1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmdomain.c
5  *
6  * defines domain join / leave apis
7  *
8  * Copyright (C) 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  *
25  */
26 
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/init.h>
32 #include <linux/spinlock.h>
33 #include <linux/delay.h>
34 #include <linux/err.h>
35 #include <linux/debugfs.h>
36 
37 #include "cluster/heartbeat.h"
38 #include "cluster/nodemanager.h"
39 #include "cluster/tcp.h"
40 
41 #include "dlmapi.h"
42 #include "dlmcommon.h"
43 #include "dlmdomain.h"
44 #include "dlmdebug.h"
45 
46 #include "dlmver.h"
47 
48 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
49 #include "cluster/masklog.h"
50 
51 /*
52  * ocfs2 node maps are array of long int, which limits to send them freely
53  * across the wire due to endianness issues. To workaround this, we convert
54  * long ints to byte arrays. Following 3 routines are helper functions to
55  * set/test/copy bits within those array of bytes
56  */
byte_set_bit(u8 nr,u8 map[])57 static inline void byte_set_bit(u8 nr, u8 map[])
58 {
59 	map[nr >> 3] |= (1UL << (nr & 7));
60 }
61 
byte_test_bit(u8 nr,u8 map[])62 static inline int byte_test_bit(u8 nr, u8 map[])
63 {
64 	return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
65 }
66 
byte_copymap(u8 dmap[],unsigned long smap[],unsigned int sz)67 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
68 			unsigned int sz)
69 {
70 	unsigned int nn;
71 
72 	if (!sz)
73 		return;
74 
75 	memset(dmap, 0, ((sz + 7) >> 3));
76 	for (nn = 0 ; nn < sz; nn++)
77 		if (test_bit(nn, smap))
78 			byte_set_bit(nn, dmap);
79 }
80 
dlm_free_pagevec(void ** vec,int pages)81 static void dlm_free_pagevec(void **vec, int pages)
82 {
83 	while (pages--)
84 		free_page((unsigned long)vec[pages]);
85 	kfree(vec);
86 }
87 
dlm_alloc_pagevec(int pages)88 static void **dlm_alloc_pagevec(int pages)
89 {
90 	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
91 	int i;
92 
93 	if (!vec)
94 		return NULL;
95 
96 	for (i = 0; i < pages; i++)
97 		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
98 			goto out_free;
99 
100 	mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
101 	     pages, (unsigned long)DLM_HASH_PAGES,
102 	     (unsigned long)DLM_BUCKETS_PER_PAGE);
103 	return vec;
104 out_free:
105 	dlm_free_pagevec(vec, i);
106 	return NULL;
107 }
108 
109 /*
110  *
111  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
112  *    dlm_domain_lock
113  *    struct dlm_ctxt->spinlock
114  *    struct dlm_lock_resource->spinlock
115  *    struct dlm_ctxt->master_lock
116  *    struct dlm_ctxt->ast_lock
117  *    dlm_master_list_entry->spinlock
118  *    dlm_lock->spinlock
119  *
120  */
121 
122 DEFINE_SPINLOCK(dlm_domain_lock);
123 LIST_HEAD(dlm_domains);
124 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
125 
126 /*
127  * The supported protocol version for DLM communication.  Running domains
128  * will have a negotiated version with the same major number and a minor
129  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
130  * be used to determine what a running domain is actually using.
131  *
132  * New in version 1.1:
133  *	- Message DLM_QUERY_REGION added to support global heartbeat
134  *	- Message DLM_QUERY_NODEINFO added to allow online node removes
135  * New in version 1.2:
136  * 	- Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain
137  */
138 static const struct dlm_protocol_version dlm_protocol = {
139 	.pv_major = 1,
140 	.pv_minor = 2,
141 };
142 
143 #define DLM_DOMAIN_BACKOFF_MS 200
144 
145 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
146 				  void **ret_data);
147 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
148 				     void **ret_data);
149 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
150 				   void **ret_data);
151 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
152 				    void *data, void **ret_data);
153 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
154 				   void **ret_data);
155 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
156 				struct dlm_protocol_version *request);
157 
158 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
159 
__dlm_unhash_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)160 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
161 {
162 	if (hlist_unhashed(&res->hash_node))
163 		return;
164 
165 	mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
166 	     res->lockname.name);
167 	hlist_del_init(&res->hash_node);
168 	dlm_lockres_put(res);
169 }
170 
__dlm_insert_lockres(struct dlm_ctxt * dlm,struct dlm_lock_resource * res)171 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
172 {
173 	struct hlist_head *bucket;
174 	struct qstr *q;
175 
176 	assert_spin_locked(&dlm->spinlock);
177 
178 	q = &res->lockname;
179 	bucket = dlm_lockres_hash(dlm, q->hash);
180 
181 	/* get a reference for our hashtable */
182 	dlm_lockres_get(res);
183 
184 	hlist_add_head(&res->hash_node, bucket);
185 
186 	mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
187 	     res->lockname.name);
188 }
189 
__dlm_lookup_lockres_full(struct dlm_ctxt * dlm,const char * name,unsigned int len,unsigned int hash)190 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
191 						     const char *name,
192 						     unsigned int len,
193 						     unsigned int hash)
194 {
195 	struct hlist_head *bucket;
196 	struct hlist_node *list;
197 
198 	mlog(0, "%.*s\n", len, name);
199 
200 	assert_spin_locked(&dlm->spinlock);
201 
202 	bucket = dlm_lockres_hash(dlm, hash);
203 
204 	hlist_for_each(list, bucket) {
205 		struct dlm_lock_resource *res = hlist_entry(list,
206 			struct dlm_lock_resource, hash_node);
207 		if (res->lockname.name[0] != name[0])
208 			continue;
209 		if (unlikely(res->lockname.len != len))
210 			continue;
211 		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
212 			continue;
213 		dlm_lockres_get(res);
214 		return res;
215 	}
216 	return NULL;
217 }
218 
219 /* intended to be called by functions which do not care about lock
220  * resources which are being purged (most net _handler functions).
221  * this will return NULL for any lock resource which is found but
222  * currently in the process of dropping its mastery reference.
223  * use __dlm_lookup_lockres_full when you need the lock resource
224  * regardless (e.g. dlm_get_lock_resource) */
__dlm_lookup_lockres(struct dlm_ctxt * dlm,const char * name,unsigned int len,unsigned int hash)225 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
226 						const char *name,
227 						unsigned int len,
228 						unsigned int hash)
229 {
230 	struct dlm_lock_resource *res = NULL;
231 
232 	mlog(0, "%.*s\n", len, name);
233 
234 	assert_spin_locked(&dlm->spinlock);
235 
236 	res = __dlm_lookup_lockres_full(dlm, name, len, hash);
237 	if (res) {
238 		spin_lock(&res->spinlock);
239 		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
240 			spin_unlock(&res->spinlock);
241 			dlm_lockres_put(res);
242 			return NULL;
243 		}
244 		spin_unlock(&res->spinlock);
245 	}
246 
247 	return res;
248 }
249 
dlm_lookup_lockres(struct dlm_ctxt * dlm,const char * name,unsigned int len)250 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
251 				    const char *name,
252 				    unsigned int len)
253 {
254 	struct dlm_lock_resource *res;
255 	unsigned int hash = dlm_lockid_hash(name, len);
256 
257 	spin_lock(&dlm->spinlock);
258 	res = __dlm_lookup_lockres(dlm, name, len, hash);
259 	spin_unlock(&dlm->spinlock);
260 	return res;
261 }
262 
__dlm_lookup_domain_full(const char * domain,int len)263 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
264 {
265 	struct dlm_ctxt *tmp = NULL;
266 	struct list_head *iter;
267 
268 	assert_spin_locked(&dlm_domain_lock);
269 
270 	/* tmp->name here is always NULL terminated,
271 	 * but domain may not be! */
272 	list_for_each(iter, &dlm_domains) {
273 		tmp = list_entry (iter, struct dlm_ctxt, list);
274 		if (strlen(tmp->name) == len &&
275 		    memcmp(tmp->name, domain, len)==0)
276 			break;
277 		tmp = NULL;
278 	}
279 
280 	return tmp;
281 }
282 
283 /* For null terminated domain strings ONLY */
__dlm_lookup_domain(const char * domain)284 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
285 {
286 	assert_spin_locked(&dlm_domain_lock);
287 
288 	return __dlm_lookup_domain_full(domain, strlen(domain));
289 }
290 
291 
292 /* returns true on one of two conditions:
293  * 1) the domain does not exist
294  * 2) the domain exists and it's state is "joined" */
dlm_wait_on_domain_helper(const char * domain)295 static int dlm_wait_on_domain_helper(const char *domain)
296 {
297 	int ret = 0;
298 	struct dlm_ctxt *tmp = NULL;
299 
300 	spin_lock(&dlm_domain_lock);
301 
302 	tmp = __dlm_lookup_domain(domain);
303 	if (!tmp)
304 		ret = 1;
305 	else if (tmp->dlm_state == DLM_CTXT_JOINED)
306 		ret = 1;
307 
308 	spin_unlock(&dlm_domain_lock);
309 	return ret;
310 }
311 
dlm_free_ctxt_mem(struct dlm_ctxt * dlm)312 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
313 {
314 	dlm_destroy_debugfs_subroot(dlm);
315 
316 	if (dlm->lockres_hash)
317 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
318 
319 	if (dlm->master_hash)
320 		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
321 
322 	if (dlm->name)
323 		kfree(dlm->name);
324 
325 	kfree(dlm);
326 }
327 
328 /* A little strange - this function will be called while holding
329  * dlm_domain_lock and is expected to be holding it on the way out. We
330  * will however drop and reacquire it multiple times */
dlm_ctxt_release(struct kref * kref)331 static void dlm_ctxt_release(struct kref *kref)
332 {
333 	struct dlm_ctxt *dlm;
334 
335 	dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
336 
337 	BUG_ON(dlm->num_joins);
338 	BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
339 
340 	/* we may still be in the list if we hit an error during join. */
341 	list_del_init(&dlm->list);
342 
343 	spin_unlock(&dlm_domain_lock);
344 
345 	mlog(0, "freeing memory from domain %s\n", dlm->name);
346 
347 	wake_up(&dlm_domain_events);
348 
349 	dlm_free_ctxt_mem(dlm);
350 
351 	spin_lock(&dlm_domain_lock);
352 }
353 
dlm_put(struct dlm_ctxt * dlm)354 void dlm_put(struct dlm_ctxt *dlm)
355 {
356 	spin_lock(&dlm_domain_lock);
357 	kref_put(&dlm->dlm_refs, dlm_ctxt_release);
358 	spin_unlock(&dlm_domain_lock);
359 }
360 
__dlm_get(struct dlm_ctxt * dlm)361 static void __dlm_get(struct dlm_ctxt *dlm)
362 {
363 	kref_get(&dlm->dlm_refs);
364 }
365 
366 /* given a questionable reference to a dlm object, gets a reference if
367  * it can find it in the list, otherwise returns NULL in which case
368  * you shouldn't trust your pointer. */
dlm_grab(struct dlm_ctxt * dlm)369 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
370 {
371 	struct list_head *iter;
372 	struct dlm_ctxt *target = NULL;
373 
374 	spin_lock(&dlm_domain_lock);
375 
376 	list_for_each(iter, &dlm_domains) {
377 		target = list_entry (iter, struct dlm_ctxt, list);
378 
379 		if (target == dlm) {
380 			__dlm_get(target);
381 			break;
382 		}
383 
384 		target = NULL;
385 	}
386 
387 	spin_unlock(&dlm_domain_lock);
388 
389 	return target;
390 }
391 
dlm_domain_fully_joined(struct dlm_ctxt * dlm)392 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
393 {
394 	int ret;
395 
396 	spin_lock(&dlm_domain_lock);
397 	ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
398 		(dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
399 	spin_unlock(&dlm_domain_lock);
400 
401 	return ret;
402 }
403 
dlm_destroy_dlm_worker(struct dlm_ctxt * dlm)404 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
405 {
406 	if (dlm->dlm_worker) {
407 		flush_workqueue(dlm->dlm_worker);
408 		destroy_workqueue(dlm->dlm_worker);
409 		dlm->dlm_worker = NULL;
410 	}
411 }
412 
dlm_complete_dlm_shutdown(struct dlm_ctxt * dlm)413 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
414 {
415 	dlm_unregister_domain_handlers(dlm);
416 	dlm_debug_shutdown(dlm);
417 	dlm_complete_thread(dlm);
418 	dlm_complete_recovery_thread(dlm);
419 	dlm_destroy_dlm_worker(dlm);
420 
421 	/* We've left the domain. Now we can take ourselves out of the
422 	 * list and allow the kref stuff to help us free the
423 	 * memory. */
424 	spin_lock(&dlm_domain_lock);
425 	list_del_init(&dlm->list);
426 	spin_unlock(&dlm_domain_lock);
427 
428 	/* Wake up anyone waiting for us to remove this domain */
429 	wake_up(&dlm_domain_events);
430 }
431 
dlm_migrate_all_locks(struct dlm_ctxt * dlm)432 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
433 {
434 	int i, num, n, ret = 0;
435 	struct dlm_lock_resource *res;
436 	struct hlist_node *iter;
437 	struct hlist_head *bucket;
438 	int dropped;
439 
440 	mlog(0, "Migrating locks from domain %s\n", dlm->name);
441 
442 	num = 0;
443 	spin_lock(&dlm->spinlock);
444 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
445 redo_bucket:
446 		n = 0;
447 		bucket = dlm_lockres_hash(dlm, i);
448 		iter = bucket->first;
449 		while (iter) {
450 			n++;
451 			res = hlist_entry(iter, struct dlm_lock_resource,
452 					  hash_node);
453 			dlm_lockres_get(res);
454 			/* migrate, if necessary.  this will drop the dlm
455 			 * spinlock and retake it if it does migration. */
456 			dropped = dlm_empty_lockres(dlm, res);
457 
458 			spin_lock(&res->spinlock);
459 			if (dropped)
460 				__dlm_lockres_calc_usage(dlm, res);
461 			else
462 				iter = res->hash_node.next;
463 			spin_unlock(&res->spinlock);
464 
465 			dlm_lockres_put(res);
466 
467 			if (dropped) {
468 				cond_resched_lock(&dlm->spinlock);
469 				goto redo_bucket;
470 			}
471 		}
472 		cond_resched_lock(&dlm->spinlock);
473 		num += n;
474 	}
475 	spin_unlock(&dlm->spinlock);
476 	wake_up(&dlm->dlm_thread_wq);
477 
478 	/* let the dlm thread take care of purging, keep scanning until
479 	 * nothing remains in the hash */
480 	if (num) {
481 		mlog(0, "%s: %d lock resources in hash last pass\n",
482 		     dlm->name, num);
483 		ret = -EAGAIN;
484 	}
485 	mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
486 	return ret;
487 }
488 
dlm_no_joining_node(struct dlm_ctxt * dlm)489 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
490 {
491 	int ret;
492 
493 	spin_lock(&dlm->spinlock);
494 	ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
495 	spin_unlock(&dlm->spinlock);
496 
497 	return ret;
498 }
499 
dlm_begin_exit_domain_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)500 static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
501 					 void *data, void **ret_data)
502 {
503 	struct dlm_ctxt *dlm = data;
504 	unsigned int node;
505 	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
506 
507 	if (!dlm_grab(dlm))
508 		return 0;
509 
510 	node = exit_msg->node_idx;
511 	mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
512 
513 	spin_lock(&dlm->spinlock);
514 	set_bit(node, dlm->exit_domain_map);
515 	spin_unlock(&dlm->spinlock);
516 
517 	dlm_put(dlm);
518 
519 	return 0;
520 }
521 
dlm_mark_domain_leaving(struct dlm_ctxt * dlm)522 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
523 {
524 	/* Yikes, a double spinlock! I need domain_lock for the dlm
525 	 * state and the dlm spinlock for join state... Sorry! */
526 again:
527 	spin_lock(&dlm_domain_lock);
528 	spin_lock(&dlm->spinlock);
529 
530 	if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
531 		mlog(0, "Node %d is joining, we wait on it.\n",
532 			  dlm->joining_node);
533 		spin_unlock(&dlm->spinlock);
534 		spin_unlock(&dlm_domain_lock);
535 
536 		wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
537 		goto again;
538 	}
539 
540 	dlm->dlm_state = DLM_CTXT_LEAVING;
541 	spin_unlock(&dlm->spinlock);
542 	spin_unlock(&dlm_domain_lock);
543 }
544 
__dlm_print_nodes(struct dlm_ctxt * dlm)545 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
546 {
547 	int node = -1, num = 0;
548 
549 	assert_spin_locked(&dlm->spinlock);
550 
551 	printk("( ");
552 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
553 				     node + 1)) < O2NM_MAX_NODES) {
554 		printk("%d ", node);
555 		++num;
556 	}
557 	printk(") %u nodes\n", num);
558 }
559 
dlm_exit_domain_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)560 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
561 				   void **ret_data)
562 {
563 	struct dlm_ctxt *dlm = data;
564 	unsigned int node;
565 	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
566 
567 	mlog(0, "%p %u %p", msg, len, data);
568 
569 	if (!dlm_grab(dlm))
570 		return 0;
571 
572 	node = exit_msg->node_idx;
573 
574 	spin_lock(&dlm->spinlock);
575 	clear_bit(node, dlm->domain_map);
576 	clear_bit(node, dlm->exit_domain_map);
577 	printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
578 	__dlm_print_nodes(dlm);
579 
580 	/* notify anything attached to the heartbeat events */
581 	dlm_hb_event_notify_attached(dlm, node, 0);
582 
583 	spin_unlock(&dlm->spinlock);
584 
585 	dlm_put(dlm);
586 
587 	return 0;
588 }
589 
dlm_send_one_domain_exit(struct dlm_ctxt * dlm,u32 msg_type,unsigned int node)590 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
591 				    unsigned int node)
592 {
593 	int status;
594 	struct dlm_exit_domain leave_msg;
595 
596 	mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
597 	     msg_type, node);
598 
599 	memset(&leave_msg, 0, sizeof(leave_msg));
600 	leave_msg.node_idx = dlm->node_num;
601 
602 	status = o2net_send_message(msg_type, dlm->key, &leave_msg,
603 				    sizeof(leave_msg), node, NULL);
604 	if (status < 0)
605 		mlog(ML_ERROR, "Error %d sending domain exit message %u "
606 		     "to node %u on domain %s\n", status, msg_type, node,
607 		     dlm->name);
608 
609 	return status;
610 }
611 
dlm_begin_exit_domain(struct dlm_ctxt * dlm)612 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
613 {
614 	int node = -1;
615 
616 	/* Support for begin exit domain was added in 1.2 */
617 	if (dlm->dlm_locking_proto.pv_major == 1 &&
618 	    dlm->dlm_locking_proto.pv_minor < 2)
619 		return;
620 
621 	/*
622 	 * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
623 	 * informational. Meaning if a node does not receive the message,
624 	 * so be it.
625 	 */
626 	spin_lock(&dlm->spinlock);
627 	while (1) {
628 		node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
629 		if (node >= O2NM_MAX_NODES)
630 			break;
631 		if (node == dlm->node_num)
632 			continue;
633 
634 		spin_unlock(&dlm->spinlock);
635 		dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
636 		spin_lock(&dlm->spinlock);
637 	}
638 	spin_unlock(&dlm->spinlock);
639 }
640 
dlm_leave_domain(struct dlm_ctxt * dlm)641 static void dlm_leave_domain(struct dlm_ctxt *dlm)
642 {
643 	int node, clear_node, status;
644 
645 	/* At this point we've migrated away all our locks and won't
646 	 * accept mastership of new ones. The dlm is responsible for
647 	 * almost nothing now. We make sure not to confuse any joining
648 	 * nodes and then commence shutdown procedure. */
649 
650 	spin_lock(&dlm->spinlock);
651 	/* Clear ourselves from the domain map */
652 	clear_bit(dlm->node_num, dlm->domain_map);
653 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
654 				     0)) < O2NM_MAX_NODES) {
655 		/* Drop the dlm spinlock. This is safe wrt the domain_map.
656 		 * -nodes cannot be added now as the
657 		 *   query_join_handlers knows to respond with OK_NO_MAP
658 		 * -we catch the right network errors if a node is
659 		 *   removed from the map while we're sending him the
660 		 *   exit message. */
661 		spin_unlock(&dlm->spinlock);
662 
663 		clear_node = 1;
664 
665 		status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
666 						  node);
667 		if (status < 0 &&
668 		    status != -ENOPROTOOPT &&
669 		    status != -ENOTCONN) {
670 			mlog(ML_NOTICE, "Error %d sending domain exit message "
671 			     "to node %d\n", status, node);
672 
673 			/* Not sure what to do here but lets sleep for
674 			 * a bit in case this was a transient
675 			 * error... */
676 			msleep(DLM_DOMAIN_BACKOFF_MS);
677 			clear_node = 0;
678 		}
679 
680 		spin_lock(&dlm->spinlock);
681 		/* If we're not clearing the node bit then we intend
682 		 * to loop back around to try again. */
683 		if (clear_node)
684 			clear_bit(node, dlm->domain_map);
685 	}
686 	spin_unlock(&dlm->spinlock);
687 }
688 
dlm_joined(struct dlm_ctxt * dlm)689 int dlm_joined(struct dlm_ctxt *dlm)
690 {
691 	int ret = 0;
692 
693 	spin_lock(&dlm_domain_lock);
694 
695 	if (dlm->dlm_state == DLM_CTXT_JOINED)
696 		ret = 1;
697 
698 	spin_unlock(&dlm_domain_lock);
699 
700 	return ret;
701 }
702 
dlm_shutting_down(struct dlm_ctxt * dlm)703 int dlm_shutting_down(struct dlm_ctxt *dlm)
704 {
705 	int ret = 0;
706 
707 	spin_lock(&dlm_domain_lock);
708 
709 	if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
710 		ret = 1;
711 
712 	spin_unlock(&dlm_domain_lock);
713 
714 	return ret;
715 }
716 
dlm_unregister_domain(struct dlm_ctxt * dlm)717 void dlm_unregister_domain(struct dlm_ctxt *dlm)
718 {
719 	int leave = 0;
720 	struct dlm_lock_resource *res;
721 
722 	spin_lock(&dlm_domain_lock);
723 	BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
724 	BUG_ON(!dlm->num_joins);
725 
726 	dlm->num_joins--;
727 	if (!dlm->num_joins) {
728 		/* We mark it "in shutdown" now so new register
729 		 * requests wait until we've completely left the
730 		 * domain. Don't use DLM_CTXT_LEAVING yet as we still
731 		 * want new domain joins to communicate with us at
732 		 * least until we've completed migration of our
733 		 * resources. */
734 		dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
735 		leave = 1;
736 	}
737 	spin_unlock(&dlm_domain_lock);
738 
739 	if (leave) {
740 		mlog(0, "shutting down domain %s\n", dlm->name);
741 		dlm_begin_exit_domain(dlm);
742 
743 		/* We changed dlm state, notify the thread */
744 		dlm_kick_thread(dlm, NULL);
745 
746 		while (dlm_migrate_all_locks(dlm)) {
747 			/* Give dlm_thread time to purge the lockres' */
748 			msleep(500);
749 			mlog(0, "%s: more migration to do\n", dlm->name);
750 		}
751 
752 		/* This list should be empty. If not, print remaining lockres */
753 		if (!list_empty(&dlm->tracking_list)) {
754 			mlog(ML_ERROR, "Following lockres' are still on the "
755 			     "tracking list:\n");
756 			list_for_each_entry(res, &dlm->tracking_list, tracking)
757 				dlm_print_one_lock_resource(res);
758 		}
759 
760 		dlm_mark_domain_leaving(dlm);
761 		dlm_leave_domain(dlm);
762 		printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
763 		dlm_force_free_mles(dlm);
764 		dlm_complete_dlm_shutdown(dlm);
765 	}
766 	dlm_put(dlm);
767 }
768 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
769 
dlm_query_join_proto_check(char * proto_type,int node,struct dlm_protocol_version * ours,struct dlm_protocol_version * request)770 static int dlm_query_join_proto_check(char *proto_type, int node,
771 				      struct dlm_protocol_version *ours,
772 				      struct dlm_protocol_version *request)
773 {
774 	int rc;
775 	struct dlm_protocol_version proto = *request;
776 
777 	if (!dlm_protocol_compare(ours, &proto)) {
778 		mlog(0,
779 		     "node %u wanted to join with %s locking protocol "
780 		     "%u.%u, we respond with %u.%u\n",
781 		     node, proto_type,
782 		     request->pv_major,
783 		     request->pv_minor,
784 		     proto.pv_major, proto.pv_minor);
785 		request->pv_minor = proto.pv_minor;
786 		rc = 0;
787 	} else {
788 		mlog(ML_NOTICE,
789 		     "Node %u wanted to join with %s locking "
790 		     "protocol %u.%u, but we have %u.%u, disallowing\n",
791 		     node, proto_type,
792 		     request->pv_major,
793 		     request->pv_minor,
794 		     ours->pv_major,
795 		     ours->pv_minor);
796 		rc = 1;
797 	}
798 
799 	return rc;
800 }
801 
802 /*
803  * struct dlm_query_join_packet is made up of four one-byte fields.  They
804  * are effectively in big-endian order already.  However, little-endian
805  * machines swap them before putting the packet on the wire (because
806  * query_join's response is a status, and that status is treated as a u32
807  * on the wire).  Thus, a big-endian and little-endian machines will treat
808  * this structure differently.
809  *
810  * The solution is to have little-endian machines swap the structure when
811  * converting from the structure to the u32 representation.  This will
812  * result in the structure having the correct format on the wire no matter
813  * the host endian format.
814  */
dlm_query_join_packet_to_wire(struct dlm_query_join_packet * packet,u32 * wire)815 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
816 					  u32 *wire)
817 {
818 	union dlm_query_join_response response;
819 
820 	response.packet = *packet;
821 	*wire = cpu_to_be32(response.intval);
822 }
823 
dlm_query_join_wire_to_packet(u32 wire,struct dlm_query_join_packet * packet)824 static void dlm_query_join_wire_to_packet(u32 wire,
825 					  struct dlm_query_join_packet *packet)
826 {
827 	union dlm_query_join_response response;
828 
829 	response.intval = cpu_to_be32(wire);
830 	*packet = response.packet;
831 }
832 
dlm_query_join_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)833 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
834 				  void **ret_data)
835 {
836 	struct dlm_query_join_request *query;
837 	struct dlm_query_join_packet packet = {
838 		.code = JOIN_DISALLOW,
839 	};
840 	struct dlm_ctxt *dlm = NULL;
841 	u32 response;
842 	u8 nodenum;
843 
844 	query = (struct dlm_query_join_request *) msg->buf;
845 
846 	mlog(0, "node %u wants to join domain %s\n", query->node_idx,
847 		  query->domain);
848 
849 	/*
850 	 * If heartbeat doesn't consider the node live, tell it
851 	 * to back off and try again.  This gives heartbeat a chance
852 	 * to catch up.
853 	 */
854 	if (!o2hb_check_node_heartbeating(query->node_idx)) {
855 		mlog(0, "node %u is not in our live map yet\n",
856 		     query->node_idx);
857 
858 		packet.code = JOIN_DISALLOW;
859 		goto respond;
860 	}
861 
862 	packet.code = JOIN_OK_NO_MAP;
863 
864 	spin_lock(&dlm_domain_lock);
865 	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
866 	if (!dlm)
867 		goto unlock_respond;
868 
869 	/*
870 	 * There is a small window where the joining node may not see the
871 	 * node(s) that just left but still part of the cluster. DISALLOW
872 	 * join request if joining node has different node map.
873 	 */
874 	nodenum=0;
875 	while (nodenum < O2NM_MAX_NODES) {
876 		if (test_bit(nodenum, dlm->domain_map)) {
877 			if (!byte_test_bit(nodenum, query->node_map)) {
878 				mlog(0, "disallow join as node %u does not "
879 				     "have node %u in its nodemap\n",
880 				     query->node_idx, nodenum);
881 				packet.code = JOIN_DISALLOW;
882 				goto unlock_respond;
883 			}
884 		}
885 		nodenum++;
886 	}
887 
888 	/* Once the dlm ctxt is marked as leaving then we don't want
889 	 * to be put in someone's domain map.
890 	 * Also, explicitly disallow joining at certain troublesome
891 	 * times (ie. during recovery). */
892 	if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
893 		int bit = query->node_idx;
894 		spin_lock(&dlm->spinlock);
895 
896 		if (dlm->dlm_state == DLM_CTXT_NEW &&
897 		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
898 			/*If this is a brand new context and we
899 			 * haven't started our join process yet, then
900 			 * the other node won the race. */
901 			packet.code = JOIN_OK_NO_MAP;
902 		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
903 			/* Disallow parallel joins. */
904 			packet.code = JOIN_DISALLOW;
905 		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
906 			mlog(0, "node %u trying to join, but recovery "
907 			     "is ongoing.\n", bit);
908 			packet.code = JOIN_DISALLOW;
909 		} else if (test_bit(bit, dlm->recovery_map)) {
910 			mlog(0, "node %u trying to join, but it "
911 			     "still needs recovery.\n", bit);
912 			packet.code = JOIN_DISALLOW;
913 		} else if (test_bit(bit, dlm->domain_map)) {
914 			mlog(0, "node %u trying to join, but it "
915 			     "is still in the domain! needs recovery?\n",
916 			     bit);
917 			packet.code = JOIN_DISALLOW;
918 		} else {
919 			/* Alright we're fully a part of this domain
920 			 * so we keep some state as to who's joining
921 			 * and indicate to him that needs to be fixed
922 			 * up. */
923 
924 			/* Make sure we speak compatible locking protocols.  */
925 			if (dlm_query_join_proto_check("DLM", bit,
926 						       &dlm->dlm_locking_proto,
927 						       &query->dlm_proto)) {
928 				packet.code = JOIN_PROTOCOL_MISMATCH;
929 			} else if (dlm_query_join_proto_check("fs", bit,
930 							      &dlm->fs_locking_proto,
931 							      &query->fs_proto)) {
932 				packet.code = JOIN_PROTOCOL_MISMATCH;
933 			} else {
934 				packet.dlm_minor = query->dlm_proto.pv_minor;
935 				packet.fs_minor = query->fs_proto.pv_minor;
936 				packet.code = JOIN_OK;
937 				__dlm_set_joining_node(dlm, query->node_idx);
938 			}
939 		}
940 
941 		spin_unlock(&dlm->spinlock);
942 	}
943 unlock_respond:
944 	spin_unlock(&dlm_domain_lock);
945 
946 respond:
947 	mlog(0, "We respond with %u\n", packet.code);
948 
949 	dlm_query_join_packet_to_wire(&packet, &response);
950 	return response;
951 }
952 
dlm_assert_joined_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)953 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
954 				     void **ret_data)
955 {
956 	struct dlm_assert_joined *assert;
957 	struct dlm_ctxt *dlm = NULL;
958 
959 	assert = (struct dlm_assert_joined *) msg->buf;
960 
961 	mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
962 		  assert->domain);
963 
964 	spin_lock(&dlm_domain_lock);
965 	dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
966 	/* XXX should we consider no dlm ctxt an error? */
967 	if (dlm) {
968 		spin_lock(&dlm->spinlock);
969 
970 		/* Alright, this node has officially joined our
971 		 * domain. Set him in the map and clean up our
972 		 * leftover join state. */
973 		BUG_ON(dlm->joining_node != assert->node_idx);
974 		set_bit(assert->node_idx, dlm->domain_map);
975 		clear_bit(assert->node_idx, dlm->exit_domain_map);
976 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
977 
978 		printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
979 		       assert->node_idx, dlm->name);
980 		__dlm_print_nodes(dlm);
981 
982 		/* notify anything attached to the heartbeat events */
983 		dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
984 
985 		spin_unlock(&dlm->spinlock);
986 	}
987 	spin_unlock(&dlm_domain_lock);
988 
989 	return 0;
990 }
991 
dlm_match_regions(struct dlm_ctxt * dlm,struct dlm_query_region * qr,char * local,int locallen)992 static int dlm_match_regions(struct dlm_ctxt *dlm,
993 			     struct dlm_query_region *qr,
994 			     char *local, int locallen)
995 {
996 	char *remote = qr->qr_regions;
997 	char *l, *r;
998 	int localnr, i, j, foundit;
999 	int status = 0;
1000 
1001 	if (!o2hb_global_heartbeat_active()) {
1002 		if (qr->qr_numregions) {
1003 			mlog(ML_ERROR, "Domain %s: Joining node %d has global "
1004 			     "heartbeat enabled but local node %d does not\n",
1005 			     qr->qr_domain, qr->qr_node, dlm->node_num);
1006 			status = -EINVAL;
1007 		}
1008 		goto bail;
1009 	}
1010 
1011 	if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
1012 		mlog(ML_ERROR, "Domain %s: Local node %d has global "
1013 		     "heartbeat enabled but joining node %d does not\n",
1014 		     qr->qr_domain, dlm->node_num, qr->qr_node);
1015 		status = -EINVAL;
1016 		goto bail;
1017 	}
1018 
1019 	r = remote;
1020 	for (i = 0; i < qr->qr_numregions; ++i) {
1021 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1022 		r += O2HB_MAX_REGION_NAME_LEN;
1023 	}
1024 
1025 	localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1026 	localnr = o2hb_get_all_regions(local, (u8)localnr);
1027 
1028 	/* compare local regions with remote */
1029 	l = local;
1030 	for (i = 0; i < localnr; ++i) {
1031 		foundit = 0;
1032 		r = remote;
1033 		for (j = 0; j <= qr->qr_numregions; ++j) {
1034 			if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1035 				foundit = 1;
1036 				break;
1037 			}
1038 			r += O2HB_MAX_REGION_NAME_LEN;
1039 		}
1040 		if (!foundit) {
1041 			status = -EINVAL;
1042 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1043 			     "in local node %d but not in joining node %d\n",
1044 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1045 			     dlm->node_num, qr->qr_node);
1046 			goto bail;
1047 		}
1048 		l += O2HB_MAX_REGION_NAME_LEN;
1049 	}
1050 
1051 	/* compare remote with local regions */
1052 	r = remote;
1053 	for (i = 0; i < qr->qr_numregions; ++i) {
1054 		foundit = 0;
1055 		l = local;
1056 		for (j = 0; j < localnr; ++j) {
1057 			if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1058 				foundit = 1;
1059 				break;
1060 			}
1061 			l += O2HB_MAX_REGION_NAME_LEN;
1062 		}
1063 		if (!foundit) {
1064 			status = -EINVAL;
1065 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1066 			     "in joining node %d but not in local node %d\n",
1067 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1068 			     qr->qr_node, dlm->node_num);
1069 			goto bail;
1070 		}
1071 		r += O2HB_MAX_REGION_NAME_LEN;
1072 	}
1073 
1074 bail:
1075 	return status;
1076 }
1077 
dlm_send_regions(struct dlm_ctxt * dlm,unsigned long * node_map)1078 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1079 {
1080 	struct dlm_query_region *qr = NULL;
1081 	int status, ret = 0, i;
1082 	char *p;
1083 
1084 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1085 		goto bail;
1086 
1087 	qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1088 	if (!qr) {
1089 		ret = -ENOMEM;
1090 		mlog_errno(ret);
1091 		goto bail;
1092 	}
1093 
1094 	qr->qr_node = dlm->node_num;
1095 	qr->qr_namelen = strlen(dlm->name);
1096 	memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1097 	/* if local hb, the numregions will be zero */
1098 	if (o2hb_global_heartbeat_active())
1099 		qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1100 							 O2NM_MAX_REGIONS);
1101 
1102 	p = qr->qr_regions;
1103 	for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1104 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1105 
1106 	i = -1;
1107 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1108 				  i + 1)) < O2NM_MAX_NODES) {
1109 		if (i == dlm->node_num)
1110 			continue;
1111 
1112 		mlog(0, "Sending regions to node %d\n", i);
1113 
1114 		ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1115 					 sizeof(struct dlm_query_region),
1116 					 i, &status);
1117 		if (ret >= 0)
1118 			ret = status;
1119 		if (ret) {
1120 			mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1121 			     ret, i);
1122 			break;
1123 		}
1124 	}
1125 
1126 bail:
1127 	kfree(qr);
1128 	return ret;
1129 }
1130 
dlm_query_region_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)1131 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1132 				    void *data, void **ret_data)
1133 {
1134 	struct dlm_query_region *qr;
1135 	struct dlm_ctxt *dlm = NULL;
1136 	char *local = NULL;
1137 	int status = 0;
1138 	int locked = 0;
1139 
1140 	qr = (struct dlm_query_region *) msg->buf;
1141 
1142 	mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1143 	     qr->qr_domain);
1144 
1145 	/* buffer used in dlm_mast_regions() */
1146 	local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1147 	if (!local) {
1148 		status = -ENOMEM;
1149 		goto bail;
1150 	}
1151 
1152 	status = -EINVAL;
1153 
1154 	spin_lock(&dlm_domain_lock);
1155 	dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1156 	if (!dlm) {
1157 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1158 		     "before join domain\n", qr->qr_node, qr->qr_domain);
1159 		goto bail;
1160 	}
1161 
1162 	spin_lock(&dlm->spinlock);
1163 	locked = 1;
1164 	if (dlm->joining_node != qr->qr_node) {
1165 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1166 		     "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1167 		     dlm->joining_node);
1168 		goto bail;
1169 	}
1170 
1171 	/* Support for global heartbeat was added in 1.1 */
1172 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1173 	    dlm->dlm_locking_proto.pv_minor == 0) {
1174 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1175 		     "but active dlm protocol is %d.%d\n", qr->qr_node,
1176 		     qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1177 		     dlm->dlm_locking_proto.pv_minor);
1178 		goto bail;
1179 	}
1180 
1181 	status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1182 
1183 bail:
1184 	if (locked)
1185 		spin_unlock(&dlm->spinlock);
1186 	spin_unlock(&dlm_domain_lock);
1187 
1188 	kfree(local);
1189 
1190 	return status;
1191 }
1192 
dlm_match_nodes(struct dlm_ctxt * dlm,struct dlm_query_nodeinfo * qn)1193 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1194 {
1195 	struct o2nm_node *local;
1196 	struct dlm_node_info *remote;
1197 	int i, j;
1198 	int status = 0;
1199 
1200 	for (j = 0; j < qn->qn_numnodes; ++j)
1201 		mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1202 		     &(qn->qn_nodes[j].ni_ipv4_address),
1203 		     ntohs(qn->qn_nodes[j].ni_ipv4_port));
1204 
1205 	for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1206 		local = o2nm_get_node_by_num(i);
1207 		remote = NULL;
1208 		for (j = 0; j < qn->qn_numnodes; ++j) {
1209 			if (qn->qn_nodes[j].ni_nodenum == i) {
1210 				remote = &(qn->qn_nodes[j]);
1211 				break;
1212 			}
1213 		}
1214 
1215 		if (!local && !remote)
1216 			continue;
1217 
1218 		if ((local && !remote) || (!local && remote))
1219 			status = -EINVAL;
1220 
1221 		if (!status &&
1222 		    ((remote->ni_nodenum != local->nd_num) ||
1223 		     (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1224 		     (remote->ni_ipv4_address != local->nd_ipv4_address)))
1225 			status = -EINVAL;
1226 
1227 		if (status) {
1228 			if (remote && !local)
1229 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1230 				     "registered in joining node %d but not in "
1231 				     "local node %d\n", qn->qn_domain,
1232 				     remote->ni_nodenum,
1233 				     &(remote->ni_ipv4_address),
1234 				     ntohs(remote->ni_ipv4_port),
1235 				     qn->qn_nodenum, dlm->node_num);
1236 			if (local && !remote)
1237 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1238 				     "registered in local node %d but not in "
1239 				     "joining node %d\n", qn->qn_domain,
1240 				     local->nd_num, &(local->nd_ipv4_address),
1241 				     ntohs(local->nd_ipv4_port),
1242 				     dlm->node_num, qn->qn_nodenum);
1243 			BUG_ON((!local && !remote));
1244 		}
1245 
1246 		if (local)
1247 			o2nm_node_put(local);
1248 	}
1249 
1250 	return status;
1251 }
1252 
dlm_send_nodeinfo(struct dlm_ctxt * dlm,unsigned long * node_map)1253 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1254 {
1255 	struct dlm_query_nodeinfo *qn = NULL;
1256 	struct o2nm_node *node;
1257 	int ret = 0, status, count, i;
1258 
1259 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1260 		goto bail;
1261 
1262 	qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1263 	if (!qn) {
1264 		ret = -ENOMEM;
1265 		mlog_errno(ret);
1266 		goto bail;
1267 	}
1268 
1269 	for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1270 		node = o2nm_get_node_by_num(i);
1271 		if (!node)
1272 			continue;
1273 		qn->qn_nodes[count].ni_nodenum = node->nd_num;
1274 		qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1275 		qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1276 		mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1277 		     &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1278 		++count;
1279 		o2nm_node_put(node);
1280 	}
1281 
1282 	qn->qn_nodenum = dlm->node_num;
1283 	qn->qn_numnodes = count;
1284 	qn->qn_namelen = strlen(dlm->name);
1285 	memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1286 
1287 	i = -1;
1288 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1289 				  i + 1)) < O2NM_MAX_NODES) {
1290 		if (i == dlm->node_num)
1291 			continue;
1292 
1293 		mlog(0, "Sending nodeinfo to node %d\n", i);
1294 
1295 		ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1296 					 qn, sizeof(struct dlm_query_nodeinfo),
1297 					 i, &status);
1298 		if (ret >= 0)
1299 			ret = status;
1300 		if (ret) {
1301 			mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1302 			break;
1303 		}
1304 	}
1305 
1306 bail:
1307 	kfree(qn);
1308 	return ret;
1309 }
1310 
dlm_query_nodeinfo_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)1311 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1312 				      void *data, void **ret_data)
1313 {
1314 	struct dlm_query_nodeinfo *qn;
1315 	struct dlm_ctxt *dlm = NULL;
1316 	int locked = 0, status = -EINVAL;
1317 
1318 	qn = (struct dlm_query_nodeinfo *) msg->buf;
1319 
1320 	mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1321 	     qn->qn_domain);
1322 
1323 	spin_lock(&dlm_domain_lock);
1324 	dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1325 	if (!dlm) {
1326 		mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1327 		     "join domain\n", qn->qn_nodenum, qn->qn_domain);
1328 		goto bail;
1329 	}
1330 
1331 	spin_lock(&dlm->spinlock);
1332 	locked = 1;
1333 	if (dlm->joining_node != qn->qn_nodenum) {
1334 		mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1335 		     "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1336 		     dlm->joining_node);
1337 		goto bail;
1338 	}
1339 
1340 	/* Support for node query was added in 1.1 */
1341 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1342 	    dlm->dlm_locking_proto.pv_minor == 0) {
1343 		mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1344 		     "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1345 		     qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1346 		     dlm->dlm_locking_proto.pv_minor);
1347 		goto bail;
1348 	}
1349 
1350 	status = dlm_match_nodes(dlm, qn);
1351 
1352 bail:
1353 	if (locked)
1354 		spin_unlock(&dlm->spinlock);
1355 	spin_unlock(&dlm_domain_lock);
1356 
1357 	return status;
1358 }
1359 
dlm_cancel_join_handler(struct o2net_msg * msg,u32 len,void * data,void ** ret_data)1360 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1361 				   void **ret_data)
1362 {
1363 	struct dlm_cancel_join *cancel;
1364 	struct dlm_ctxt *dlm = NULL;
1365 
1366 	cancel = (struct dlm_cancel_join *) msg->buf;
1367 
1368 	mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1369 		  cancel->domain);
1370 
1371 	spin_lock(&dlm_domain_lock);
1372 	dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1373 
1374 	if (dlm) {
1375 		spin_lock(&dlm->spinlock);
1376 
1377 		/* Yikes, this guy wants to cancel his join. No
1378 		 * problem, we simply cleanup our join state. */
1379 		BUG_ON(dlm->joining_node != cancel->node_idx);
1380 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1381 
1382 		spin_unlock(&dlm->spinlock);
1383 	}
1384 	spin_unlock(&dlm_domain_lock);
1385 
1386 	return 0;
1387 }
1388 
dlm_send_one_join_cancel(struct dlm_ctxt * dlm,unsigned int node)1389 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1390 				    unsigned int node)
1391 {
1392 	int status;
1393 	struct dlm_cancel_join cancel_msg;
1394 
1395 	memset(&cancel_msg, 0, sizeof(cancel_msg));
1396 	cancel_msg.node_idx = dlm->node_num;
1397 	cancel_msg.name_len = strlen(dlm->name);
1398 	memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1399 
1400 	status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1401 				    &cancel_msg, sizeof(cancel_msg), node,
1402 				    NULL);
1403 	if (status < 0) {
1404 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1405 		     "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1406 		     node);
1407 		goto bail;
1408 	}
1409 
1410 bail:
1411 	return status;
1412 }
1413 
1414 /* map_size should be in bytes. */
dlm_send_join_cancels(struct dlm_ctxt * dlm,unsigned long * node_map,unsigned int map_size)1415 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1416 				 unsigned long *node_map,
1417 				 unsigned int map_size)
1418 {
1419 	int status, tmpstat;
1420 	unsigned int node;
1421 
1422 	if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1423 			 sizeof(unsigned long))) {
1424 		mlog(ML_ERROR,
1425 		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1426 		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1427 		return -EINVAL;
1428 	}
1429 
1430 	status = 0;
1431 	node = -1;
1432 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1433 				     node + 1)) < O2NM_MAX_NODES) {
1434 		if (node == dlm->node_num)
1435 			continue;
1436 
1437 		tmpstat = dlm_send_one_join_cancel(dlm, node);
1438 		if (tmpstat) {
1439 			mlog(ML_ERROR, "Error return %d cancelling join on "
1440 			     "node %d\n", tmpstat, node);
1441 			if (!status)
1442 				status = tmpstat;
1443 		}
1444 	}
1445 
1446 	if (status)
1447 		mlog_errno(status);
1448 	return status;
1449 }
1450 
dlm_request_join(struct dlm_ctxt * dlm,int node,enum dlm_query_join_response_code * response)1451 static int dlm_request_join(struct dlm_ctxt *dlm,
1452 			    int node,
1453 			    enum dlm_query_join_response_code *response)
1454 {
1455 	int status;
1456 	struct dlm_query_join_request join_msg;
1457 	struct dlm_query_join_packet packet;
1458 	u32 join_resp;
1459 
1460 	mlog(0, "querying node %d\n", node);
1461 
1462 	memset(&join_msg, 0, sizeof(join_msg));
1463 	join_msg.node_idx = dlm->node_num;
1464 	join_msg.name_len = strlen(dlm->name);
1465 	memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1466 	join_msg.dlm_proto = dlm->dlm_locking_proto;
1467 	join_msg.fs_proto = dlm->fs_locking_proto;
1468 
1469 	/* copy live node map to join message */
1470 	byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1471 
1472 	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1473 				    sizeof(join_msg), node, &join_resp);
1474 	if (status < 0 && status != -ENOPROTOOPT) {
1475 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1476 		     "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1477 		     node);
1478 		goto bail;
1479 	}
1480 	dlm_query_join_wire_to_packet(join_resp, &packet);
1481 
1482 	/* -ENOPROTOOPT from the net code means the other side isn't
1483 	    listening for our message type -- that's fine, it means
1484 	    his dlm isn't up, so we can consider him a 'yes' but not
1485 	    joined into the domain.  */
1486 	if (status == -ENOPROTOOPT) {
1487 		status = 0;
1488 		*response = JOIN_OK_NO_MAP;
1489 	} else if (packet.code == JOIN_DISALLOW ||
1490 		   packet.code == JOIN_OK_NO_MAP) {
1491 		*response = packet.code;
1492 	} else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1493 		mlog(ML_NOTICE,
1494 		     "This node requested DLM locking protocol %u.%u and "
1495 		     "filesystem locking protocol %u.%u.  At least one of "
1496 		     "the protocol versions on node %d is not compatible, "
1497 		     "disconnecting\n",
1498 		     dlm->dlm_locking_proto.pv_major,
1499 		     dlm->dlm_locking_proto.pv_minor,
1500 		     dlm->fs_locking_proto.pv_major,
1501 		     dlm->fs_locking_proto.pv_minor,
1502 		     node);
1503 		status = -EPROTO;
1504 		*response = packet.code;
1505 	} else if (packet.code == JOIN_OK) {
1506 		*response = packet.code;
1507 		/* Use the same locking protocol as the remote node */
1508 		dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1509 		dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1510 		mlog(0,
1511 		     "Node %d responds JOIN_OK with DLM locking protocol "
1512 		     "%u.%u and fs locking protocol %u.%u\n",
1513 		     node,
1514 		     dlm->dlm_locking_proto.pv_major,
1515 		     dlm->dlm_locking_proto.pv_minor,
1516 		     dlm->fs_locking_proto.pv_major,
1517 		     dlm->fs_locking_proto.pv_minor);
1518 	} else {
1519 		status = -EINVAL;
1520 		mlog(ML_ERROR, "invalid response %d from node %u\n",
1521 		     packet.code, node);
1522 	}
1523 
1524 	mlog(0, "status %d, node %d response is %d\n", status, node,
1525 	     *response);
1526 
1527 bail:
1528 	return status;
1529 }
1530 
dlm_send_one_join_assert(struct dlm_ctxt * dlm,unsigned int node)1531 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1532 				    unsigned int node)
1533 {
1534 	int status;
1535 	struct dlm_assert_joined assert_msg;
1536 
1537 	mlog(0, "Sending join assert to node %u\n", node);
1538 
1539 	memset(&assert_msg, 0, sizeof(assert_msg));
1540 	assert_msg.node_idx = dlm->node_num;
1541 	assert_msg.name_len = strlen(dlm->name);
1542 	memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1543 
1544 	status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1545 				    &assert_msg, sizeof(assert_msg), node,
1546 				    NULL);
1547 	if (status < 0)
1548 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1549 		     "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1550 		     node);
1551 
1552 	return status;
1553 }
1554 
dlm_send_join_asserts(struct dlm_ctxt * dlm,unsigned long * node_map)1555 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1556 				  unsigned long *node_map)
1557 {
1558 	int status, node, live;
1559 
1560 	status = 0;
1561 	node = -1;
1562 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1563 				     node + 1)) < O2NM_MAX_NODES) {
1564 		if (node == dlm->node_num)
1565 			continue;
1566 
1567 		do {
1568 			/* It is very important that this message be
1569 			 * received so we spin until either the node
1570 			 * has died or it gets the message. */
1571 			status = dlm_send_one_join_assert(dlm, node);
1572 
1573 			spin_lock(&dlm->spinlock);
1574 			live = test_bit(node, dlm->live_nodes_map);
1575 			spin_unlock(&dlm->spinlock);
1576 
1577 			if (status) {
1578 				mlog(ML_ERROR, "Error return %d asserting "
1579 				     "join on node %d\n", status, node);
1580 
1581 				/* give us some time between errors... */
1582 				if (live)
1583 					msleep(DLM_DOMAIN_BACKOFF_MS);
1584 			}
1585 		} while (status && live);
1586 	}
1587 }
1588 
1589 struct domain_join_ctxt {
1590 	unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1591 	unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1592 };
1593 
dlm_should_restart_join(struct dlm_ctxt * dlm,struct domain_join_ctxt * ctxt,enum dlm_query_join_response_code response)1594 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1595 				   struct domain_join_ctxt *ctxt,
1596 				   enum dlm_query_join_response_code response)
1597 {
1598 	int ret;
1599 
1600 	if (response == JOIN_DISALLOW) {
1601 		mlog(0, "Latest response of disallow -- should restart\n");
1602 		return 1;
1603 	}
1604 
1605 	spin_lock(&dlm->spinlock);
1606 	/* For now, we restart the process if the node maps have
1607 	 * changed at all */
1608 	ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1609 		     sizeof(dlm->live_nodes_map));
1610 	spin_unlock(&dlm->spinlock);
1611 
1612 	if (ret)
1613 		mlog(0, "Node maps changed -- should restart\n");
1614 
1615 	return ret;
1616 }
1617 
dlm_try_to_join_domain(struct dlm_ctxt * dlm)1618 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1619 {
1620 	int status = 0, tmpstat, node;
1621 	struct domain_join_ctxt *ctxt;
1622 	enum dlm_query_join_response_code response = JOIN_DISALLOW;
1623 
1624 	mlog(0, "%p", dlm);
1625 
1626 	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1627 	if (!ctxt) {
1628 		status = -ENOMEM;
1629 		mlog_errno(status);
1630 		goto bail;
1631 	}
1632 
1633 	/* group sem locking should work for us here -- we're already
1634 	 * registered for heartbeat events so filling this should be
1635 	 * atomic wrt getting those handlers called. */
1636 	o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1637 
1638 	spin_lock(&dlm->spinlock);
1639 	memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1640 
1641 	__dlm_set_joining_node(dlm, dlm->node_num);
1642 
1643 	spin_unlock(&dlm->spinlock);
1644 
1645 	node = -1;
1646 	while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1647 				     node + 1)) < O2NM_MAX_NODES) {
1648 		if (node == dlm->node_num)
1649 			continue;
1650 
1651 		status = dlm_request_join(dlm, node, &response);
1652 		if (status < 0) {
1653 			mlog_errno(status);
1654 			goto bail;
1655 		}
1656 
1657 		/* Ok, either we got a response or the node doesn't have a
1658 		 * dlm up. */
1659 		if (response == JOIN_OK)
1660 			set_bit(node, ctxt->yes_resp_map);
1661 
1662 		if (dlm_should_restart_join(dlm, ctxt, response)) {
1663 			status = -EAGAIN;
1664 			goto bail;
1665 		}
1666 	}
1667 
1668 	mlog(0, "Yay, done querying nodes!\n");
1669 
1670 	/* Yay, everyone agree's we can join the domain. My domain is
1671 	 * comprised of all nodes who were put in the
1672 	 * yes_resp_map. Copy that into our domain map and send a join
1673 	 * assert message to clean up everyone elses state. */
1674 	spin_lock(&dlm->spinlock);
1675 	memcpy(dlm->domain_map, ctxt->yes_resp_map,
1676 	       sizeof(ctxt->yes_resp_map));
1677 	set_bit(dlm->node_num, dlm->domain_map);
1678 	spin_unlock(&dlm->spinlock);
1679 
1680 	/* Support for global heartbeat and node info was added in 1.1 */
1681 	if (dlm->dlm_locking_proto.pv_major > 1 ||
1682 	    dlm->dlm_locking_proto.pv_minor > 0) {
1683 		status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1684 		if (status) {
1685 			mlog_errno(status);
1686 			goto bail;
1687 		}
1688 		status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1689 		if (status) {
1690 			mlog_errno(status);
1691 			goto bail;
1692 		}
1693 	}
1694 
1695 	dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1696 
1697 	/* Joined state *must* be set before the joining node
1698 	 * information, otherwise the query_join handler may read no
1699 	 * current joiner but a state of NEW and tell joining nodes
1700 	 * we're not in the domain. */
1701 	spin_lock(&dlm_domain_lock);
1702 	dlm->dlm_state = DLM_CTXT_JOINED;
1703 	dlm->num_joins++;
1704 	spin_unlock(&dlm_domain_lock);
1705 
1706 bail:
1707 	spin_lock(&dlm->spinlock);
1708 	__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1709 	if (!status) {
1710 		printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1711 		__dlm_print_nodes(dlm);
1712 	}
1713 	spin_unlock(&dlm->spinlock);
1714 
1715 	if (ctxt) {
1716 		/* Do we need to send a cancel message to any nodes? */
1717 		if (status < 0) {
1718 			tmpstat = dlm_send_join_cancels(dlm,
1719 							ctxt->yes_resp_map,
1720 							sizeof(ctxt->yes_resp_map));
1721 			if (tmpstat < 0)
1722 				mlog_errno(tmpstat);
1723 		}
1724 		kfree(ctxt);
1725 	}
1726 
1727 	mlog(0, "returning %d\n", status);
1728 	return status;
1729 }
1730 
dlm_unregister_domain_handlers(struct dlm_ctxt * dlm)1731 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1732 {
1733 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1734 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1735 	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1736 }
1737 
dlm_register_domain_handlers(struct dlm_ctxt * dlm)1738 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1739 {
1740 	int status;
1741 
1742 	mlog(0, "registering handlers.\n");
1743 
1744 	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1745 			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1746 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1747 	if (status)
1748 		goto bail;
1749 
1750 	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1751 			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1752 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1753 	if (status)
1754 		goto bail;
1755 
1756 	status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1757 					sizeof(struct dlm_master_request),
1758 					dlm_master_request_handler,
1759 					dlm, NULL, &dlm->dlm_domain_handlers);
1760 	if (status)
1761 		goto bail;
1762 
1763 	status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1764 					sizeof(struct dlm_assert_master),
1765 					dlm_assert_master_handler,
1766 					dlm, dlm_assert_master_post_handler,
1767 					&dlm->dlm_domain_handlers);
1768 	if (status)
1769 		goto bail;
1770 
1771 	status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1772 					sizeof(struct dlm_create_lock),
1773 					dlm_create_lock_handler,
1774 					dlm, NULL, &dlm->dlm_domain_handlers);
1775 	if (status)
1776 		goto bail;
1777 
1778 	status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1779 					DLM_CONVERT_LOCK_MAX_LEN,
1780 					dlm_convert_lock_handler,
1781 					dlm, NULL, &dlm->dlm_domain_handlers);
1782 	if (status)
1783 		goto bail;
1784 
1785 	status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1786 					DLM_UNLOCK_LOCK_MAX_LEN,
1787 					dlm_unlock_lock_handler,
1788 					dlm, NULL, &dlm->dlm_domain_handlers);
1789 	if (status)
1790 		goto bail;
1791 
1792 	status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1793 					DLM_PROXY_AST_MAX_LEN,
1794 					dlm_proxy_ast_handler,
1795 					dlm, NULL, &dlm->dlm_domain_handlers);
1796 	if (status)
1797 		goto bail;
1798 
1799 	status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1800 					sizeof(struct dlm_exit_domain),
1801 					dlm_exit_domain_handler,
1802 					dlm, NULL, &dlm->dlm_domain_handlers);
1803 	if (status)
1804 		goto bail;
1805 
1806 	status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1807 					sizeof(struct dlm_deref_lockres),
1808 					dlm_deref_lockres_handler,
1809 					dlm, NULL, &dlm->dlm_domain_handlers);
1810 	if (status)
1811 		goto bail;
1812 
1813 	status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1814 					sizeof(struct dlm_migrate_request),
1815 					dlm_migrate_request_handler,
1816 					dlm, NULL, &dlm->dlm_domain_handlers);
1817 	if (status)
1818 		goto bail;
1819 
1820 	status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1821 					DLM_MIG_LOCKRES_MAX_LEN,
1822 					dlm_mig_lockres_handler,
1823 					dlm, NULL, &dlm->dlm_domain_handlers);
1824 	if (status)
1825 		goto bail;
1826 
1827 	status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1828 					sizeof(struct dlm_master_requery),
1829 					dlm_master_requery_handler,
1830 					dlm, NULL, &dlm->dlm_domain_handlers);
1831 	if (status)
1832 		goto bail;
1833 
1834 	status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1835 					sizeof(struct dlm_lock_request),
1836 					dlm_request_all_locks_handler,
1837 					dlm, NULL, &dlm->dlm_domain_handlers);
1838 	if (status)
1839 		goto bail;
1840 
1841 	status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1842 					sizeof(struct dlm_reco_data_done),
1843 					dlm_reco_data_done_handler,
1844 					dlm, NULL, &dlm->dlm_domain_handlers);
1845 	if (status)
1846 		goto bail;
1847 
1848 	status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1849 					sizeof(struct dlm_begin_reco),
1850 					dlm_begin_reco_handler,
1851 					dlm, NULL, &dlm->dlm_domain_handlers);
1852 	if (status)
1853 		goto bail;
1854 
1855 	status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1856 					sizeof(struct dlm_finalize_reco),
1857 					dlm_finalize_reco_handler,
1858 					dlm, NULL, &dlm->dlm_domain_handlers);
1859 	if (status)
1860 		goto bail;
1861 
1862 	status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1863 					sizeof(struct dlm_exit_domain),
1864 					dlm_begin_exit_domain_handler,
1865 					dlm, NULL, &dlm->dlm_domain_handlers);
1866 	if (status)
1867 		goto bail;
1868 
1869 bail:
1870 	if (status)
1871 		dlm_unregister_domain_handlers(dlm);
1872 
1873 	return status;
1874 }
1875 
dlm_join_domain(struct dlm_ctxt * dlm)1876 static int dlm_join_domain(struct dlm_ctxt *dlm)
1877 {
1878 	int status;
1879 	unsigned int backoff;
1880 	unsigned int total_backoff = 0;
1881 
1882 	BUG_ON(!dlm);
1883 
1884 	mlog(0, "Join domain %s\n", dlm->name);
1885 
1886 	status = dlm_register_domain_handlers(dlm);
1887 	if (status) {
1888 		mlog_errno(status);
1889 		goto bail;
1890 	}
1891 
1892 	status = dlm_debug_init(dlm);
1893 	if (status < 0) {
1894 		mlog_errno(status);
1895 		goto bail;
1896 	}
1897 
1898 	status = dlm_launch_thread(dlm);
1899 	if (status < 0) {
1900 		mlog_errno(status);
1901 		goto bail;
1902 	}
1903 
1904 	status = dlm_launch_recovery_thread(dlm);
1905 	if (status < 0) {
1906 		mlog_errno(status);
1907 		goto bail;
1908 	}
1909 
1910 	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1911 	if (!dlm->dlm_worker) {
1912 		status = -ENOMEM;
1913 		mlog_errno(status);
1914 		goto bail;
1915 	}
1916 
1917 	do {
1918 		status = dlm_try_to_join_domain(dlm);
1919 
1920 		/* If we're racing another node to the join, then we
1921 		 * need to back off temporarily and let them
1922 		 * complete. */
1923 #define	DLM_JOIN_TIMEOUT_MSECS	90000
1924 		if (status == -EAGAIN) {
1925 			if (signal_pending(current)) {
1926 				status = -ERESTARTSYS;
1927 				goto bail;
1928 			}
1929 
1930 			if (total_backoff >
1931 			    msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1932 				status = -ERESTARTSYS;
1933 				mlog(ML_NOTICE, "Timed out joining dlm domain "
1934 				     "%s after %u msecs\n", dlm->name,
1935 				     jiffies_to_msecs(total_backoff));
1936 				goto bail;
1937 			}
1938 
1939 			/*
1940 			 * <chip> After you!
1941 			 * <dale> No, after you!
1942 			 * <chip> I insist!
1943 			 * <dale> But you first!
1944 			 * ...
1945 			 */
1946 			backoff = (unsigned int)(jiffies & 0x3);
1947 			backoff *= DLM_DOMAIN_BACKOFF_MS;
1948 			total_backoff += backoff;
1949 			mlog(0, "backoff %d\n", backoff);
1950 			msleep(backoff);
1951 		}
1952 	} while (status == -EAGAIN);
1953 
1954 	if (status < 0) {
1955 		mlog_errno(status);
1956 		goto bail;
1957 	}
1958 
1959 	status = 0;
1960 bail:
1961 	wake_up(&dlm_domain_events);
1962 
1963 	if (status) {
1964 		dlm_unregister_domain_handlers(dlm);
1965 		dlm_debug_shutdown(dlm);
1966 		dlm_complete_thread(dlm);
1967 		dlm_complete_recovery_thread(dlm);
1968 		dlm_destroy_dlm_worker(dlm);
1969 	}
1970 
1971 	return status;
1972 }
1973 
dlm_alloc_ctxt(const char * domain,u32 key)1974 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1975 				u32 key)
1976 {
1977 	int i;
1978 	int ret;
1979 	struct dlm_ctxt *dlm = NULL;
1980 
1981 	dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1982 	if (!dlm) {
1983 		mlog_errno(-ENOMEM);
1984 		goto leave;
1985 	}
1986 
1987 	dlm->name = kstrdup(domain, GFP_KERNEL);
1988 	if (dlm->name == NULL) {
1989 		mlog_errno(-ENOMEM);
1990 		kfree(dlm);
1991 		dlm = NULL;
1992 		goto leave;
1993 	}
1994 
1995 	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1996 	if (!dlm->lockres_hash) {
1997 		mlog_errno(-ENOMEM);
1998 		kfree(dlm->name);
1999 		kfree(dlm);
2000 		dlm = NULL;
2001 		goto leave;
2002 	}
2003 
2004 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
2005 		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
2006 
2007 	dlm->master_hash = (struct hlist_head **)
2008 				dlm_alloc_pagevec(DLM_HASH_PAGES);
2009 	if (!dlm->master_hash) {
2010 		mlog_errno(-ENOMEM);
2011 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
2012 		kfree(dlm->name);
2013 		kfree(dlm);
2014 		dlm = NULL;
2015 		goto leave;
2016 	}
2017 
2018 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
2019 		INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2020 
2021 	dlm->key = key;
2022 	dlm->node_num = o2nm_this_node();
2023 
2024 	ret = dlm_create_debugfs_subroot(dlm);
2025 	if (ret < 0) {
2026 		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
2027 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
2028 		kfree(dlm->name);
2029 		kfree(dlm);
2030 		dlm = NULL;
2031 		goto leave;
2032 	}
2033 
2034 	spin_lock_init(&dlm->spinlock);
2035 	spin_lock_init(&dlm->master_lock);
2036 	spin_lock_init(&dlm->ast_lock);
2037 	spin_lock_init(&dlm->track_lock);
2038 	INIT_LIST_HEAD(&dlm->list);
2039 	INIT_LIST_HEAD(&dlm->dirty_list);
2040 	INIT_LIST_HEAD(&dlm->reco.resources);
2041 	INIT_LIST_HEAD(&dlm->reco.received);
2042 	INIT_LIST_HEAD(&dlm->reco.node_data);
2043 	INIT_LIST_HEAD(&dlm->purge_list);
2044 	INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2045 	INIT_LIST_HEAD(&dlm->tracking_list);
2046 	dlm->reco.state = 0;
2047 
2048 	INIT_LIST_HEAD(&dlm->pending_asts);
2049 	INIT_LIST_HEAD(&dlm->pending_basts);
2050 
2051 	mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2052 		  dlm->recovery_map, &(dlm->recovery_map[0]));
2053 
2054 	memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2055 	memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2056 	memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2057 
2058 	dlm->dlm_thread_task = NULL;
2059 	dlm->dlm_reco_thread_task = NULL;
2060 	dlm->dlm_worker = NULL;
2061 	init_waitqueue_head(&dlm->dlm_thread_wq);
2062 	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2063 	init_waitqueue_head(&dlm->reco.event);
2064 	init_waitqueue_head(&dlm->ast_wq);
2065 	init_waitqueue_head(&dlm->migration_wq);
2066 	INIT_LIST_HEAD(&dlm->mle_hb_events);
2067 
2068 	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2069 	init_waitqueue_head(&dlm->dlm_join_events);
2070 
2071 	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2072 	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2073 
2074 	atomic_set(&dlm->res_tot_count, 0);
2075 	atomic_set(&dlm->res_cur_count, 0);
2076 	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2077 		atomic_set(&dlm->mle_tot_count[i], 0);
2078 		atomic_set(&dlm->mle_cur_count[i], 0);
2079 	}
2080 
2081 	spin_lock_init(&dlm->work_lock);
2082 	INIT_LIST_HEAD(&dlm->work_list);
2083 	INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2084 
2085 	kref_init(&dlm->dlm_refs);
2086 	dlm->dlm_state = DLM_CTXT_NEW;
2087 
2088 	INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2089 
2090 	mlog(0, "context init: refcount %u\n",
2091 		  atomic_read(&dlm->dlm_refs.refcount));
2092 
2093 leave:
2094 	return dlm;
2095 }
2096 
2097 /*
2098  * Compare a requested locking protocol version against the current one.
2099  *
2100  * If the major numbers are different, they are incompatible.
2101  * If the current minor is greater than the request, they are incompatible.
2102  * If the current minor is less than or equal to the request, they are
2103  * compatible, and the requester should run at the current minor version.
2104  */
dlm_protocol_compare(struct dlm_protocol_version * existing,struct dlm_protocol_version * request)2105 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2106 				struct dlm_protocol_version *request)
2107 {
2108 	if (existing->pv_major != request->pv_major)
2109 		return 1;
2110 
2111 	if (existing->pv_minor > request->pv_minor)
2112 		return 1;
2113 
2114 	if (existing->pv_minor < request->pv_minor)
2115 		request->pv_minor = existing->pv_minor;
2116 
2117 	return 0;
2118 }
2119 
2120 /*
2121  * dlm_register_domain: one-time setup per "domain".
2122  *
2123  * The filesystem passes in the requested locking version via proto.
2124  * If registration was successful, proto will contain the negotiated
2125  * locking protocol.
2126  */
dlm_register_domain(const char * domain,u32 key,struct dlm_protocol_version * fs_proto)2127 struct dlm_ctxt * dlm_register_domain(const char *domain,
2128 			       u32 key,
2129 			       struct dlm_protocol_version *fs_proto)
2130 {
2131 	int ret;
2132 	struct dlm_ctxt *dlm = NULL;
2133 	struct dlm_ctxt *new_ctxt = NULL;
2134 
2135 	if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2136 		ret = -ENAMETOOLONG;
2137 		mlog(ML_ERROR, "domain name length too long\n");
2138 		goto leave;
2139 	}
2140 
2141 	mlog(0, "register called for domain \"%s\"\n", domain);
2142 
2143 retry:
2144 	dlm = NULL;
2145 	if (signal_pending(current)) {
2146 		ret = -ERESTARTSYS;
2147 		mlog_errno(ret);
2148 		goto leave;
2149 	}
2150 
2151 	spin_lock(&dlm_domain_lock);
2152 
2153 	dlm = __dlm_lookup_domain(domain);
2154 	if (dlm) {
2155 		if (dlm->dlm_state != DLM_CTXT_JOINED) {
2156 			spin_unlock(&dlm_domain_lock);
2157 
2158 			mlog(0, "This ctxt is not joined yet!\n");
2159 			wait_event_interruptible(dlm_domain_events,
2160 						 dlm_wait_on_domain_helper(
2161 							 domain));
2162 			goto retry;
2163 		}
2164 
2165 		if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2166 			spin_unlock(&dlm_domain_lock);
2167 			mlog(ML_ERROR,
2168 			     "Requested locking protocol version is not "
2169 			     "compatible with already registered domain "
2170 			     "\"%s\"\n", domain);
2171 			ret = -EPROTO;
2172 			goto leave;
2173 		}
2174 
2175 		__dlm_get(dlm);
2176 		dlm->num_joins++;
2177 
2178 		spin_unlock(&dlm_domain_lock);
2179 
2180 		ret = 0;
2181 		goto leave;
2182 	}
2183 
2184 	/* doesn't exist */
2185 	if (!new_ctxt) {
2186 		spin_unlock(&dlm_domain_lock);
2187 
2188 		new_ctxt = dlm_alloc_ctxt(domain, key);
2189 		if (new_ctxt)
2190 			goto retry;
2191 
2192 		ret = -ENOMEM;
2193 		mlog_errno(ret);
2194 		goto leave;
2195 	}
2196 
2197 	/* a little variable switch-a-roo here... */
2198 	dlm = new_ctxt;
2199 	new_ctxt = NULL;
2200 
2201 	/* add the new domain */
2202 	list_add_tail(&dlm->list, &dlm_domains);
2203 	spin_unlock(&dlm_domain_lock);
2204 
2205 	/*
2206 	 * Pass the locking protocol version into the join.  If the join
2207 	 * succeeds, it will have the negotiated protocol set.
2208 	 */
2209 	dlm->dlm_locking_proto = dlm_protocol;
2210 	dlm->fs_locking_proto = *fs_proto;
2211 
2212 	ret = dlm_join_domain(dlm);
2213 	if (ret) {
2214 		mlog_errno(ret);
2215 		dlm_put(dlm);
2216 		goto leave;
2217 	}
2218 
2219 	/* Tell the caller what locking protocol we negotiated */
2220 	*fs_proto = dlm->fs_locking_proto;
2221 
2222 	ret = 0;
2223 leave:
2224 	if (new_ctxt)
2225 		dlm_free_ctxt_mem(new_ctxt);
2226 
2227 	if (ret < 0)
2228 		dlm = ERR_PTR(ret);
2229 
2230 	return dlm;
2231 }
2232 EXPORT_SYMBOL_GPL(dlm_register_domain);
2233 
2234 static LIST_HEAD(dlm_join_handlers);
2235 
dlm_unregister_net_handlers(void)2236 static void dlm_unregister_net_handlers(void)
2237 {
2238 	o2net_unregister_handler_list(&dlm_join_handlers);
2239 }
2240 
dlm_register_net_handlers(void)2241 static int dlm_register_net_handlers(void)
2242 {
2243 	int status = 0;
2244 
2245 	status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2246 					sizeof(struct dlm_query_join_request),
2247 					dlm_query_join_handler,
2248 					NULL, NULL, &dlm_join_handlers);
2249 	if (status)
2250 		goto bail;
2251 
2252 	status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2253 					sizeof(struct dlm_assert_joined),
2254 					dlm_assert_joined_handler,
2255 					NULL, NULL, &dlm_join_handlers);
2256 	if (status)
2257 		goto bail;
2258 
2259 	status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2260 					sizeof(struct dlm_cancel_join),
2261 					dlm_cancel_join_handler,
2262 					NULL, NULL, &dlm_join_handlers);
2263 	if (status)
2264 		goto bail;
2265 
2266 	status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2267 					sizeof(struct dlm_query_region),
2268 					dlm_query_region_handler,
2269 					NULL, NULL, &dlm_join_handlers);
2270 
2271 	if (status)
2272 		goto bail;
2273 
2274 	status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2275 					sizeof(struct dlm_query_nodeinfo),
2276 					dlm_query_nodeinfo_handler,
2277 					NULL, NULL, &dlm_join_handlers);
2278 bail:
2279 	if (status < 0)
2280 		dlm_unregister_net_handlers();
2281 
2282 	return status;
2283 }
2284 
2285 /* Domain eviction callback handling.
2286  *
2287  * The file system requires notification of node death *before* the
2288  * dlm completes it's recovery work, otherwise it may be able to
2289  * acquire locks on resources requiring recovery. Since the dlm can
2290  * evict a node from it's domain *before* heartbeat fires, a similar
2291  * mechanism is required. */
2292 
2293 /* Eviction is not expected to happen often, so a per-domain lock is
2294  * not necessary. Eviction callbacks are allowed to sleep for short
2295  * periods of time. */
2296 static DECLARE_RWSEM(dlm_callback_sem);
2297 
dlm_fire_domain_eviction_callbacks(struct dlm_ctxt * dlm,int node_num)2298 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2299 					int node_num)
2300 {
2301 	struct list_head *iter;
2302 	struct dlm_eviction_cb *cb;
2303 
2304 	down_read(&dlm_callback_sem);
2305 	list_for_each(iter, &dlm->dlm_eviction_callbacks) {
2306 		cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
2307 
2308 		cb->ec_func(node_num, cb->ec_data);
2309 	}
2310 	up_read(&dlm_callback_sem);
2311 }
2312 
dlm_setup_eviction_cb(struct dlm_eviction_cb * cb,dlm_eviction_func * f,void * data)2313 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2314 			   dlm_eviction_func *f,
2315 			   void *data)
2316 {
2317 	INIT_LIST_HEAD(&cb->ec_item);
2318 	cb->ec_func = f;
2319 	cb->ec_data = data;
2320 }
2321 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2322 
dlm_register_eviction_cb(struct dlm_ctxt * dlm,struct dlm_eviction_cb * cb)2323 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2324 			      struct dlm_eviction_cb *cb)
2325 {
2326 	down_write(&dlm_callback_sem);
2327 	list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2328 	up_write(&dlm_callback_sem);
2329 }
2330 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2331 
dlm_unregister_eviction_cb(struct dlm_eviction_cb * cb)2332 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2333 {
2334 	down_write(&dlm_callback_sem);
2335 	list_del_init(&cb->ec_item);
2336 	up_write(&dlm_callback_sem);
2337 }
2338 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2339 
dlm_init(void)2340 static int __init dlm_init(void)
2341 {
2342 	int status;
2343 
2344 	dlm_print_version();
2345 
2346 	status = dlm_init_mle_cache();
2347 	if (status) {
2348 		mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2349 		goto error;
2350 	}
2351 
2352 	status = dlm_init_master_caches();
2353 	if (status) {
2354 		mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2355 		     "o2dlm_lockname slabcaches\n");
2356 		goto error;
2357 	}
2358 
2359 	status = dlm_init_lock_cache();
2360 	if (status) {
2361 		mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2362 		goto error;
2363 	}
2364 
2365 	status = dlm_register_net_handlers();
2366 	if (status) {
2367 		mlog(ML_ERROR, "Unable to register network handlers\n");
2368 		goto error;
2369 	}
2370 
2371 	status = dlm_create_debugfs_root();
2372 	if (status)
2373 		goto error;
2374 
2375 	return 0;
2376 error:
2377 	dlm_unregister_net_handlers();
2378 	dlm_destroy_lock_cache();
2379 	dlm_destroy_master_caches();
2380 	dlm_destroy_mle_cache();
2381 	return -1;
2382 }
2383 
dlm_exit(void)2384 static void __exit dlm_exit (void)
2385 {
2386 	dlm_destroy_debugfs_root();
2387 	dlm_unregister_net_handlers();
2388 	dlm_destroy_lock_cache();
2389 	dlm_destroy_master_caches();
2390 	dlm_destroy_mle_cache();
2391 }
2392 
2393 MODULE_AUTHOR("Oracle");
2394 MODULE_LICENSE("GPL");
2395 
2396 module_init(dlm_init);
2397 module_exit(dlm_exit);
2398