMEM;
1055 				goto accept_err;
1056 			}
1057 
1058 			dlm_con_init(othercon, nodeid);
1059 			lockdep_set_subclass(&othercon->sock_lock, 1);
1060 			newcon->othercon = othercon;
1061 			set_bit(CF_IS_OTHERCON, &othercon->flags);
1062 		} else {
1063 			/* close other sock con if we have something new */
1064 			close_connection(othercon, false);
1065 		}
1066 
1067 		down_write(&othercon->sock_lock);
1068 		add_sock(newsock, othercon);
1069 
1070 		/* check if we receved something while adding */
1071 		lock_sock(othercon->sock->sk);
1072 		lowcomms_queue_rwork(othercon);
1073 		release_sock(othercon->sock->sk);
1074 		up_write(&othercon->sock_lock);
1075 	}
1076 	else {
1077 		/* accept copies the sk after we've saved the callbacks, so we
1078 		   don't want to save them a second time or comm errors will
1079 		   result in calling sk_error_report recursively. */
1080 		add_sock(newsock, newcon);
1081 
1082 		/* check if we receved something while adding */
1083 		lock_sock(newcon->sock->sk);
1084 		lowcomms_queue_rwork(newcon);
1085 		release_sock(newcon->sock->sk);
1086 	}
1087 	up_write(&newcon->sock_lock);
1088 	srcu_read_unlock(&connections_srcu, idx);
1089 
1090 	return DLM_IO_SUCCESS;
1091 
1092 accept_err:
1093 	if (newsock)
1094 		sock_release(newsock);
1095 
1096 	return result;
1097 }
1098 
1099 /*
1100  * writequeue_entry_complete - try to delete and free write queue entry
1101  * @e: write queue entry to try to delete
1102  * @completed: bytes completed
1103  *
1104  * writequeue_lock must be held.
1105  */
writequeue_entry_complete(struct writequeue_entry * e,int completed)1106 static void writequeue_entry_complete(struct writequeue_entry *e, int completed)
1107 {
1108 	e->offset += completed;
1109 	e->len -= completed;
1110 	/* signal that page was half way transmitted */
1111 	e->dirty = true;
1112 
1113 	if (e->len == 0 && e->users == 0)
1114 		free_entry(e);
1115 }
1116 
1117 /*
1118  * sctp_bind_addrs - bind a SCTP socket to all our addresses
1119  */
sctp_bind_addrs(struct socket * sock,uint16_t port)1120 static int sctp_bind_addrs(struct socket *sock, uint16_t port)
1121 {
1122 	struct sockaddr_storage localaddr;
1123 	struct sockaddr *addr = (struct sockaddr *)&localaddr;
1124 	int i, addr_len, result = 0;
1125 
1126 	for (i = 0; i < dlm_local_count; i++) {
1127 		memcpy(&localaddr, &dlm_local_addr[i], sizeof(localaddr));
1128 		make_sockaddr(&localaddr, port, &addr_len);
1129 
1130 		if (!i)
1131 			result = kernel_bind(sock, addr, addr_len);
1132 		else
1133 			result = sock_bind_add(sock->sk, addr, addr_len);
1134 
1135 		if (result < 0) {
1136 			log_print("Can't bind to %d addr number %d, %d.\n",
1137 				  port, i + 1, result);
1138 			break;
1139 		}
1140 	}
1141 	return result;
1142 }
1143 
1144 /* Get local addresses */
init_local(void)1145 static void init_local(void)
1146 {
1147 	struct sockaddr_storage sas;
1148 	int i;
1149 
1150 	dlm_local_count = 0;
1151 	for (i = 0; i < DLM_MAX_ADDR_COUNT; i++) {
1152 		if (dlm_our_addr(&sas, i))
1153 			break;
1154 
1155 		memcpy(&dlm_local_addr[dlm_local_count++], &sas, sizeof(sas));
1156 	}
1157 }
1158 
new_writequeue_entry(struct connection * con)1159 static struct writequeue_entry *new_writequeue_entry(struct connection *con)
1160 {
1161 	struct writequeue_entry *entry;
1162 
1163 	entry = dlm_allocate_writequeue();
1164 	if (!entry)
1165 		return NULL;
1166 
1167 	entry->page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
1168 	if (!entry->page) {
1169 		dlm_free_writequeue(entry);
1170 		return NULL;
1171 	}
1172 
1173 	entry->offset = 0;
1174 	entry->len = 0;
1175 	entry->end = 0;
1176 	entry->dirty = false;
1177 	entry->con = con;
1178 	entry->users = 1;
1179 	kref_init(&entry->ref);
1180 	return entry;
1181 }
1182 
new_wq_entry(struct connection * con,int len,char ** ppc,void (* cb)(void * data),void * data)1183 static struct writequeue_entry *new_wq_entry(struct connection *con, int len,
1184 					     char **ppc, void (*cb)(void *data),
1185 					     void *data)
1186 {
1187 	struct writequeue_entry *e;
1188 
1189 	spin_lock_bh(&con->writequeue_lock);
1190 	if (!list_empty(&con->writequeue)) {
1191 		e = list_last_entry(&con->writequeue, struct writequeue_entry, list);
1192 		if (DLM_WQ_REMAIN_BYTES(e) >= len) {
1193 			kref_get(&e->ref);
1194 
1195 			*ppc = page_address(e->page) + e->end;
1196 			if (cb)
1197 				cb(data);
1198 
1199 			e->end += len;
1200 			e->users++;
1201 			goto out;
1202 		}
1203 	}
1204 
1205 	e = new_writequeue_entry(con);
1206 	if (!e)
1207 		goto out;
1208 
1209 	kref_get(&e->ref);
1210 	*ppc = page_address(e->page);
1211 	e->end += len;
1212 	if (cb)
1213 		cb(data);
1214 
1215 	list_add_tail(&e->list, &con->writequeue);
1216 
1217 out:
1218 	spin_unlock_bh(&con->writequeue_lock);
1219 	return e;
1220 };
1221 
dlm_lowcomms_new_msg_con(struct connection * con,int len,gfp_t allocation,char ** ppc,void (* cb)(void * data),void * data)1222 static struct dlm_msg *dlm_lowcomms_new_msg_con(struct connection *con, int len,
1223 						gfp_t allocation, char **ppc,
1224 						void (*cb)(void *data),
1225 						void *data)
1226 {
1227 	struct writequeue_entry *e;
1228 	struct dlm_msg *msg;
1229 
1230 	msg = dlm_allocate_msg(allocation);
1231 	if (!msg)
1232 		return NULL;
1233 
1234 	kref_init(&msg->ref);
1235 
1236 	e = new_wq_entry(con, len, ppc, cb, data);
1237 	if (!e) {
1238 		dlm_free_msg(msg);
1239 		return NULL;
1240 	}
1241 
1242 	msg->retransmit = false;
1243 	msg->orig_msg = NULL;
1244 	msg->ppc = *ppc;
1245 	msg->len = len;
1246 	msg->entry = e;
1247 
1248 	return msg;
1249 }
1250 
1251 /* avoid false positive for nodes_srcu, unlock happens in
1252  * dlm_lowcomms_commit_msg which is a must call if success
1253  */
1254 #ifndef __CHECKER__
dlm_lowcomms_new_msg(int nodeid,int len,gfp_t allocation,char ** ppc,void (* cb)(void * data),void * data)1255 struct dlm_msg *dlm_lowcomms_new_msg(int nodeid, int len, gfp_t allocation,
1256 				     char **ppc, void (*cb)(void *data),
1257 				     void *data)
1258 {
1259 	struct connection *con;
1260 	struct dlm_msg *msg;
1261 	int idx;
1262 
1263 	if (len > DLM_MAX_SOCKET_BUFSIZE ||
1264 	    len < sizeof(struct dlm_header)) {
1265 		BUILD_BUG_ON(PAGE_SIZE < DLM_MAX_SOCKET_BUFSIZE);
1266 		log_print("failed to allocate a buffer of size %d", len);
1267 		WARN_ON_ONCE(1);
1268 		return NULL;
1269 	}
1270 
1271 	idx = srcu_read_lock(&connections_srcu);
1272 	con = nodeid2con(nodeid, 0);
1273 	if (WARN_ON_ONCE(!con)) {
1274 		srcu_read_unlock(&connections_srcu, idx);
1275 		return NULL;
1276 	}
1277 
1278 	msg = dlm_lowcomms_new_msg_con(con, len, allocation, ppc, cb, data);
1279 	if (!msg) {
1280 		srcu_read_unlock(&connections_srcu, idx);
1281 		return NULL;
1282 	}
1283 
1284 	/* for dlm_lowcomms_commit_msg() */
1285 	kref_get(&msg->ref);
1286 	/* we assume if successful commit must called */
1287 	msg->idx = idx;
1288 	return msg;
1289 }
1290 #endif
1291 
_dlm_lowcomms_commit_msg(struct dlm_msg * msg)1292 static void _dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1293 {
1294 	struct writequeue_entry *e = msg->entry;
1295 	struct connection *con = e->con;
1296 	int users;
1297 
1298 	spin_lock_bh(&con->writequeue_lock);
1299 	kref_get(&msg->ref);
1300 	list_add(&msg->list, &e->msgs);
1301 
1302 	users = --e->users;
1303 	if (users)
1304 		goto out;
1305 
1306 	e->len = DLM_WQ_LENGTH_BYTES(e);
1307 
1308 	lowcomms_queue_swork(con);
1309 
1310 out:
1311 	spin_unlock_bh(&con->writequeue_lock);
1312 	return;
1313 }
1314 
1315 /* avoid false positive for nodes_srcu, lock was happen in
1316  * dlm_lowcomms_new_msg
1317  */
1318 #ifndef __CHECKER__
dlm_lowcomms_commit_msg(struct dlm_msg * msg)1319 void dlm_lowcomms_commit_msg(struct dlm_msg *msg)
1320 {
1321 	_dlm_lowcomms_commit_msg(msg);
1322 	srcu_read_unlock(&connections_srcu, msg->idx);
1323 	/* because dlm_lowcomms_new_msg() */
1324 	kref_put(&msg->ref, dlm_msg_release);
1325 }
1326 #endif
1327 
dlm_lowcomms_put_msg(struct dlm_msg * msg)1328 void dlm_lowcomms_put_msg(struct dlm_msg *msg)
1329 {
1330 	kref_put(&msg->ref, dlm_msg_release);
1331 }
1332 
1333 /* does not held connections_srcu, usage lowcomms_error_report only */
dlm_lowcomms_resend_msg(struct dlm_msg * msg)1334 int dlm_lowcomms_resend_msg(struct dlm_msg *msg)
1335 {
1336 	struct dlm_msg *msg_resend;
1337 	char *ppc;
1338 
1339 	if (msg->retransmit)
1340 		return 1;
1341 
1342 	msg_resend = dlm_lowcomms_new_msg_con(msg->entry->con, msg->len,
1343 					      GFP_ATOMIC, &ppc, NULL, NULL);
1344 	if (!msg_resend)
1345 		return -ENOMEM;
1346 
1347 	msg->retransmit = true;
1348 	kref_get(&msg->ref);
1349 	msg_resend->orig_msg = msg;
1350 
1351 	memcpy(ppc, msg->ppc, msg->len);
1352 	_dlm_lowcomms_commit_msg(msg_resend);
1353 	dlm_lowcomms_put_msg(msg_resend);
1354 
1355 	return 0;
1356 }
1357 
1358 /* Send a message */
send_to_sock(struct connection * con)1359 static int send_to_sock(struct connection *con)
1360 {
1361 	struct writequeue_entry *e;
1362 	struct bio_vec bvec;
1363 	struct msghdr msg = {
1364 		.msg_flags = MSG_SPLICE_PAGES | MSG_DONTWAIT | MSG_NOSIGNAL,
1365 	};
1366 	int len, offset, ret;
1367 
1368 	spin_lock_bh(&con->writequeue_lock);
1369 	e = con_next_wq(con);
1370 	if (!e) {
1371 		clear_bit(CF_SEND_PENDING, &con->flags);
1372 		spin_unlock_bh(&con->writequeue_lock);
1373 		return DLM_IO_END;
1374 	}
1375 
1376 	len = e->len;
1377 	offset = e->offset;
1378 	WARN_ON_ONCE(len == 0 && e->users == 0);
1379 	spin_unlock_bh(&con->writequeue_lock);
1380 
1381 	bvec_set_page(&bvec, e->page, len, offset);
1382 	iov_iter_bvec(&msg.msg_iter, ITER_SOURCE, &bvec, 1, len);
1383 	ret = sock_sendmsg(con->sock, &msg);
1384 	trace_dlm_send(con->nodeid, ret);
1385 	if (ret == -EAGAIN || ret == 0) {
1386 		lock_sock(con->sock->sk);
1387 		spin_lock_bh(&con->writequeue_lock);
1388 		if (test_bit(SOCKWQ_ASYNC_NOSPACE, &con->sock->flags) &&
1389 		    !test_and_set_bit(CF_APP_LIMITED, &con->flags)) {
1390 			/* Notify TCP that we're limited by the
1391 			 * application window size.
1392 			 */
1393 			set_bit(SOCK_NOSPACE, &con->sock->sk->sk_socket->flags);
1394 			con->sock->sk->sk_write_pending++;
1395 
1396 			clear_bit(CF_SEND_PENDING, &con->flags);
1397 			spin_unlock_bh(&con->writequeue_lock);
1398 			release_sock(con->sock->sk);
1399 
1400 			/* wait for write_space() event */
1401 			return DLM_IO_END;
1402 		}
1403 		spin_unlock_bh(&con->writequeue_lock);
1404 		release_sock(con->sock->sk);
1405 
1406 		return DLM_IO_RESCHED;
1407 	} else if (ret < 0) {
1408 		return ret;
1409 	}
1410 
1411 	spin_lock_bh(&con->writequeue_lock);
1412 	writequeue_entry_complete(e, ret);
1413 	spin_unlock_bh(&con->writequeue_lock);
1414 
1415 	return DLM_IO_SUCCESS;
1416 }
1417 
clean_one_writequeue(struct connection * con)1418 static void clean_one_writequeue(struct connection *con)
1419 {
1420 	struct writequeue_entry *e, *safe;
1421 
1422 	spin_lock_bh(&con->writequeue_lock);
1423 	list_for_each_entry_safe(e, safe, &con->writequeue, list) {
1424 		free_entry(e);
1425 	}
1426 	spin_unlock_bh(&con->writequeue_lock);
1427 }
1428 
connection_release(struct rcu_head * rcu)1429 static void connection_release(struct rcu_head *rcu)
1430 {
1431 	struct connection *con = container_of(rcu, struct connection, rcu);
1432 
1433 	WARN_ON_ONCE(!list_empty(&con->writequeue));
1434 	WARN_ON_ONCE(con->sock);
1435 	kfree(con);
1436 }
1437 
1438 /* Called from recovery when it knows that a node has
1439    left the cluster */
dlm_lowcomms_close(int nodeid)1440 int dlm_lowcomms_close(int nodeid)
1441 {
1442 	struct connection *con;
1443 	int idx;
1444 
1445 	log_print("closing connection to node %d", nodeid);
1446 
1447 	idx = srcu_read_lock(&connections_srcu);
1448 	con = nodeid2con(nodeid, 0);
1449 	if (WARN_ON_ONCE(!con)) {
1450 		srcu_read_unlock(&connections_srcu, idx);
1451 		return -ENOENT;
1452 	}
1453 
1454 	stop_connection_io(con);
1455 	log_print("io handling for node: %d stopped", nodeid);
1456 	close_connection(con, true);
1457 
1458 	spin_lock(&connections_lock);
1459 	hlist_del_rcu(&con->list);
1460 	spin_unlock(&connections_lock);
1461 
1462 	clean_one_writequeue(con);
1463 	call_srcu(&connections_srcu, &con->rcu, connection_release);
1464 	if (con->othercon) {
1465 		clean_one_writequeue(con->othercon);
1466 		call_srcu(&connections_srcu, &con->othercon->rcu, connection_release);
1467 	}
1468 	srcu_read_unlock(&connections_srcu, idx);
1469 
1470 	/* for debugging we print when we are done to compare with other
1471 	 * messages in between. This function need to be correctly synchronized
1472 	 * with io handling
1473 	 */
1474 	log_print("closing connection to node %d done", nodeid);
1475 
1476 	return 0;
1477 }
1478 
1479 /* Receive worker function */
process_recv_sockets(struct work_struct * work)1480 static void process_recv_sockets(struct work_struct *work)
1481 {
1482 	struct connection *con = container_of(work, struct connection, rwork);
1483 	int ret, buflen;
1484 
1485 	down_read(&con->sock_lock);
1486 	if (!con->sock) {
1487 		up_read(&con->sock_lock);
1488 		return;
1489 	}
1490 
1491 	buflen = READ_ONCE(dlm_config.ci_buffer_size);
1492 	do {
1493 		ret = receive_from_sock(con, buflen);
1494 	} while (ret == DLM_IO_SUCCESS);
1495 	up_read(&con->sock_lock);
1496 
1497 	switch (ret) {
1498 	case DLM_IO_END:
1499 		/* CF_RECV_PENDING cleared */
1500 		break;
1501 	case DLM_IO_EOF:
1502 		close_connection(con, false);
1503 		wake_up(&con->shutdown_wait);
1504 		/* CF_RECV_PENDING cleared */
1505 		break;
1506 	case DLM_IO_RESCHED:
1507 		cond_resched();
1508 		queue_work(io_workqueue, &con->rwork);
1509 		/* CF_RECV_PENDING not cleared */
1510 		break;
1511 	default:
1512 		if (ret < 0) {
1513 			if (test_bit(CF_IS_OTHERCON, &con->flags)) {
1514 				close_connection(con, false);
1515 			} else {
1516 				spin_lock_bh(&con->writequeue_lock);
1517 				lowcomms_queue_swork(con);
1518 				spin_unlock_bh(&con->writequeue_lock);
1519 			}
1520 
1521 			/* CF_RECV_PENDING cleared for othercon
1522 			 * we trigger send queue if not already done
1523 			 * and process_send_sockets will handle it
1524 			 */
1525 			break;
1526 		}
1527 
1528 		WARN_ON_ONCE(1);
1529 		break;
1530 	}
1531 }
1532 
process_listen_recv_socket(struct work_struct * work)1533 static void process_listen_recv_socket(struct work_struct *work)
1534 {
1535 	int ret;
1536 
1537 	if (WARN_ON_ONCE(!listen_con.sock))
1538 		return;
1539 
1540 	do {
1541 		ret = accept_from_sock();
1542 	} while (ret == DLM_IO_SUCCESS);
1543 
1544 	if (ret < 0)
1545 		log_print("critical error accepting connection: %d", ret);
1546 }
1547 
dlm_connect(struct connection * con)1548 static int dlm_connect(struct connection *con)
1549 {
1550 	struct sockaddr_storage addr;
1551 	int result, addr_len;
1552 	struct socket *sock;
1553 	unsigned int mark;
1554 
1555 	memset(&addr, 0, sizeof(addr));
1556 	result = nodeid_to_addr(con->nodeid, &addr, NULL,
1557 				dlm_proto_ops->try_new_addr, &mark);
1558 	if (result < 0) {
1559 		log_print("no address for nodeid %d", con->nodeid);
1560 		return result;
1561 	}
1562 
1563 	/* Create a socket to communicate with */
1564 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1565 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1566 	if (result < 0)
1567 		return result;
1568 
1569 	sock_set_mark(sock->sk, mark);
1570 	dlm_proto_ops->sockopts(sock);
1571 
1572 	result = dlm_proto_ops->bind(sock);
1573 	if (result < 0) {
1574 		sock_release(sock);
1575 		return result;
1576 	}
1577 
1578 	add_sock(sock, con);
1579 
1580 	log_print_ratelimited("connecting to %d", con->nodeid);
1581 	make_sockaddr(&addr, dlm_config.ci_tcp_port, &addr_len);
1582 	result = dlm_proto_ops->connect(con, sock, (struct sockaddr *)&addr,
1583 					addr_len);
1584 	switch (result) {
1585 	case -EINPROGRESS:
1586 		/* not an error */
1587 		fallthrough;
1588 	case 0:
1589 		break;
1590 	default:
1591 		if (result < 0)
1592 			dlm_close_sock(&con->sock);
1593 
1594 		break;
1595 	}
1596 
1597 	return result;
1598 }
1599 
1600 /* Send worker function */
process_send_sockets(struct work_struct * work)1601 static void process_send_sockets(struct work_struct *work)
1602 {
1603 	struct connection *con = container_of(work, struct connection, swork);
1604 	int ret;
1605 
1606 	WARN_ON_ONCE(test_bit(CF_IS_OTHERCON, &con->flags));
1607 
1608 	down_read(&con->sock_lock);
1609 	if (!con->sock) {
1610 		up_read(&con->sock_lock);
1611 		down_write(&con->sock_lock);
1612 		if (!con->sock) {
1613 			ret = dlm_connect(con);
1614 			switch (ret) {
1615 			case 0:
1616 				break;
1617 			case -EINPROGRESS:
1618 				/* avoid spamming resched on connection
1619 				 * we might can switch to a state_change
1620 				 * event based mechanism if established
1621 				 */
1622 				msleep(100);
1623 				break;
1624 			default:
1625 				/* CF_SEND_PENDING not cleared */
1626 				up_write(&con->sock_lock);
1627 				log_print("connect to node %d try %d error %d",
1628 					  con->nodeid, con->retries++, ret);
1629 				msleep(1000);
1630 				/* For now we try forever to reconnect. In
1631 				 * future we should send a event to cluster
1632 				 * manager to fence itself after certain amount
1633 				 * of retries.
1634 				 */
1635 				queue_work(io_workqueue, &con->swork);
1636 				return;
1637 			}
1638 		}
1639 		downgrade_write(&con->sock_lock);
1640 	}
1641 
1642 	do {
1643 		ret = send_to_sock(con);
1644 	} while (ret == DLM_IO_SUCCESS);
1645 	up_read(&con->sock_lock);
1646 
1647 	switch (ret) {
1648 	case DLM_IO_END:
1649 		/* CF_SEND_PENDING cleared */
1650 		break;
1651 	case DLM_IO_RESCHED:
1652 		/* CF_SEND_PENDING not cleared */
1653 		cond_resched();
1654 		queue_work(io_workqueue, &con->swork);
1655 		break;
1656 	default:
1657 		if (ret < 0) {
1658 			close_connection(con, false);
1659 
1660 			/* CF_SEND_PENDING cleared */
1661 			spin_lock_bh(&con->writequeue_lock);
1662 			lowcomms_queue_swork(con);
1663 			spin_unlock_bh(&con->writequeue_lock);
1664 			break;
1665 		}
1666 
1667 		WARN_ON_ONCE(1);
1668 		break;
1669 	}
1670 }
1671 
work_stop(void)1672 static void work_stop(void)
1673 {
1674 	if (io_workqueue) {
1675 		destroy_workqueue(io_workqueue);
1676 		io_workqueue = NULL;
1677 	}
1678 
1679 	if (process_workqueue) {
1680 		destroy_workqueue(process_workqueue);
1681 		process_workqueue = NULL;
1682 	}
1683 }
1684 
work_start(void)1685 static int work_start(void)
1686 {
1687 	io_workqueue = alloc_workqueue("dlm_io", WQ_HIGHPRI | WQ_MEM_RECLAIM |
1688 				       WQ_UNBOUND, 0);
1689 	if (!io_workqueue) {
1690 		log_print("can't start dlm_io");
1691 		return -ENOMEM;
1692 	}
1693 
1694 	/* ordered dlm message process queue,
1695 	 * should be converted to a tasklet
1696 	 */
1697 	process_workqueue = alloc_ordered_workqueue("dlm_process",
1698 						    WQ_HIGHPRI | WQ_MEM_RECLAIM);
1699 	if (!process_workqueue) {
1700 		log_print("can't start dlm_process");
1701 		destroy_workqueue(io_workqueue);
1702 		io_workqueue = NULL;
1703 		return -ENOMEM;
1704 	}
1705 
1706 	return 0;
1707 }
1708 
dlm_lowcomms_shutdown(void)1709 void dlm_lowcomms_shutdown(void)
1710 {
1711 	struct connection *con;
1712 	int i, idx;
1713 
1714 	/* stop lowcomms_listen_data_ready calls */
1715 	lock_sock(listen_con.sock->sk);
1716 	listen_con.sock->sk->sk_data_ready = listen_sock.sk_data_ready;
1717 	release_sock(listen_con.sock->sk);
1718 
1719 	cancel_work_sync(&listen_con.rwork);
1720 	dlm_close_sock(&listen_con.sock);
1721 
1722 	idx = srcu_read_lock(&connections_srcu);
1723 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1724 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
1725 			shutdown_connection(con, true);
1726 			stop_connection_io(con);
1727 			flush_workqueue(process_workqueue);
1728 			close_connection(con, true);
1729 
1730 			clean_one_writequeue(con);
1731 			if (con->othercon)
1732 				clean_one_writequeue(con->othercon);
1733 			allow_connection_io(con);
1734 		}
1735 	}
1736 	srcu_read_unlock(&connections_srcu, idx);
1737 }
1738 
dlm_lowcomms_stop(void)1739 void dlm_lowcomms_stop(void)
1740 {
1741 	work_stop();
1742 	dlm_proto_ops = NULL;
1743 }
1744 
dlm_listen_for_all(void)1745 static int dlm_listen_for_all(void)
1746 {
1747 	struct socket *sock;
1748 	int result;
1749 
1750 	log_print("Using %s for communications",
1751 		  dlm_proto_ops->name);
1752 
1753 	result = dlm_proto_ops->listen_validate();
1754 	if (result < 0)
1755 		return result;
1756 
1757 	result = sock_create_kern(&init_net, dlm_local_addr[0].ss_family,
1758 				  SOCK_STREAM, dlm_proto_ops->proto, &sock);
1759 	if (result < 0) {
1760 		log_print("Can't create comms socket: %d", result);
1761 		return result;
1762 	}
1763 
1764 	sock_set_mark(sock->sk, dlm_config.ci_mark);
1765 	dlm_proto_ops->listen_sockopts(sock);
1766 
1767 	result = dlm_proto_ops->listen_bind(sock);
1768 	if (result < 0)
1769 		goto out;
1770 
1771 	lock_sock(sock->sk);
1772 	listen_sock.sk_data_ready = sock->sk->sk_data_ready;
1773 	listen_sock.sk_write_space = sock->sk->sk_write_space;
1774 	listen_sock.sk_error_report = sock->sk->sk_error_report;
1775 	listen_sock.sk_state_change = sock->sk->sk_state_change;
1776 
1777 	listen_con.sock = sock;
1778 
1779 	sock->sk->sk_allocation = GFP_NOFS;
1780 	sock->sk->sk_use_task_frag = false;
1781 	sock->sk->sk_data_ready = lowcomms_listen_data_ready;
1782 	release_sock(sock->sk);
1783 
1784 	result = sock->ops->listen(sock, 128);
1785 	if (result < 0) {
1786 		dlm_close_sock(&listen_con.sock);
1787 		return result;
1788 	}
1789 
1790 	return 0;
1791 
1792 out:
1793 	sock_release(sock);
1794 	return result;
1795 }
1796 
dlm_tcp_bind(struct socket * sock)1797 static int dlm_tcp_bind(struct socket *sock)
1798 {
1799 	struct sockaddr_storage src_addr;
1800 	int result, addr_len;
1801 
1802 	/* Bind to our cluster-known address connecting to avoid
1803 	 * routing problems.
1804 	 */
1805 	memcpy(&src_addr, &dlm_local_addr[0], sizeof(src_addr));
1806 	make_sockaddr(&src_addr, 0, &addr_len);
1807 
1808 	result = kernel_bind(sock, (struct sockaddr *)&src_addr,
1809 			     addr_len);
1810 	if (result < 0) {
1811 		/* This *may* not indicate a critical error */
1812 		log_print("could not bind for connect: %d", result);
1813 	}
1814 
1815 	return 0;
1816 }
1817 
dlm_tcp_connect(struct connection * con,struct socket * sock,struct sockaddr * addr,int addr_len)1818 static int dlm_tcp_connect(struct connection *con, struct socket *sock,
1819 			   struct sockaddr *addr, int addr_len)
1820 {
1821 	return kernel_connect(sock, addr, addr_len, O_NONBLOCK);
1822 }
1823 
dlm_tcp_listen_validate(void)1824 static int dlm_tcp_listen_validate(void)
1825 {
1826 	/* We don't support multi-homed hosts */
1827 	if (dlm_local_count > 1) {
1828 		log_print("TCP protocol can't handle multi-homed hosts, try SCTP");
1829 		return -EINVAL;
1830 	}
1831 
1832 	return 0;
1833 }
1834 
dlm_tcp_sockopts(struct socket * sock)1835 static void dlm_tcp_sockopts(struct socket *sock)
1836 {
1837 	/* Turn off Nagle's algorithm */
1838 	tcp_sock_set_nodelay(sock->sk);
1839 }
1840 
dlm_tcp_listen_sockopts(struct socket * sock)1841 static void dlm_tcp_listen_sockopts(struct socket *sock)
1842 {
1843 	dlm_tcp_sockopts(sock);
1844 	sock_set_reuseaddr(sock->sk);
1845 }
1846 
dlm_tcp_listen_bind(struct socket * sock)1847 static int dlm_tcp_listen_bind(struct socket *sock)
1848 {
1849 	int addr_len;
1850 
1851 	/* Bind to our port */
1852 	make_sockaddr(&dlm_local_addr[0], dlm_config.ci_tcp_port, &addr_len);
1853 	return kernel_bind(sock, (struct sockaddr *)&dlm_local_addr[0],
1854 			   addr_len);
1855 }
1856 
1857 static const struct dlm_proto_ops dlm_tcp_ops = {
1858 	.name = "TCP",
1859 	.proto = IPPROTO_TCP,
1860 	.connect = dlm_tcp_connect,
1861 	.sockopts = dlm_tcp_sockopts,
1862 	.bind = dlm_tcp_bind,
1863 	.listen_validate = dlm_tcp_listen_validate,
1864 	.listen_sockopts = dlm_tcp_listen_sockopts,
1865 	.listen_bind = dlm_tcp_listen_bind,
1866 };
1867 
dlm_sctp_bind(struct socket * sock)1868 static int dlm_sctp_bind(struct socket *sock)
1869 {
1870 	return sctp_bind_addrs(sock, 0);
1871 }
1872 
dlm_sctp_connect(struct connection * con,struct socket * sock,struct sockaddr * addr,int addr_len)1873 static int dlm_sctp_connect(struct connection *con, struct socket *sock,
1874 			    struct sockaddr *addr, int addr_len)
1875 {
1876 	int ret;
1877 
1878 	/*
1879 	 * Make kernel_connect() function return in specified time,
1880 	 * since O_NONBLOCK argument in connect() function does not work here,
1881 	 * then, we should restore the default value of this attribute.
1882 	 */
1883 	sock_set_sndtimeo(sock->sk, 5);
1884 	ret = kernel_connect(sock, addr, addr_len, 0);
1885 	sock_set_sndtimeo(sock->sk, 0);
1886 	return ret;
1887 }
1888 
dlm_sctp_listen_validate(void)1889 static int dlm_sctp_listen_validate(void)
1890 {
1891 	if (!IS_ENABLED(CONFIG_IP_SCTP)) {
1892 		log_print("SCTP is not enabled by this kernel");
1893 		return -EOPNOTSUPP;
1894 	}
1895 
1896 	request_module("sctp");
1897 	return 0;
1898 }
1899 
dlm_sctp_bind_listen(struct socket * sock)1900 static int dlm_sctp_bind_listen(struct socket *sock)
1901 {
1902 	return sctp_bind_addrs(sock, dlm_config.ci_tcp_port);
1903 }
1904 
dlm_sctp_sockopts(struct socket * sock)1905 static void dlm_sctp_sockopts(struct socket *sock)
1906 {
1907 	/* Turn off Nagle's algorithm */
1908 	sctp_sock_set_nodelay(sock->sk);
1909 	sock_set_rcvbuf(sock->sk, NEEDED_RMEM);
1910 }
1911 
1912 static const struct dlm_proto_ops dlm_sctp_ops = {
1913 	.name = "SCTP",
1914 	.proto = IPPROTO_SCTP,
1915 	.try_new_addr = true,
1916 	.connect = dlm_sctp_connect,
1917 	.sockopts = dlm_sctp_sockopts,
1918 	.bind = dlm_sctp_bind,
1919 	.listen_validate = dlm_sctp_listen_validate,
1920 	.listen_sockopts = dlm_sctp_sockopts,
1921 	.listen_bind = dlm_sctp_bind_listen,
1922 };
1923 
dlm_lowcomms_start(void)1924 int dlm_lowcomms_start(void)
1925 {
1926 	int error;
1927 
1928 	init_local();
1929 	if (!dlm_local_count) {
1930 		error = -ENOTCONN;
1931 		log_print("no local IP address has been set");
1932 		goto fail;
1933 	}
1934 
1935 	error = work_start();
1936 	if (error)
1937 		goto fail;
1938 
1939 	/* Start listening */
1940 	switch (dlm_config.ci_protocol) {
1941 	case DLM_PROTO_TCP:
1942 		dlm_proto_ops = &dlm_tcp_ops;
1943 		break;
1944 	case DLM_PROTO_SCTP:
1945 		dlm_proto_ops = &dlm_sctp_ops;
1946 		break;
1947 	default:
1948 		log_print("Invalid protocol identifier %d set",
1949 			  dlm_config.ci_protocol);
1950 		error = -EINVAL;
1951 		goto fail_proto_ops;
1952 	}
1953 
1954 	error = dlm_listen_for_all();
1955 	if (error)
1956 		goto fail_listen;
1957 
1958 	return 0;
1959 
1960 fail_listen:
1961 	dlm_proto_ops = NULL;
1962 fail_proto_ops:
1963 	work_stop();
1964 fail:
1965 	return error;
1966 }
1967 
dlm_lowcomms_init(void)1968 void dlm_lowcomms_init(void)
1969 {
1970 	int i;
1971 
1972 	for (i = 0; i < CONN_HASH_SIZE; i++)
1973 		INIT_HLIST_HEAD(&connection_hash[i]);
1974 
1975 	INIT_WORK(&listen_con.rwork, process_listen_recv_socket);
1976 }
1977 
dlm_lowcomms_exit(void)1978 void dlm_lowcomms_exit(void)
1979 {
1980 	struct connection *con;
1981 	int i, idx;
1982 
1983 	idx = srcu_read_lock(&connections_srcu);
1984 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1985 		hlist_for_each_entry_rcu(con, &connection_hash[i], list) {
1986 			spin_lock(&connections_lock);
1987 			hlist_del_rcu(&con->list);
1988 			spin_unlock(&connections_lock);
1989 
1990 			if (con->othercon)
1991 				call_srcu(&connections_srcu, &con->othercon->rcu,
1992 					  connection_release);
1993 			call_srcu(&connections_srcu, &con->rcu, connection_release);
1994 		}
1995 	}
1996 	srcu_read_unlock(&connections_srcu, idx);
1997 }
1998