xref: /openbmc/linux/fs/dlm/midcomms.c (revision 808643ea)
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) 2004-2021 Red Hat, Inc.  All rights reserved.
7 **
8 **
9 *******************************************************************************
10 ******************************************************************************/
11 
12 /*
13  * midcomms.c
14  *
15  * This is the appallingly named "mid-level" comms layer. It takes care about
16  * deliver an on application layer "reliable" communication above the used
17  * lowcomms transport layer.
18  *
19  * How it works:
20  *
21  * Each nodes keeps track of all send DLM messages in send_queue with a sequence
22  * number. The receive will send an DLM_ACK message back for every DLM message
23  * received at the other side. If a reconnect happens in lowcomms we will send
24  * all unacknowledged dlm messages again. The receiving side might drop any already
25  * received message by comparing sequence numbers.
26  *
27  * How version detection works:
28  *
29  * Due the fact that dlm has pre-configured node addresses on every side
30  * it is in it's nature that every side connects at starts to transmit
31  * dlm messages which ends in a race. However DLM_RCOM_NAMES, DLM_RCOM_STATUS
32  * and their replies are the first messages which are exchanges. Due backwards
33  * compatibility these messages are not covered by the midcomms re-transmission
34  * layer. These messages have their own re-transmission handling in the dlm
35  * application layer. The version field of every node will be set on these RCOM
36  * messages as soon as they arrived and the node isn't yet part of the nodes
37  * hash. There exists also logic to detect version mismatched if something weird
38  * going on or the first messages isn't an expected one.
39  *
40  * Termination:
41  *
42  * The midcomms layer does a 4 way handshake for termination on DLM protocol
43  * like TCP supports it with half-closed socket support. SCTP doesn't support
44  * half-closed socket, so we do it on DLM layer. Also socket shutdown() can be
45  * interrupted by .e.g. tcp reset itself. Additional there exists the othercon
46  * paradigm in lowcomms which cannot be easily without breaking backwards
47  * compatibility. A node cannot send anything to another node when a DLM_FIN
48  * message was send. There exists additional logic to print a warning if
49  * DLM wants to do it. There exists a state handling like RFC 793 but reduced
50  * to termination only. The event "member removal event" describes the cluster
51  * manager removed the node from internal lists, at this point DLM does not
52  * send any message to the other node. There exists two cases:
53  *
54  * 1. The cluster member was removed and we received a FIN
55  * OR
56  * 2. We received a FIN but the member was not removed yet
57  *
58  * One of these cases will do the CLOSE_WAIT to LAST_ACK change.
59  *
60  *
61  *                              +---------+
62  *                              | CLOSED  |
63  *                              +---------+
64  *                                   | add member/receive RCOM version
65  *                                   |            detection msg
66  *                                   V
67  *                              +---------+
68  *                              |  ESTAB  |
69  *                              +---------+
70  *                       CLOSE    |     |    rcv FIN
71  *                      -------   |     |    -------
72  * +---------+          snd FIN  /       \   snd ACK          +---------+
73  * |  FIN    |<-----------------           ------------------>|  CLOSE  |
74  * | WAIT-1  |------------------                              |   WAIT  |
75  * +---------+          rcv FIN  \                            +---------+
76  * | rcv ACK of FIN   -------   |                            CLOSE  | member
77  * | --------------   snd ACK   |                           ------- | removal
78  * V        x                   V                           snd FIN V event
79  * +---------+                  +---------+                   +---------+
80  * |FINWAIT-2|                  | CLOSING |                   | LAST-ACK|
81  * +---------+                  +---------+                   +---------+
82  * |                rcv ACK of FIN |                 rcv ACK of FIN |
83  * |  rcv FIN       -------------- |                 -------------- |
84  * |  -------              x       V                        x       V
85  *  \ snd ACK                 +---------+                   +---------+
86  *   ------------------------>| CLOSED  |                   | CLOSED  |
87  *                            +---------+                   +---------+
88  *
89  * NOTE: any state can interrupted by midcomms_close() and state will be
90  * switched to CLOSED in case of fencing. There exists also some timeout
91  * handling when we receive the version detection RCOM messages which is
92  * made by observation.
93  *
94  * Future improvements:
95  *
96  * There exists some known issues/improvements of the dlm handling. Some
97  * of them should be done in a next major dlm version bump which makes
98  * it incompatible with previous versions.
99  *
100  * Unaligned memory access:
101  *
102  * There exists cases when the dlm message buffer length is not aligned
103  * to 8 byte. However seems nobody detected any problem with it. This
104  * can be fixed in the next major version bump of dlm.
105  *
106  * Version detection:
107  *
108  * The version detection and how it's done is related to backwards
109  * compatibility. There exists better ways to make a better handling.
110  * However this should be changed in the next major version bump of dlm.
111  *
112  * Ack handling:
113  *
114  * Currently we send an ack message for every dlm message. However we
115  * can ack multiple dlm messages with one ack by just delaying the ack
116  * message. Will reduce some traffic but makes the drop detection slower.
117  *
118  * Tail Size checking:
119  *
120  * There exists a message tail payload in e.g. DLM_MSG however we don't
121  * check it against the message length yet regarding to the receive buffer
122  * length. That need to be validated.
123  *
124  * Fencing bad nodes:
125  *
126  * At timeout places or weird sequence number behaviours we should send
127  * a fencing request to the cluster manager.
128  */
129 
130 /* Debug switch to enable a 5 seconds sleep waiting of a termination.
131  * This can be useful to test fencing while termination is running.
132  * This requires a setup with only gfs2 as dlm user, so that the
133  * last umount will terminate the connection.
134  *
135  * However it became useful to test, while the 5 seconds block in umount
136  * just press the reset button. In a lot of dropping the termination
137  * process can could take several seconds.
138  */
139 #define DLM_DEBUG_FENCE_TERMINATION	0
140 
141 #include <net/tcp.h>
142 
143 #include "dlm_internal.h"
144 #include "lowcomms.h"
145 #include "config.h"
146 #include "lock.h"
147 #include "util.h"
148 #include "midcomms.h"
149 
150 /* init value for sequence numbers for testing purpose only e.g. overflows */
151 #define DLM_SEQ_INIT		0
152 /* 3 minutes wait to sync ending of dlm */
153 #define DLM_SHUTDOWN_TIMEOUT	msecs_to_jiffies(3 * 60 * 1000)
154 #define DLM_VERSION_NOT_SET	0
155 
156 struct midcomms_node {
157 	int nodeid;
158 	uint32_t version;
159 	uint32_t seq_send;
160 	uint32_t seq_next;
161 	/* These queues are unbound because we cannot drop any message in dlm.
162 	 * We could send a fence signal for a specific node to the cluster
163 	 * manager if queues hits some maximum value, however this handling
164 	 * not supported yet.
165 	 */
166 	struct list_head send_queue;
167 	spinlock_t send_queue_lock;
168 	atomic_t send_queue_cnt;
169 #define DLM_NODE_FLAG_CLOSE	1
170 #define DLM_NODE_FLAG_STOP_TX	2
171 #define DLM_NODE_FLAG_STOP_RX	3
172 	unsigned long flags;
173 	wait_queue_head_t shutdown_wait;
174 
175 	/* dlm tcp termination state */
176 #define DLM_CLOSED	1
177 #define DLM_ESTABLISHED	2
178 #define DLM_FIN_WAIT1	3
179 #define DLM_FIN_WAIT2	4
180 #define DLM_CLOSE_WAIT	5
181 #define DLM_LAST_ACK	6
182 #define DLM_CLOSING	7
183 	int state;
184 	spinlock_t state_lock;
185 
186 	/* counts how many lockspaces are using this node
187 	 * this refcount is necessary to determine if the
188 	 * node wants to disconnect.
189 	 */
190 	int users;
191 
192 	/* not protected by srcu, node_hash lifetime */
193 	void *debugfs;
194 
195 	struct hlist_node hlist;
196 	struct rcu_head rcu;
197 };
198 
199 struct dlm_mhandle {
200 	const struct dlm_header *inner_hd;
201 	struct midcomms_node *node;
202 	struct dlm_opts *opts;
203 	struct dlm_msg *msg;
204 	bool committed;
205 	uint32_t seq;
206 
207 	void (*ack_rcv)(struct midcomms_node *node);
208 
209 	/* get_mhandle/commit srcu idx exchange */
210 	int idx;
211 
212 	struct list_head list;
213 	struct rcu_head rcu;
214 };
215 
216 static struct hlist_head node_hash[CONN_HASH_SIZE];
217 static DEFINE_SPINLOCK(nodes_lock);
218 DEFINE_STATIC_SRCU(nodes_srcu);
219 
220 /* This mutex prevents that midcomms_close() is running while
221  * stop() or remove(). As I experienced invalid memory access
222  * behaviours when DLM_DEBUG_FENCE_TERMINATION is enabled and
223  * resetting machines. I will end in some double deletion in nodes
224  * datastructure.
225  */
226 static DEFINE_MUTEX(close_lock);
227 
228 static inline const char *dlm_state_str(int state)
229 {
230 	switch (state) {
231 	case DLM_CLOSED:
232 		return "CLOSED";
233 	case DLM_ESTABLISHED:
234 		return "ESTABLISHED";
235 	case DLM_FIN_WAIT1:
236 		return "FIN_WAIT1";
237 	case DLM_FIN_WAIT2:
238 		return "FIN_WAIT2";
239 	case DLM_CLOSE_WAIT:
240 		return "CLOSE_WAIT";
241 	case DLM_LAST_ACK:
242 		return "LAST_ACK";
243 	case DLM_CLOSING:
244 		return "CLOSING";
245 	default:
246 		return "UNKNOWN";
247 	}
248 }
249 
250 const char *dlm_midcomms_state(struct midcomms_node *node)
251 {
252 	return dlm_state_str(node->state);
253 }
254 
255 unsigned long dlm_midcomms_flags(struct midcomms_node *node)
256 {
257 	return node->flags;
258 }
259 
260 int dlm_midcomms_send_queue_cnt(struct midcomms_node *node)
261 {
262 	return atomic_read(&node->send_queue_cnt);
263 }
264 
265 uint32_t dlm_midcomms_version(struct midcomms_node *node)
266 {
267 	return node->version;
268 }
269 
270 static struct midcomms_node *__find_node(int nodeid, int r)
271 {
272 	struct midcomms_node *node;
273 
274 	hlist_for_each_entry_rcu(node, &node_hash[r], hlist) {
275 		if (node->nodeid == nodeid)
276 			return node;
277 	}
278 
279 	return NULL;
280 }
281 
282 static void dlm_mhandle_release(struct rcu_head *rcu)
283 {
284 	struct dlm_mhandle *mh = container_of(rcu, struct dlm_mhandle, rcu);
285 
286 	dlm_lowcomms_put_msg(mh->msg);
287 	kfree(mh);
288 }
289 
290 static void dlm_mhandle_delete(struct midcomms_node *node,
291 			       struct dlm_mhandle *mh)
292 {
293 	list_del_rcu(&mh->list);
294 	atomic_dec(&node->send_queue_cnt);
295 	call_rcu(&mh->rcu, dlm_mhandle_release);
296 }
297 
298 static void dlm_send_queue_flush(struct midcomms_node *node)
299 {
300 	struct dlm_mhandle *mh;
301 
302 	pr_debug("flush midcomms send queue of node %d\n", node->nodeid);
303 
304 	rcu_read_lock();
305 	spin_lock(&node->send_queue_lock);
306 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
307 		dlm_mhandle_delete(node, mh);
308 	}
309 	spin_unlock(&node->send_queue_lock);
310 	rcu_read_unlock();
311 }
312 
313 static void midcomms_node_reset(struct midcomms_node *node)
314 {
315 	pr_debug("reset node %d\n", node->nodeid);
316 
317 	node->seq_next = DLM_SEQ_INIT;
318 	node->seq_send = DLM_SEQ_INIT;
319 	node->version = DLM_VERSION_NOT_SET;
320 	node->flags = 0;
321 
322 	dlm_send_queue_flush(node);
323 	node->state = DLM_CLOSED;
324 	wake_up(&node->shutdown_wait);
325 }
326 
327 static struct midcomms_node *nodeid2node(int nodeid, gfp_t alloc)
328 {
329 	struct midcomms_node *node, *tmp;
330 	int r = nodeid_hash(nodeid);
331 
332 	node = __find_node(nodeid, r);
333 	if (node || !alloc)
334 		return node;
335 
336 	node = kmalloc(sizeof(*node), alloc);
337 	if (!node)
338 		return NULL;
339 
340 	node->nodeid = nodeid;
341 	spin_lock_init(&node->state_lock);
342 	spin_lock_init(&node->send_queue_lock);
343 	atomic_set(&node->send_queue_cnt, 0);
344 	INIT_LIST_HEAD(&node->send_queue);
345 	init_waitqueue_head(&node->shutdown_wait);
346 	node->users = 0;
347 	midcomms_node_reset(node);
348 
349 	spin_lock(&nodes_lock);
350 	/* check again if there was somebody else
351 	 * earlier here to add the node
352 	 */
353 	tmp = __find_node(nodeid, r);
354 	if (tmp) {
355 		spin_unlock(&nodes_lock);
356 		kfree(node);
357 		return tmp;
358 	}
359 
360 	hlist_add_head_rcu(&node->hlist, &node_hash[r]);
361 	spin_unlock(&nodes_lock);
362 
363 	node->debugfs = dlm_create_debug_comms_file(nodeid, node);
364 	return node;
365 }
366 
367 static int dlm_send_ack(int nodeid, uint32_t seq)
368 {
369 	int mb_len = sizeof(struct dlm_header);
370 	struct dlm_header *m_header;
371 	struct dlm_msg *msg;
372 	char *ppc;
373 
374 	msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_NOFS, &ppc,
375 				   NULL, NULL);
376 	if (!msg)
377 		return -ENOMEM;
378 
379 	m_header = (struct dlm_header *)ppc;
380 
381 	m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
382 	m_header->h_nodeid = dlm_our_nodeid();
383 	m_header->h_length = mb_len;
384 	m_header->h_cmd = DLM_ACK;
385 	m_header->u.h_seq = seq;
386 
387 	header_out(m_header);
388 	dlm_lowcomms_commit_msg(msg);
389 	dlm_lowcomms_put_msg(msg);
390 
391 	return 0;
392 }
393 
394 static int dlm_send_fin(struct midcomms_node *node,
395 			void (*ack_rcv)(struct midcomms_node *node))
396 {
397 	int mb_len = sizeof(struct dlm_header);
398 	struct dlm_header *m_header;
399 	struct dlm_mhandle *mh;
400 	char *ppc;
401 
402 	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_NOFS, &ppc);
403 	if (!mh)
404 		return -ENOMEM;
405 
406 	mh->ack_rcv = ack_rcv;
407 
408 	m_header = (struct dlm_header *)ppc;
409 
410 	m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
411 	m_header->h_nodeid = dlm_our_nodeid();
412 	m_header->h_length = mb_len;
413 	m_header->h_cmd = DLM_FIN;
414 
415 	header_out(m_header);
416 
417 	pr_debug("sending fin msg to node %d\n", node->nodeid);
418 	dlm_midcomms_commit_mhandle(mh);
419 	set_bit(DLM_NODE_FLAG_STOP_TX, &node->flags);
420 
421 	return 0;
422 }
423 
424 static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
425 {
426 	struct dlm_mhandle *mh;
427 
428 	rcu_read_lock();
429 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
430 		if (before(mh->seq, seq)) {
431 			if (mh->ack_rcv)
432 				mh->ack_rcv(node);
433 		} else {
434 			/* send queue should be ordered */
435 			break;
436 		}
437 	}
438 
439 	spin_lock(&node->send_queue_lock);
440 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
441 		if (before(mh->seq, seq)) {
442 			dlm_mhandle_delete(node, mh);
443 		} else {
444 			/* send queue should be ordered */
445 			break;
446 		}
447 	}
448 	spin_unlock(&node->send_queue_lock);
449 	rcu_read_unlock();
450 }
451 
452 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
453 {
454 	spin_lock(&node->state_lock);
455 	pr_debug("receive passive fin ack from node %d with state %s\n",
456 		 node->nodeid, dlm_state_str(node->state));
457 
458 	switch (node->state) {
459 	case DLM_LAST_ACK:
460 		/* DLM_CLOSED */
461 		midcomms_node_reset(node);
462 		break;
463 	case DLM_CLOSED:
464 		/* not valid but somehow we got what we want */
465 		wake_up(&node->shutdown_wait);
466 		break;
467 	default:
468 		spin_unlock(&node->state_lock);
469 		log_print("%s: unexpected state: %d\n",
470 			  __func__, node->state);
471 		WARN_ON(1);
472 		return;
473 	}
474 	spin_unlock(&node->state_lock);
475 }
476 
477 static void dlm_midcomms_receive_buffer(union dlm_packet *p,
478 					struct midcomms_node *node,
479 					uint32_t seq)
480 {
481 	if (seq == node->seq_next) {
482 		node->seq_next++;
483 		/* send ack before fin */
484 		dlm_send_ack(node->nodeid, node->seq_next);
485 
486 		switch (p->header.h_cmd) {
487 		case DLM_FIN:
488 			spin_lock(&node->state_lock);
489 			pr_debug("receive fin msg from node %d with state %s\n",
490 				 node->nodeid, dlm_state_str(node->state));
491 
492 			switch (node->state) {
493 			case DLM_ESTABLISHED:
494 				node->state = DLM_CLOSE_WAIT;
495 				pr_debug("switch node %d to state %s\n",
496 					 node->nodeid, dlm_state_str(node->state));
497 				/* passive shutdown DLM_LAST_ACK case 1
498 				 * additional we check if the node is used by
499 				 * cluster manager events at all.
500 				 */
501 				if (node->users == 0) {
502 					node->state = DLM_LAST_ACK;
503 					pr_debug("switch node %d to state %s case 1\n",
504 						 node->nodeid, dlm_state_str(node->state));
505 					spin_unlock(&node->state_lock);
506 					goto send_fin;
507 				}
508 				break;
509 			case DLM_FIN_WAIT1:
510 				node->state = DLM_CLOSING;
511 				pr_debug("switch node %d to state %s\n",
512 					 node->nodeid, dlm_state_str(node->state));
513 				break;
514 			case DLM_FIN_WAIT2:
515 				midcomms_node_reset(node);
516 				pr_debug("switch node %d to state %s\n",
517 					 node->nodeid, dlm_state_str(node->state));
518 				wake_up(&node->shutdown_wait);
519 				break;
520 			case DLM_LAST_ACK:
521 				/* probably remove_member caught it, do nothing */
522 				break;
523 			default:
524 				spin_unlock(&node->state_lock);
525 				log_print("%s: unexpected state: %d\n",
526 					  __func__, node->state);
527 				WARN_ON(1);
528 				return;
529 			}
530 			spin_unlock(&node->state_lock);
531 
532 			set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
533 			break;
534 		default:
535 			WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
536 			dlm_receive_buffer(p, node->nodeid);
537 			break;
538 		}
539 	} else {
540 		/* retry to ack message which we already have by sending back
541 		 * current node->seq_next number as ack.
542 		 */
543 		if (seq < node->seq_next)
544 			dlm_send_ack(node->nodeid, node->seq_next);
545 
546 		log_print_ratelimited("ignore dlm msg because seq mismatch, seq: %u, expected: %u, nodeid: %d",
547 				      seq, node->seq_next, node->nodeid);
548 	}
549 
550 	return;
551 
552 send_fin:
553 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
554 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
555 }
556 
557 static struct midcomms_node *
558 dlm_midcomms_recv_node_lookup(int nodeid, const union dlm_packet *p,
559 			      uint16_t msglen, int (*cb)(struct midcomms_node *node))
560 {
561 	struct midcomms_node *node = NULL;
562 	gfp_t allocation = 0;
563 	int ret;
564 
565 	switch (p->header.h_cmd) {
566 	case DLM_RCOM:
567 		if (msglen < sizeof(struct dlm_rcom)) {
568 			log_print("rcom msg too small: %u, will skip this message from node %d",
569 				  msglen, nodeid);
570 			return NULL;
571 		}
572 
573 		switch (le32_to_cpu(p->rcom.rc_type)) {
574 		case DLM_RCOM_NAMES:
575 			fallthrough;
576 		case DLM_RCOM_NAMES_REPLY:
577 			fallthrough;
578 		case DLM_RCOM_STATUS:
579 			fallthrough;
580 		case DLM_RCOM_STATUS_REPLY:
581 			node = nodeid2node(nodeid, 0);
582 			if (node) {
583 				spin_lock(&node->state_lock);
584 				if (node->state != DLM_ESTABLISHED)
585 					pr_debug("receive begin RCOM msg from node %d with state %s\n",
586 						 node->nodeid, dlm_state_str(node->state));
587 
588 				switch (node->state) {
589 				case DLM_CLOSED:
590 					node->state = DLM_ESTABLISHED;
591 					pr_debug("switch node %d to state %s\n",
592 						 node->nodeid, dlm_state_str(node->state));
593 					break;
594 				case DLM_ESTABLISHED:
595 					break;
596 				default:
597 					/* some invalid state passive shutdown
598 					 * was failed, we try to reset and
599 					 * hope it will go on.
600 					 */
601 					log_print("reset node %d because shutdown stuck",
602 						  node->nodeid);
603 
604 					midcomms_node_reset(node);
605 					node->state = DLM_ESTABLISHED;
606 					break;
607 				}
608 				spin_unlock(&node->state_lock);
609 			}
610 
611 			allocation = GFP_NOFS;
612 			break;
613 		default:
614 			break;
615 		}
616 
617 		break;
618 	default:
619 		break;
620 	}
621 
622 	node = nodeid2node(nodeid, allocation);
623 	if (!node) {
624 		switch (p->header.h_cmd) {
625 		case DLM_OPTS:
626 			if (msglen < sizeof(struct dlm_opts)) {
627 				log_print("opts msg too small: %u, will skip this message from node %d",
628 					  msglen, nodeid);
629 				return NULL;
630 			}
631 
632 			log_print_ratelimited("received dlm opts message nextcmd %d from node %d in an invalid sequence",
633 					      p->opts.o_nextcmd, nodeid);
634 			break;
635 		default:
636 			log_print_ratelimited("received dlm message cmd %d from node %d in an invalid sequence",
637 					      p->header.h_cmd, nodeid);
638 			break;
639 		}
640 
641 		return NULL;
642 	}
643 
644 	ret = cb(node);
645 	if (ret < 0)
646 		return NULL;
647 
648 	return node;
649 }
650 
651 static int dlm_midcomms_version_check_3_2(struct midcomms_node *node)
652 {
653 	switch (node->version) {
654 	case DLM_VERSION_NOT_SET:
655 		node->version = DLM_VERSION_3_2;
656 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_2,
657 			  node->nodeid);
658 		break;
659 	case DLM_VERSION_3_2:
660 		break;
661 	default:
662 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
663 				      DLM_VERSION_3_2, node->nodeid, node->version);
664 		return -1;
665 	}
666 
667 	return 0;
668 }
669 
670 static int dlm_opts_check_msglen(union dlm_packet *p, uint16_t msglen, int nodeid)
671 {
672 	int len = msglen;
673 
674 	/* we only trust outer header msglen because
675 	 * it's checked against receive buffer length.
676 	 */
677 	if (len < sizeof(struct dlm_opts))
678 		return -1;
679 	len -= sizeof(struct dlm_opts);
680 
681 	if (len < le16_to_cpu(p->opts.o_optlen))
682 		return -1;
683 	len -= le16_to_cpu(p->opts.o_optlen);
684 
685 	switch (p->opts.o_nextcmd) {
686 	case DLM_FIN:
687 		if (len < sizeof(struct dlm_header)) {
688 			log_print("fin too small: %d, will skip this message from node %d",
689 				  len, nodeid);
690 			return -1;
691 		}
692 
693 		break;
694 	case DLM_MSG:
695 		if (len < sizeof(struct dlm_message)) {
696 			log_print("msg too small: %d, will skip this message from node %d",
697 				  msglen, nodeid);
698 			return -1;
699 		}
700 
701 		break;
702 	case DLM_RCOM:
703 		if (len < sizeof(struct dlm_rcom)) {
704 			log_print("rcom msg too small: %d, will skip this message from node %d",
705 				  len, nodeid);
706 			return -1;
707 		}
708 
709 		break;
710 	default:
711 		log_print("unsupported o_nextcmd received: %u, will skip this message from node %d",
712 			  p->opts.o_nextcmd, nodeid);
713 		return -1;
714 	}
715 
716 	return 0;
717 }
718 
719 static void dlm_midcomms_receive_buffer_3_2(union dlm_packet *p, int nodeid)
720 {
721 	uint16_t msglen = le16_to_cpu(p->header.h_length);
722 	struct midcomms_node *node;
723 	uint32_t seq;
724 	int ret, idx;
725 
726 	idx = srcu_read_lock(&nodes_srcu);
727 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
728 					     dlm_midcomms_version_check_3_2);
729 	if (!node)
730 		goto out;
731 
732 	switch (p->header.h_cmd) {
733 	case DLM_RCOM:
734 		/* these rcom message we use to determine version.
735 		 * they have their own retransmission handling and
736 		 * are the first messages of dlm.
737 		 *
738 		 * length already checked.
739 		 */
740 		switch (le32_to_cpu(p->rcom.rc_type)) {
741 		case DLM_RCOM_NAMES:
742 			fallthrough;
743 		case DLM_RCOM_NAMES_REPLY:
744 			fallthrough;
745 		case DLM_RCOM_STATUS:
746 			fallthrough;
747 		case DLM_RCOM_STATUS_REPLY:
748 			break;
749 		default:
750 			log_print("unsupported rcom type received: %u, will skip this message from node %d",
751 				  le32_to_cpu(p->rcom.rc_type), nodeid);
752 			goto out;
753 		}
754 
755 		WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
756 		dlm_receive_buffer(p, nodeid);
757 		break;
758 	case DLM_OPTS:
759 		seq = le32_to_cpu(p->header.u.h_seq);
760 
761 		ret = dlm_opts_check_msglen(p, msglen, nodeid);
762 		if (ret < 0) {
763 			log_print("opts msg too small: %u, will skip this message from node %d",
764 				  msglen, nodeid);
765 			goto out;
766 		}
767 
768 		p = (union dlm_packet *)((unsigned char *)p->opts.o_opts +
769 					 le16_to_cpu(p->opts.o_optlen));
770 
771 		/* recheck inner msglen just if it's not garbage */
772 		msglen = le16_to_cpu(p->header.h_length);
773 		switch (p->header.h_cmd) {
774 		case DLM_RCOM:
775 			if (msglen < sizeof(struct dlm_rcom)) {
776 				log_print("inner rcom msg too small: %u, will skip this message from node %d",
777 					  msglen, nodeid);
778 				goto out;
779 			}
780 
781 			break;
782 		case DLM_MSG:
783 			if (msglen < sizeof(struct dlm_message)) {
784 				log_print("inner msg too small: %u, will skip this message from node %d",
785 					  msglen, nodeid);
786 				goto out;
787 			}
788 
789 			break;
790 		case DLM_FIN:
791 			if (msglen < sizeof(struct dlm_header)) {
792 				log_print("inner fin too small: %u, will skip this message from node %d",
793 					  msglen, nodeid);
794 				goto out;
795 			}
796 
797 			break;
798 		default:
799 			log_print("unsupported inner h_cmd received: %u, will skip this message from node %d",
800 				  msglen, nodeid);
801 			goto out;
802 		}
803 
804 		dlm_midcomms_receive_buffer(p, node, seq);
805 		break;
806 	case DLM_ACK:
807 		seq = le32_to_cpu(p->header.u.h_seq);
808 		dlm_receive_ack(node, seq);
809 		break;
810 	default:
811 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
812 			  p->header.h_cmd, nodeid);
813 		break;
814 	}
815 
816 out:
817 	srcu_read_unlock(&nodes_srcu, idx);
818 }
819 
820 static int dlm_midcomms_version_check_3_1(struct midcomms_node *node)
821 {
822 	switch (node->version) {
823 	case DLM_VERSION_NOT_SET:
824 		node->version = DLM_VERSION_3_1;
825 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_1,
826 			  node->nodeid);
827 		break;
828 	case DLM_VERSION_3_1:
829 		break;
830 	default:
831 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
832 				      DLM_VERSION_3_1, node->nodeid, node->version);
833 		return -1;
834 	}
835 
836 	return 0;
837 }
838 
839 static void dlm_midcomms_receive_buffer_3_1(union dlm_packet *p, int nodeid)
840 {
841 	uint16_t msglen = le16_to_cpu(p->header.h_length);
842 	struct midcomms_node *node;
843 	int idx;
844 
845 	idx = srcu_read_lock(&nodes_srcu);
846 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
847 					     dlm_midcomms_version_check_3_1);
848 	if (!node) {
849 		srcu_read_unlock(&nodes_srcu, idx);
850 		return;
851 	}
852 	srcu_read_unlock(&nodes_srcu, idx);
853 
854 	switch (p->header.h_cmd) {
855 	case DLM_RCOM:
856 		/* length already checked */
857 		break;
858 	case DLM_MSG:
859 		if (msglen < sizeof(struct dlm_message)) {
860 			log_print("msg too small: %u, will skip this message from node %d",
861 				  msglen, nodeid);
862 			return;
863 		}
864 
865 		break;
866 	default:
867 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
868 			  p->header.h_cmd, nodeid);
869 		return;
870 	}
871 
872 	dlm_receive_buffer(p, nodeid);
873 }
874 
875 /*
876  * Called from the low-level comms layer to process a buffer of
877  * commands.
878  */
879 
880 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int len)
881 {
882 	const unsigned char *ptr = buf;
883 	const struct dlm_header *hd;
884 	uint16_t msglen;
885 	int ret = 0;
886 
887 	while (len >= sizeof(struct dlm_header)) {
888 		hd = (struct dlm_header *)ptr;
889 
890 		/* no message should be more than DLM_MAX_SOCKET_BUFSIZE or
891 		 * less than dlm_header size.
892 		 *
893 		 * Some messages does not have a 8 byte length boundary yet
894 		 * which can occur in a unaligned memory access of some dlm
895 		 * messages. However this problem need to be fixed at the
896 		 * sending side, for now it seems nobody run into architecture
897 		 * related issues yet but it slows down some processing.
898 		 * Fixing this issue should be scheduled in future by doing
899 		 * the next major version bump.
900 		 */
901 		msglen = le16_to_cpu(hd->h_length);
902 		if (msglen > DLM_MAX_SOCKET_BUFSIZE ||
903 		    msglen < sizeof(struct dlm_header)) {
904 			log_print("received invalid length header: %u from node %d, will abort message parsing",
905 				  msglen, nodeid);
906 			return -EBADMSG;
907 		}
908 
909 		/* caller will take care that leftover
910 		 * will be parsed next call with more data
911 		 */
912 		if (msglen > len)
913 			break;
914 
915 		switch (le32_to_cpu(hd->h_version)) {
916 		case DLM_VERSION_3_1:
917 			dlm_midcomms_receive_buffer_3_1((union dlm_packet *)ptr, nodeid);
918 			break;
919 		case DLM_VERSION_3_2:
920 			dlm_midcomms_receive_buffer_3_2((union dlm_packet *)ptr, nodeid);
921 			break;
922 		default:
923 			log_print("received invalid version header: %u from node %d, will skip this message",
924 				  le32_to_cpu(hd->h_version), nodeid);
925 			break;
926 		}
927 
928 		ret += msglen;
929 		len -= msglen;
930 		ptr += msglen;
931 	}
932 
933 	return ret;
934 }
935 
936 void dlm_midcomms_unack_msg_resend(int nodeid)
937 {
938 	struct midcomms_node *node;
939 	struct dlm_mhandle *mh;
940 	int idx, ret;
941 
942 	idx = srcu_read_lock(&nodes_srcu);
943 	node = nodeid2node(nodeid, 0);
944 	if (!node) {
945 		srcu_read_unlock(&nodes_srcu, idx);
946 		return;
947 	}
948 
949 	/* old protocol, we don't support to retransmit on failure */
950 	switch (node->version) {
951 	case DLM_VERSION_3_2:
952 		break;
953 	default:
954 		srcu_read_unlock(&nodes_srcu, idx);
955 		return;
956 	}
957 
958 	rcu_read_lock();
959 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
960 		if (!mh->committed)
961 			continue;
962 
963 		ret = dlm_lowcomms_resend_msg(mh->msg);
964 		if (!ret)
965 			log_print_ratelimited("retransmit dlm msg, seq %u, nodeid %d",
966 					      mh->seq, node->nodeid);
967 	}
968 	rcu_read_unlock();
969 	srcu_read_unlock(&nodes_srcu, idx);
970 }
971 
972 static void dlm_fill_opts_header(struct dlm_opts *opts, uint16_t inner_len,
973 				 uint32_t seq)
974 {
975 	opts->o_header.h_cmd = DLM_OPTS;
976 	opts->o_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
977 	opts->o_header.h_nodeid = dlm_our_nodeid();
978 	opts->o_header.h_length = DLM_MIDCOMMS_OPT_LEN + inner_len;
979 	opts->o_header.u.h_seq = seq;
980 	header_out(&opts->o_header);
981 }
982 
983 static void midcomms_new_msg_cb(struct dlm_mhandle *mh)
984 {
985 	atomic_inc(&mh->node->send_queue_cnt);
986 
987 	spin_lock(&mh->node->send_queue_lock);
988 	list_add_tail_rcu(&mh->list, &mh->node->send_queue);
989 	spin_unlock(&mh->node->send_queue_lock);
990 
991 	mh->seq = mh->node->seq_send++;
992 }
993 
994 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
995 						int len, gfp_t allocation, char **ppc)
996 {
997 	struct dlm_opts *opts;
998 	struct dlm_msg *msg;
999 
1000 	msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
1001 				   allocation, ppc, midcomms_new_msg_cb, mh);
1002 	if (!msg)
1003 		return NULL;
1004 
1005 	opts = (struct dlm_opts *)*ppc;
1006 	mh->opts = opts;
1007 
1008 	/* add possible options here */
1009 	dlm_fill_opts_header(opts, len, mh->seq);
1010 
1011 	*ppc += sizeof(*opts);
1012 	mh->inner_hd = (const struct dlm_header *)*ppc;
1013 	return msg;
1014 }
1015 
1016 struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
1017 					     gfp_t allocation, char **ppc)
1018 {
1019 	struct midcomms_node *node;
1020 	struct dlm_mhandle *mh;
1021 	struct dlm_msg *msg;
1022 	int idx;
1023 
1024 	idx = srcu_read_lock(&nodes_srcu);
1025 	node = nodeid2node(nodeid, 0);
1026 	if (!node) {
1027 		WARN_ON_ONCE(1);
1028 		goto err;
1029 	}
1030 
1031 	/* this is a bug, however we going on and hope it will be resolved */
1032 	WARN_ON(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
1033 
1034 	mh = kzalloc(sizeof(*mh), GFP_NOFS);
1035 	if (!mh)
1036 		goto err;
1037 
1038 	mh->idx = idx;
1039 	mh->node = node;
1040 
1041 	switch (node->version) {
1042 	case DLM_VERSION_3_1:
1043 		msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
1044 					   NULL, NULL);
1045 		if (!msg) {
1046 			kfree(mh);
1047 			goto err;
1048 		}
1049 
1050 		break;
1051 	case DLM_VERSION_3_2:
1052 		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
1053 					       ppc);
1054 		if (!msg) {
1055 			kfree(mh);
1056 			goto err;
1057 		}
1058 
1059 		break;
1060 	default:
1061 		kfree(mh);
1062 		WARN_ON(1);
1063 		goto err;
1064 	}
1065 
1066 	mh->msg = msg;
1067 
1068 	/* keep in mind that is a must to call
1069 	 * dlm_midcomms_commit_msg() which releases
1070 	 * nodes_srcu using mh->idx which is assumed
1071 	 * here that the application will call it.
1072 	 */
1073 	return mh;
1074 
1075 err:
1076 	srcu_read_unlock(&nodes_srcu, idx);
1077 	return NULL;
1078 }
1079 
1080 static void dlm_midcomms_commit_msg_3_2(struct dlm_mhandle *mh)
1081 {
1082 	/* nexthdr chain for fast lookup */
1083 	mh->opts->o_nextcmd = mh->inner_hd->h_cmd;
1084 	mh->committed = true;
1085 	dlm_lowcomms_commit_msg(mh->msg);
1086 }
1087 
1088 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh)
1089 {
1090 	switch (mh->node->version) {
1091 	case DLM_VERSION_3_1:
1092 		srcu_read_unlock(&nodes_srcu, mh->idx);
1093 
1094 		dlm_lowcomms_commit_msg(mh->msg);
1095 		dlm_lowcomms_put_msg(mh->msg);
1096 		/* mh is not part of rcu list in this case */
1097 		kfree(mh);
1098 		break;
1099 	case DLM_VERSION_3_2:
1100 		dlm_midcomms_commit_msg_3_2(mh);
1101 		srcu_read_unlock(&nodes_srcu, mh->idx);
1102 		break;
1103 	default:
1104 		srcu_read_unlock(&nodes_srcu, mh->idx);
1105 		WARN_ON(1);
1106 		break;
1107 	}
1108 }
1109 
1110 int dlm_midcomms_start(void)
1111 {
1112 	int i;
1113 
1114 	for (i = 0; i < CONN_HASH_SIZE; i++)
1115 		INIT_HLIST_HEAD(&node_hash[i]);
1116 
1117 	return dlm_lowcomms_start();
1118 }
1119 
1120 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
1121 {
1122 	spin_lock(&node->state_lock);
1123 	pr_debug("receive active fin ack from node %d with state %s\n",
1124 		 node->nodeid, dlm_state_str(node->state));
1125 
1126 	switch (node->state) {
1127 	case DLM_FIN_WAIT1:
1128 		node->state = DLM_FIN_WAIT2;
1129 		pr_debug("switch node %d to state %s\n",
1130 			 node->nodeid, dlm_state_str(node->state));
1131 		break;
1132 	case DLM_CLOSING:
1133 		midcomms_node_reset(node);
1134 		pr_debug("switch node %d to state %s\n",
1135 			 node->nodeid, dlm_state_str(node->state));
1136 		wake_up(&node->shutdown_wait);
1137 		break;
1138 	case DLM_CLOSED:
1139 		/* not valid but somehow we got what we want */
1140 		wake_up(&node->shutdown_wait);
1141 		break;
1142 	default:
1143 		spin_unlock(&node->state_lock);
1144 		log_print("%s: unexpected state: %d\n",
1145 			  __func__, node->state);
1146 		WARN_ON(1);
1147 		return;
1148 	}
1149 	spin_unlock(&node->state_lock);
1150 }
1151 
1152 void dlm_midcomms_add_member(int nodeid)
1153 {
1154 	struct midcomms_node *node;
1155 	int idx;
1156 
1157 	if (nodeid == dlm_our_nodeid())
1158 		return;
1159 
1160 	idx = srcu_read_lock(&nodes_srcu);
1161 	node = nodeid2node(nodeid, GFP_NOFS);
1162 	if (!node) {
1163 		srcu_read_unlock(&nodes_srcu, idx);
1164 		return;
1165 	}
1166 
1167 	spin_lock(&node->state_lock);
1168 	if (!node->users) {
1169 		pr_debug("receive add member from node %d with state %s\n",
1170 			 node->nodeid, dlm_state_str(node->state));
1171 		switch (node->state) {
1172 		case DLM_ESTABLISHED:
1173 			break;
1174 		case DLM_CLOSED:
1175 			node->state = DLM_ESTABLISHED;
1176 			pr_debug("switch node %d to state %s\n",
1177 				 node->nodeid, dlm_state_str(node->state));
1178 			break;
1179 		default:
1180 			/* some invalid state passive shutdown
1181 			 * was failed, we try to reset and
1182 			 * hope it will go on.
1183 			 */
1184 			log_print("reset node %d because shutdown stuck",
1185 				  node->nodeid);
1186 
1187 			midcomms_node_reset(node);
1188 			node->state = DLM_ESTABLISHED;
1189 			break;
1190 		}
1191 	}
1192 
1193 	node->users++;
1194 	pr_debug("users inc count %d\n", node->users);
1195 	spin_unlock(&node->state_lock);
1196 
1197 	srcu_read_unlock(&nodes_srcu, idx);
1198 }
1199 
1200 void dlm_midcomms_remove_member(int nodeid)
1201 {
1202 	struct midcomms_node *node;
1203 	int idx;
1204 
1205 	if (nodeid == dlm_our_nodeid())
1206 		return;
1207 
1208 	idx = srcu_read_lock(&nodes_srcu);
1209 	node = nodeid2node(nodeid, 0);
1210 	if (!node) {
1211 		srcu_read_unlock(&nodes_srcu, idx);
1212 		return;
1213 	}
1214 
1215 	spin_lock(&node->state_lock);
1216 	node->users--;
1217 	pr_debug("users dec count %d\n", node->users);
1218 
1219 	/* hitting users count to zero means the
1220 	 * other side is running dlm_midcomms_stop()
1221 	 * we meet us to have a clean disconnect.
1222 	 */
1223 	if (node->users == 0) {
1224 		pr_debug("receive remove member from node %d with state %s\n",
1225 			 node->nodeid, dlm_state_str(node->state));
1226 		switch (node->state) {
1227 		case DLM_ESTABLISHED:
1228 			break;
1229 		case DLM_CLOSE_WAIT:
1230 			/* passive shutdown DLM_LAST_ACK case 2 */
1231 			node->state = DLM_LAST_ACK;
1232 			spin_unlock(&node->state_lock);
1233 
1234 			pr_debug("switch node %d to state %s case 2\n",
1235 				 node->nodeid, dlm_state_str(node->state));
1236 			goto send_fin;
1237 		case DLM_LAST_ACK:
1238 			/* probably receive fin caught it, do nothing */
1239 			break;
1240 		case DLM_CLOSED:
1241 			/* already gone, do nothing */
1242 			break;
1243 		default:
1244 			log_print("%s: unexpected state: %d\n",
1245 				  __func__, node->state);
1246 			break;
1247 		}
1248 	}
1249 	spin_unlock(&node->state_lock);
1250 
1251 	srcu_read_unlock(&nodes_srcu, idx);
1252 	return;
1253 
1254 send_fin:
1255 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
1256 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
1257 	srcu_read_unlock(&nodes_srcu, idx);
1258 }
1259 
1260 static void midcomms_node_release(struct rcu_head *rcu)
1261 {
1262 	struct midcomms_node *node = container_of(rcu, struct midcomms_node, rcu);
1263 
1264 	WARN_ON(atomic_read(&node->send_queue_cnt));
1265 	kfree(node);
1266 }
1267 
1268 static void midcomms_shutdown(struct midcomms_node *node)
1269 {
1270 	int ret;
1271 
1272 	/* old protocol, we don't wait for pending operations */
1273 	switch (node->version) {
1274 	case DLM_VERSION_3_2:
1275 		break;
1276 	default:
1277 		return;
1278 	}
1279 
1280 	spin_lock(&node->state_lock);
1281 	pr_debug("receive active shutdown for node %d with state %s\n",
1282 		 node->nodeid, dlm_state_str(node->state));
1283 	switch (node->state) {
1284 	case DLM_ESTABLISHED:
1285 		node->state = DLM_FIN_WAIT1;
1286 		pr_debug("switch node %d to state %s case 2\n",
1287 			 node->nodeid, dlm_state_str(node->state));
1288 		break;
1289 	case DLM_CLOSED:
1290 		/* we have what we want */
1291 		spin_unlock(&node->state_lock);
1292 		return;
1293 	default:
1294 		/* busy to enter DLM_FIN_WAIT1, wait until passive
1295 		 * done in shutdown_wait to enter DLM_CLOSED.
1296 		 */
1297 		break;
1298 	}
1299 	spin_unlock(&node->state_lock);
1300 
1301 	if (node->state == DLM_FIN_WAIT1) {
1302 		dlm_send_fin(node, dlm_act_fin_ack_rcv);
1303 
1304 		if (DLM_DEBUG_FENCE_TERMINATION)
1305 			msleep(5000);
1306 	}
1307 
1308 	/* wait for other side dlm + fin */
1309 	ret = wait_event_timeout(node->shutdown_wait,
1310 				 node->state == DLM_CLOSED ||
1311 				 test_bit(DLM_NODE_FLAG_CLOSE, &node->flags),
1312 				 DLM_SHUTDOWN_TIMEOUT);
1313 	if (!ret || test_bit(DLM_NODE_FLAG_CLOSE, &node->flags)) {
1314 		pr_debug("active shutdown timed out for node %d with state %s\n",
1315 			 node->nodeid, dlm_state_str(node->state));
1316 		midcomms_node_reset(node);
1317 		return;
1318 	}
1319 
1320 	pr_debug("active shutdown done for node %d with state %s\n",
1321 		 node->nodeid, dlm_state_str(node->state));
1322 }
1323 
1324 void dlm_midcomms_shutdown(void)
1325 {
1326 	struct midcomms_node *node;
1327 	int i, idx;
1328 
1329 	mutex_lock(&close_lock);
1330 	idx = srcu_read_lock(&nodes_srcu);
1331 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1332 		hlist_for_each_entry_rcu(node, &node_hash[i], hlist) {
1333 			midcomms_shutdown(node);
1334 
1335 			dlm_delete_debug_comms_file(node->debugfs);
1336 
1337 			spin_lock(&nodes_lock);
1338 			hlist_del_rcu(&node->hlist);
1339 			spin_unlock(&nodes_lock);
1340 
1341 			call_srcu(&nodes_srcu, &node->rcu, midcomms_node_release);
1342 		}
1343 	}
1344 	srcu_read_unlock(&nodes_srcu, idx);
1345 	mutex_unlock(&close_lock);
1346 
1347 	dlm_lowcomms_shutdown();
1348 }
1349 
1350 int dlm_midcomms_close(int nodeid)
1351 {
1352 	struct midcomms_node *node;
1353 	int idx, ret;
1354 
1355 	if (nodeid == dlm_our_nodeid())
1356 		return 0;
1357 
1358 	idx = srcu_read_lock(&nodes_srcu);
1359 	/* Abort pending close/remove operation */
1360 	node = nodeid2node(nodeid, 0);
1361 	if (node) {
1362 		/* let shutdown waiters leave */
1363 		set_bit(DLM_NODE_FLAG_CLOSE, &node->flags);
1364 		wake_up(&node->shutdown_wait);
1365 	}
1366 	srcu_read_unlock(&nodes_srcu, idx);
1367 
1368 	synchronize_srcu(&nodes_srcu);
1369 
1370 	idx = srcu_read_lock(&nodes_srcu);
1371 	mutex_lock(&close_lock);
1372 	node = nodeid2node(nodeid, 0);
1373 	if (!node) {
1374 		mutex_unlock(&close_lock);
1375 		srcu_read_unlock(&nodes_srcu, idx);
1376 		return dlm_lowcomms_close(nodeid);
1377 	}
1378 
1379 	ret = dlm_lowcomms_close(nodeid);
1380 	spin_lock(&node->state_lock);
1381 	midcomms_node_reset(node);
1382 	spin_unlock(&node->state_lock);
1383 	srcu_read_unlock(&nodes_srcu, idx);
1384 	mutex_unlock(&close_lock);
1385 
1386 	return ret;
1387 }
1388