xref: /openbmc/linux/fs/dlm/midcomms.c (revision 3f58ff6b)
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 <trace/events/dlm.h>
136 #include <net/tcp.h>
137 
138 #include "dlm_internal.h"
139 #include "lockspace.h"
140 #include "lowcomms.h"
141 #include "config.h"
142 #include "memory.h"
143 #include "lock.h"
144 #include "util.h"
145 #include "midcomms.h"
146 
147 /* init value for sequence numbers for testing purpose only e.g. overflows */
148 #define DLM_SEQ_INIT		0
149 /* 3 minutes wait to sync ending of dlm */
150 #define DLM_SHUTDOWN_TIMEOUT	msecs_to_jiffies(3 * 60 * 1000)
151 #define DLM_VERSION_NOT_SET	0
152 
153 struct midcomms_node {
154 	int nodeid;
155 	uint32_t version;
156 	uint32_t seq_send;
157 	uint32_t seq_next;
158 	/* These queues are unbound because we cannot drop any message in dlm.
159 	 * We could send a fence signal for a specific node to the cluster
160 	 * manager if queues hits some maximum value, however this handling
161 	 * not supported yet.
162 	 */
163 	struct list_head send_queue;
164 	spinlock_t send_queue_lock;
165 	atomic_t send_queue_cnt;
166 #define DLM_NODE_FLAG_CLOSE	1
167 #define DLM_NODE_FLAG_STOP_TX	2
168 #define DLM_NODE_FLAG_STOP_RX	3
169 #define DLM_NODE_ULP_DELIVERED	4
170 	unsigned long flags;
171 	wait_queue_head_t shutdown_wait;
172 
173 	/* dlm tcp termination state */
174 #define DLM_CLOSED	1
175 #define DLM_ESTABLISHED	2
176 #define DLM_FIN_WAIT1	3
177 #define DLM_FIN_WAIT2	4
178 #define DLM_CLOSE_WAIT	5
179 #define DLM_LAST_ACK	6
180 #define DLM_CLOSING	7
181 	int state;
182 	spinlock_t state_lock;
183 
184 	/* counts how many lockspaces are using this node
185 	 * this refcount is necessary to determine if the
186 	 * node wants to disconnect.
187 	 */
188 	int users;
189 
190 	/* not protected by srcu, node_hash lifetime */
191 	void *debugfs;
192 
193 	struct hlist_node hlist;
194 	struct rcu_head rcu;
195 };
196 
197 struct dlm_mhandle {
198 	const union dlm_packet *inner_p;
199 	struct midcomms_node *node;
200 	struct dlm_opts *opts;
201 	struct dlm_msg *msg;
202 	bool committed;
203 	uint32_t seq;
204 
205 	void (*ack_rcv)(struct midcomms_node *node);
206 
207 	/* get_mhandle/commit srcu idx exchange */
208 	int idx;
209 
210 	struct list_head list;
211 	struct rcu_head rcu;
212 };
213 
214 static struct hlist_head node_hash[CONN_HASH_SIZE];
215 static DEFINE_SPINLOCK(nodes_lock);
216 DEFINE_STATIC_SRCU(nodes_srcu);
217 
218 /* This mutex prevents that midcomms_close() is running while
219  * stop() or remove(). As I experienced invalid memory access
220  * behaviours when DLM_DEBUG_FENCE_TERMINATION is enabled and
221  * resetting machines. I will end in some double deletion in nodes
222  * datastructure.
223  */
224 static DEFINE_MUTEX(close_lock);
225 
226 struct kmem_cache *dlm_midcomms_cache_create(void)
227 {
228 	return kmem_cache_create("dlm_mhandle", sizeof(struct dlm_mhandle),
229 				 0, 0, NULL);
230 }
231 
232 static inline const char *dlm_state_str(int state)
233 {
234 	switch (state) {
235 	case DLM_CLOSED:
236 		return "CLOSED";
237 	case DLM_ESTABLISHED:
238 		return "ESTABLISHED";
239 	case DLM_FIN_WAIT1:
240 		return "FIN_WAIT1";
241 	case DLM_FIN_WAIT2:
242 		return "FIN_WAIT2";
243 	case DLM_CLOSE_WAIT:
244 		return "CLOSE_WAIT";
245 	case DLM_LAST_ACK:
246 		return "LAST_ACK";
247 	case DLM_CLOSING:
248 		return "CLOSING";
249 	default:
250 		return "UNKNOWN";
251 	}
252 }
253 
254 const char *dlm_midcomms_state(struct midcomms_node *node)
255 {
256 	return dlm_state_str(node->state);
257 }
258 
259 unsigned long dlm_midcomms_flags(struct midcomms_node *node)
260 {
261 	return node->flags;
262 }
263 
264 int dlm_midcomms_send_queue_cnt(struct midcomms_node *node)
265 {
266 	return atomic_read(&node->send_queue_cnt);
267 }
268 
269 uint32_t dlm_midcomms_version(struct midcomms_node *node)
270 {
271 	return node->version;
272 }
273 
274 static struct midcomms_node *__find_node(int nodeid, int r)
275 {
276 	struct midcomms_node *node;
277 
278 	hlist_for_each_entry_rcu(node, &node_hash[r], hlist) {
279 		if (node->nodeid == nodeid)
280 			return node;
281 	}
282 
283 	return NULL;
284 }
285 
286 static void dlm_mhandle_release(struct rcu_head *rcu)
287 {
288 	struct dlm_mhandle *mh = container_of(rcu, struct dlm_mhandle, rcu);
289 
290 	dlm_lowcomms_put_msg(mh->msg);
291 	dlm_free_mhandle(mh);
292 }
293 
294 static void dlm_mhandle_delete(struct midcomms_node *node,
295 			       struct dlm_mhandle *mh)
296 {
297 	list_del_rcu(&mh->list);
298 	atomic_dec(&node->send_queue_cnt);
299 	call_rcu(&mh->rcu, dlm_mhandle_release);
300 }
301 
302 static void dlm_send_queue_flush(struct midcomms_node *node)
303 {
304 	struct dlm_mhandle *mh;
305 
306 	pr_debug("flush midcomms send queue of node %d\n", node->nodeid);
307 
308 	rcu_read_lock();
309 	spin_lock_bh(&node->send_queue_lock);
310 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
311 		dlm_mhandle_delete(node, mh);
312 	}
313 	spin_unlock_bh(&node->send_queue_lock);
314 	rcu_read_unlock();
315 }
316 
317 static void midcomms_node_reset(struct midcomms_node *node)
318 {
319 	pr_debug("reset node %d\n", node->nodeid);
320 
321 	node->seq_next = DLM_SEQ_INIT;
322 	node->seq_send = DLM_SEQ_INIT;
323 	node->version = DLM_VERSION_NOT_SET;
324 	node->flags = 0;
325 
326 	dlm_send_queue_flush(node);
327 	node->state = DLM_CLOSED;
328 	wake_up(&node->shutdown_wait);
329 }
330 
331 static struct midcomms_node *nodeid2node(int nodeid, gfp_t alloc)
332 {
333 	struct midcomms_node *node, *tmp;
334 	int r = nodeid_hash(nodeid);
335 
336 	node = __find_node(nodeid, r);
337 	if (node || !alloc)
338 		return node;
339 
340 	node = kmalloc(sizeof(*node), alloc);
341 	if (!node)
342 		return NULL;
343 
344 	node->nodeid = nodeid;
345 	spin_lock_init(&node->state_lock);
346 	spin_lock_init(&node->send_queue_lock);
347 	atomic_set(&node->send_queue_cnt, 0);
348 	INIT_LIST_HEAD(&node->send_queue);
349 	init_waitqueue_head(&node->shutdown_wait);
350 	node->users = 0;
351 	midcomms_node_reset(node);
352 
353 	spin_lock(&nodes_lock);
354 	/* check again if there was somebody else
355 	 * earlier here to add the node
356 	 */
357 	tmp = __find_node(nodeid, r);
358 	if (tmp) {
359 		spin_unlock(&nodes_lock);
360 		kfree(node);
361 		return tmp;
362 	}
363 
364 	hlist_add_head_rcu(&node->hlist, &node_hash[r]);
365 	spin_unlock(&nodes_lock);
366 
367 	node->debugfs = dlm_create_debug_comms_file(nodeid, node);
368 	return node;
369 }
370 
371 static int dlm_send_ack(int nodeid, uint32_t seq)
372 {
373 	int mb_len = sizeof(struct dlm_header);
374 	struct dlm_header *m_header;
375 	struct dlm_msg *msg;
376 	char *ppc;
377 
378 	msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_NOFS, &ppc,
379 				   NULL, NULL);
380 	if (!msg)
381 		return -ENOMEM;
382 
383 	m_header = (struct dlm_header *)ppc;
384 
385 	m_header->h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
386 	m_header->h_nodeid = cpu_to_le32(dlm_our_nodeid());
387 	m_header->h_length = cpu_to_le16(mb_len);
388 	m_header->h_cmd = DLM_ACK;
389 	m_header->u.h_seq = cpu_to_le32(seq);
390 
391 	dlm_lowcomms_commit_msg(msg);
392 	dlm_lowcomms_put_msg(msg);
393 
394 	return 0;
395 }
396 
397 static int dlm_send_fin(struct midcomms_node *node,
398 			void (*ack_rcv)(struct midcomms_node *node))
399 {
400 	int mb_len = sizeof(struct dlm_header);
401 	struct dlm_header *m_header;
402 	struct dlm_mhandle *mh;
403 	char *ppc;
404 
405 	mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_NOFS, &ppc);
406 	if (!mh)
407 		return -ENOMEM;
408 
409 	mh->ack_rcv = ack_rcv;
410 
411 	m_header = (struct dlm_header *)ppc;
412 
413 	m_header->h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
414 	m_header->h_nodeid = cpu_to_le32(dlm_our_nodeid());
415 	m_header->h_length = cpu_to_le16(mb_len);
416 	m_header->h_cmd = DLM_FIN;
417 
418 	pr_debug("sending fin msg to node %d\n", node->nodeid);
419 	dlm_midcomms_commit_mhandle(mh, NULL, 0);
420 	set_bit(DLM_NODE_FLAG_STOP_TX, &node->flags);
421 
422 	return 0;
423 }
424 
425 static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
426 {
427 	struct dlm_mhandle *mh;
428 
429 	rcu_read_lock();
430 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
431 		if (before(mh->seq, seq)) {
432 			if (mh->ack_rcv)
433 				mh->ack_rcv(node);
434 		} else {
435 			/* send queue should be ordered */
436 			break;
437 		}
438 	}
439 
440 	spin_lock_bh(&node->send_queue_lock);
441 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
442 		if (before(mh->seq, seq)) {
443 			dlm_mhandle_delete(node, mh);
444 		} else {
445 			/* send queue should be ordered */
446 			break;
447 		}
448 	}
449 	spin_unlock_bh(&node->send_queue_lock);
450 	rcu_read_unlock();
451 }
452 
453 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
454 {
455 	spin_lock(&node->state_lock);
456 	pr_debug("receive passive fin ack from node %d with state %s\n",
457 		 node->nodeid, dlm_state_str(node->state));
458 
459 	switch (node->state) {
460 	case DLM_LAST_ACK:
461 		/* DLM_CLOSED */
462 		midcomms_node_reset(node);
463 		break;
464 	case DLM_CLOSED:
465 		/* not valid but somehow we got what we want */
466 		wake_up(&node->shutdown_wait);
467 		break;
468 	default:
469 		spin_unlock(&node->state_lock);
470 		log_print("%s: unexpected state: %d\n",
471 			  __func__, node->state);
472 		WARN_ON_ONCE(1);
473 		return;
474 	}
475 	spin_unlock(&node->state_lock);
476 }
477 
478 static void dlm_receive_buffer_3_2_trace(uint32_t seq, union dlm_packet *p)
479 {
480 	switch (p->header.h_cmd) {
481 	case DLM_MSG:
482 		trace_dlm_recv_message(dlm_our_nodeid(), seq, &p->message);
483 		break;
484 	case DLM_RCOM:
485 		trace_dlm_recv_rcom(dlm_our_nodeid(), seq, &p->rcom);
486 		break;
487 	default:
488 		break;
489 	}
490 }
491 
492 static void dlm_midcomms_receive_buffer(union dlm_packet *p,
493 					struct midcomms_node *node,
494 					uint32_t seq)
495 {
496 	if (seq == node->seq_next) {
497 		node->seq_next++;
498 
499 		switch (p->header.h_cmd) {
500 		case DLM_FIN:
501 			/* send ack before fin */
502 			dlm_send_ack(node->nodeid, node->seq_next);
503 
504 			spin_lock(&node->state_lock);
505 			pr_debug("receive fin msg from node %d with state %s\n",
506 				 node->nodeid, dlm_state_str(node->state));
507 
508 			switch (node->state) {
509 			case DLM_ESTABLISHED:
510 				node->state = DLM_CLOSE_WAIT;
511 				pr_debug("switch node %d to state %s\n",
512 					 node->nodeid, dlm_state_str(node->state));
513 				/* passive shutdown DLM_LAST_ACK case 1
514 				 * additional we check if the node is used by
515 				 * cluster manager events at all.
516 				 */
517 				if (node->users == 0) {
518 					node->state = DLM_LAST_ACK;
519 					pr_debug("switch node %d to state %s case 1\n",
520 						 node->nodeid, dlm_state_str(node->state));
521 					spin_unlock(&node->state_lock);
522 					goto send_fin;
523 				}
524 				break;
525 			case DLM_FIN_WAIT1:
526 				node->state = DLM_CLOSING;
527 				pr_debug("switch node %d to state %s\n",
528 					 node->nodeid, dlm_state_str(node->state));
529 				break;
530 			case DLM_FIN_WAIT2:
531 				midcomms_node_reset(node);
532 				pr_debug("switch node %d to state %s\n",
533 					 node->nodeid, dlm_state_str(node->state));
534 				wake_up(&node->shutdown_wait);
535 				break;
536 			case DLM_LAST_ACK:
537 				/* probably remove_member caught it, do nothing */
538 				break;
539 			default:
540 				spin_unlock(&node->state_lock);
541 				log_print("%s: unexpected state: %d\n",
542 					  __func__, node->state);
543 				WARN_ON_ONCE(1);
544 				return;
545 			}
546 			spin_unlock(&node->state_lock);
547 
548 			set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
549 			break;
550 		default:
551 			WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
552 			dlm_receive_buffer_3_2_trace(seq, p);
553 			dlm_receive_buffer(p, node->nodeid);
554 			set_bit(DLM_NODE_ULP_DELIVERED, &node->flags);
555 			break;
556 		}
557 	} else {
558 		/* retry to ack message which we already have by sending back
559 		 * current node->seq_next number as ack.
560 		 */
561 		if (seq < node->seq_next)
562 			dlm_send_ack(node->nodeid, node->seq_next);
563 
564 		log_print_ratelimited("ignore dlm msg because seq mismatch, seq: %u, expected: %u, nodeid: %d",
565 				      seq, node->seq_next, node->nodeid);
566 	}
567 
568 	return;
569 
570 send_fin:
571 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
572 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
573 }
574 
575 static struct midcomms_node *
576 dlm_midcomms_recv_node_lookup(int nodeid, const union dlm_packet *p,
577 			      uint16_t msglen, int (*cb)(struct midcomms_node *node))
578 {
579 	struct midcomms_node *node = NULL;
580 	gfp_t allocation = 0;
581 	int ret;
582 
583 	switch (p->header.h_cmd) {
584 	case DLM_RCOM:
585 		if (msglen < sizeof(struct dlm_rcom)) {
586 			log_print("rcom msg too small: %u, will skip this message from node %d",
587 				  msglen, nodeid);
588 			return NULL;
589 		}
590 
591 		switch (p->rcom.rc_type) {
592 		case cpu_to_le32(DLM_RCOM_NAMES):
593 			fallthrough;
594 		case cpu_to_le32(DLM_RCOM_NAMES_REPLY):
595 			fallthrough;
596 		case cpu_to_le32(DLM_RCOM_STATUS):
597 			fallthrough;
598 		case cpu_to_le32(DLM_RCOM_STATUS_REPLY):
599 			node = nodeid2node(nodeid, 0);
600 			if (node) {
601 				spin_lock(&node->state_lock);
602 				if (node->state != DLM_ESTABLISHED)
603 					pr_debug("receive begin RCOM msg from node %d with state %s\n",
604 						 node->nodeid, dlm_state_str(node->state));
605 
606 				switch (node->state) {
607 				case DLM_CLOSED:
608 					node->state = DLM_ESTABLISHED;
609 					pr_debug("switch node %d to state %s\n",
610 						 node->nodeid, dlm_state_str(node->state));
611 					break;
612 				case DLM_ESTABLISHED:
613 					break;
614 				default:
615 					/* some invalid state passive shutdown
616 					 * was failed, we try to reset and
617 					 * hope it will go on.
618 					 */
619 					log_print("reset node %d because shutdown stuck",
620 						  node->nodeid);
621 
622 					midcomms_node_reset(node);
623 					node->state = DLM_ESTABLISHED;
624 					break;
625 				}
626 				spin_unlock(&node->state_lock);
627 			}
628 
629 			allocation = GFP_NOFS;
630 			break;
631 		default:
632 			break;
633 		}
634 
635 		break;
636 	default:
637 		break;
638 	}
639 
640 	node = nodeid2node(nodeid, allocation);
641 	if (!node) {
642 		switch (p->header.h_cmd) {
643 		case DLM_OPTS:
644 			if (msglen < sizeof(struct dlm_opts)) {
645 				log_print("opts msg too small: %u, will skip this message from node %d",
646 					  msglen, nodeid);
647 				return NULL;
648 			}
649 
650 			log_print_ratelimited("received dlm opts message nextcmd %d from node %d in an invalid sequence",
651 					      p->opts.o_nextcmd, nodeid);
652 			break;
653 		default:
654 			log_print_ratelimited("received dlm message cmd %d from node %d in an invalid sequence",
655 					      p->header.h_cmd, nodeid);
656 			break;
657 		}
658 
659 		return NULL;
660 	}
661 
662 	ret = cb(node);
663 	if (ret < 0)
664 		return NULL;
665 
666 	return node;
667 }
668 
669 static int dlm_midcomms_version_check_3_2(struct midcomms_node *node)
670 {
671 	switch (node->version) {
672 	case DLM_VERSION_NOT_SET:
673 		node->version = DLM_VERSION_3_2;
674 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_2,
675 			  node->nodeid);
676 		break;
677 	case DLM_VERSION_3_2:
678 		break;
679 	default:
680 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
681 				      DLM_VERSION_3_2, node->nodeid, node->version);
682 		return -1;
683 	}
684 
685 	return 0;
686 }
687 
688 static int dlm_opts_check_msglen(union dlm_packet *p, uint16_t msglen, int nodeid)
689 {
690 	int len = msglen;
691 
692 	/* we only trust outer header msglen because
693 	 * it's checked against receive buffer length.
694 	 */
695 	if (len < sizeof(struct dlm_opts))
696 		return -1;
697 	len -= sizeof(struct dlm_opts);
698 
699 	if (len < le16_to_cpu(p->opts.o_optlen))
700 		return -1;
701 	len -= le16_to_cpu(p->opts.o_optlen);
702 
703 	switch (p->opts.o_nextcmd) {
704 	case DLM_FIN:
705 		if (len < sizeof(struct dlm_header)) {
706 			log_print("fin too small: %d, will skip this message from node %d",
707 				  len, nodeid);
708 			return -1;
709 		}
710 
711 		break;
712 	case DLM_MSG:
713 		if (len < sizeof(struct dlm_message)) {
714 			log_print("msg too small: %d, will skip this message from node %d",
715 				  msglen, nodeid);
716 			return -1;
717 		}
718 
719 		break;
720 	case DLM_RCOM:
721 		if (len < sizeof(struct dlm_rcom)) {
722 			log_print("rcom msg too small: %d, will skip this message from node %d",
723 				  len, nodeid);
724 			return -1;
725 		}
726 
727 		break;
728 	default:
729 		log_print("unsupported o_nextcmd received: %u, will skip this message from node %d",
730 			  p->opts.o_nextcmd, nodeid);
731 		return -1;
732 	}
733 
734 	return 0;
735 }
736 
737 static void dlm_midcomms_receive_buffer_3_2(union dlm_packet *p, int nodeid)
738 {
739 	uint16_t msglen = le16_to_cpu(p->header.h_length);
740 	struct midcomms_node *node;
741 	uint32_t seq;
742 	int ret, idx;
743 
744 	idx = srcu_read_lock(&nodes_srcu);
745 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
746 					     dlm_midcomms_version_check_3_2);
747 	if (!node)
748 		goto out;
749 
750 	switch (p->header.h_cmd) {
751 	case DLM_RCOM:
752 		/* these rcom message we use to determine version.
753 		 * they have their own retransmission handling and
754 		 * are the first messages of dlm.
755 		 *
756 		 * length already checked.
757 		 */
758 		switch (p->rcom.rc_type) {
759 		case cpu_to_le32(DLM_RCOM_NAMES):
760 			fallthrough;
761 		case cpu_to_le32(DLM_RCOM_NAMES_REPLY):
762 			fallthrough;
763 		case cpu_to_le32(DLM_RCOM_STATUS):
764 			fallthrough;
765 		case cpu_to_le32(DLM_RCOM_STATUS_REPLY):
766 			break;
767 		default:
768 			log_print("unsupported rcom type received: %u, will skip this message from node %d",
769 				  le32_to_cpu(p->rcom.rc_type), nodeid);
770 			goto out;
771 		}
772 
773 		WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
774 		dlm_receive_buffer(p, nodeid);
775 		break;
776 	case DLM_OPTS:
777 		seq = le32_to_cpu(p->header.u.h_seq);
778 
779 		ret = dlm_opts_check_msglen(p, msglen, nodeid);
780 		if (ret < 0) {
781 			log_print("opts msg too small: %u, will skip this message from node %d",
782 				  msglen, nodeid);
783 			goto out;
784 		}
785 
786 		p = (union dlm_packet *)((unsigned char *)p->opts.o_opts +
787 					 le16_to_cpu(p->opts.o_optlen));
788 
789 		/* recheck inner msglen just if it's not garbage */
790 		msglen = le16_to_cpu(p->header.h_length);
791 		switch (p->header.h_cmd) {
792 		case DLM_RCOM:
793 			if (msglen < sizeof(struct dlm_rcom)) {
794 				log_print("inner rcom msg too small: %u, will skip this message from node %d",
795 					  msglen, nodeid);
796 				goto out;
797 			}
798 
799 			break;
800 		case DLM_MSG:
801 			if (msglen < sizeof(struct dlm_message)) {
802 				log_print("inner msg too small: %u, will skip this message from node %d",
803 					  msglen, nodeid);
804 				goto out;
805 			}
806 
807 			break;
808 		case DLM_FIN:
809 			if (msglen < sizeof(struct dlm_header)) {
810 				log_print("inner fin too small: %u, will skip this message from node %d",
811 					  msglen, nodeid);
812 				goto out;
813 			}
814 
815 			break;
816 		default:
817 			log_print("unsupported inner h_cmd received: %u, will skip this message from node %d",
818 				  msglen, nodeid);
819 			goto out;
820 		}
821 
822 		dlm_midcomms_receive_buffer(p, node, seq);
823 		break;
824 	case DLM_ACK:
825 		seq = le32_to_cpu(p->header.u.h_seq);
826 		dlm_receive_ack(node, seq);
827 		break;
828 	default:
829 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
830 			  p->header.h_cmd, nodeid);
831 		break;
832 	}
833 
834 out:
835 	srcu_read_unlock(&nodes_srcu, idx);
836 }
837 
838 static int dlm_midcomms_version_check_3_1(struct midcomms_node *node)
839 {
840 	switch (node->version) {
841 	case DLM_VERSION_NOT_SET:
842 		node->version = DLM_VERSION_3_1;
843 		log_print("version 0x%08x for node %d detected", DLM_VERSION_3_1,
844 			  node->nodeid);
845 		break;
846 	case DLM_VERSION_3_1:
847 		break;
848 	default:
849 		log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
850 				      DLM_VERSION_3_1, node->nodeid, node->version);
851 		return -1;
852 	}
853 
854 	return 0;
855 }
856 
857 static void dlm_midcomms_receive_buffer_3_1(union dlm_packet *p, int nodeid)
858 {
859 	uint16_t msglen = le16_to_cpu(p->header.h_length);
860 	struct midcomms_node *node;
861 	int idx;
862 
863 	idx = srcu_read_lock(&nodes_srcu);
864 	node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
865 					     dlm_midcomms_version_check_3_1);
866 	if (!node) {
867 		srcu_read_unlock(&nodes_srcu, idx);
868 		return;
869 	}
870 	srcu_read_unlock(&nodes_srcu, idx);
871 
872 	switch (p->header.h_cmd) {
873 	case DLM_RCOM:
874 		/* length already checked */
875 		break;
876 	case DLM_MSG:
877 		if (msglen < sizeof(struct dlm_message)) {
878 			log_print("msg too small: %u, will skip this message from node %d",
879 				  msglen, nodeid);
880 			return;
881 		}
882 
883 		break;
884 	default:
885 		log_print("unsupported h_cmd received: %u, will skip this message from node %d",
886 			  p->header.h_cmd, nodeid);
887 		return;
888 	}
889 
890 	dlm_receive_buffer(p, nodeid);
891 }
892 
893 int dlm_validate_incoming_buffer(int nodeid, unsigned char *buf, int len)
894 {
895 	const unsigned char *ptr = buf;
896 	const struct dlm_header *hd;
897 	uint16_t msglen;
898 	int ret = 0;
899 
900 	while (len >= sizeof(struct dlm_header)) {
901 		hd = (struct dlm_header *)ptr;
902 
903 		/* no message should be more than DLM_MAX_SOCKET_BUFSIZE or
904 		 * less than dlm_header size.
905 		 *
906 		 * Some messages does not have a 8 byte length boundary yet
907 		 * which can occur in a unaligned memory access of some dlm
908 		 * messages. However this problem need to be fixed at the
909 		 * sending side, for now it seems nobody run into architecture
910 		 * related issues yet but it slows down some processing.
911 		 * Fixing this issue should be scheduled in future by doing
912 		 * the next major version bump.
913 		 */
914 		msglen = le16_to_cpu(hd->h_length);
915 		if (msglen > DLM_MAX_SOCKET_BUFSIZE ||
916 		    msglen < sizeof(struct dlm_header)) {
917 			log_print("received invalid length header: %u from node %d, will abort message parsing",
918 				  msglen, nodeid);
919 			return -EBADMSG;
920 		}
921 
922 		/* caller will take care that leftover
923 		 * will be parsed next call with more data
924 		 */
925 		if (msglen > len)
926 			break;
927 
928 		ret += msglen;
929 		len -= msglen;
930 		ptr += msglen;
931 	}
932 
933 	return ret;
934 }
935 
936 /*
937  * Called from the low-level comms layer to process a buffer of
938  * commands.
939  */
940 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int len)
941 {
942 	const unsigned char *ptr = buf;
943 	const struct dlm_header *hd;
944 	uint16_t msglen;
945 	int ret = 0;
946 
947 	while (len >= sizeof(struct dlm_header)) {
948 		hd = (struct dlm_header *)ptr;
949 
950 		msglen = le16_to_cpu(hd->h_length);
951 		if (msglen > len)
952 			break;
953 
954 		switch (hd->h_version) {
955 		case cpu_to_le32(DLM_VERSION_3_1):
956 			dlm_midcomms_receive_buffer_3_1((union dlm_packet *)ptr, nodeid);
957 			break;
958 		case cpu_to_le32(DLM_VERSION_3_2):
959 			dlm_midcomms_receive_buffer_3_2((union dlm_packet *)ptr, nodeid);
960 			break;
961 		default:
962 			log_print("received invalid version header: %u from node %d, will skip this message",
963 				  le32_to_cpu(hd->h_version), nodeid);
964 			break;
965 		}
966 
967 		ret += msglen;
968 		len -= msglen;
969 		ptr += msglen;
970 	}
971 
972 	return ret;
973 }
974 
975 void dlm_midcomms_receive_done(int nodeid)
976 {
977 	struct midcomms_node *node;
978 	int idx;
979 
980 	idx = srcu_read_lock(&nodes_srcu);
981 	node = nodeid2node(nodeid, 0);
982 	if (!node) {
983 		srcu_read_unlock(&nodes_srcu, idx);
984 		return;
985 	}
986 
987 	/* old protocol, we do nothing */
988 	switch (node->version) {
989 	case DLM_VERSION_3_2:
990 		break;
991 	default:
992 		srcu_read_unlock(&nodes_srcu, idx);
993 		return;
994 	}
995 
996 	/* do nothing if we didn't delivered stateful to ulp */
997 	if (!test_and_clear_bit(DLM_NODE_ULP_DELIVERED,
998 				&node->flags)) {
999 		srcu_read_unlock(&nodes_srcu, idx);
1000 		return;
1001 	}
1002 
1003 	spin_lock(&node->state_lock);
1004 	/* we only ack if state is ESTABLISHED */
1005 	switch (node->state) {
1006 	case DLM_ESTABLISHED:
1007 		spin_unlock(&node->state_lock);
1008 		dlm_send_ack(node->nodeid, node->seq_next);
1009 		break;
1010 	default:
1011 		spin_unlock(&node->state_lock);
1012 		/* do nothing FIN has it's own ack send */
1013 		break;
1014 	}
1015 	srcu_read_unlock(&nodes_srcu, idx);
1016 }
1017 
1018 void dlm_midcomms_unack_msg_resend(int nodeid)
1019 {
1020 	struct midcomms_node *node;
1021 	struct dlm_mhandle *mh;
1022 	int idx, ret;
1023 
1024 	idx = srcu_read_lock(&nodes_srcu);
1025 	node = nodeid2node(nodeid, 0);
1026 	if (!node) {
1027 		srcu_read_unlock(&nodes_srcu, idx);
1028 		return;
1029 	}
1030 
1031 	/* old protocol, we don't support to retransmit on failure */
1032 	switch (node->version) {
1033 	case DLM_VERSION_3_2:
1034 		break;
1035 	default:
1036 		srcu_read_unlock(&nodes_srcu, idx);
1037 		return;
1038 	}
1039 
1040 	rcu_read_lock();
1041 	list_for_each_entry_rcu(mh, &node->send_queue, list) {
1042 		if (!mh->committed)
1043 			continue;
1044 
1045 		ret = dlm_lowcomms_resend_msg(mh->msg);
1046 		if (!ret)
1047 			log_print_ratelimited("retransmit dlm msg, seq %u, nodeid %d",
1048 					      mh->seq, node->nodeid);
1049 	}
1050 	rcu_read_unlock();
1051 	srcu_read_unlock(&nodes_srcu, idx);
1052 }
1053 
1054 static void dlm_fill_opts_header(struct dlm_opts *opts, uint16_t inner_len,
1055 				 uint32_t seq)
1056 {
1057 	opts->o_header.h_cmd = DLM_OPTS;
1058 	opts->o_header.h_version = cpu_to_le32(DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
1059 	opts->o_header.h_nodeid = cpu_to_le32(dlm_our_nodeid());
1060 	opts->o_header.h_length = cpu_to_le16(DLM_MIDCOMMS_OPT_LEN + inner_len);
1061 	opts->o_header.u.h_seq = cpu_to_le32(seq);
1062 }
1063 
1064 static void midcomms_new_msg_cb(void *data)
1065 {
1066 	struct dlm_mhandle *mh = data;
1067 
1068 	atomic_inc(&mh->node->send_queue_cnt);
1069 
1070 	spin_lock_bh(&mh->node->send_queue_lock);
1071 	list_add_tail_rcu(&mh->list, &mh->node->send_queue);
1072 	spin_unlock_bh(&mh->node->send_queue_lock);
1073 
1074 	mh->seq = mh->node->seq_send++;
1075 }
1076 
1077 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
1078 						int len, gfp_t allocation, char **ppc)
1079 {
1080 	struct dlm_opts *opts;
1081 	struct dlm_msg *msg;
1082 
1083 	msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
1084 				   allocation, ppc, midcomms_new_msg_cb, mh);
1085 	if (!msg)
1086 		return NULL;
1087 
1088 	opts = (struct dlm_opts *)*ppc;
1089 	mh->opts = opts;
1090 
1091 	/* add possible options here */
1092 	dlm_fill_opts_header(opts, len, mh->seq);
1093 
1094 	*ppc += sizeof(*opts);
1095 	mh->inner_p = (const union dlm_packet *)*ppc;
1096 	return msg;
1097 }
1098 
1099 /* avoid false positive for nodes_srcu, unlock happens in
1100  * dlm_midcomms_commit_mhandle which is a must call if success
1101  */
1102 #ifndef __CHECKER__
1103 struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
1104 					     gfp_t allocation, char **ppc)
1105 {
1106 	struct midcomms_node *node;
1107 	struct dlm_mhandle *mh;
1108 	struct dlm_msg *msg;
1109 	int idx;
1110 
1111 	idx = srcu_read_lock(&nodes_srcu);
1112 	node = nodeid2node(nodeid, 0);
1113 	if (!node) {
1114 		WARN_ON_ONCE(1);
1115 		goto err;
1116 	}
1117 
1118 	/* this is a bug, however we going on and hope it will be resolved */
1119 	WARN_ON_ONCE(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
1120 
1121 	mh = dlm_allocate_mhandle(allocation);
1122 	if (!mh)
1123 		goto err;
1124 
1125 	mh->committed = false;
1126 	mh->ack_rcv = NULL;
1127 	mh->idx = idx;
1128 	mh->node = node;
1129 
1130 	switch (node->version) {
1131 	case DLM_VERSION_3_1:
1132 		msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
1133 					   NULL, NULL);
1134 		if (!msg) {
1135 			dlm_free_mhandle(mh);
1136 			goto err;
1137 		}
1138 
1139 		break;
1140 	case DLM_VERSION_3_2:
1141 		msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
1142 					       ppc);
1143 		if (!msg) {
1144 			dlm_free_mhandle(mh);
1145 			goto err;
1146 		}
1147 
1148 		break;
1149 	default:
1150 		dlm_free_mhandle(mh);
1151 		WARN_ON_ONCE(1);
1152 		goto err;
1153 	}
1154 
1155 	mh->msg = msg;
1156 
1157 	/* keep in mind that is a must to call
1158 	 * dlm_midcomms_commit_msg() which releases
1159 	 * nodes_srcu using mh->idx which is assumed
1160 	 * here that the application will call it.
1161 	 */
1162 	return mh;
1163 
1164 err:
1165 	srcu_read_unlock(&nodes_srcu, idx);
1166 	return NULL;
1167 }
1168 #endif
1169 
1170 static void dlm_midcomms_commit_msg_3_2_trace(const struct dlm_mhandle *mh,
1171 					      const void *name, int namelen)
1172 {
1173 	switch (mh->inner_p->header.h_cmd) {
1174 	case DLM_MSG:
1175 		trace_dlm_send_message(mh->node->nodeid, mh->seq,
1176 				       &mh->inner_p->message,
1177 				       name, namelen);
1178 		break;
1179 	case DLM_RCOM:
1180 		trace_dlm_send_rcom(mh->node->nodeid, mh->seq,
1181 				    &mh->inner_p->rcom);
1182 		break;
1183 	default:
1184 		/* nothing to trace */
1185 		break;
1186 	}
1187 }
1188 
1189 static void dlm_midcomms_commit_msg_3_2(struct dlm_mhandle *mh,
1190 					const void *name, int namelen)
1191 {
1192 	/* nexthdr chain for fast lookup */
1193 	mh->opts->o_nextcmd = mh->inner_p->header.h_cmd;
1194 	mh->committed = true;
1195 	dlm_midcomms_commit_msg_3_2_trace(mh, name, namelen);
1196 	dlm_lowcomms_commit_msg(mh->msg);
1197 }
1198 
1199 /* avoid false positive for nodes_srcu, lock was happen in
1200  * dlm_midcomms_get_mhandle
1201  */
1202 #ifndef __CHECKER__
1203 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh,
1204 				 const void *name, int namelen)
1205 {
1206 
1207 	switch (mh->node->version) {
1208 	case DLM_VERSION_3_1:
1209 		srcu_read_unlock(&nodes_srcu, mh->idx);
1210 
1211 		dlm_lowcomms_commit_msg(mh->msg);
1212 		dlm_lowcomms_put_msg(mh->msg);
1213 		/* mh is not part of rcu list in this case */
1214 		dlm_free_mhandle(mh);
1215 		break;
1216 	case DLM_VERSION_3_2:
1217 		dlm_midcomms_commit_msg_3_2(mh, name, namelen);
1218 		srcu_read_unlock(&nodes_srcu, mh->idx);
1219 		break;
1220 	default:
1221 		srcu_read_unlock(&nodes_srcu, mh->idx);
1222 		WARN_ON_ONCE(1);
1223 		break;
1224 	}
1225 }
1226 #endif
1227 
1228 int dlm_midcomms_start(void)
1229 {
1230 	return dlm_lowcomms_start();
1231 }
1232 
1233 void dlm_midcomms_stop(void)
1234 {
1235 	dlm_lowcomms_stop();
1236 }
1237 
1238 void dlm_midcomms_init(void)
1239 {
1240 	int i;
1241 
1242 	for (i = 0; i < CONN_HASH_SIZE; i++)
1243 		INIT_HLIST_HEAD(&node_hash[i]);
1244 
1245 	dlm_lowcomms_init();
1246 }
1247 
1248 void dlm_midcomms_exit(void)
1249 {
1250 	dlm_lowcomms_exit();
1251 }
1252 
1253 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
1254 {
1255 	spin_lock(&node->state_lock);
1256 	pr_debug("receive active fin ack from node %d with state %s\n",
1257 		 node->nodeid, dlm_state_str(node->state));
1258 
1259 	switch (node->state) {
1260 	case DLM_FIN_WAIT1:
1261 		node->state = DLM_FIN_WAIT2;
1262 		pr_debug("switch node %d to state %s\n",
1263 			 node->nodeid, dlm_state_str(node->state));
1264 		break;
1265 	case DLM_CLOSING:
1266 		midcomms_node_reset(node);
1267 		pr_debug("switch node %d to state %s\n",
1268 			 node->nodeid, dlm_state_str(node->state));
1269 		wake_up(&node->shutdown_wait);
1270 		break;
1271 	case DLM_CLOSED:
1272 		/* not valid but somehow we got what we want */
1273 		wake_up(&node->shutdown_wait);
1274 		break;
1275 	default:
1276 		spin_unlock(&node->state_lock);
1277 		log_print("%s: unexpected state: %d\n",
1278 			  __func__, node->state);
1279 		WARN_ON_ONCE(1);
1280 		return;
1281 	}
1282 	spin_unlock(&node->state_lock);
1283 }
1284 
1285 void dlm_midcomms_add_member(int nodeid)
1286 {
1287 	struct midcomms_node *node;
1288 	int idx;
1289 
1290 	if (nodeid == dlm_our_nodeid())
1291 		return;
1292 
1293 	idx = srcu_read_lock(&nodes_srcu);
1294 	node = nodeid2node(nodeid, GFP_NOFS);
1295 	if (!node) {
1296 		srcu_read_unlock(&nodes_srcu, idx);
1297 		return;
1298 	}
1299 
1300 	spin_lock(&node->state_lock);
1301 	if (!node->users) {
1302 		pr_debug("receive add member from node %d with state %s\n",
1303 			 node->nodeid, dlm_state_str(node->state));
1304 		switch (node->state) {
1305 		case DLM_ESTABLISHED:
1306 			break;
1307 		case DLM_CLOSED:
1308 			node->state = DLM_ESTABLISHED;
1309 			pr_debug("switch node %d to state %s\n",
1310 				 node->nodeid, dlm_state_str(node->state));
1311 			break;
1312 		default:
1313 			/* some invalid state passive shutdown
1314 			 * was failed, we try to reset and
1315 			 * hope it will go on.
1316 			 */
1317 			log_print("reset node %d because shutdown stuck",
1318 				  node->nodeid);
1319 
1320 			midcomms_node_reset(node);
1321 			node->state = DLM_ESTABLISHED;
1322 			break;
1323 		}
1324 	}
1325 
1326 	node->users++;
1327 	pr_debug("node %d users inc count %d\n", nodeid, node->users);
1328 	spin_unlock(&node->state_lock);
1329 
1330 	srcu_read_unlock(&nodes_srcu, idx);
1331 }
1332 
1333 void dlm_midcomms_remove_member(int nodeid)
1334 {
1335 	struct midcomms_node *node;
1336 	int idx;
1337 
1338 	if (nodeid == dlm_our_nodeid())
1339 		return;
1340 
1341 	idx = srcu_read_lock(&nodes_srcu);
1342 	node = nodeid2node(nodeid, 0);
1343 	if (!node) {
1344 		srcu_read_unlock(&nodes_srcu, idx);
1345 		return;
1346 	}
1347 
1348 	spin_lock(&node->state_lock);
1349 	node->users--;
1350 	pr_debug("node %d users dec count %d\n", nodeid, node->users);
1351 
1352 	/* hitting users count to zero means the
1353 	 * other side is running dlm_midcomms_stop()
1354 	 * we meet us to have a clean disconnect.
1355 	 */
1356 	if (node->users == 0) {
1357 		pr_debug("receive remove member from node %d with state %s\n",
1358 			 node->nodeid, dlm_state_str(node->state));
1359 		switch (node->state) {
1360 		case DLM_ESTABLISHED:
1361 			break;
1362 		case DLM_CLOSE_WAIT:
1363 			/* passive shutdown DLM_LAST_ACK case 2 */
1364 			node->state = DLM_LAST_ACK;
1365 			spin_unlock(&node->state_lock);
1366 
1367 			pr_debug("switch node %d to state %s case 2\n",
1368 				 node->nodeid, dlm_state_str(node->state));
1369 			goto send_fin;
1370 		case DLM_LAST_ACK:
1371 			/* probably receive fin caught it, do nothing */
1372 			break;
1373 		case DLM_CLOSED:
1374 			/* already gone, do nothing */
1375 			break;
1376 		default:
1377 			log_print("%s: unexpected state: %d\n",
1378 				  __func__, node->state);
1379 			break;
1380 		}
1381 	}
1382 	spin_unlock(&node->state_lock);
1383 
1384 	srcu_read_unlock(&nodes_srcu, idx);
1385 	return;
1386 
1387 send_fin:
1388 	set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
1389 	dlm_send_fin(node, dlm_pas_fin_ack_rcv);
1390 	srcu_read_unlock(&nodes_srcu, idx);
1391 }
1392 
1393 static void midcomms_node_release(struct rcu_head *rcu)
1394 {
1395 	struct midcomms_node *node = container_of(rcu, struct midcomms_node, rcu);
1396 
1397 	WARN_ON_ONCE(atomic_read(&node->send_queue_cnt));
1398 	kfree(node);
1399 }
1400 
1401 static void midcomms_shutdown(struct midcomms_node *node)
1402 {
1403 	int ret;
1404 
1405 	/* old protocol, we don't wait for pending operations */
1406 	switch (node->version) {
1407 	case DLM_VERSION_3_2:
1408 		break;
1409 	default:
1410 		return;
1411 	}
1412 
1413 	spin_lock(&node->state_lock);
1414 	pr_debug("receive active shutdown for node %d with state %s\n",
1415 		 node->nodeid, dlm_state_str(node->state));
1416 	switch (node->state) {
1417 	case DLM_ESTABLISHED:
1418 		node->state = DLM_FIN_WAIT1;
1419 		pr_debug("switch node %d to state %s case 2\n",
1420 			 node->nodeid, dlm_state_str(node->state));
1421 		break;
1422 	case DLM_CLOSED:
1423 		/* we have what we want */
1424 		spin_unlock(&node->state_lock);
1425 		return;
1426 	default:
1427 		/* busy to enter DLM_FIN_WAIT1, wait until passive
1428 		 * done in shutdown_wait to enter DLM_CLOSED.
1429 		 */
1430 		break;
1431 	}
1432 	spin_unlock(&node->state_lock);
1433 
1434 	if (node->state == DLM_FIN_WAIT1) {
1435 		dlm_send_fin(node, dlm_act_fin_ack_rcv);
1436 
1437 		if (DLM_DEBUG_FENCE_TERMINATION)
1438 			msleep(5000);
1439 	}
1440 
1441 	/* wait for other side dlm + fin */
1442 	ret = wait_event_timeout(node->shutdown_wait,
1443 				 node->state == DLM_CLOSED ||
1444 				 test_bit(DLM_NODE_FLAG_CLOSE, &node->flags),
1445 				 DLM_SHUTDOWN_TIMEOUT);
1446 	if (!ret || test_bit(DLM_NODE_FLAG_CLOSE, &node->flags)) {
1447 		pr_debug("active shutdown timed out for node %d with state %s\n",
1448 			 node->nodeid, dlm_state_str(node->state));
1449 		midcomms_node_reset(node);
1450 		dlm_lowcomms_shutdown_node(node->nodeid, true);
1451 		return;
1452 	}
1453 
1454 	pr_debug("active shutdown done for node %d with state %s\n",
1455 		 node->nodeid, dlm_state_str(node->state));
1456 	dlm_lowcomms_shutdown_node(node->nodeid, false);
1457 }
1458 
1459 void dlm_midcomms_shutdown(void)
1460 {
1461 	struct midcomms_node *node;
1462 	int i, idx;
1463 
1464 	dlm_lowcomms_shutdown();
1465 
1466 	mutex_lock(&close_lock);
1467 	idx = srcu_read_lock(&nodes_srcu);
1468 	for (i = 0; i < CONN_HASH_SIZE; i++) {
1469 		hlist_for_each_entry_rcu(node, &node_hash[i], hlist) {
1470 			midcomms_shutdown(node);
1471 
1472 			dlm_delete_debug_comms_file(node->debugfs);
1473 
1474 			spin_lock(&nodes_lock);
1475 			hlist_del_rcu(&node->hlist);
1476 			spin_unlock(&nodes_lock);
1477 
1478 			call_srcu(&nodes_srcu, &node->rcu, midcomms_node_release);
1479 		}
1480 	}
1481 	srcu_read_unlock(&nodes_srcu, idx);
1482 	mutex_unlock(&close_lock);
1483 }
1484 
1485 int dlm_midcomms_close(int nodeid)
1486 {
1487 	struct midcomms_node *node;
1488 	int idx, ret;
1489 
1490 	if (nodeid == dlm_our_nodeid())
1491 		return 0;
1492 
1493 	dlm_stop_lockspaces_check();
1494 
1495 	idx = srcu_read_lock(&nodes_srcu);
1496 	/* Abort pending close/remove operation */
1497 	node = nodeid2node(nodeid, 0);
1498 	if (node) {
1499 		/* let shutdown waiters leave */
1500 		set_bit(DLM_NODE_FLAG_CLOSE, &node->flags);
1501 		wake_up(&node->shutdown_wait);
1502 	}
1503 	srcu_read_unlock(&nodes_srcu, idx);
1504 
1505 	synchronize_srcu(&nodes_srcu);
1506 
1507 	idx = srcu_read_lock(&nodes_srcu);
1508 	mutex_lock(&close_lock);
1509 	node = nodeid2node(nodeid, 0);
1510 	if (!node) {
1511 		mutex_unlock(&close_lock);
1512 		srcu_read_unlock(&nodes_srcu, idx);
1513 		return dlm_lowcomms_close(nodeid);
1514 	}
1515 
1516 	ret = dlm_lowcomms_close(nodeid);
1517 	spin_lock(&node->state_lock);
1518 	midcomms_node_reset(node);
1519 	spin_unlock(&node->state_lock);
1520 	srcu_read_unlock(&nodes_srcu, idx);
1521 	mutex_unlock(&close_lock);
1522 
1523 	return ret;
1524 }
1525 
1526 /* debug functionality to send raw dlm msg from user space */
1527 struct dlm_rawmsg_data {
1528 	struct midcomms_node *node;
1529 	void *buf;
1530 };
1531 
1532 static void midcomms_new_rawmsg_cb(void *data)
1533 {
1534 	struct dlm_rawmsg_data *rd = data;
1535 	struct dlm_header *h = rd->buf;
1536 
1537 	switch (h->h_version) {
1538 	case cpu_to_le32(DLM_VERSION_3_1):
1539 		break;
1540 	default:
1541 		switch (h->h_cmd) {
1542 		case DLM_OPTS:
1543 			if (!h->u.h_seq)
1544 				h->u.h_seq = cpu_to_le32(rd->node->seq_send++);
1545 			break;
1546 		default:
1547 			break;
1548 		}
1549 		break;
1550 	}
1551 }
1552 
1553 int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
1554 			     int buflen)
1555 {
1556 	struct dlm_rawmsg_data rd;
1557 	struct dlm_msg *msg;
1558 	char *msgbuf;
1559 
1560 	rd.node = node;
1561 	rd.buf = buf;
1562 
1563 	msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
1564 				   &msgbuf, midcomms_new_rawmsg_cb, &rd);
1565 	if (!msg)
1566 		return -ENOMEM;
1567 
1568 	memcpy(msgbuf, buf, buflen);
1569 	dlm_lowcomms_commit_msg(msg);
1570 	return 0;
1571 }
1572 
1573