xref: /openbmc/linux/fs/dlm/midcomms.c (revision 2427f03f)
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  * Tail Size checking:
113  *
114  * There exists a message tail payload in e.g. DLM_MSG however we don't
115  * check it against the message length yet regarding to the receive buffer
116  * length. That need to be validated.
117  *
118  * Fencing bad nodes:
119  *
120  * At timeout places or weird sequence number behaviours we should send
121  * a fencing request to the cluster manager.
122  */
123 
124 /* Debug switch to enable a 5 seconds sleep waiting of a termination.
125  * This can be useful to test fencing while termination is running.
126  * This requires a setup with only gfs2 as dlm user, so that the
127  * last umount will terminate the connection.
128  *
129  * However it became useful to test, while the 5 seconds block in umount
130  * just press the reset button. In a lot of dropping the termination
131  * process can could take several seconds.
132  */
133 #define DLM_DEBUG_FENCE_TERMINATION	0
134 
135 #include <net/tcp.h>
136 
137 #include "dlm_internal.h"
138 #include "lowcomms.h"
139 #include "config.h"
140 #include "memory.h"
141 #include "lock.h"
142 #include "util.h"
143 #include "midcomms.h"
144 
145 /* init value for sequence numbers for testing purpose only e.g. overflows */
146 #define DLM_SEQ_INIT		0
147 /* 3 minutes wait to sync ending of dlm */
148 #define DLM_SHUTDOWN_TIMEOUT	msecs_to_jiffies(3 * 60 * 1000)
149 #define DLM_VERSION_NOT_SET	0
150 
151 struct midcomms_node {
152 	int nodeid;
153 	uint32_t version;
154 	uint32_t seq_send;
155 	uint32_t seq_next;
156 	/* These queues are unbound because we cannot drop any message in dlm.
157 	 * We could send a fence signal for a specific node to the cluster
158 	 * manager if queues hits some maximum value, however this handling
159 	 * not supported yet.
160 	 */
161 	struct list_head send_queue;
162 	spinlock_t send_queue_lock;
163 	atomic_t send_queue_cnt;
164 #define DLM_NODE_FLAG_CLOSE	1
165 #define DLM_NODE_FLAG_STOP_TX	2
166 #define DLM_NODE_FLAG_STOP_RX	3
167 #define DLM_NODE_ULP_DELIVERED	4
168 	unsigned long flags;
169 	wait_queue_head_t shutdown_wait;
170 
171 	/* dlm tcp termination state */
172 #define DLM_CLOSED	1
173 #define DLM_ESTABLISHED	2
174 #define DLM_FIN_WAIT1	3
175 #define DLM_FIN_WAIT2	4
176 #define DLM_CLOSE_WAIT	5
177 #define DLM_LAST_ACK	6
178 #define DLM_CLOSING	7
179 	int state;
180 	spinlock_t state_lock;
181 
182 	/* counts how many lockspaces are using this node
183 	 * this refcount is necessary to determine if the
184 	 * node wants to disconnect.
185 	 */
186 	int users;
187 
188 	/* not protected by srcu, node_hash lifetime */
189 	void *debugfs;
190 
191 	struct hlist_node hlist;
192 	struct rcu_head rcu;
193 };
194 
195 struct dlm_mhandle {
196 	const struct dlm_header *inner_hd;
197 	struct midcomms_node *node;
198 	struct dlm_opts *opts;
199 	struct dlm_msg *msg;
200 	bool committed;
201 	uint32_t seq;
202 
203 	void (*ack_rcv)(struct midcomms_node *node);
204 
205 	/* get_mhandle/commit srcu idx exchange */
206 	int idx;
207 
208 	struct list_head list;
209 	struct rcu_head rcu;
210 };
211 
212 static struct hlist_head node_hash[CONN_HASH_SIZE];
213 static DEFINE_SPINLOCK(nodes_lock);
214 DEFINE_STATIC_SRCU(nodes_srcu);
215 
216 /* This mutex prevents that midcomms_close() is running while
217  * stop() or remove(). As I experienced invalid memory access
218  * behaviours when DLM_DEBUG_FENCE_TERMINATION is enabled and
219  * resetting machines. I will end in some double deletion in nodes
220  * datastructure.
221  */
222 static DEFINE_MUTEX(close_lock);
223 
224 struct kmem_cache *dlm_midcomms_cache_create(void)
225 {
226 	return kmem_cache_create("dlm_mhandle", sizeof(struct dlm_mhandle),
227 				 0, 0, NULL);
228 }
229 
230 static inline const char *dlm_state_str(int state)
231 {
232 	switch (state) {
233 	case DLM_CLOSED:
234 		return "CLOSED";
235 	case DLM_ESTABLISHED:
236 		return "ESTABLISHED";
237 	case DLM_FIN_WAIT1:
238 		return "FIN_WAIT1";
239 	case DLM_FIN_WAIT2:
240 		return "FIN_WAIT2";
241 	case DLM_CLOSE_WAIT:
242 		return "CLOSE_WAIT";
243 	case DLM_LAST_ACK:
244 		return "LAST_ACK";
245 	case DLM_CLOSING:
246 		return "CLOSING";
247 	default:
248 		return "UNKNOWN";
249 	}
250 }
251 
252 const char *dlm_midcomms_state(struct midcomms_node *node)
253 {
254 	return dlm_state_str(node->state);
255 }
256 
257 unsigned long dlm_midcomms_flags(struct midcomms_node *node)
258 {
259 	return node->flags;
260 }
261 
262 int dlm_midcomms_send_queue_cnt(struct midcomms_node *node)
263 {
264 	return atomic_read(&node->send_queue_cnt);
265 }
266 
267 uint32_t dlm_midcomms_version(struct midcomms_node *node)
268 {
269 	return node->version;
270 }
271 
272 static struct midcomms_node *__find_node(int nodeid, int r)
273 {
274 	struct midcomms_node *node;
275 
276 	hlist_for_each_entry_rcu(node, &node_hash[r], hlist) {
277 		if (node->nodeid == nodeid)
278 			return node;
279 	}
280 
281 	return NULL;
282 }
283 
284 static void dlm_mhandle_release(struct rcu_head *rcu)
285 {
286 	struct dlm_mhandle *mh = container_of(rcu, struct dlm_mhandle, rcu);
287 
288 	dlm_lowcomms_put_msg(mh->msg);
289 	dlm_free_mhandle(mh);
290 }
291 
292 static void dlm_mhandle_delete(struct midcomms_node *node,
293 			       struct dlm_mhandle *mh)
294 {
295 	list_del_rcu(&mh->list);
296 	atomic_dec(&node->send_queue_cnt);
297 	call_rcu(&mh->rcu, dlm_mhandle_release);
298 }
299 
300 static void dlm_send_queue_flush(struct midcomms_node *node)
301 {
302 	struct dlm_mhandle *mh;
303 
304 	pr_debug("flush midcomms send queue of node %d\n", node->nodeid);
305 
306 	rcu_read_lock();
307 	spin_lock(&node->send_queue_lock);
308 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
309 		dlm_mhandle_delete(node, mh);
310 	}
311 	spin_unlock(&node->send_queue_lock);
312 	rcu_read_unlock();
313 }
314 
315 static void midcomms_node_reset(struct midcomms_node *node)
316 {
317 	pr_debug("reset node %d\n", node->nodeid);
318 
319 	node->seq_next = DLM_SEQ_INIT;
320 	node->seq_send = DLM_SEQ_INIT;
321 	node->version = DLM_VERSION_NOT_SET;
322 	node->flags = 0;
323 
324 	dlm_send_queue_flush(node);
325 	node->state = DLM_CLOSED;
326 	wake_up(&node->shutdown_wait);
327 }
328 
329 static struct midcomms_node *nodeid2node(int nodeid, gfp_t alloc)
330 {
331 	struct midcomms_node *node, *tmp;
332 	int r = nodeid_hash(nodeid);
333 
334 	node = __find_node(nodeid, r);
335 	if (node || !alloc)
336 		return node;
337 
338 	node = kmalloc(sizeof(*node), alloc);
339 	if (!node)
340 		return NULL;
341 
342 	node->nodeid = nodeid;
343 	spin_lock_init(&node->state_lock);
344 	spin_lock_init(&node->send_queue_lock);
345 	atomic_set(&node->send_queue_cnt, 0);
346 	INIT_LIST_HEAD(&node->send_queue);
347 	init_waitqueue_head(&node->shutdown_wait);
348 	node->users = 0;
349 	midcomms_node_reset(node);
350 
351 	spin_lock(&nodes_lock);
352 	/* check again if there was somebody else
353 	 * earlier here to add the node
354 	 */
355 	tmp = __find_node(nodeid, r);
356 	if (tmp) {
357 		spin_unlock(&nodes_lock);
358 		kfree(node);
359 		return tmp;
360 	}
361 
362 	hlist_add_head_rcu(&node->hlist, &node_hash[r]);
363 	spin_unlock(&nodes_lock);
364 
365 	node->debugfs = dlm_create_debug_comms_file(nodeid, node);
366 	return node;
367 }
368 
369 static int dlm_send_ack(int nodeid, uint32_t seq)
370 {
371 	int mb_len = sizeof(struct dlm_header);
372 	struct dlm_header *m_header;
373 	struct dlm_msg *msg;
374 	char *ppc;
375 
376 	msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_NOFS, &ppc,
377 				   NULL, NULL);
378 	if (!msg)
379 		return -ENOMEM;
380 
381 	m_header = (struct dlm_header *)ppc;
382 
383 	m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
384 	m_header->h_nodeid = dlm_our_nodeid();
385 	m_header->h_length = mb_len;
386 	m_header->h_cmd = DLM_ACK;
387 	m_header->u.h_seq = seq;
388 
389 	header_out(m_header);
390 	dlm_lowcomms_commit_msg(msg);
391 	dlm_lowcomms_put_msg(msg);
392 
393 	return 0;
394 }
395 
396 static int dlm_send_fin(struct midcomms_node *node,
397 			void (*ack_rcv)(struct midcomms_node *node))
398 {
399 	int mb_len = sizeof(struct dlm_header);
400 	struct dlm_header *m_header;
401 	struct dlm_mhandle *mh;
402 	char *ppc;
403 
404 	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_NOFS, &ppc);
405 	if (!mh)
406 		return -ENOMEM;
407 
408 	mh->ack_rcv = ack_rcv;
409 
410 	m_header = (struct dlm_header *)ppc;
411 
412 	m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
413 	m_header->h_nodeid = dlm_our_nodeid();
414 	m_header->h_length = mb_len;
415 	m_header->h_cmd = DLM_FIN;
416 
417 	header_out(m_header);
418 
419 	pr_debug("sending fin msg to node %d\n", node->nodeid);
420 	dlm_midcomms_commit_mhandle(mh);
421 	set_bit(DLM_NODE_FLAG_STOP_TX, &node->flags);
422 
423 	return 0;
424 }
425 
426 static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
427 {
428 	struct dlm_mhandle *mh;
429 
430 	rcu_read_lock();
431 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
432 		if (before(mh->seq, seq)) {
433 			if (mh->ack_rcv)
434 				mh->ack_rcv(node);
435 		} else {
436 			/* send queue should be ordered */
437 			break;
438 		}
439 	}
440 
441 	spin_lock(&node->send_queue_lock);
442 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
443 		if (before(mh->seq, seq)) {
444 			dlm_mhandle_delete(node, mh);
445 		} else {
446 			/* send queue should be ordered */
447 			break;
448 		}
449 	}
450 	spin_unlock(&node->send_queue_lock);
451 	rcu_read_unlock();
452 }
453 
454 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
455 {
456 	spin_lock(&node->state_lock);
457 	pr_debug("receive passive fin ack from node %d with state %s\n",
458 		 node->nodeid, dlm_state_str(node->state));
459 
460 	switch (node->state) {
461 	case DLM_LAST_ACK:
462 		/* DLM_CLOSED */
463 		midcomms_node_reset(node);
464 		break;
465 	case DLM_CLOSED:
466 		/* not valid but somehow we got what we want */
467 		wake_up(&node->shutdown_wait);
468 		break;
469 	default:
470 		spin_unlock(&node->state_lock);
471 		log_print("%s: unexpected state: %d\n",
472 			  __func__, node->state);
473 		WARN_ON(1);
474 		return;
475 	}
476 	spin_unlock(&node->state_lock);
477 }
478 
479 static void dlm_midcomms_receive_buffer(union dlm_packet *p,
480 					struct midcomms_node *node,
481 					uint32_t seq)
482 {
483 	if (seq == node->seq_next) {
484 		node->seq_next++;
485 
486 		switch (p->header.h_cmd) {
487 		case DLM_FIN:
488 			/* send ack before fin */
489 			dlm_send_ack(node->nodeid, node->seq_next);
490 
491 			spin_lock(&node->state_lock);
492 			pr_debug("receive fin msg from node %d with state %s\n",
493 				 node->nodeid, dlm_state_str(node->state));
494 
495 			switch (node->state) {
496 			case DLM_ESTABLISHED:
497 				node->state = DLM_CLOSE_WAIT;
498 				pr_debug("switch node %d to state %s\n",
499 					 node->nodeid, dlm_state_str(node->state));
500 				/* passive shutdown DLM_LAST_ACK case 1
501 				 * additional we check if the node is used by
502 				 * cluster manager events at all.
503 				 */
504 				if (node->users == 0) {
505 					node->state = DLM_LAST_ACK;
506 					pr_debug("switch node %d to state %s case 1\n",
507 						 node->nodeid, dlm_state_str(node->state));
508 					spin_unlock(&node->state_lock);
509 					goto send_fin;
510 				}
511 				break;
512 			case DLM_FIN_WAIT1:
513 				node->state = DLM_CLOSING;
514 				pr_debug("switch node %d to state %s\n",
515 					 node->nodeid, dlm_state_str(node->state));
516 				break;
517 			case DLM_FIN_WAIT2:
518 				midcomms_node_reset(node);
519 				pr_debug("switch node %d to state %s\n",
520 					 node->nodeid, dlm_state_str(node->state));
521 				wake_up(&node->shutdown_wait);
522 				break;
523 			case DLM_LAST_ACK:
524 				/* probably remove_member caught it, do nothing */
525 				break;
526 			default:
527 				spin_unlock(&node->state_lock);
528 				log_print("%s: unexpected state: %d\n",
529 					  __func__, node->state);
530 				WARN_ON(1);
531 				return;
532 			}
533 			spin_unlock(&node->state_lock);
534 
535 			set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
536 			break;
537 		default:
538 			WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
539 			dlm_receive_buffer(p, node->nodeid);
540 			set_bit(DLM_NODE_ULP_DELIVERED, &node->flags);
541 			break;
542 		}
543 	} else {
544 		/* retry to ack message which we already have by sending back
545 		 * current node->seq_next number as ack.
546 		 */
547 		if (seq < node->seq_next)
548 			dlm_send_ack(node->nodeid, node->seq_next);
549 
550 		log_print_ratelimited("ignore dlm msg because seq mismatch, seq: %u, expected: %u, nodeid: %d",
551 				      seq, node->seq_next, node->nodeid);
552 	}
553 
554 	return;
555 
556 send_fin:
557 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
558 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
559 }
560 
561 static struct midcomms_node *
562 dlm_midcomms_recv_node_lookup(int nodeid, const union dlm_packet *p,
563 			      uint16_t msglen, int (*cb)(struct midcomms_node *node))
564 {
565 	struct midcomms_node *node = NULL;
566 	gfp_t allocation = 0;
567 	int ret;
568 
569 	switch (p->header.h_cmd) {
570 	case DLM_RCOM:
571 		if (msglen < sizeof(struct dlm_rcom)) {
572 			log_print("rcom msg too small: %u, will skip this message from node %d",
573 				  msglen, nodeid);
574 			return NULL;
575 		}
576 
577 		switch (le32_to_cpu(p->rcom.rc_type)) {
578 		case DLM_RCOM_NAMES:
579 			fallthrough;
580 		case DLM_RCOM_NAMES_REPLY:
581 			fallthrough;
582 		case DLM_RCOM_STATUS:
583 			fallthrough;
584 		case DLM_RCOM_STATUS_REPLY:
585 			node = nodeid2node(nodeid, 0);
586 			if (node) {
587 				spin_lock(&node->state_lock);
588 				if (node->state != DLM_ESTABLISHED)
589 					pr_debug("receive begin RCOM msg from node %d with state %s\n",
590 						 node->nodeid, dlm_state_str(node->state));
591 
592 				switch (node->state) {
593 				case DLM_CLOSED:
594 					node->state = DLM_ESTABLISHED;
595 					pr_debug("switch node %d to state %s\n",
596 						 node->nodeid, dlm_state_str(node->state));
597 					break;
598 				case DLM_ESTABLISHED:
599 					break;
600 				default:
601 					/* some invalid state passive shutdown
602 					 * was failed, we try to reset and
603 					 * hope it will go on.
604 					 */
605 					log_print("reset node %d because shutdown stuck",
606 						  node->nodeid);
607 
608 					midcomms_node_reset(node);
609 					node->state = DLM_ESTABLISHED;
610 					break;
611 				}
612 				spin_unlock(&node->state_lock);
613 			}
614 
615 			allocation = GFP_NOFS;
616 			break;
617 		default:
618 			break;
619 		}
620 
621 		break;
622 	default:
623 		break;
624 	}
625 
626 	node = nodeid2node(nodeid, allocation);
627 	if (!node) {
628 		switch (p->header.h_cmd) {
629 		case DLM_OPTS:
630 			if (msglen < sizeof(struct dlm_opts)) {
631 				log_print("opts msg too small: %u, will skip this message from node %d",
632 					  msglen, nodeid);
633 				return NULL;
634 			}
635 
636 			log_print_ratelimited("received dlm opts message nextcmd %d from node %d in an invalid sequence",
637 					      p->opts.o_nextcmd, nodeid);
638 			break;
639 		default:
640 			log_print_ratelimited("received dlm message cmd %d from node %d in an invalid sequence",
641 					      p->header.h_cmd, nodeid);
642 			break;
643 		}
644 
645 		return NULL;
646 	}
647 
648 	ret = cb(node);
649 	if (ret < 0)
650 		return NULL;
651 
652 	return node;
653 }
654 
655 static int dlm_midcomms_version_check_3_2(struct midcomms_node *node)
656 {
657 	switch (node->version) {
658 	case DLM_VERSION_NOT_SET:
659 		node->version = DLM_VERSION_3_2;
660 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_2,
661 			  node->nodeid);
662 		break;
663 	case DLM_VERSION_3_2:
664 		break;
665 	default:
666 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
667 				      DLM_VERSION_3_2, node->nodeid, node->version);
668 		return -1;
669 	}
670 
671 	return 0;
672 }
673 
674 static int dlm_opts_check_msglen(union dlm_packet *p, uint16_t msglen, int nodeid)
675 {
676 	int len = msglen;
677 
678 	/* we only trust outer header msglen because
679 	 * it's checked against receive buffer length.
680 	 */
681 	if (len < sizeof(struct dlm_opts))
682 		return -1;
683 	len -= sizeof(struct dlm_opts);
684 
685 	if (len < le16_to_cpu(p->opts.o_optlen))
686 		return -1;
687 	len -= le16_to_cpu(p->opts.o_optlen);
688 
689 	switch (p->opts.o_nextcmd) {
690 	case DLM_FIN:
691 		if (len < sizeof(struct dlm_header)) {
692 			log_print("fin too small: %d, will skip this message from node %d",
693 				  len, nodeid);
694 			return -1;
695 		}
696 
697 		break;
698 	case DLM_MSG:
699 		if (len < sizeof(struct dlm_message)) {
700 			log_print("msg too small: %d, will skip this message from node %d",
701 				  msglen, nodeid);
702 			return -1;
703 		}
704 
705 		break;
706 	case DLM_RCOM:
707 		if (len < sizeof(struct dlm_rcom)) {
708 			log_print("rcom msg too small: %d, will skip this message from node %d",
709 				  len, nodeid);
710 			return -1;
711 		}
712 
713 		break;
714 	default:
715 		log_print("unsupported o_nextcmd received: %u, will skip this message from node %d",
716 			  p->opts.o_nextcmd, nodeid);
717 		return -1;
718 	}
719 
720 	return 0;
721 }
722 
723 static void dlm_midcomms_receive_buffer_3_2(union dlm_packet *p, int nodeid)
724 {
725 	uint16_t msglen = le16_to_cpu(p->header.h_length);
726 	struct midcomms_node *node;
727 	uint32_t seq;
728 	int ret, idx;
729 
730 	idx = srcu_read_lock(&nodes_srcu);
731 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
732 					     dlm_midcomms_version_check_3_2);
733 	if (!node)
734 		goto out;
735 
736 	switch (p->header.h_cmd) {
737 	case DLM_RCOM:
738 		/* these rcom message we use to determine version.
739 		 * they have their own retransmission handling and
740 		 * are the first messages of dlm.
741 		 *
742 		 * length already checked.
743 		 */
744 		switch (le32_to_cpu(p->rcom.rc_type)) {
745 		case DLM_RCOM_NAMES:
746 			fallthrough;
747 		case DLM_RCOM_NAMES_REPLY:
748 			fallthrough;
749 		case DLM_RCOM_STATUS:
750 			fallthrough;
751 		case DLM_RCOM_STATUS_REPLY:
752 			break;
753 		default:
754 			log_print("unsupported rcom type received: %u, will skip this message from node %d",
755 				  le32_to_cpu(p->rcom.rc_type), nodeid);
756 			goto out;
757 		}
758 
759 		WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
760 		dlm_receive_buffer(p, nodeid);
761 		break;
762 	case DLM_OPTS:
763 		seq = le32_to_cpu(p->header.u.h_seq);
764 
765 		ret = dlm_opts_check_msglen(p, msglen, nodeid);
766 		if (ret < 0) {
767 			log_print("opts msg too small: %u, will skip this message from node %d",
768 				  msglen, nodeid);
769 			goto out;
770 		}
771 
772 		p = (union dlm_packet *)((unsigned char *)p->opts.o_opts +
773 					 le16_to_cpu(p->opts.o_optlen));
774 
775 		/* recheck inner msglen just if it's not garbage */
776 		msglen = le16_to_cpu(p->header.h_length);
777 		switch (p->header.h_cmd) {
778 		case DLM_RCOM:
779 			if (msglen < sizeof(struct dlm_rcom)) {
780 				log_print("inner rcom msg too small: %u, will skip this message from node %d",
781 					  msglen, nodeid);
782 				goto out;
783 			}
784 
785 			break;
786 		case DLM_MSG:
787 			if (msglen < sizeof(struct dlm_message)) {
788 				log_print("inner msg too small: %u, will skip this message from node %d",
789 					  msglen, nodeid);
790 				goto out;
791 			}
792 
793 			break;
794 		case DLM_FIN:
795 			if (msglen < sizeof(struct dlm_header)) {
796 				log_print("inner fin too small: %u, will skip this message from node %d",
797 					  msglen, nodeid);
798 				goto out;
799 			}
800 
801 			break;
802 		default:
803 			log_print("unsupported inner h_cmd received: %u, will skip this message from node %d",
804 				  msglen, nodeid);
805 			goto out;
806 		}
807 
808 		dlm_midcomms_receive_buffer(p, node, seq);
809 		break;
810 	case DLM_ACK:
811 		seq = le32_to_cpu(p->header.u.h_seq);
812 		dlm_receive_ack(node, seq);
813 		break;
814 	default:
815 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
816 			  p->header.h_cmd, nodeid);
817 		break;
818 	}
819 
820 out:
821 	srcu_read_unlock(&nodes_srcu, idx);
822 }
823 
824 static int dlm_midcomms_version_check_3_1(struct midcomms_node *node)
825 {
826 	switch (node->version) {
827 	case DLM_VERSION_NOT_SET:
828 		node->version = DLM_VERSION_3_1;
829 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_1,
830 			  node->nodeid);
831 		break;
832 	case DLM_VERSION_3_1:
833 		break;
834 	default:
835 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
836 				      DLM_VERSION_3_1, node->nodeid, node->version);
837 		return -1;
838 	}
839 
840 	return 0;
841 }
842 
843 static void dlm_midcomms_receive_buffer_3_1(union dlm_packet *p, int nodeid)
844 {
845 	uint16_t msglen = le16_to_cpu(p->header.h_length);
846 	struct midcomms_node *node;
847 	int idx;
848 
849 	idx = srcu_read_lock(&nodes_srcu);
850 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
851 					     dlm_midcomms_version_check_3_1);
852 	if (!node) {
853 		srcu_read_unlock(&nodes_srcu, idx);
854 		return;
855 	}
856 	srcu_read_unlock(&nodes_srcu, idx);
857 
858 	switch (p->header.h_cmd) {
859 	case DLM_RCOM:
860 		/* length already checked */
861 		break;
862 	case DLM_MSG:
863 		if (msglen < sizeof(struct dlm_message)) {
864 			log_print("msg too small: %u, will skip this message from node %d",
865 				  msglen, nodeid);
866 			return;
867 		}
868 
869 		break;
870 	default:
871 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
872 			  p->header.h_cmd, nodeid);
873 		return;
874 	}
875 
876 	dlm_receive_buffer(p, nodeid);
877 }
878 
879 /*
880  * Called from the low-level comms layer to process a buffer of
881  * commands.
882  */
883 
884 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int len)
885 {
886 	const unsigned char *ptr = buf;
887 	const struct dlm_header *hd;
888 	uint16_t msglen;
889 	int ret = 0;
890 
891 	while (len >= sizeof(struct dlm_header)) {
892 		hd = (struct dlm_header *)ptr;
893 
894 		/* no message should be more than DLM_MAX_SOCKET_BUFSIZE or
895 		 * less than dlm_header size.
896 		 *
897 		 * Some messages does not have a 8 byte length boundary yet
898 		 * which can occur in a unaligned memory access of some dlm
899 		 * messages. However this problem need to be fixed at the
900 		 * sending side, for now it seems nobody run into architecture
901 		 * related issues yet but it slows down some processing.
902 		 * Fixing this issue should be scheduled in future by doing
903 		 * the next major version bump.
904 		 */
905 		msglen = le16_to_cpu(hd->h_length);
906 		if (msglen > DLM_MAX_SOCKET_BUFSIZE ||
907 		    msglen < sizeof(struct dlm_header)) {
908 			log_print("received invalid length header: %u from node %d, will abort message parsing",
909 				  msglen, nodeid);
910 			return -EBADMSG;
911 		}
912 
913 		/* caller will take care that leftover
914 		 * will be parsed next call with more data
915 		 */
916 		if (msglen > len)
917 			break;
918 
919 		switch (hd->h_version) {
920 		case cpu_to_le32(DLM_VERSION_3_1):
921 			dlm_midcomms_receive_buffer_3_1((union dlm_packet *)ptr, nodeid);
922 			break;
923 		case cpu_to_le32(DLM_VERSION_3_2):
924 			dlm_midcomms_receive_buffer_3_2((union dlm_packet *)ptr, nodeid);
925 			break;
926 		default:
927 			log_print("received invalid version header: %u from node %d, will skip this message",
928 				  le32_to_cpu(hd->h_version), nodeid);
929 			break;
930 		}
931 
932 		ret += msglen;
933 		len -= msglen;
934 		ptr += msglen;
935 	}
936 
937 	return ret;
938 }
939 
940 void dlm_midcomms_receive_done(int nodeid)
941 {
942 	struct midcomms_node *node;
943 	int idx;
944 
945 	idx = srcu_read_lock(&nodes_srcu);
946 	node = nodeid2node(nodeid, 0);
947 	if (!node) {
948 		srcu_read_unlock(&nodes_srcu, idx);
949 		return;
950 	}
951 
952 	/* old protocol, we do nothing */
953 	switch (node->version) {
954 	case DLM_VERSION_3_2:
955 		break;
956 	default:
957 		srcu_read_unlock(&nodes_srcu, idx);
958 		return;
959 	}
960 
961 	/* do nothing if we didn't delivered stateful to ulp */
962 	if (!test_and_clear_bit(DLM_NODE_ULP_DELIVERED,
963 				&node->flags)) {
964 		srcu_read_unlock(&nodes_srcu, idx);
965 		return;
966 	}
967 
968 	spin_lock(&node->state_lock);
969 	/* we only ack if state is ESTABLISHED */
970 	switch (node->state) {
971 	case DLM_ESTABLISHED:
972 		spin_unlock(&node->state_lock);
973 		dlm_send_ack(node->nodeid, node->seq_next);
974 		break;
975 	default:
976 		spin_unlock(&node->state_lock);
977 		/* do nothing FIN has it's own ack send */
978 		break;
979 	}
980 	srcu_read_unlock(&nodes_srcu, idx);
981 }
982 
983 void dlm_midcomms_unack_msg_resend(int nodeid)
984 {
985 	struct midcomms_node *node;
986 	struct dlm_mhandle *mh;
987 	int idx, ret;
988 
989 	idx = srcu_read_lock(&nodes_srcu);
990 	node = nodeid2node(nodeid, 0);
991 	if (!node) {
992 		srcu_read_unlock(&nodes_srcu, idx);
993 		return;
994 	}
995 
996 	/* old protocol, we don't support to retransmit on failure */
997 	switch (node->version) {
998 	case DLM_VERSION_3_2:
999 		break;
1000 	default:
1001 		srcu_read_unlock(&nodes_srcu, idx);
1002 		return;
1003 	}
1004 
1005 	rcu_read_lock();
1006 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
1007 		if (!mh->committed)
1008 			continue;
1009 
1010 		ret = dlm_lowcomms_resend_msg(mh->msg);
1011 		if (!ret)
1012 			log_print_ratelimited("retransmit dlm msg, seq %u, nodeid %d",
1013 					      mh->seq, node->nodeid);
1014 	}
1015 	rcu_read_unlock();
1016 	srcu_read_unlock(&nodes_srcu, idx);
1017 }
1018 
1019 static void dlm_fill_opts_header(struct dlm_opts *opts, uint16_t inner_len,
1020 				 uint32_t seq)
1021 {
1022 	opts->o_header.h_cmd = DLM_OPTS;
1023 	opts->o_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
1024 	opts->o_header.h_nodeid = dlm_our_nodeid();
1025 	opts->o_header.h_length = DLM_MIDCOMMS_OPT_LEN + inner_len;
1026 	opts->o_header.u.h_seq = seq;
1027 	header_out(&opts->o_header);
1028 }
1029 
1030 static void midcomms_new_msg_cb(void *data)
1031 {
1032 	struct dlm_mhandle *mh = data;
1033 
1034 	atomic_inc(&mh->node->send_queue_cnt);
1035 
1036 	spin_lock(&mh->node->send_queue_lock);
1037 	list_add_tail_rcu(&mh->list, &mh->node->send_queue);
1038 	spin_unlock(&mh->node->send_queue_lock);
1039 
1040 	mh->seq = mh->node->seq_send++;
1041 }
1042 
1043 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
1044 						int len, gfp_t allocation, char **ppc)
1045 {
1046 	struct dlm_opts *opts;
1047 	struct dlm_msg *msg;
1048 
1049 	msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
1050 				   allocation, ppc, midcomms_new_msg_cb, mh);
1051 	if (!msg)
1052 		return NULL;
1053 
1054 	opts = (struct dlm_opts *)*ppc;
1055 	mh->opts = opts;
1056 
1057 	/* add possible options here */
1058 	dlm_fill_opts_header(opts, len, mh->seq);
1059 
1060 	*ppc += sizeof(*opts);
1061 	mh->inner_hd = (const struct dlm_header *)*ppc;
1062 	return msg;
1063 }
1064 
1065 struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
1066 					     gfp_t allocation, char **ppc)
1067 {
1068 	struct midcomms_node *node;
1069 	struct dlm_mhandle *mh;
1070 	struct dlm_msg *msg;
1071 	int idx;
1072 
1073 	idx = srcu_read_lock(&nodes_srcu);
1074 	node = nodeid2node(nodeid, 0);
1075 	if (!node) {
1076 		WARN_ON_ONCE(1);
1077 		goto err;
1078 	}
1079 
1080 	/* this is a bug, however we going on and hope it will be resolved */
1081 	WARN_ON(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
1082 
1083 	mh = dlm_allocate_mhandle();
1084 	if (!mh)
1085 		goto err;
1086 
1087 	mh->committed = false;
1088 	mh->ack_rcv = NULL;
1089 	mh->idx = idx;
1090 	mh->node = node;
1091 
1092 	switch (node->version) {
1093 	case DLM_VERSION_3_1:
1094 		msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
1095 					   NULL, NULL);
1096 		if (!msg) {
1097 			dlm_free_mhandle(mh);
1098 			goto err;
1099 		}
1100 
1101 		break;
1102 	case DLM_VERSION_3_2:
1103 		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
1104 					       ppc);
1105 		if (!msg) {
1106 			dlm_free_mhandle(mh);
1107 			goto err;
1108 		}
1109 
1110 		break;
1111 	default:
1112 		dlm_free_mhandle(mh);
1113 		WARN_ON(1);
1114 		goto err;
1115 	}
1116 
1117 	mh->msg = msg;
1118 
1119 	/* keep in mind that is a must to call
1120 	 * dlm_midcomms_commit_msg() which releases
1121 	 * nodes_srcu using mh->idx which is assumed
1122 	 * here that the application will call it.
1123 	 */
1124 	return mh;
1125 
1126 err:
1127 	srcu_read_unlock(&nodes_srcu, idx);
1128 	return NULL;
1129 }
1130 
1131 static void dlm_midcomms_commit_msg_3_2(struct dlm_mhandle *mh)
1132 {
1133 	/* nexthdr chain for fast lookup */
1134 	mh->opts->o_nextcmd = mh->inner_hd->h_cmd;
1135 	mh->committed = true;
1136 	dlm_lowcomms_commit_msg(mh->msg);
1137 }
1138 
1139 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh)
1140 {
1141 	switch (mh->node->version) {
1142 	case DLM_VERSION_3_1:
1143 		srcu_read_unlock(&nodes_srcu, mh->idx);
1144 
1145 		dlm_lowcomms_commit_msg(mh->msg);
1146 		dlm_lowcomms_put_msg(mh->msg);
1147 		/* mh is not part of rcu list in this case */
1148 		dlm_free_mhandle(mh);
1149 		break;
1150 	case DLM_VERSION_3_2:
1151 		dlm_midcomms_commit_msg_3_2(mh);
1152 		srcu_read_unlock(&nodes_srcu, mh->idx);
1153 		break;
1154 	default:
1155 		srcu_read_unlock(&nodes_srcu, mh->idx);
1156 		WARN_ON(1);
1157 		break;
1158 	}
1159 }
1160 
1161 int dlm_midcomms_start(void)
1162 {
1163 	int i;
1164 
1165 	for (i = 0; i < CONN_HASH_SIZE; i++)
1166 		INIT_HLIST_HEAD(&node_hash[i]);
1167 
1168 	return dlm_lowcomms_start();
1169 }
1170 
1171 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
1172 {
1173 	spin_lock(&node->state_lock);
1174 	pr_debug("receive active fin ack from node %d with state %s\n",
1175 		 node->nodeid, dlm_state_str(node->state));
1176 
1177 	switch (node->state) {
1178 	case DLM_FIN_WAIT1:
1179 		node->state = DLM_FIN_WAIT2;
1180 		pr_debug("switch node %d to state %s\n",
1181 			 node->nodeid, dlm_state_str(node->state));
1182 		break;
1183 	case DLM_CLOSING:
1184 		midcomms_node_reset(node);
1185 		pr_debug("switch node %d to state %s\n",
1186 			 node->nodeid, dlm_state_str(node->state));
1187 		wake_up(&node->shutdown_wait);
1188 		break;
1189 	case DLM_CLOSED:
1190 		/* not valid but somehow we got what we want */
1191 		wake_up(&node->shutdown_wait);
1192 		break;
1193 	default:
1194 		spin_unlock(&node->state_lock);
1195 		log_print("%s: unexpected state: %d\n",
1196 			  __func__, node->state);
1197 		WARN_ON(1);
1198 		return;
1199 	}
1200 	spin_unlock(&node->state_lock);
1201 }
1202 
1203 void dlm_midcomms_add_member(int nodeid)
1204 {
1205 	struct midcomms_node *node;
1206 	int idx;
1207 
1208 	if (nodeid == dlm_our_nodeid())
1209 		return;
1210 
1211 	idx = srcu_read_lock(&nodes_srcu);
1212 	node = nodeid2node(nodeid, GFP_NOFS);
1213 	if (!node) {
1214 		srcu_read_unlock(&nodes_srcu, idx);
1215 		return;
1216 	}
1217 
1218 	spin_lock(&node->state_lock);
1219 	if (!node->users) {
1220 		pr_debug("receive add member from node %d with state %s\n",
1221 			 node->nodeid, dlm_state_str(node->state));
1222 		switch (node->state) {
1223 		case DLM_ESTABLISHED:
1224 			break;
1225 		case DLM_CLOSED:
1226 			node->state = DLM_ESTABLISHED;
1227 			pr_debug("switch node %d to state %s\n",
1228 				 node->nodeid, dlm_state_str(node->state));
1229 			break;
1230 		default:
1231 			/* some invalid state passive shutdown
1232 			 * was failed, we try to reset and
1233 			 * hope it will go on.
1234 			 */
1235 			log_print("reset node %d because shutdown stuck",
1236 				  node->nodeid);
1237 
1238 			midcomms_node_reset(node);
1239 			node->state = DLM_ESTABLISHED;
1240 			break;
1241 		}
1242 	}
1243 
1244 	node->users++;
1245 	pr_debug("node %d users inc count %d\n", nodeid, node->users);
1246 	spin_unlock(&node->state_lock);
1247 
1248 	srcu_read_unlock(&nodes_srcu, idx);
1249 }
1250 
1251 void dlm_midcomms_remove_member(int nodeid)
1252 {
1253 	struct midcomms_node *node;
1254 	int idx;
1255 
1256 	if (nodeid == dlm_our_nodeid())
1257 		return;
1258 
1259 	idx = srcu_read_lock(&nodes_srcu);
1260 	node = nodeid2node(nodeid, 0);
1261 	if (!node) {
1262 		srcu_read_unlock(&nodes_srcu, idx);
1263 		return;
1264 	}
1265 
1266 	spin_lock(&node->state_lock);
1267 	node->users--;
1268 	pr_debug("node %d users dec count %d\n", nodeid, node->users);
1269 
1270 	/* hitting users count to zero means the
1271 	 * other side is running dlm_midcomms_stop()
1272 	 * we meet us to have a clean disconnect.
1273 	 */
1274 	if (node->users == 0) {
1275 		pr_debug("receive remove member from node %d with state %s\n",
1276 			 node->nodeid, dlm_state_str(node->state));
1277 		switch (node->state) {
1278 		case DLM_ESTABLISHED:
1279 			break;
1280 		case DLM_CLOSE_WAIT:
1281 			/* passive shutdown DLM_LAST_ACK case 2 */
1282 			node->state = DLM_LAST_ACK;
1283 			spin_unlock(&node->state_lock);
1284 
1285 			pr_debug("switch node %d to state %s case 2\n",
1286 				 node->nodeid, dlm_state_str(node->state));
1287 			goto send_fin;
1288 		case DLM_LAST_ACK:
1289 			/* probably receive fin caught it, do nothing */
1290 			break;
1291 		case DLM_CLOSED:
1292 			/* already gone, do nothing */
1293 			break;
1294 		default:
1295 			log_print("%s: unexpected state: %d\n",
1296 				  __func__, node->state);
1297 			break;
1298 		}
1299 	}
1300 	spin_unlock(&node->state_lock);
1301 
1302 	srcu_read_unlock(&nodes_srcu, idx);
1303 	return;
1304 
1305 send_fin:
1306 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
1307 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
1308 	srcu_read_unlock(&nodes_srcu, idx);
1309 }
1310 
1311 static void midcomms_node_release(struct rcu_head *rcu)
1312 {
1313 	struct midcomms_node *node = container_of(rcu, struct midcomms_node, rcu);
1314 
1315 	WARN_ON(atomic_read(&node->send_queue_cnt));
1316 	kfree(node);
1317 }
1318 
1319 static void midcomms_shutdown(struct midcomms_node *node)
1320 {
1321 	int ret;
1322 
1323 	/* old protocol, we don't wait for pending operations */
1324 	switch (node->version) {
1325 	case DLM_VERSION_3_2:
1326 		break;
1327 	default:
1328 		return;
1329 	}
1330 
1331 	spin_lock(&node->state_lock);
1332 	pr_debug("receive active shutdown for node %d with state %s\n",
1333 		 node->nodeid, dlm_state_str(node->state));
1334 	switch (node->state) {
1335 	case DLM_ESTABLISHED:
1336 		node->state = DLM_FIN_WAIT1;
1337 		pr_debug("switch node %d to state %s case 2\n",
1338 			 node->nodeid, dlm_state_str(node->state));
1339 		break;
1340 	case DLM_CLOSED:
1341 		/* we have what we want */
1342 		spin_unlock(&node->state_lock);
1343 		return;
1344 	default:
1345 		/* busy to enter DLM_FIN_WAIT1, wait until passive
1346 		 * done in shutdown_wait to enter DLM_CLOSED.
1347 		 */
1348 		break;
1349 	}
1350 	spin_unlock(&node->state_lock);
1351 
1352 	if (node->state == DLM_FIN_WAIT1) {
1353 		dlm_send_fin(node, dlm_act_fin_ack_rcv);
1354 
1355 		if (DLM_DEBUG_FENCE_TERMINATION)
1356 			msleep(5000);
1357 	}
1358 
1359 	/* wait for other side dlm + fin */
1360 	ret = wait_event_timeout(node->shutdown_wait,
1361 				 node->state == DLM_CLOSED ||
1362 				 test_bit(DLM_NODE_FLAG_CLOSE, &node->flags),
1363 				 DLM_SHUTDOWN_TIMEOUT);
1364 	if (!ret || test_bit(DLM_NODE_FLAG_CLOSE, &node->flags)) {
1365 		pr_debug("active shutdown timed out for node %d with state %s\n",
1366 			 node->nodeid, dlm_state_str(node->state));
1367 		midcomms_node_reset(node);
1368 		return;
1369 	}
1370 
1371 	pr_debug("active shutdown done for node %d with state %s\n",
1372 		 node->nodeid, dlm_state_str(node->state));
1373 }
1374 
1375 void dlm_midcomms_shutdown(void)
1376 {
1377 	struct midcomms_node *node;
1378 	int i, idx;
1379 
1380 	mutex_lock(&close_lock);
1381 	idx = srcu_read_lock(&nodes_srcu);
1382 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1383 		hlist_for_each_entry_rcu(node, &node_hash[i], hlist) {
1384 			midcomms_shutdown(node);
1385 
1386 			dlm_delete_debug_comms_file(node->debugfs);
1387 
1388 			spin_lock(&nodes_lock);
1389 			hlist_del_rcu(&node->hlist);
1390 			spin_unlock(&nodes_lock);
1391 
1392 			call_srcu(&nodes_srcu, &node->rcu, midcomms_node_release);
1393 		}
1394 	}
1395 	srcu_read_unlock(&nodes_srcu, idx);
1396 	mutex_unlock(&close_lock);
1397 
1398 	dlm_lowcomms_shutdown();
1399 }
1400 
1401 int dlm_midcomms_close(int nodeid)
1402 {
1403 	struct midcomms_node *node;
1404 	int idx, ret;
1405 
1406 	if (nodeid == dlm_our_nodeid())
1407 		return 0;
1408 
1409 	idx = srcu_read_lock(&nodes_srcu);
1410 	/* Abort pending close/remove operation */
1411 	node = nodeid2node(nodeid, 0);
1412 	if (node) {
1413 		/* let shutdown waiters leave */
1414 		set_bit(DLM_NODE_FLAG_CLOSE, &node->flags);
1415 		wake_up(&node->shutdown_wait);
1416 	}
1417 	srcu_read_unlock(&nodes_srcu, idx);
1418 
1419 	synchronize_srcu(&nodes_srcu);
1420 
1421 	idx = srcu_read_lock(&nodes_srcu);
1422 	mutex_lock(&close_lock);
1423 	node = nodeid2node(nodeid, 0);
1424 	if (!node) {
1425 		mutex_unlock(&close_lock);
1426 		srcu_read_unlock(&nodes_srcu, idx);
1427 		return dlm_lowcomms_close(nodeid);
1428 	}
1429 
1430 	ret = dlm_lowcomms_close(nodeid);
1431 	spin_lock(&node->state_lock);
1432 	midcomms_node_reset(node);
1433 	spin_unlock(&node->state_lock);
1434 	srcu_read_unlock(&nodes_srcu, idx);
1435 	mutex_unlock(&close_lock);
1436 
1437 	return ret;
1438 }
1439 
1440 /* debug functionality to send raw dlm msg from user space */
1441 struct dlm_rawmsg_data {
1442 	struct midcomms_node *node;
1443 	void *buf;
1444 };
1445 
1446 static void midcomms_new_rawmsg_cb(void *data)
1447 {
1448 	struct dlm_rawmsg_data *rd = data;
1449 	struct dlm_header *h = rd->buf;
1450 
1451 	switch (h->h_version) {
1452 	case cpu_to_le32(DLM_VERSION_3_1):
1453 		break;
1454 	default:
1455 		switch (h->h_cmd) {
1456 		case DLM_OPTS:
1457 			if (!h->u.h_seq)
1458 				h->u.h_seq = rd->node->seq_send++;
1459 			break;
1460 		default:
1461 			break;
1462 		}
1463 		break;
1464 	}
1465 }
1466 
1467 int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
1468 			     int buflen)
1469 {
1470 	struct dlm_rawmsg_data rd;
1471 	struct dlm_msg *msg;
1472 	char *msgbuf;
1473 
1474 	rd.node = node;
1475 	rd.buf = buf;
1476 
1477 	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
1478 				   &msgbuf, midcomms_new_rawmsg_cb, &rd);
1479 	if (!msg)
1480 		return -ENOMEM;
1481 
1482 	memcpy(msgbuf, buf, buflen);
1483 	dlm_lowcomms_commit_msg(msg);
1484 	return 0;
1485 }
1486 
1487