xref: /openbmc/linux/fs/ocfs2/dlm/dlmdomain.c (revision df2634f43f5106947f3735a0b61a6527a4b278cd)
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  */
57 static inline void byte_set_bit(u8 nr, u8 map[])
58 {
59 	map[nr >> 3] |= (1UL << (nr & 7));
60 }
61 
62 static inline int byte_test_bit(u8 nr, u8 map[])
63 {
64 	return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
65 }
66 
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 
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 
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  */
136 static const struct dlm_protocol_version dlm_protocol = {
137 	.pv_major = 1,
138 	.pv_minor = 1,
139 };
140 
141 #define DLM_DOMAIN_BACKOFF_MS 200
142 
143 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
144 				  void **ret_data);
145 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
146 				     void **ret_data);
147 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
148 				   void **ret_data);
149 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
150 				    void *data, void **ret_data);
151 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
152 				   void **ret_data);
153 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
154 				struct dlm_protocol_version *request);
155 
156 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
157 
158 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
159 {
160 	if (!hlist_unhashed(&lockres->hash_node)) {
161 		hlist_del_init(&lockres->hash_node);
162 		dlm_lockres_put(lockres);
163 	}
164 }
165 
166 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
167 		       struct dlm_lock_resource *res)
168 {
169 	struct hlist_head *bucket;
170 	struct qstr *q;
171 
172 	assert_spin_locked(&dlm->spinlock);
173 
174 	q = &res->lockname;
175 	bucket = dlm_lockres_hash(dlm, q->hash);
176 
177 	/* get a reference for our hashtable */
178 	dlm_lockres_get(res);
179 
180 	hlist_add_head(&res->hash_node, bucket);
181 }
182 
183 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
184 						     const char *name,
185 						     unsigned int len,
186 						     unsigned int hash)
187 {
188 	struct hlist_head *bucket;
189 	struct hlist_node *list;
190 
191 	mlog_entry("%.*s\n", len, name);
192 
193 	assert_spin_locked(&dlm->spinlock);
194 
195 	bucket = dlm_lockres_hash(dlm, hash);
196 
197 	hlist_for_each(list, bucket) {
198 		struct dlm_lock_resource *res = hlist_entry(list,
199 			struct dlm_lock_resource, hash_node);
200 		if (res->lockname.name[0] != name[0])
201 			continue;
202 		if (unlikely(res->lockname.len != len))
203 			continue;
204 		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
205 			continue;
206 		dlm_lockres_get(res);
207 		return res;
208 	}
209 	return NULL;
210 }
211 
212 /* intended to be called by functions which do not care about lock
213  * resources which are being purged (most net _handler functions).
214  * this will return NULL for any lock resource which is found but
215  * currently in the process of dropping its mastery reference.
216  * use __dlm_lookup_lockres_full when you need the lock resource
217  * regardless (e.g. dlm_get_lock_resource) */
218 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
219 						const char *name,
220 						unsigned int len,
221 						unsigned int hash)
222 {
223 	struct dlm_lock_resource *res = NULL;
224 
225 	mlog_entry("%.*s\n", len, name);
226 
227 	assert_spin_locked(&dlm->spinlock);
228 
229 	res = __dlm_lookup_lockres_full(dlm, name, len, hash);
230 	if (res) {
231 		spin_lock(&res->spinlock);
232 		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
233 			spin_unlock(&res->spinlock);
234 			dlm_lockres_put(res);
235 			return NULL;
236 		}
237 		spin_unlock(&res->spinlock);
238 	}
239 
240 	return res;
241 }
242 
243 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
244 				    const char *name,
245 				    unsigned int len)
246 {
247 	struct dlm_lock_resource *res;
248 	unsigned int hash = dlm_lockid_hash(name, len);
249 
250 	spin_lock(&dlm->spinlock);
251 	res = __dlm_lookup_lockres(dlm, name, len, hash);
252 	spin_unlock(&dlm->spinlock);
253 	return res;
254 }
255 
256 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
257 {
258 	struct dlm_ctxt *tmp = NULL;
259 	struct list_head *iter;
260 
261 	assert_spin_locked(&dlm_domain_lock);
262 
263 	/* tmp->name here is always NULL terminated,
264 	 * but domain may not be! */
265 	list_for_each(iter, &dlm_domains) {
266 		tmp = list_entry (iter, struct dlm_ctxt, list);
267 		if (strlen(tmp->name) == len &&
268 		    memcmp(tmp->name, domain, len)==0)
269 			break;
270 		tmp = NULL;
271 	}
272 
273 	return tmp;
274 }
275 
276 /* For null terminated domain strings ONLY */
277 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
278 {
279 	assert_spin_locked(&dlm_domain_lock);
280 
281 	return __dlm_lookup_domain_full(domain, strlen(domain));
282 }
283 
284 
285 /* returns true on one of two conditions:
286  * 1) the domain does not exist
287  * 2) the domain exists and it's state is "joined" */
288 static int dlm_wait_on_domain_helper(const char *domain)
289 {
290 	int ret = 0;
291 	struct dlm_ctxt *tmp = NULL;
292 
293 	spin_lock(&dlm_domain_lock);
294 
295 	tmp = __dlm_lookup_domain(domain);
296 	if (!tmp)
297 		ret = 1;
298 	else if (tmp->dlm_state == DLM_CTXT_JOINED)
299 		ret = 1;
300 
301 	spin_unlock(&dlm_domain_lock);
302 	return ret;
303 }
304 
305 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
306 {
307 	dlm_destroy_debugfs_subroot(dlm);
308 
309 	if (dlm->lockres_hash)
310 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
311 
312 	if (dlm->master_hash)
313 		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
314 
315 	if (dlm->name)
316 		kfree(dlm->name);
317 
318 	kfree(dlm);
319 }
320 
321 /* A little strange - this function will be called while holding
322  * dlm_domain_lock and is expected to be holding it on the way out. We
323  * will however drop and reacquire it multiple times */
324 static void dlm_ctxt_release(struct kref *kref)
325 {
326 	struct dlm_ctxt *dlm;
327 
328 	dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
329 
330 	BUG_ON(dlm->num_joins);
331 	BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
332 
333 	/* we may still be in the list if we hit an error during join. */
334 	list_del_init(&dlm->list);
335 
336 	spin_unlock(&dlm_domain_lock);
337 
338 	mlog(0, "freeing memory from domain %s\n", dlm->name);
339 
340 	wake_up(&dlm_domain_events);
341 
342 	dlm_free_ctxt_mem(dlm);
343 
344 	spin_lock(&dlm_domain_lock);
345 }
346 
347 void dlm_put(struct dlm_ctxt *dlm)
348 {
349 	spin_lock(&dlm_domain_lock);
350 	kref_put(&dlm->dlm_refs, dlm_ctxt_release);
351 	spin_unlock(&dlm_domain_lock);
352 }
353 
354 static void __dlm_get(struct dlm_ctxt *dlm)
355 {
356 	kref_get(&dlm->dlm_refs);
357 }
358 
359 /* given a questionable reference to a dlm object, gets a reference if
360  * it can find it in the list, otherwise returns NULL in which case
361  * you shouldn't trust your pointer. */
362 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
363 {
364 	struct list_head *iter;
365 	struct dlm_ctxt *target = NULL;
366 
367 	spin_lock(&dlm_domain_lock);
368 
369 	list_for_each(iter, &dlm_domains) {
370 		target = list_entry (iter, struct dlm_ctxt, list);
371 
372 		if (target == dlm) {
373 			__dlm_get(target);
374 			break;
375 		}
376 
377 		target = NULL;
378 	}
379 
380 	spin_unlock(&dlm_domain_lock);
381 
382 	return target;
383 }
384 
385 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
386 {
387 	int ret;
388 
389 	spin_lock(&dlm_domain_lock);
390 	ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
391 		(dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
392 	spin_unlock(&dlm_domain_lock);
393 
394 	return ret;
395 }
396 
397 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
398 {
399 	if (dlm->dlm_worker) {
400 		flush_workqueue(dlm->dlm_worker);
401 		destroy_workqueue(dlm->dlm_worker);
402 		dlm->dlm_worker = NULL;
403 	}
404 }
405 
406 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
407 {
408 	dlm_unregister_domain_handlers(dlm);
409 	dlm_debug_shutdown(dlm);
410 	dlm_complete_thread(dlm);
411 	dlm_complete_recovery_thread(dlm);
412 	dlm_destroy_dlm_worker(dlm);
413 
414 	/* We've left the domain. Now we can take ourselves out of the
415 	 * list and allow the kref stuff to help us free the
416 	 * memory. */
417 	spin_lock(&dlm_domain_lock);
418 	list_del_init(&dlm->list);
419 	spin_unlock(&dlm_domain_lock);
420 
421 	/* Wake up anyone waiting for us to remove this domain */
422 	wake_up(&dlm_domain_events);
423 }
424 
425 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
426 {
427 	int i, num, n, ret = 0;
428 	struct dlm_lock_resource *res;
429 	struct hlist_node *iter;
430 	struct hlist_head *bucket;
431 	int dropped;
432 
433 	mlog(0, "Migrating locks from domain %s\n", dlm->name);
434 
435 	num = 0;
436 	spin_lock(&dlm->spinlock);
437 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
438 redo_bucket:
439 		n = 0;
440 		bucket = dlm_lockres_hash(dlm, i);
441 		iter = bucket->first;
442 		while (iter) {
443 			n++;
444 			res = hlist_entry(iter, struct dlm_lock_resource,
445 					  hash_node);
446 			dlm_lockres_get(res);
447 			/* migrate, if necessary.  this will drop the dlm
448 			 * spinlock and retake it if it does migration. */
449 			dropped = dlm_empty_lockres(dlm, res);
450 
451 			spin_lock(&res->spinlock);
452 			__dlm_lockres_calc_usage(dlm, res);
453 			iter = res->hash_node.next;
454 			spin_unlock(&res->spinlock);
455 
456 			dlm_lockres_put(res);
457 
458 			if (dropped)
459 				goto redo_bucket;
460 		}
461 		cond_resched_lock(&dlm->spinlock);
462 		num += n;
463 	}
464 	spin_unlock(&dlm->spinlock);
465 	wake_up(&dlm->dlm_thread_wq);
466 
467 	/* let the dlm thread take care of purging, keep scanning until
468 	 * nothing remains in the hash */
469 	if (num) {
470 		mlog(0, "%s: %d lock resources in hash last pass\n",
471 		     dlm->name, num);
472 		ret = -EAGAIN;
473 	}
474 	mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
475 	return ret;
476 }
477 
478 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
479 {
480 	int ret;
481 
482 	spin_lock(&dlm->spinlock);
483 	ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
484 	spin_unlock(&dlm->spinlock);
485 
486 	return ret;
487 }
488 
489 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
490 {
491 	/* Yikes, a double spinlock! I need domain_lock for the dlm
492 	 * state and the dlm spinlock for join state... Sorry! */
493 again:
494 	spin_lock(&dlm_domain_lock);
495 	spin_lock(&dlm->spinlock);
496 
497 	if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
498 		mlog(0, "Node %d is joining, we wait on it.\n",
499 			  dlm->joining_node);
500 		spin_unlock(&dlm->spinlock);
501 		spin_unlock(&dlm_domain_lock);
502 
503 		wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
504 		goto again;
505 	}
506 
507 	dlm->dlm_state = DLM_CTXT_LEAVING;
508 	spin_unlock(&dlm->spinlock);
509 	spin_unlock(&dlm_domain_lock);
510 }
511 
512 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
513 {
514 	int node = -1;
515 
516 	assert_spin_locked(&dlm->spinlock);
517 
518 	printk(KERN_NOTICE "o2dlm: Nodes in domain %s: ", dlm->name);
519 
520 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
521 				     node + 1)) < O2NM_MAX_NODES) {
522 		printk("%d ", node);
523 	}
524 	printk("\n");
525 }
526 
527 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
528 				   void **ret_data)
529 {
530 	struct dlm_ctxt *dlm = data;
531 	unsigned int node;
532 	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
533 
534 	mlog_entry("%p %u %p", msg, len, data);
535 
536 	if (!dlm_grab(dlm))
537 		return 0;
538 
539 	node = exit_msg->node_idx;
540 
541 	printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s\n", node, dlm->name);
542 
543 	spin_lock(&dlm->spinlock);
544 	clear_bit(node, dlm->domain_map);
545 	__dlm_print_nodes(dlm);
546 
547 	/* notify anything attached to the heartbeat events */
548 	dlm_hb_event_notify_attached(dlm, node, 0);
549 
550 	spin_unlock(&dlm->spinlock);
551 
552 	dlm_put(dlm);
553 
554 	return 0;
555 }
556 
557 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
558 				    unsigned int node)
559 {
560 	int status;
561 	struct dlm_exit_domain leave_msg;
562 
563 	mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
564 		  node, dlm->name, dlm->node_num);
565 
566 	memset(&leave_msg, 0, sizeof(leave_msg));
567 	leave_msg.node_idx = dlm->node_num;
568 
569 	status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
570 				    &leave_msg, sizeof(leave_msg), node,
571 				    NULL);
572 	if (status < 0)
573 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
574 		     "node %u\n", status, DLM_EXIT_DOMAIN_MSG, dlm->key, node);
575 	mlog(0, "status return %d from o2net_send_message\n", status);
576 
577 	return status;
578 }
579 
580 
581 static void dlm_leave_domain(struct dlm_ctxt *dlm)
582 {
583 	int node, clear_node, status;
584 
585 	/* At this point we've migrated away all our locks and won't
586 	 * accept mastership of new ones. The dlm is responsible for
587 	 * almost nothing now. We make sure not to confuse any joining
588 	 * nodes and then commence shutdown procedure. */
589 
590 	spin_lock(&dlm->spinlock);
591 	/* Clear ourselves from the domain map */
592 	clear_bit(dlm->node_num, dlm->domain_map);
593 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
594 				     0)) < O2NM_MAX_NODES) {
595 		/* Drop the dlm spinlock. This is safe wrt the domain_map.
596 		 * -nodes cannot be added now as the
597 		 *   query_join_handlers knows to respond with OK_NO_MAP
598 		 * -we catch the right network errors if a node is
599 		 *   removed from the map while we're sending him the
600 		 *   exit message. */
601 		spin_unlock(&dlm->spinlock);
602 
603 		clear_node = 1;
604 
605 		status = dlm_send_one_domain_exit(dlm, node);
606 		if (status < 0 &&
607 		    status != -ENOPROTOOPT &&
608 		    status != -ENOTCONN) {
609 			mlog(ML_NOTICE, "Error %d sending domain exit message "
610 			     "to node %d\n", status, node);
611 
612 			/* Not sure what to do here but lets sleep for
613 			 * a bit in case this was a transient
614 			 * error... */
615 			msleep(DLM_DOMAIN_BACKOFF_MS);
616 			clear_node = 0;
617 		}
618 
619 		spin_lock(&dlm->spinlock);
620 		/* If we're not clearing the node bit then we intend
621 		 * to loop back around to try again. */
622 		if (clear_node)
623 			clear_bit(node, dlm->domain_map);
624 	}
625 	spin_unlock(&dlm->spinlock);
626 }
627 
628 int dlm_joined(struct dlm_ctxt *dlm)
629 {
630 	int ret = 0;
631 
632 	spin_lock(&dlm_domain_lock);
633 
634 	if (dlm->dlm_state == DLM_CTXT_JOINED)
635 		ret = 1;
636 
637 	spin_unlock(&dlm_domain_lock);
638 
639 	return ret;
640 }
641 
642 int dlm_shutting_down(struct dlm_ctxt *dlm)
643 {
644 	int ret = 0;
645 
646 	spin_lock(&dlm_domain_lock);
647 
648 	if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
649 		ret = 1;
650 
651 	spin_unlock(&dlm_domain_lock);
652 
653 	return ret;
654 }
655 
656 void dlm_unregister_domain(struct dlm_ctxt *dlm)
657 {
658 	int leave = 0;
659 	struct dlm_lock_resource *res;
660 
661 	spin_lock(&dlm_domain_lock);
662 	BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
663 	BUG_ON(!dlm->num_joins);
664 
665 	dlm->num_joins--;
666 	if (!dlm->num_joins) {
667 		/* We mark it "in shutdown" now so new register
668 		 * requests wait until we've completely left the
669 		 * domain. Don't use DLM_CTXT_LEAVING yet as we still
670 		 * want new domain joins to communicate with us at
671 		 * least until we've completed migration of our
672 		 * resources. */
673 		dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
674 		leave = 1;
675 	}
676 	spin_unlock(&dlm_domain_lock);
677 
678 	if (leave) {
679 		mlog(0, "shutting down domain %s\n", dlm->name);
680 
681 		/* We changed dlm state, notify the thread */
682 		dlm_kick_thread(dlm, NULL);
683 
684 		while (dlm_migrate_all_locks(dlm)) {
685 			/* Give dlm_thread time to purge the lockres' */
686 			msleep(500);
687 			mlog(0, "%s: more migration to do\n", dlm->name);
688 		}
689 
690 		/* This list should be empty. If not, print remaining lockres */
691 		if (!list_empty(&dlm->tracking_list)) {
692 			mlog(ML_ERROR, "Following lockres' are still on the "
693 			     "tracking list:\n");
694 			list_for_each_entry(res, &dlm->tracking_list, tracking)
695 				dlm_print_one_lock_resource(res);
696 		}
697 
698 		dlm_mark_domain_leaving(dlm);
699 		dlm_leave_domain(dlm);
700 		dlm_force_free_mles(dlm);
701 		dlm_complete_dlm_shutdown(dlm);
702 	}
703 	dlm_put(dlm);
704 }
705 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
706 
707 static int dlm_query_join_proto_check(char *proto_type, int node,
708 				      struct dlm_protocol_version *ours,
709 				      struct dlm_protocol_version *request)
710 {
711 	int rc;
712 	struct dlm_protocol_version proto = *request;
713 
714 	if (!dlm_protocol_compare(ours, &proto)) {
715 		mlog(0,
716 		     "node %u wanted to join with %s locking protocol "
717 		     "%u.%u, we respond with %u.%u\n",
718 		     node, proto_type,
719 		     request->pv_major,
720 		     request->pv_minor,
721 		     proto.pv_major, proto.pv_minor);
722 		request->pv_minor = proto.pv_minor;
723 		rc = 0;
724 	} else {
725 		mlog(ML_NOTICE,
726 		     "Node %u wanted to join with %s locking "
727 		     "protocol %u.%u, but we have %u.%u, disallowing\n",
728 		     node, proto_type,
729 		     request->pv_major,
730 		     request->pv_minor,
731 		     ours->pv_major,
732 		     ours->pv_minor);
733 		rc = 1;
734 	}
735 
736 	return rc;
737 }
738 
739 /*
740  * struct dlm_query_join_packet is made up of four one-byte fields.  They
741  * are effectively in big-endian order already.  However, little-endian
742  * machines swap them before putting the packet on the wire (because
743  * query_join's response is a status, and that status is treated as a u32
744  * on the wire).  Thus, a big-endian and little-endian machines will treat
745  * this structure differently.
746  *
747  * The solution is to have little-endian machines swap the structure when
748  * converting from the structure to the u32 representation.  This will
749  * result in the structure having the correct format on the wire no matter
750  * the host endian format.
751  */
752 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
753 					  u32 *wire)
754 {
755 	union dlm_query_join_response response;
756 
757 	response.packet = *packet;
758 	*wire = cpu_to_be32(response.intval);
759 }
760 
761 static void dlm_query_join_wire_to_packet(u32 wire,
762 					  struct dlm_query_join_packet *packet)
763 {
764 	union dlm_query_join_response response;
765 
766 	response.intval = cpu_to_be32(wire);
767 	*packet = response.packet;
768 }
769 
770 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
771 				  void **ret_data)
772 {
773 	struct dlm_query_join_request *query;
774 	struct dlm_query_join_packet packet = {
775 		.code = JOIN_DISALLOW,
776 	};
777 	struct dlm_ctxt *dlm = NULL;
778 	u32 response;
779 	u8 nodenum;
780 
781 	query = (struct dlm_query_join_request *) msg->buf;
782 
783 	mlog(0, "node %u wants to join domain %s\n", query->node_idx,
784 		  query->domain);
785 
786 	/*
787 	 * If heartbeat doesn't consider the node live, tell it
788 	 * to back off and try again.  This gives heartbeat a chance
789 	 * to catch up.
790 	 */
791 	if (!o2hb_check_node_heartbeating(query->node_idx)) {
792 		mlog(0, "node %u is not in our live map yet\n",
793 		     query->node_idx);
794 
795 		packet.code = JOIN_DISALLOW;
796 		goto respond;
797 	}
798 
799 	packet.code = JOIN_OK_NO_MAP;
800 
801 	spin_lock(&dlm_domain_lock);
802 	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
803 	if (!dlm)
804 		goto unlock_respond;
805 
806 	/*
807 	 * There is a small window where the joining node may not see the
808 	 * node(s) that just left but still part of the cluster. DISALLOW
809 	 * join request if joining node has different node map.
810 	 */
811 	nodenum=0;
812 	while (nodenum < O2NM_MAX_NODES) {
813 		if (test_bit(nodenum, dlm->domain_map)) {
814 			if (!byte_test_bit(nodenum, query->node_map)) {
815 				mlog(0, "disallow join as node %u does not "
816 				     "have node %u in its nodemap\n",
817 				     query->node_idx, nodenum);
818 				packet.code = JOIN_DISALLOW;
819 				goto unlock_respond;
820 			}
821 		}
822 		nodenum++;
823 	}
824 
825 	/* Once the dlm ctxt is marked as leaving then we don't want
826 	 * to be put in someone's domain map.
827 	 * Also, explicitly disallow joining at certain troublesome
828 	 * times (ie. during recovery). */
829 	if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
830 		int bit = query->node_idx;
831 		spin_lock(&dlm->spinlock);
832 
833 		if (dlm->dlm_state == DLM_CTXT_NEW &&
834 		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
835 			/*If this is a brand new context and we
836 			 * haven't started our join process yet, then
837 			 * the other node won the race. */
838 			packet.code = JOIN_OK_NO_MAP;
839 		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
840 			/* Disallow parallel joins. */
841 			packet.code = JOIN_DISALLOW;
842 		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
843 			mlog(0, "node %u trying to join, but recovery "
844 			     "is ongoing.\n", bit);
845 			packet.code = JOIN_DISALLOW;
846 		} else if (test_bit(bit, dlm->recovery_map)) {
847 			mlog(0, "node %u trying to join, but it "
848 			     "still needs recovery.\n", bit);
849 			packet.code = JOIN_DISALLOW;
850 		} else if (test_bit(bit, dlm->domain_map)) {
851 			mlog(0, "node %u trying to join, but it "
852 			     "is still in the domain! needs recovery?\n",
853 			     bit);
854 			packet.code = JOIN_DISALLOW;
855 		} else {
856 			/* Alright we're fully a part of this domain
857 			 * so we keep some state as to who's joining
858 			 * and indicate to him that needs to be fixed
859 			 * up. */
860 
861 			/* Make sure we speak compatible locking protocols.  */
862 			if (dlm_query_join_proto_check("DLM", bit,
863 						       &dlm->dlm_locking_proto,
864 						       &query->dlm_proto)) {
865 				packet.code = JOIN_PROTOCOL_MISMATCH;
866 			} else if (dlm_query_join_proto_check("fs", bit,
867 							      &dlm->fs_locking_proto,
868 							      &query->fs_proto)) {
869 				packet.code = JOIN_PROTOCOL_MISMATCH;
870 			} else {
871 				packet.dlm_minor = query->dlm_proto.pv_minor;
872 				packet.fs_minor = query->fs_proto.pv_minor;
873 				packet.code = JOIN_OK;
874 				__dlm_set_joining_node(dlm, query->node_idx);
875 			}
876 		}
877 
878 		spin_unlock(&dlm->spinlock);
879 	}
880 unlock_respond:
881 	spin_unlock(&dlm_domain_lock);
882 
883 respond:
884 	mlog(0, "We respond with %u\n", packet.code);
885 
886 	dlm_query_join_packet_to_wire(&packet, &response);
887 	return response;
888 }
889 
890 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
891 				     void **ret_data)
892 {
893 	struct dlm_assert_joined *assert;
894 	struct dlm_ctxt *dlm = NULL;
895 
896 	assert = (struct dlm_assert_joined *) msg->buf;
897 
898 	mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
899 		  assert->domain);
900 
901 	spin_lock(&dlm_domain_lock);
902 	dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
903 	/* XXX should we consider no dlm ctxt an error? */
904 	if (dlm) {
905 		spin_lock(&dlm->spinlock);
906 
907 		/* Alright, this node has officially joined our
908 		 * domain. Set him in the map and clean up our
909 		 * leftover join state. */
910 		BUG_ON(dlm->joining_node != assert->node_idx);
911 		set_bit(assert->node_idx, dlm->domain_map);
912 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
913 
914 		printk(KERN_NOTICE "o2dlm: Node %u joins domain %s\n",
915 		       assert->node_idx, dlm->name);
916 		__dlm_print_nodes(dlm);
917 
918 		/* notify anything attached to the heartbeat events */
919 		dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
920 
921 		spin_unlock(&dlm->spinlock);
922 	}
923 	spin_unlock(&dlm_domain_lock);
924 
925 	return 0;
926 }
927 
928 static int dlm_match_regions(struct dlm_ctxt *dlm,
929 			     struct dlm_query_region *qr)
930 {
931 	char *local = NULL, *remote = qr->qr_regions;
932 	char *l, *r;
933 	int localnr, i, j, foundit;
934 	int status = 0;
935 
936 	if (!o2hb_global_heartbeat_active()) {
937 		if (qr->qr_numregions) {
938 			mlog(ML_ERROR, "Domain %s: Joining node %d has global "
939 			     "heartbeat enabled but local node %d does not\n",
940 			     qr->qr_domain, qr->qr_node, dlm->node_num);
941 			status = -EINVAL;
942 		}
943 		goto bail;
944 	}
945 
946 	if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
947 		mlog(ML_ERROR, "Domain %s: Local node %d has global "
948 		     "heartbeat enabled but joining node %d does not\n",
949 		     qr->qr_domain, dlm->node_num, qr->qr_node);
950 		status = -EINVAL;
951 		goto bail;
952 	}
953 
954 	r = remote;
955 	for (i = 0; i < qr->qr_numregions; ++i) {
956 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
957 		r += O2HB_MAX_REGION_NAME_LEN;
958 	}
959 
960 	local = kmalloc(sizeof(qr->qr_regions), GFP_ATOMIC);
961 	if (!local) {
962 		status = -ENOMEM;
963 		goto bail;
964 	}
965 
966 	localnr = o2hb_get_all_regions(local, O2NM_MAX_REGIONS);
967 
968 	/* compare local regions with remote */
969 	l = local;
970 	for (i = 0; i < localnr; ++i) {
971 		foundit = 0;
972 		r = remote;
973 		for (j = 0; j <= qr->qr_numregions; ++j) {
974 			if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
975 				foundit = 1;
976 				break;
977 			}
978 			r += O2HB_MAX_REGION_NAME_LEN;
979 		}
980 		if (!foundit) {
981 			status = -EINVAL;
982 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
983 			     "in local node %d but not in joining node %d\n",
984 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
985 			     dlm->node_num, qr->qr_node);
986 			goto bail;
987 		}
988 		l += O2HB_MAX_REGION_NAME_LEN;
989 	}
990 
991 	/* compare remote with local regions */
992 	r = remote;
993 	for (i = 0; i < qr->qr_numregions; ++i) {
994 		foundit = 0;
995 		l = local;
996 		for (j = 0; j < localnr; ++j) {
997 			if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
998 				foundit = 1;
999 				break;
1000 			}
1001 			l += O2HB_MAX_REGION_NAME_LEN;
1002 		}
1003 		if (!foundit) {
1004 			status = -EINVAL;
1005 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1006 			     "in joining node %d but not in local node %d\n",
1007 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1008 			     qr->qr_node, dlm->node_num);
1009 			goto bail;
1010 		}
1011 		r += O2HB_MAX_REGION_NAME_LEN;
1012 	}
1013 
1014 bail:
1015 	kfree(local);
1016 
1017 	return status;
1018 }
1019 
1020 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1021 {
1022 	struct dlm_query_region *qr = NULL;
1023 	int status, ret = 0, i;
1024 	char *p;
1025 
1026 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1027 		goto bail;
1028 
1029 	qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1030 	if (!qr) {
1031 		ret = -ENOMEM;
1032 		mlog_errno(ret);
1033 		goto bail;
1034 	}
1035 
1036 	qr->qr_node = dlm->node_num;
1037 	qr->qr_namelen = strlen(dlm->name);
1038 	memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1039 	/* if local hb, the numregions will be zero */
1040 	if (o2hb_global_heartbeat_active())
1041 		qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1042 							 O2NM_MAX_REGIONS);
1043 
1044 	p = qr->qr_regions;
1045 	for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1046 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1047 
1048 	i = -1;
1049 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1050 				  i + 1)) < O2NM_MAX_NODES) {
1051 		if (i == dlm->node_num)
1052 			continue;
1053 
1054 		mlog(0, "Sending regions to node %d\n", i);
1055 
1056 		ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1057 					 sizeof(struct dlm_query_region),
1058 					 i, &status);
1059 		if (ret >= 0)
1060 			ret = status;
1061 		if (ret) {
1062 			mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1063 			     ret, i);
1064 			break;
1065 		}
1066 	}
1067 
1068 bail:
1069 	kfree(qr);
1070 	return ret;
1071 }
1072 
1073 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1074 				    void *data, void **ret_data)
1075 {
1076 	struct dlm_query_region *qr;
1077 	struct dlm_ctxt *dlm = NULL;
1078 	int status = 0;
1079 	int locked = 0;
1080 
1081 	qr = (struct dlm_query_region *) msg->buf;
1082 
1083 	mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1084 	     qr->qr_domain);
1085 
1086 	status = -EINVAL;
1087 
1088 	spin_lock(&dlm_domain_lock);
1089 	dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1090 	if (!dlm) {
1091 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1092 		     "before join domain\n", qr->qr_node, qr->qr_domain);
1093 		goto bail;
1094 	}
1095 
1096 	spin_lock(&dlm->spinlock);
1097 	locked = 1;
1098 	if (dlm->joining_node != qr->qr_node) {
1099 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1100 		     "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1101 		     dlm->joining_node);
1102 		goto bail;
1103 	}
1104 
1105 	/* Support for global heartbeat was added in 1.1 */
1106 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1107 	    dlm->dlm_locking_proto.pv_minor == 0) {
1108 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1109 		     "but active dlm protocol is %d.%d\n", qr->qr_node,
1110 		     qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1111 		     dlm->dlm_locking_proto.pv_minor);
1112 		goto bail;
1113 	}
1114 
1115 	status = dlm_match_regions(dlm, qr);
1116 
1117 bail:
1118 	if (locked)
1119 		spin_unlock(&dlm->spinlock);
1120 	spin_unlock(&dlm_domain_lock);
1121 
1122 	return status;
1123 }
1124 
1125 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1126 {
1127 	struct o2nm_node *local;
1128 	struct dlm_node_info *remote;
1129 	int i, j;
1130 	int status = 0;
1131 
1132 	for (j = 0; j < qn->qn_numnodes; ++j)
1133 		mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1134 		     &(qn->qn_nodes[j].ni_ipv4_address),
1135 		     ntohs(qn->qn_nodes[j].ni_ipv4_port));
1136 
1137 	for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1138 		local = o2nm_get_node_by_num(i);
1139 		remote = NULL;
1140 		for (j = 0; j < qn->qn_numnodes; ++j) {
1141 			if (qn->qn_nodes[j].ni_nodenum == i) {
1142 				remote = &(qn->qn_nodes[j]);
1143 				break;
1144 			}
1145 		}
1146 
1147 		if (!local && !remote)
1148 			continue;
1149 
1150 		if ((local && !remote) || (!local && remote))
1151 			status = -EINVAL;
1152 
1153 		if (!status &&
1154 		    ((remote->ni_nodenum != local->nd_num) ||
1155 		     (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1156 		     (remote->ni_ipv4_address != local->nd_ipv4_address)))
1157 			status = -EINVAL;
1158 
1159 		if (status) {
1160 			if (remote && !local)
1161 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1162 				     "registered in joining node %d but not in "
1163 				     "local node %d\n", qn->qn_domain,
1164 				     remote->ni_nodenum,
1165 				     &(remote->ni_ipv4_address),
1166 				     ntohs(remote->ni_ipv4_port),
1167 				     qn->qn_nodenum, dlm->node_num);
1168 			if (local && !remote)
1169 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1170 				     "registered in local node %d but not in "
1171 				     "joining node %d\n", qn->qn_domain,
1172 				     local->nd_num, &(local->nd_ipv4_address),
1173 				     ntohs(local->nd_ipv4_port),
1174 				     dlm->node_num, qn->qn_nodenum);
1175 			BUG_ON((!local && !remote));
1176 		}
1177 
1178 		if (local)
1179 			o2nm_node_put(local);
1180 	}
1181 
1182 	return status;
1183 }
1184 
1185 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1186 {
1187 	struct dlm_query_nodeinfo *qn = NULL;
1188 	struct o2nm_node *node;
1189 	int ret = 0, status, count, i;
1190 
1191 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1192 		goto bail;
1193 
1194 	qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1195 	if (!qn) {
1196 		ret = -ENOMEM;
1197 		mlog_errno(ret);
1198 		goto bail;
1199 	}
1200 
1201 	for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1202 		node = o2nm_get_node_by_num(i);
1203 		if (!node)
1204 			continue;
1205 		qn->qn_nodes[count].ni_nodenum = node->nd_num;
1206 		qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1207 		qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1208 		mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1209 		     &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1210 		++count;
1211 		o2nm_node_put(node);
1212 	}
1213 
1214 	qn->qn_nodenum = dlm->node_num;
1215 	qn->qn_numnodes = count;
1216 	qn->qn_namelen = strlen(dlm->name);
1217 	memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1218 
1219 	i = -1;
1220 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1221 				  i + 1)) < O2NM_MAX_NODES) {
1222 		if (i == dlm->node_num)
1223 			continue;
1224 
1225 		mlog(0, "Sending nodeinfo to node %d\n", i);
1226 
1227 		ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1228 					 qn, sizeof(struct dlm_query_nodeinfo),
1229 					 i, &status);
1230 		if (ret >= 0)
1231 			ret = status;
1232 		if (ret) {
1233 			mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1234 			break;
1235 		}
1236 	}
1237 
1238 bail:
1239 	kfree(qn);
1240 	return ret;
1241 }
1242 
1243 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1244 				      void *data, void **ret_data)
1245 {
1246 	struct dlm_query_nodeinfo *qn;
1247 	struct dlm_ctxt *dlm = NULL;
1248 	int locked = 0, status = -EINVAL;
1249 
1250 	qn = (struct dlm_query_nodeinfo *) msg->buf;
1251 
1252 	mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1253 	     qn->qn_domain);
1254 
1255 	spin_lock(&dlm_domain_lock);
1256 	dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1257 	if (!dlm) {
1258 		mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1259 		     "join domain\n", qn->qn_nodenum, qn->qn_domain);
1260 		goto bail;
1261 	}
1262 
1263 	spin_lock(&dlm->spinlock);
1264 	locked = 1;
1265 	if (dlm->joining_node != qn->qn_nodenum) {
1266 		mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1267 		     "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1268 		     dlm->joining_node);
1269 		goto bail;
1270 	}
1271 
1272 	/* Support for node query was added in 1.1 */
1273 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1274 	    dlm->dlm_locking_proto.pv_minor == 0) {
1275 		mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1276 		     "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1277 		     qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1278 		     dlm->dlm_locking_proto.pv_minor);
1279 		goto bail;
1280 	}
1281 
1282 	status = dlm_match_nodes(dlm, qn);
1283 
1284 bail:
1285 	if (locked)
1286 		spin_unlock(&dlm->spinlock);
1287 	spin_unlock(&dlm_domain_lock);
1288 
1289 	return status;
1290 }
1291 
1292 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1293 				   void **ret_data)
1294 {
1295 	struct dlm_cancel_join *cancel;
1296 	struct dlm_ctxt *dlm = NULL;
1297 
1298 	cancel = (struct dlm_cancel_join *) msg->buf;
1299 
1300 	mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1301 		  cancel->domain);
1302 
1303 	spin_lock(&dlm_domain_lock);
1304 	dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1305 
1306 	if (dlm) {
1307 		spin_lock(&dlm->spinlock);
1308 
1309 		/* Yikes, this guy wants to cancel his join. No
1310 		 * problem, we simply cleanup our join state. */
1311 		BUG_ON(dlm->joining_node != cancel->node_idx);
1312 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1313 
1314 		spin_unlock(&dlm->spinlock);
1315 	}
1316 	spin_unlock(&dlm_domain_lock);
1317 
1318 	return 0;
1319 }
1320 
1321 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1322 				    unsigned int node)
1323 {
1324 	int status;
1325 	struct dlm_cancel_join cancel_msg;
1326 
1327 	memset(&cancel_msg, 0, sizeof(cancel_msg));
1328 	cancel_msg.node_idx = dlm->node_num;
1329 	cancel_msg.name_len = strlen(dlm->name);
1330 	memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1331 
1332 	status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1333 				    &cancel_msg, sizeof(cancel_msg), node,
1334 				    NULL);
1335 	if (status < 0) {
1336 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1337 		     "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1338 		     node);
1339 		goto bail;
1340 	}
1341 
1342 bail:
1343 	return status;
1344 }
1345 
1346 /* map_size should be in bytes. */
1347 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1348 				 unsigned long *node_map,
1349 				 unsigned int map_size)
1350 {
1351 	int status, tmpstat;
1352 	unsigned int node;
1353 
1354 	if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1355 			 sizeof(unsigned long))) {
1356 		mlog(ML_ERROR,
1357 		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1358 		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1359 		return -EINVAL;
1360 	}
1361 
1362 	status = 0;
1363 	node = -1;
1364 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1365 				     node + 1)) < O2NM_MAX_NODES) {
1366 		if (node == dlm->node_num)
1367 			continue;
1368 
1369 		tmpstat = dlm_send_one_join_cancel(dlm, node);
1370 		if (tmpstat) {
1371 			mlog(ML_ERROR, "Error return %d cancelling join on "
1372 			     "node %d\n", tmpstat, node);
1373 			if (!status)
1374 				status = tmpstat;
1375 		}
1376 	}
1377 
1378 	if (status)
1379 		mlog_errno(status);
1380 	return status;
1381 }
1382 
1383 static int dlm_request_join(struct dlm_ctxt *dlm,
1384 			    int node,
1385 			    enum dlm_query_join_response_code *response)
1386 {
1387 	int status;
1388 	struct dlm_query_join_request join_msg;
1389 	struct dlm_query_join_packet packet;
1390 	u32 join_resp;
1391 
1392 	mlog(0, "querying node %d\n", node);
1393 
1394 	memset(&join_msg, 0, sizeof(join_msg));
1395 	join_msg.node_idx = dlm->node_num;
1396 	join_msg.name_len = strlen(dlm->name);
1397 	memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1398 	join_msg.dlm_proto = dlm->dlm_locking_proto;
1399 	join_msg.fs_proto = dlm->fs_locking_proto;
1400 
1401 	/* copy live node map to join message */
1402 	byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1403 
1404 	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1405 				    sizeof(join_msg), node, &join_resp);
1406 	if (status < 0 && status != -ENOPROTOOPT) {
1407 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1408 		     "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1409 		     node);
1410 		goto bail;
1411 	}
1412 	dlm_query_join_wire_to_packet(join_resp, &packet);
1413 
1414 	/* -ENOPROTOOPT from the net code means the other side isn't
1415 	    listening for our message type -- that's fine, it means
1416 	    his dlm isn't up, so we can consider him a 'yes' but not
1417 	    joined into the domain.  */
1418 	if (status == -ENOPROTOOPT) {
1419 		status = 0;
1420 		*response = JOIN_OK_NO_MAP;
1421 	} else if (packet.code == JOIN_DISALLOW ||
1422 		   packet.code == JOIN_OK_NO_MAP) {
1423 		*response = packet.code;
1424 	} else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1425 		mlog(ML_NOTICE,
1426 		     "This node requested DLM locking protocol %u.%u and "
1427 		     "filesystem locking protocol %u.%u.  At least one of "
1428 		     "the protocol versions on node %d is not compatible, "
1429 		     "disconnecting\n",
1430 		     dlm->dlm_locking_proto.pv_major,
1431 		     dlm->dlm_locking_proto.pv_minor,
1432 		     dlm->fs_locking_proto.pv_major,
1433 		     dlm->fs_locking_proto.pv_minor,
1434 		     node);
1435 		status = -EPROTO;
1436 		*response = packet.code;
1437 	} else if (packet.code == JOIN_OK) {
1438 		*response = packet.code;
1439 		/* Use the same locking protocol as the remote node */
1440 		dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1441 		dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1442 		mlog(0,
1443 		     "Node %d responds JOIN_OK with DLM locking protocol "
1444 		     "%u.%u and fs locking protocol %u.%u\n",
1445 		     node,
1446 		     dlm->dlm_locking_proto.pv_major,
1447 		     dlm->dlm_locking_proto.pv_minor,
1448 		     dlm->fs_locking_proto.pv_major,
1449 		     dlm->fs_locking_proto.pv_minor);
1450 	} else {
1451 		status = -EINVAL;
1452 		mlog(ML_ERROR, "invalid response %d from node %u\n",
1453 		     packet.code, node);
1454 	}
1455 
1456 	mlog(0, "status %d, node %d response is %d\n", status, node,
1457 	     *response);
1458 
1459 bail:
1460 	return status;
1461 }
1462 
1463 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1464 				    unsigned int node)
1465 {
1466 	int status;
1467 	struct dlm_assert_joined assert_msg;
1468 
1469 	mlog(0, "Sending join assert to node %u\n", node);
1470 
1471 	memset(&assert_msg, 0, sizeof(assert_msg));
1472 	assert_msg.node_idx = dlm->node_num;
1473 	assert_msg.name_len = strlen(dlm->name);
1474 	memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1475 
1476 	status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1477 				    &assert_msg, sizeof(assert_msg), node,
1478 				    NULL);
1479 	if (status < 0)
1480 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1481 		     "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1482 		     node);
1483 
1484 	return status;
1485 }
1486 
1487 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1488 				  unsigned long *node_map)
1489 {
1490 	int status, node, live;
1491 
1492 	status = 0;
1493 	node = -1;
1494 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1495 				     node + 1)) < O2NM_MAX_NODES) {
1496 		if (node == dlm->node_num)
1497 			continue;
1498 
1499 		do {
1500 			/* It is very important that this message be
1501 			 * received so we spin until either the node
1502 			 * has died or it gets the message. */
1503 			status = dlm_send_one_join_assert(dlm, node);
1504 
1505 			spin_lock(&dlm->spinlock);
1506 			live = test_bit(node, dlm->live_nodes_map);
1507 			spin_unlock(&dlm->spinlock);
1508 
1509 			if (status) {
1510 				mlog(ML_ERROR, "Error return %d asserting "
1511 				     "join on node %d\n", status, node);
1512 
1513 				/* give us some time between errors... */
1514 				if (live)
1515 					msleep(DLM_DOMAIN_BACKOFF_MS);
1516 			}
1517 		} while (status && live);
1518 	}
1519 }
1520 
1521 struct domain_join_ctxt {
1522 	unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1523 	unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1524 };
1525 
1526 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1527 				   struct domain_join_ctxt *ctxt,
1528 				   enum dlm_query_join_response_code response)
1529 {
1530 	int ret;
1531 
1532 	if (response == JOIN_DISALLOW) {
1533 		mlog(0, "Latest response of disallow -- should restart\n");
1534 		return 1;
1535 	}
1536 
1537 	spin_lock(&dlm->spinlock);
1538 	/* For now, we restart the process if the node maps have
1539 	 * changed at all */
1540 	ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1541 		     sizeof(dlm->live_nodes_map));
1542 	spin_unlock(&dlm->spinlock);
1543 
1544 	if (ret)
1545 		mlog(0, "Node maps changed -- should restart\n");
1546 
1547 	return ret;
1548 }
1549 
1550 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1551 {
1552 	int status = 0, tmpstat, node;
1553 	struct domain_join_ctxt *ctxt;
1554 	enum dlm_query_join_response_code response = JOIN_DISALLOW;
1555 
1556 	mlog_entry("%p", dlm);
1557 
1558 	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1559 	if (!ctxt) {
1560 		status = -ENOMEM;
1561 		mlog_errno(status);
1562 		goto bail;
1563 	}
1564 
1565 	/* group sem locking should work for us here -- we're already
1566 	 * registered for heartbeat events so filling this should be
1567 	 * atomic wrt getting those handlers called. */
1568 	o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1569 
1570 	spin_lock(&dlm->spinlock);
1571 	memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1572 
1573 	__dlm_set_joining_node(dlm, dlm->node_num);
1574 
1575 	spin_unlock(&dlm->spinlock);
1576 
1577 	node = -1;
1578 	while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1579 				     node + 1)) < O2NM_MAX_NODES) {
1580 		if (node == dlm->node_num)
1581 			continue;
1582 
1583 		status = dlm_request_join(dlm, node, &response);
1584 		if (status < 0) {
1585 			mlog_errno(status);
1586 			goto bail;
1587 		}
1588 
1589 		/* Ok, either we got a response or the node doesn't have a
1590 		 * dlm up. */
1591 		if (response == JOIN_OK)
1592 			set_bit(node, ctxt->yes_resp_map);
1593 
1594 		if (dlm_should_restart_join(dlm, ctxt, response)) {
1595 			status = -EAGAIN;
1596 			goto bail;
1597 		}
1598 	}
1599 
1600 	mlog(0, "Yay, done querying nodes!\n");
1601 
1602 	/* Yay, everyone agree's we can join the domain. My domain is
1603 	 * comprised of all nodes who were put in the
1604 	 * yes_resp_map. Copy that into our domain map and send a join
1605 	 * assert message to clean up everyone elses state. */
1606 	spin_lock(&dlm->spinlock);
1607 	memcpy(dlm->domain_map, ctxt->yes_resp_map,
1608 	       sizeof(ctxt->yes_resp_map));
1609 	set_bit(dlm->node_num, dlm->domain_map);
1610 	spin_unlock(&dlm->spinlock);
1611 
1612 	/* Support for global heartbeat and node info was added in 1.1 */
1613 	if (dlm_protocol.pv_major > 1 || dlm_protocol.pv_minor > 0) {
1614 		status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1615 		if (status) {
1616 			mlog_errno(status);
1617 			goto bail;
1618 		}
1619 		status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1620 		if (status) {
1621 			mlog_errno(status);
1622 			goto bail;
1623 		}
1624 	}
1625 
1626 	dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1627 
1628 	/* Joined state *must* be set before the joining node
1629 	 * information, otherwise the query_join handler may read no
1630 	 * current joiner but a state of NEW and tell joining nodes
1631 	 * we're not in the domain. */
1632 	spin_lock(&dlm_domain_lock);
1633 	dlm->dlm_state = DLM_CTXT_JOINED;
1634 	dlm->num_joins++;
1635 	spin_unlock(&dlm_domain_lock);
1636 
1637 bail:
1638 	spin_lock(&dlm->spinlock);
1639 	__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1640 	if (!status)
1641 		__dlm_print_nodes(dlm);
1642 	spin_unlock(&dlm->spinlock);
1643 
1644 	if (ctxt) {
1645 		/* Do we need to send a cancel message to any nodes? */
1646 		if (status < 0) {
1647 			tmpstat = dlm_send_join_cancels(dlm,
1648 							ctxt->yes_resp_map,
1649 							sizeof(ctxt->yes_resp_map));
1650 			if (tmpstat < 0)
1651 				mlog_errno(tmpstat);
1652 		}
1653 		kfree(ctxt);
1654 	}
1655 
1656 	mlog(0, "returning %d\n", status);
1657 	return status;
1658 }
1659 
1660 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1661 {
1662 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1663 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1664 	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1665 }
1666 
1667 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1668 {
1669 	int status;
1670 
1671 	mlog(0, "registering handlers.\n");
1672 
1673 	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1674 			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1675 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1676 	if (status)
1677 		goto bail;
1678 
1679 	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1680 			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1681 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1682 	if (status)
1683 		goto bail;
1684 
1685 	status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1686 					sizeof(struct dlm_master_request),
1687 					dlm_master_request_handler,
1688 					dlm, NULL, &dlm->dlm_domain_handlers);
1689 	if (status)
1690 		goto bail;
1691 
1692 	status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1693 					sizeof(struct dlm_assert_master),
1694 					dlm_assert_master_handler,
1695 					dlm, dlm_assert_master_post_handler,
1696 					&dlm->dlm_domain_handlers);
1697 	if (status)
1698 		goto bail;
1699 
1700 	status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1701 					sizeof(struct dlm_create_lock),
1702 					dlm_create_lock_handler,
1703 					dlm, NULL, &dlm->dlm_domain_handlers);
1704 	if (status)
1705 		goto bail;
1706 
1707 	status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1708 					DLM_CONVERT_LOCK_MAX_LEN,
1709 					dlm_convert_lock_handler,
1710 					dlm, NULL, &dlm->dlm_domain_handlers);
1711 	if (status)
1712 		goto bail;
1713 
1714 	status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1715 					DLM_UNLOCK_LOCK_MAX_LEN,
1716 					dlm_unlock_lock_handler,
1717 					dlm, NULL, &dlm->dlm_domain_handlers);
1718 	if (status)
1719 		goto bail;
1720 
1721 	status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1722 					DLM_PROXY_AST_MAX_LEN,
1723 					dlm_proxy_ast_handler,
1724 					dlm, NULL, &dlm->dlm_domain_handlers);
1725 	if (status)
1726 		goto bail;
1727 
1728 	status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1729 					sizeof(struct dlm_exit_domain),
1730 					dlm_exit_domain_handler,
1731 					dlm, NULL, &dlm->dlm_domain_handlers);
1732 	if (status)
1733 		goto bail;
1734 
1735 	status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1736 					sizeof(struct dlm_deref_lockres),
1737 					dlm_deref_lockres_handler,
1738 					dlm, NULL, &dlm->dlm_domain_handlers);
1739 	if (status)
1740 		goto bail;
1741 
1742 	status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1743 					sizeof(struct dlm_migrate_request),
1744 					dlm_migrate_request_handler,
1745 					dlm, NULL, &dlm->dlm_domain_handlers);
1746 	if (status)
1747 		goto bail;
1748 
1749 	status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1750 					DLM_MIG_LOCKRES_MAX_LEN,
1751 					dlm_mig_lockres_handler,
1752 					dlm, NULL, &dlm->dlm_domain_handlers);
1753 	if (status)
1754 		goto bail;
1755 
1756 	status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1757 					sizeof(struct dlm_master_requery),
1758 					dlm_master_requery_handler,
1759 					dlm, NULL, &dlm->dlm_domain_handlers);
1760 	if (status)
1761 		goto bail;
1762 
1763 	status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1764 					sizeof(struct dlm_lock_request),
1765 					dlm_request_all_locks_handler,
1766 					dlm, NULL, &dlm->dlm_domain_handlers);
1767 	if (status)
1768 		goto bail;
1769 
1770 	status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1771 					sizeof(struct dlm_reco_data_done),
1772 					dlm_reco_data_done_handler,
1773 					dlm, NULL, &dlm->dlm_domain_handlers);
1774 	if (status)
1775 		goto bail;
1776 
1777 	status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1778 					sizeof(struct dlm_begin_reco),
1779 					dlm_begin_reco_handler,
1780 					dlm, NULL, &dlm->dlm_domain_handlers);
1781 	if (status)
1782 		goto bail;
1783 
1784 	status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1785 					sizeof(struct dlm_finalize_reco),
1786 					dlm_finalize_reco_handler,
1787 					dlm, NULL, &dlm->dlm_domain_handlers);
1788 	if (status)
1789 		goto bail;
1790 
1791 bail:
1792 	if (status)
1793 		dlm_unregister_domain_handlers(dlm);
1794 
1795 	return status;
1796 }
1797 
1798 static int dlm_join_domain(struct dlm_ctxt *dlm)
1799 {
1800 	int status;
1801 	unsigned int backoff;
1802 	unsigned int total_backoff = 0;
1803 
1804 	BUG_ON(!dlm);
1805 
1806 	mlog(0, "Join domain %s\n", dlm->name);
1807 
1808 	status = dlm_register_domain_handlers(dlm);
1809 	if (status) {
1810 		mlog_errno(status);
1811 		goto bail;
1812 	}
1813 
1814 	status = dlm_debug_init(dlm);
1815 	if (status < 0) {
1816 		mlog_errno(status);
1817 		goto bail;
1818 	}
1819 
1820 	status = dlm_launch_thread(dlm);
1821 	if (status < 0) {
1822 		mlog_errno(status);
1823 		goto bail;
1824 	}
1825 
1826 	status = dlm_launch_recovery_thread(dlm);
1827 	if (status < 0) {
1828 		mlog_errno(status);
1829 		goto bail;
1830 	}
1831 
1832 	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1833 	if (!dlm->dlm_worker) {
1834 		status = -ENOMEM;
1835 		mlog_errno(status);
1836 		goto bail;
1837 	}
1838 
1839 	do {
1840 		status = dlm_try_to_join_domain(dlm);
1841 
1842 		/* If we're racing another node to the join, then we
1843 		 * need to back off temporarily and let them
1844 		 * complete. */
1845 #define	DLM_JOIN_TIMEOUT_MSECS	90000
1846 		if (status == -EAGAIN) {
1847 			if (signal_pending(current)) {
1848 				status = -ERESTARTSYS;
1849 				goto bail;
1850 			}
1851 
1852 			if (total_backoff >
1853 			    msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1854 				status = -ERESTARTSYS;
1855 				mlog(ML_NOTICE, "Timed out joining dlm domain "
1856 				     "%s after %u msecs\n", dlm->name,
1857 				     jiffies_to_msecs(total_backoff));
1858 				goto bail;
1859 			}
1860 
1861 			/*
1862 			 * <chip> After you!
1863 			 * <dale> No, after you!
1864 			 * <chip> I insist!
1865 			 * <dale> But you first!
1866 			 * ...
1867 			 */
1868 			backoff = (unsigned int)(jiffies & 0x3);
1869 			backoff *= DLM_DOMAIN_BACKOFF_MS;
1870 			total_backoff += backoff;
1871 			mlog(0, "backoff %d\n", backoff);
1872 			msleep(backoff);
1873 		}
1874 	} while (status == -EAGAIN);
1875 
1876 	if (status < 0) {
1877 		mlog_errno(status);
1878 		goto bail;
1879 	}
1880 
1881 	status = 0;
1882 bail:
1883 	wake_up(&dlm_domain_events);
1884 
1885 	if (status) {
1886 		dlm_unregister_domain_handlers(dlm);
1887 		dlm_debug_shutdown(dlm);
1888 		dlm_complete_thread(dlm);
1889 		dlm_complete_recovery_thread(dlm);
1890 		dlm_destroy_dlm_worker(dlm);
1891 	}
1892 
1893 	return status;
1894 }
1895 
1896 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1897 				u32 key)
1898 {
1899 	int i;
1900 	int ret;
1901 	struct dlm_ctxt *dlm = NULL;
1902 
1903 	dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1904 	if (!dlm) {
1905 		mlog_errno(-ENOMEM);
1906 		goto leave;
1907 	}
1908 
1909 	dlm->name = kstrdup(domain, GFP_KERNEL);
1910 	if (dlm->name == NULL) {
1911 		mlog_errno(-ENOMEM);
1912 		kfree(dlm);
1913 		dlm = NULL;
1914 		goto leave;
1915 	}
1916 
1917 	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1918 	if (!dlm->lockres_hash) {
1919 		mlog_errno(-ENOMEM);
1920 		kfree(dlm->name);
1921 		kfree(dlm);
1922 		dlm = NULL;
1923 		goto leave;
1924 	}
1925 
1926 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
1927 		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1928 
1929 	dlm->master_hash = (struct hlist_head **)
1930 				dlm_alloc_pagevec(DLM_HASH_PAGES);
1931 	if (!dlm->master_hash) {
1932 		mlog_errno(-ENOMEM);
1933 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1934 		kfree(dlm->name);
1935 		kfree(dlm);
1936 		dlm = NULL;
1937 		goto leave;
1938 	}
1939 
1940 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
1941 		INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1942 
1943 	dlm->key = key;
1944 	dlm->node_num = o2nm_this_node();
1945 
1946 	ret = dlm_create_debugfs_subroot(dlm);
1947 	if (ret < 0) {
1948 		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
1949 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1950 		kfree(dlm->name);
1951 		kfree(dlm);
1952 		dlm = NULL;
1953 		goto leave;
1954 	}
1955 
1956 	spin_lock_init(&dlm->spinlock);
1957 	spin_lock_init(&dlm->master_lock);
1958 	spin_lock_init(&dlm->ast_lock);
1959 	spin_lock_init(&dlm->track_lock);
1960 	INIT_LIST_HEAD(&dlm->list);
1961 	INIT_LIST_HEAD(&dlm->dirty_list);
1962 	INIT_LIST_HEAD(&dlm->reco.resources);
1963 	INIT_LIST_HEAD(&dlm->reco.received);
1964 	INIT_LIST_HEAD(&dlm->reco.node_data);
1965 	INIT_LIST_HEAD(&dlm->purge_list);
1966 	INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1967 	INIT_LIST_HEAD(&dlm->tracking_list);
1968 	dlm->reco.state = 0;
1969 
1970 	INIT_LIST_HEAD(&dlm->pending_asts);
1971 	INIT_LIST_HEAD(&dlm->pending_basts);
1972 
1973 	mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1974 		  dlm->recovery_map, &(dlm->recovery_map[0]));
1975 
1976 	memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1977 	memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1978 	memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1979 
1980 	dlm->dlm_thread_task = NULL;
1981 	dlm->dlm_reco_thread_task = NULL;
1982 	dlm->dlm_worker = NULL;
1983 	init_waitqueue_head(&dlm->dlm_thread_wq);
1984 	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1985 	init_waitqueue_head(&dlm->reco.event);
1986 	init_waitqueue_head(&dlm->ast_wq);
1987 	init_waitqueue_head(&dlm->migration_wq);
1988 	INIT_LIST_HEAD(&dlm->mle_hb_events);
1989 
1990 	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1991 	init_waitqueue_head(&dlm->dlm_join_events);
1992 
1993 	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1994 	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1995 
1996 	atomic_set(&dlm->res_tot_count, 0);
1997 	atomic_set(&dlm->res_cur_count, 0);
1998 	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
1999 		atomic_set(&dlm->mle_tot_count[i], 0);
2000 		atomic_set(&dlm->mle_cur_count[i], 0);
2001 	}
2002 
2003 	spin_lock_init(&dlm->work_lock);
2004 	INIT_LIST_HEAD(&dlm->work_list);
2005 	INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2006 
2007 	kref_init(&dlm->dlm_refs);
2008 	dlm->dlm_state = DLM_CTXT_NEW;
2009 
2010 	INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2011 
2012 	mlog(0, "context init: refcount %u\n",
2013 		  atomic_read(&dlm->dlm_refs.refcount));
2014 
2015 leave:
2016 	return dlm;
2017 }
2018 
2019 /*
2020  * Compare a requested locking protocol version against the current one.
2021  *
2022  * If the major numbers are different, they are incompatible.
2023  * If the current minor is greater than the request, they are incompatible.
2024  * If the current minor is less than or equal to the request, they are
2025  * compatible, and the requester should run at the current minor version.
2026  */
2027 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2028 				struct dlm_protocol_version *request)
2029 {
2030 	if (existing->pv_major != request->pv_major)
2031 		return 1;
2032 
2033 	if (existing->pv_minor > request->pv_minor)
2034 		return 1;
2035 
2036 	if (existing->pv_minor < request->pv_minor)
2037 		request->pv_minor = existing->pv_minor;
2038 
2039 	return 0;
2040 }
2041 
2042 /*
2043  * dlm_register_domain: one-time setup per "domain".
2044  *
2045  * The filesystem passes in the requested locking version via proto.
2046  * If registration was successful, proto will contain the negotiated
2047  * locking protocol.
2048  */
2049 struct dlm_ctxt * dlm_register_domain(const char *domain,
2050 			       u32 key,
2051 			       struct dlm_protocol_version *fs_proto)
2052 {
2053 	int ret;
2054 	struct dlm_ctxt *dlm = NULL;
2055 	struct dlm_ctxt *new_ctxt = NULL;
2056 
2057 	if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2058 		ret = -ENAMETOOLONG;
2059 		mlog(ML_ERROR, "domain name length too long\n");
2060 		goto leave;
2061 	}
2062 
2063 	if (!o2hb_check_local_node_heartbeating()) {
2064 		mlog(ML_ERROR, "the local node has not been configured, or is "
2065 		     "not heartbeating\n");
2066 		ret = -EPROTO;
2067 		goto leave;
2068 	}
2069 
2070 	mlog(0, "register called for domain \"%s\"\n", domain);
2071 
2072 retry:
2073 	dlm = NULL;
2074 	if (signal_pending(current)) {
2075 		ret = -ERESTARTSYS;
2076 		mlog_errno(ret);
2077 		goto leave;
2078 	}
2079 
2080 	spin_lock(&dlm_domain_lock);
2081 
2082 	dlm = __dlm_lookup_domain(domain);
2083 	if (dlm) {
2084 		if (dlm->dlm_state != DLM_CTXT_JOINED) {
2085 			spin_unlock(&dlm_domain_lock);
2086 
2087 			mlog(0, "This ctxt is not joined yet!\n");
2088 			wait_event_interruptible(dlm_domain_events,
2089 						 dlm_wait_on_domain_helper(
2090 							 domain));
2091 			goto retry;
2092 		}
2093 
2094 		if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2095 			spin_unlock(&dlm_domain_lock);
2096 			mlog(ML_ERROR,
2097 			     "Requested locking protocol version is not "
2098 			     "compatible with already registered domain "
2099 			     "\"%s\"\n", domain);
2100 			ret = -EPROTO;
2101 			goto leave;
2102 		}
2103 
2104 		__dlm_get(dlm);
2105 		dlm->num_joins++;
2106 
2107 		spin_unlock(&dlm_domain_lock);
2108 
2109 		ret = 0;
2110 		goto leave;
2111 	}
2112 
2113 	/* doesn't exist */
2114 	if (!new_ctxt) {
2115 		spin_unlock(&dlm_domain_lock);
2116 
2117 		new_ctxt = dlm_alloc_ctxt(domain, key);
2118 		if (new_ctxt)
2119 			goto retry;
2120 
2121 		ret = -ENOMEM;
2122 		mlog_errno(ret);
2123 		goto leave;
2124 	}
2125 
2126 	/* a little variable switch-a-roo here... */
2127 	dlm = new_ctxt;
2128 	new_ctxt = NULL;
2129 
2130 	/* add the new domain */
2131 	list_add_tail(&dlm->list, &dlm_domains);
2132 	spin_unlock(&dlm_domain_lock);
2133 
2134 	/*
2135 	 * Pass the locking protocol version into the join.  If the join
2136 	 * succeeds, it will have the negotiated protocol set.
2137 	 */
2138 	dlm->dlm_locking_proto = dlm_protocol;
2139 	dlm->fs_locking_proto = *fs_proto;
2140 
2141 	ret = dlm_join_domain(dlm);
2142 	if (ret) {
2143 		mlog_errno(ret);
2144 		dlm_put(dlm);
2145 		goto leave;
2146 	}
2147 
2148 	/* Tell the caller what locking protocol we negotiated */
2149 	*fs_proto = dlm->fs_locking_proto;
2150 
2151 	ret = 0;
2152 leave:
2153 	if (new_ctxt)
2154 		dlm_free_ctxt_mem(new_ctxt);
2155 
2156 	if (ret < 0)
2157 		dlm = ERR_PTR(ret);
2158 
2159 	return dlm;
2160 }
2161 EXPORT_SYMBOL_GPL(dlm_register_domain);
2162 
2163 static LIST_HEAD(dlm_join_handlers);
2164 
2165 static void dlm_unregister_net_handlers(void)
2166 {
2167 	o2net_unregister_handler_list(&dlm_join_handlers);
2168 }
2169 
2170 static int dlm_register_net_handlers(void)
2171 {
2172 	int status = 0;
2173 
2174 	status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2175 					sizeof(struct dlm_query_join_request),
2176 					dlm_query_join_handler,
2177 					NULL, NULL, &dlm_join_handlers);
2178 	if (status)
2179 		goto bail;
2180 
2181 	status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2182 					sizeof(struct dlm_assert_joined),
2183 					dlm_assert_joined_handler,
2184 					NULL, NULL, &dlm_join_handlers);
2185 	if (status)
2186 		goto bail;
2187 
2188 	status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2189 					sizeof(struct dlm_cancel_join),
2190 					dlm_cancel_join_handler,
2191 					NULL, NULL, &dlm_join_handlers);
2192 	if (status)
2193 		goto bail;
2194 
2195 	status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2196 					sizeof(struct dlm_query_region),
2197 					dlm_query_region_handler,
2198 					NULL, NULL, &dlm_join_handlers);
2199 
2200 	if (status)
2201 		goto bail;
2202 
2203 	status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2204 					sizeof(struct dlm_query_nodeinfo),
2205 					dlm_query_nodeinfo_handler,
2206 					NULL, NULL, &dlm_join_handlers);
2207 bail:
2208 	if (status < 0)
2209 		dlm_unregister_net_handlers();
2210 
2211 	return status;
2212 }
2213 
2214 /* Domain eviction callback handling.
2215  *
2216  * The file system requires notification of node death *before* the
2217  * dlm completes it's recovery work, otherwise it may be able to
2218  * acquire locks on resources requiring recovery. Since the dlm can
2219  * evict a node from it's domain *before* heartbeat fires, a similar
2220  * mechanism is required. */
2221 
2222 /* Eviction is not expected to happen often, so a per-domain lock is
2223  * not necessary. Eviction callbacks are allowed to sleep for short
2224  * periods of time. */
2225 static DECLARE_RWSEM(dlm_callback_sem);
2226 
2227 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2228 					int node_num)
2229 {
2230 	struct list_head *iter;
2231 	struct dlm_eviction_cb *cb;
2232 
2233 	down_read(&dlm_callback_sem);
2234 	list_for_each(iter, &dlm->dlm_eviction_callbacks) {
2235 		cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
2236 
2237 		cb->ec_func(node_num, cb->ec_data);
2238 	}
2239 	up_read(&dlm_callback_sem);
2240 }
2241 
2242 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2243 			   dlm_eviction_func *f,
2244 			   void *data)
2245 {
2246 	INIT_LIST_HEAD(&cb->ec_item);
2247 	cb->ec_func = f;
2248 	cb->ec_data = data;
2249 }
2250 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2251 
2252 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2253 			      struct dlm_eviction_cb *cb)
2254 {
2255 	down_write(&dlm_callback_sem);
2256 	list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2257 	up_write(&dlm_callback_sem);
2258 }
2259 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2260 
2261 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2262 {
2263 	down_write(&dlm_callback_sem);
2264 	list_del_init(&cb->ec_item);
2265 	up_write(&dlm_callback_sem);
2266 }
2267 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2268 
2269 static int __init dlm_init(void)
2270 {
2271 	int status;
2272 
2273 	dlm_print_version();
2274 
2275 	status = dlm_init_mle_cache();
2276 	if (status) {
2277 		mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2278 		goto error;
2279 	}
2280 
2281 	status = dlm_init_master_caches();
2282 	if (status) {
2283 		mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2284 		     "o2dlm_lockname slabcaches\n");
2285 		goto error;
2286 	}
2287 
2288 	status = dlm_init_lock_cache();
2289 	if (status) {
2290 		mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2291 		goto error;
2292 	}
2293 
2294 	status = dlm_register_net_handlers();
2295 	if (status) {
2296 		mlog(ML_ERROR, "Unable to register network handlers\n");
2297 		goto error;
2298 	}
2299 
2300 	status = dlm_create_debugfs_root();
2301 	if (status)
2302 		goto error;
2303 
2304 	return 0;
2305 error:
2306 	dlm_unregister_net_handlers();
2307 	dlm_destroy_lock_cache();
2308 	dlm_destroy_master_caches();
2309 	dlm_destroy_mle_cache();
2310 	return -1;
2311 }
2312 
2313 static void __exit dlm_exit (void)
2314 {
2315 	dlm_destroy_debugfs_root();
2316 	dlm_unregister_net_handlers();
2317 	dlm_destroy_lock_cache();
2318 	dlm_destroy_master_caches();
2319 	dlm_destroy_mle_cache();
2320 }
2321 
2322 MODULE_AUTHOR("Oracle");
2323 MODULE_LICENSE("GPL");
2324 
2325 module_init(dlm_init);
2326 module_exit(dlm_exit);
2327