xref: /openbmc/linux/fs/ocfs2/dlm/dlmdomain.c (revision 293d5b43)
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 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
47 #include "cluster/masklog.h"
48 
49 /*
50  * ocfs2 node maps are array of long int, which limits to send them freely
51  * across the wire due to endianness issues. To workaround this, we convert
52  * long ints to byte arrays. Following 3 routines are helper functions to
53  * set/test/copy bits within those array of bytes
54  */
55 static inline void byte_set_bit(u8 nr, u8 map[])
56 {
57 	map[nr >> 3] |= (1UL << (nr & 7));
58 }
59 
60 static inline int byte_test_bit(u8 nr, u8 map[])
61 {
62 	return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
63 }
64 
65 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
66 			unsigned int sz)
67 {
68 	unsigned int nn;
69 
70 	if (!sz)
71 		return;
72 
73 	memset(dmap, 0, ((sz + 7) >> 3));
74 	for (nn = 0 ; nn < sz; nn++)
75 		if (test_bit(nn, smap))
76 			byte_set_bit(nn, dmap);
77 }
78 
79 static void dlm_free_pagevec(void **vec, int pages)
80 {
81 	while (pages--)
82 		free_page((unsigned long)vec[pages]);
83 	kfree(vec);
84 }
85 
86 static void **dlm_alloc_pagevec(int pages)
87 {
88 	void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
89 	int i;
90 
91 	if (!vec)
92 		return NULL;
93 
94 	for (i = 0; i < pages; i++)
95 		if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
96 			goto out_free;
97 
98 	mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
99 	     pages, (unsigned long)DLM_HASH_PAGES,
100 	     (unsigned long)DLM_BUCKETS_PER_PAGE);
101 	return vec;
102 out_free:
103 	dlm_free_pagevec(vec, i);
104 	return NULL;
105 }
106 
107 /*
108  *
109  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
110  *    dlm_domain_lock
111  *    struct dlm_ctxt->spinlock
112  *    struct dlm_lock_resource->spinlock
113  *    struct dlm_ctxt->master_lock
114  *    struct dlm_ctxt->ast_lock
115  *    dlm_master_list_entry->spinlock
116  *    dlm_lock->spinlock
117  *
118  */
119 
120 DEFINE_SPINLOCK(dlm_domain_lock);
121 LIST_HEAD(dlm_domains);
122 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
123 
124 /*
125  * The supported protocol version for DLM communication.  Running domains
126  * will have a negotiated version with the same major number and a minor
127  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
128  * be used to determine what a running domain is actually using.
129  *
130  * New in version 1.1:
131  *	- Message DLM_QUERY_REGION added to support global heartbeat
132  *	- Message DLM_QUERY_NODEINFO added to allow online node removes
133  * New in version 1.2:
134  * 	- Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain
135  * New in version 1.3:
136  *	- Message DLM_DEREF_LOCKRES_DONE added to inform non-master that the
137  *	  refmap is cleared
138  */
139 static const struct dlm_protocol_version dlm_protocol = {
140 	.pv_major = 1,
141 	.pv_minor = 3,
142 };
143 
144 #define DLM_DOMAIN_BACKOFF_MS 200
145 
146 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
147 				  void **ret_data);
148 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
149 				     void **ret_data);
150 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
151 				   void **ret_data);
152 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
153 				    void *data, void **ret_data);
154 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
155 				   void **ret_data);
156 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
157 				struct dlm_protocol_version *request);
158 
159 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
160 
161 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
162 {
163 	if (hlist_unhashed(&res->hash_node))
164 		return;
165 
166 	mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
167 	     res->lockname.name);
168 	hlist_del_init(&res->hash_node);
169 	dlm_lockres_put(res);
170 }
171 
172 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
173 {
174 	struct hlist_head *bucket;
175 
176 	assert_spin_locked(&dlm->spinlock);
177 
178 	bucket = dlm_lockres_hash(dlm, res->lockname.hash);
179 
180 	/* get a reference for our hashtable */
181 	dlm_lockres_get(res);
182 
183 	hlist_add_head(&res->hash_node, bucket);
184 
185 	mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
186 	     res->lockname.name);
187 }
188 
189 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
190 						     const char *name,
191 						     unsigned int len,
192 						     unsigned int hash)
193 {
194 	struct hlist_head *bucket;
195 	struct dlm_lock_resource *res;
196 
197 	mlog(0, "%.*s\n", len, name);
198 
199 	assert_spin_locked(&dlm->spinlock);
200 
201 	bucket = dlm_lockres_hash(dlm, hash);
202 
203 	hlist_for_each_entry(res, bucket, hash_node) {
204 		if (res->lockname.name[0] != name[0])
205 			continue;
206 		if (unlikely(res->lockname.len != len))
207 			continue;
208 		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
209 			continue;
210 		dlm_lockres_get(res);
211 		return res;
212 	}
213 	return NULL;
214 }
215 
216 /* intended to be called by functions which do not care about lock
217  * resources which are being purged (most net _handler functions).
218  * this will return NULL for any lock resource which is found but
219  * currently in the process of dropping its mastery reference.
220  * use __dlm_lookup_lockres_full when you need the lock resource
221  * regardless (e.g. dlm_get_lock_resource) */
222 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
223 						const char *name,
224 						unsigned int len,
225 						unsigned int hash)
226 {
227 	struct dlm_lock_resource *res = NULL;
228 
229 	mlog(0, "%.*s\n", len, name);
230 
231 	assert_spin_locked(&dlm->spinlock);
232 
233 	res = __dlm_lookup_lockres_full(dlm, name, len, hash);
234 	if (res) {
235 		spin_lock(&res->spinlock);
236 		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
237 			spin_unlock(&res->spinlock);
238 			dlm_lockres_put(res);
239 			return NULL;
240 		}
241 		spin_unlock(&res->spinlock);
242 	}
243 
244 	return res;
245 }
246 
247 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
248 				    const char *name,
249 				    unsigned int len)
250 {
251 	struct dlm_lock_resource *res;
252 	unsigned int hash = dlm_lockid_hash(name, len);
253 
254 	spin_lock(&dlm->spinlock);
255 	res = __dlm_lookup_lockres(dlm, name, len, hash);
256 	spin_unlock(&dlm->spinlock);
257 	return res;
258 }
259 
260 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
261 {
262 	struct dlm_ctxt *tmp;
263 
264 	assert_spin_locked(&dlm_domain_lock);
265 
266 	/* tmp->name here is always NULL terminated,
267 	 * but domain may not be! */
268 	list_for_each_entry(tmp, &dlm_domains, list) {
269 		if (strlen(tmp->name) == len &&
270 		    memcmp(tmp->name, domain, len)==0)
271 			return tmp;
272 	}
273 
274 	return NULL;
275 }
276 
277 /* For null terminated domain strings ONLY */
278 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
279 {
280 	assert_spin_locked(&dlm_domain_lock);
281 
282 	return __dlm_lookup_domain_full(domain, strlen(domain));
283 }
284 
285 
286 /* returns true on one of two conditions:
287  * 1) the domain does not exist
288  * 2) the domain exists and it's state is "joined" */
289 static int dlm_wait_on_domain_helper(const char *domain)
290 {
291 	int ret = 0;
292 	struct dlm_ctxt *tmp = NULL;
293 
294 	spin_lock(&dlm_domain_lock);
295 
296 	tmp = __dlm_lookup_domain(domain);
297 	if (!tmp)
298 		ret = 1;
299 	else if (tmp->dlm_state == DLM_CTXT_JOINED)
300 		ret = 1;
301 
302 	spin_unlock(&dlm_domain_lock);
303 	return ret;
304 }
305 
306 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
307 {
308 	dlm_destroy_debugfs_subroot(dlm);
309 
310 	if (dlm->lockres_hash)
311 		dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
312 
313 	if (dlm->master_hash)
314 		dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
315 
316 	kfree(dlm->name);
317 	kfree(dlm);
318 }
319 
320 /* A little strange - this function will be called while holding
321  * dlm_domain_lock and is expected to be holding it on the way out. We
322  * will however drop and reacquire it multiple times */
323 static void dlm_ctxt_release(struct kref *kref)
324 {
325 	struct dlm_ctxt *dlm;
326 
327 	dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
328 
329 	BUG_ON(dlm->num_joins);
330 	BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
331 
332 	/* we may still be in the list if we hit an error during join. */
333 	list_del_init(&dlm->list);
334 
335 	spin_unlock(&dlm_domain_lock);
336 
337 	mlog(0, "freeing memory from domain %s\n", dlm->name);
338 
339 	wake_up(&dlm_domain_events);
340 
341 	dlm_free_ctxt_mem(dlm);
342 
343 	spin_lock(&dlm_domain_lock);
344 }
345 
346 void dlm_put(struct dlm_ctxt *dlm)
347 {
348 	spin_lock(&dlm_domain_lock);
349 	kref_put(&dlm->dlm_refs, dlm_ctxt_release);
350 	spin_unlock(&dlm_domain_lock);
351 }
352 
353 static void __dlm_get(struct dlm_ctxt *dlm)
354 {
355 	kref_get(&dlm->dlm_refs);
356 }
357 
358 /* given a questionable reference to a dlm object, gets a reference if
359  * it can find it in the list, otherwise returns NULL in which case
360  * you shouldn't trust your pointer. */
361 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
362 {
363 	struct dlm_ctxt *target;
364 	struct dlm_ctxt *ret = NULL;
365 
366 	spin_lock(&dlm_domain_lock);
367 
368 	list_for_each_entry(target, &dlm_domains, list) {
369 		if (target == dlm) {
370 			__dlm_get(target);
371 			ret = target;
372 			break;
373 		}
374 	}
375 
376 	spin_unlock(&dlm_domain_lock);
377 
378 	return ret;
379 }
380 
381 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
382 {
383 	int ret;
384 
385 	spin_lock(&dlm_domain_lock);
386 	ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
387 		(dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
388 	spin_unlock(&dlm_domain_lock);
389 
390 	return ret;
391 }
392 
393 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
394 {
395 	if (dlm->dlm_worker) {
396 		flush_workqueue(dlm->dlm_worker);
397 		destroy_workqueue(dlm->dlm_worker);
398 		dlm->dlm_worker = NULL;
399 	}
400 }
401 
402 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
403 {
404 	dlm_unregister_domain_handlers(dlm);
405 	dlm_debug_shutdown(dlm);
406 	dlm_complete_thread(dlm);
407 	dlm_complete_recovery_thread(dlm);
408 	dlm_destroy_dlm_worker(dlm);
409 
410 	/* We've left the domain. Now we can take ourselves out of the
411 	 * list and allow the kref stuff to help us free the
412 	 * memory. */
413 	spin_lock(&dlm_domain_lock);
414 	list_del_init(&dlm->list);
415 	spin_unlock(&dlm_domain_lock);
416 
417 	/* Wake up anyone waiting for us to remove this domain */
418 	wake_up(&dlm_domain_events);
419 }
420 
421 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
422 {
423 	int i, num, n, ret = 0;
424 	struct dlm_lock_resource *res;
425 	struct hlist_node *iter;
426 	struct hlist_head *bucket;
427 	int dropped;
428 
429 	mlog(0, "Migrating locks from domain %s\n", dlm->name);
430 
431 	num = 0;
432 	spin_lock(&dlm->spinlock);
433 	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
434 redo_bucket:
435 		n = 0;
436 		bucket = dlm_lockres_hash(dlm, i);
437 		iter = bucket->first;
438 		while (iter) {
439 			n++;
440 			res = hlist_entry(iter, struct dlm_lock_resource,
441 					  hash_node);
442 			dlm_lockres_get(res);
443 			/* migrate, if necessary.  this will drop the dlm
444 			 * spinlock and retake it if it does migration. */
445 			dropped = dlm_empty_lockres(dlm, res);
446 
447 			spin_lock(&res->spinlock);
448 			if (dropped)
449 				__dlm_lockres_calc_usage(dlm, res);
450 			else
451 				iter = res->hash_node.next;
452 			spin_unlock(&res->spinlock);
453 
454 			dlm_lockres_put(res);
455 
456 			if (dropped) {
457 				cond_resched_lock(&dlm->spinlock);
458 				goto redo_bucket;
459 			}
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 int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
490 					 void *data, void **ret_data)
491 {
492 	struct dlm_ctxt *dlm = data;
493 	unsigned int node;
494 	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
495 
496 	if (!dlm_grab(dlm))
497 		return 0;
498 
499 	node = exit_msg->node_idx;
500 	mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
501 
502 	spin_lock(&dlm->spinlock);
503 	set_bit(node, dlm->exit_domain_map);
504 	spin_unlock(&dlm->spinlock);
505 
506 	dlm_put(dlm);
507 
508 	return 0;
509 }
510 
511 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
512 {
513 	/* Yikes, a double spinlock! I need domain_lock for the dlm
514 	 * state and the dlm spinlock for join state... Sorry! */
515 again:
516 	spin_lock(&dlm_domain_lock);
517 	spin_lock(&dlm->spinlock);
518 
519 	if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
520 		mlog(0, "Node %d is joining, we wait on it.\n",
521 			  dlm->joining_node);
522 		spin_unlock(&dlm->spinlock);
523 		spin_unlock(&dlm_domain_lock);
524 
525 		wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
526 		goto again;
527 	}
528 
529 	dlm->dlm_state = DLM_CTXT_LEAVING;
530 	spin_unlock(&dlm->spinlock);
531 	spin_unlock(&dlm_domain_lock);
532 }
533 
534 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
535 {
536 	int node = -1, num = 0;
537 
538 	assert_spin_locked(&dlm->spinlock);
539 
540 	printk("( ");
541 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
542 				     node + 1)) < O2NM_MAX_NODES) {
543 		printk("%d ", node);
544 		++num;
545 	}
546 	printk(") %u nodes\n", num);
547 }
548 
549 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
550 				   void **ret_data)
551 {
552 	struct dlm_ctxt *dlm = data;
553 	unsigned int node;
554 	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
555 
556 	mlog(0, "%p %u %p", msg, len, data);
557 
558 	if (!dlm_grab(dlm))
559 		return 0;
560 
561 	node = exit_msg->node_idx;
562 
563 	spin_lock(&dlm->spinlock);
564 	clear_bit(node, dlm->domain_map);
565 	clear_bit(node, dlm->exit_domain_map);
566 	printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
567 	__dlm_print_nodes(dlm);
568 
569 	/* notify anything attached to the heartbeat events */
570 	dlm_hb_event_notify_attached(dlm, node, 0);
571 
572 	spin_unlock(&dlm->spinlock);
573 
574 	dlm_put(dlm);
575 
576 	return 0;
577 }
578 
579 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
580 				    unsigned int node)
581 {
582 	int status;
583 	struct dlm_exit_domain leave_msg;
584 
585 	mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
586 	     msg_type, node);
587 
588 	memset(&leave_msg, 0, sizeof(leave_msg));
589 	leave_msg.node_idx = dlm->node_num;
590 
591 	status = o2net_send_message(msg_type, dlm->key, &leave_msg,
592 				    sizeof(leave_msg), node, NULL);
593 	if (status < 0)
594 		mlog(ML_ERROR, "Error %d sending domain exit message %u "
595 		     "to node %u on domain %s\n", status, msg_type, node,
596 		     dlm->name);
597 
598 	return status;
599 }
600 
601 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
602 {
603 	int node = -1;
604 
605 	/* Support for begin exit domain was added in 1.2 */
606 	if (dlm->dlm_locking_proto.pv_major == 1 &&
607 	    dlm->dlm_locking_proto.pv_minor < 2)
608 		return;
609 
610 	/*
611 	 * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
612 	 * informational. Meaning if a node does not receive the message,
613 	 * so be it.
614 	 */
615 	spin_lock(&dlm->spinlock);
616 	while (1) {
617 		node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
618 		if (node >= O2NM_MAX_NODES)
619 			break;
620 		if (node == dlm->node_num)
621 			continue;
622 
623 		spin_unlock(&dlm->spinlock);
624 		dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
625 		spin_lock(&dlm->spinlock);
626 	}
627 	spin_unlock(&dlm->spinlock);
628 }
629 
630 static void dlm_leave_domain(struct dlm_ctxt *dlm)
631 {
632 	int node, clear_node, status;
633 
634 	/* At this point we've migrated away all our locks and won't
635 	 * accept mastership of new ones. The dlm is responsible for
636 	 * almost nothing now. We make sure not to confuse any joining
637 	 * nodes and then commence shutdown procedure. */
638 
639 	spin_lock(&dlm->spinlock);
640 	/* Clear ourselves from the domain map */
641 	clear_bit(dlm->node_num, dlm->domain_map);
642 	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
643 				     0)) < O2NM_MAX_NODES) {
644 		/* Drop the dlm spinlock. This is safe wrt the domain_map.
645 		 * -nodes cannot be added now as the
646 		 *   query_join_handlers knows to respond with OK_NO_MAP
647 		 * -we catch the right network errors if a node is
648 		 *   removed from the map while we're sending him the
649 		 *   exit message. */
650 		spin_unlock(&dlm->spinlock);
651 
652 		clear_node = 1;
653 
654 		status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
655 						  node);
656 		if (status < 0 &&
657 		    status != -ENOPROTOOPT &&
658 		    status != -ENOTCONN) {
659 			mlog(ML_NOTICE, "Error %d sending domain exit message "
660 			     "to node %d\n", status, node);
661 
662 			/* Not sure what to do here but lets sleep for
663 			 * a bit in case this was a transient
664 			 * error... */
665 			msleep(DLM_DOMAIN_BACKOFF_MS);
666 			clear_node = 0;
667 		}
668 
669 		spin_lock(&dlm->spinlock);
670 		/* If we're not clearing the node bit then we intend
671 		 * to loop back around to try again. */
672 		if (clear_node)
673 			clear_bit(node, dlm->domain_map);
674 	}
675 	spin_unlock(&dlm->spinlock);
676 }
677 
678 int dlm_shutting_down(struct dlm_ctxt *dlm)
679 {
680 	int ret = 0;
681 
682 	spin_lock(&dlm_domain_lock);
683 
684 	if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
685 		ret = 1;
686 
687 	spin_unlock(&dlm_domain_lock);
688 
689 	return ret;
690 }
691 
692 void dlm_unregister_domain(struct dlm_ctxt *dlm)
693 {
694 	int leave = 0;
695 	struct dlm_lock_resource *res;
696 
697 	spin_lock(&dlm_domain_lock);
698 	BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
699 	BUG_ON(!dlm->num_joins);
700 
701 	dlm->num_joins--;
702 	if (!dlm->num_joins) {
703 		/* We mark it "in shutdown" now so new register
704 		 * requests wait until we've completely left the
705 		 * domain. Don't use DLM_CTXT_LEAVING yet as we still
706 		 * want new domain joins to communicate with us at
707 		 * least until we've completed migration of our
708 		 * resources. */
709 		dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
710 		leave = 1;
711 	}
712 	spin_unlock(&dlm_domain_lock);
713 
714 	if (leave) {
715 		mlog(0, "shutting down domain %s\n", dlm->name);
716 		dlm_begin_exit_domain(dlm);
717 
718 		/* We changed dlm state, notify the thread */
719 		dlm_kick_thread(dlm, NULL);
720 
721 		while (dlm_migrate_all_locks(dlm)) {
722 			/* Give dlm_thread time to purge the lockres' */
723 			msleep(500);
724 			mlog(0, "%s: more migration to do\n", dlm->name);
725 		}
726 
727 		/* This list should be empty. If not, print remaining lockres */
728 		if (!list_empty(&dlm->tracking_list)) {
729 			mlog(ML_ERROR, "Following lockres' are still on the "
730 			     "tracking list:\n");
731 			list_for_each_entry(res, &dlm->tracking_list, tracking)
732 				dlm_print_one_lock_resource(res);
733 		}
734 
735 		dlm_mark_domain_leaving(dlm);
736 		dlm_leave_domain(dlm);
737 		printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
738 		dlm_force_free_mles(dlm);
739 		dlm_complete_dlm_shutdown(dlm);
740 	}
741 	dlm_put(dlm);
742 }
743 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
744 
745 static int dlm_query_join_proto_check(char *proto_type, int node,
746 				      struct dlm_protocol_version *ours,
747 				      struct dlm_protocol_version *request)
748 {
749 	int rc;
750 	struct dlm_protocol_version proto = *request;
751 
752 	if (!dlm_protocol_compare(ours, &proto)) {
753 		mlog(0,
754 		     "node %u wanted to join with %s locking protocol "
755 		     "%u.%u, we respond with %u.%u\n",
756 		     node, proto_type,
757 		     request->pv_major,
758 		     request->pv_minor,
759 		     proto.pv_major, proto.pv_minor);
760 		request->pv_minor = proto.pv_minor;
761 		rc = 0;
762 	} else {
763 		mlog(ML_NOTICE,
764 		     "Node %u wanted to join with %s locking "
765 		     "protocol %u.%u, but we have %u.%u, disallowing\n",
766 		     node, proto_type,
767 		     request->pv_major,
768 		     request->pv_minor,
769 		     ours->pv_major,
770 		     ours->pv_minor);
771 		rc = 1;
772 	}
773 
774 	return rc;
775 }
776 
777 /*
778  * struct dlm_query_join_packet is made up of four one-byte fields.  They
779  * are effectively in big-endian order already.  However, little-endian
780  * machines swap them before putting the packet on the wire (because
781  * query_join's response is a status, and that status is treated as a u32
782  * on the wire).  Thus, a big-endian and little-endian machines will treat
783  * this structure differently.
784  *
785  * The solution is to have little-endian machines swap the structure when
786  * converting from the structure to the u32 representation.  This will
787  * result in the structure having the correct format on the wire no matter
788  * the host endian format.
789  */
790 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
791 					  u32 *wire)
792 {
793 	union dlm_query_join_response response;
794 
795 	response.packet = *packet;
796 	*wire = be32_to_cpu(response.intval);
797 }
798 
799 static void dlm_query_join_wire_to_packet(u32 wire,
800 					  struct dlm_query_join_packet *packet)
801 {
802 	union dlm_query_join_response response;
803 
804 	response.intval = cpu_to_be32(wire);
805 	*packet = response.packet;
806 }
807 
808 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
809 				  void **ret_data)
810 {
811 	struct dlm_query_join_request *query;
812 	struct dlm_query_join_packet packet = {
813 		.code = JOIN_DISALLOW,
814 	};
815 	struct dlm_ctxt *dlm = NULL;
816 	u32 response;
817 	u8 nodenum;
818 
819 	query = (struct dlm_query_join_request *) msg->buf;
820 
821 	mlog(0, "node %u wants to join domain %s\n", query->node_idx,
822 		  query->domain);
823 
824 	/*
825 	 * If heartbeat doesn't consider the node live, tell it
826 	 * to back off and try again.  This gives heartbeat a chance
827 	 * to catch up.
828 	 */
829 	if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
830 		mlog(0, "node %u is not in our live map yet\n",
831 		     query->node_idx);
832 
833 		packet.code = JOIN_DISALLOW;
834 		goto respond;
835 	}
836 
837 	packet.code = JOIN_OK_NO_MAP;
838 
839 	spin_lock(&dlm_domain_lock);
840 	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
841 	if (!dlm)
842 		goto unlock_respond;
843 
844 	/*
845 	 * There is a small window where the joining node may not see the
846 	 * node(s) that just left but still part of the cluster. DISALLOW
847 	 * join request if joining node has different node map.
848 	 */
849 	nodenum=0;
850 	while (nodenum < O2NM_MAX_NODES) {
851 		if (test_bit(nodenum, dlm->domain_map)) {
852 			if (!byte_test_bit(nodenum, query->node_map)) {
853 				mlog(0, "disallow join as node %u does not "
854 				     "have node %u in its nodemap\n",
855 				     query->node_idx, nodenum);
856 				packet.code = JOIN_DISALLOW;
857 				goto unlock_respond;
858 			}
859 		}
860 		nodenum++;
861 	}
862 
863 	/* Once the dlm ctxt is marked as leaving then we don't want
864 	 * to be put in someone's domain map.
865 	 * Also, explicitly disallow joining at certain troublesome
866 	 * times (ie. during recovery). */
867 	if (dlm->dlm_state != DLM_CTXT_LEAVING) {
868 		int bit = query->node_idx;
869 		spin_lock(&dlm->spinlock);
870 
871 		if (dlm->dlm_state == DLM_CTXT_NEW &&
872 		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
873 			/*If this is a brand new context and we
874 			 * haven't started our join process yet, then
875 			 * the other node won the race. */
876 			packet.code = JOIN_OK_NO_MAP;
877 		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
878 			/* Disallow parallel joins. */
879 			packet.code = JOIN_DISALLOW;
880 		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
881 			mlog(0, "node %u trying to join, but recovery "
882 			     "is ongoing.\n", bit);
883 			packet.code = JOIN_DISALLOW;
884 		} else if (test_bit(bit, dlm->recovery_map)) {
885 			mlog(0, "node %u trying to join, but it "
886 			     "still needs recovery.\n", bit);
887 			packet.code = JOIN_DISALLOW;
888 		} else if (test_bit(bit, dlm->domain_map)) {
889 			mlog(0, "node %u trying to join, but it "
890 			     "is still in the domain! needs recovery?\n",
891 			     bit);
892 			packet.code = JOIN_DISALLOW;
893 		} else {
894 			/* Alright we're fully a part of this domain
895 			 * so we keep some state as to who's joining
896 			 * and indicate to him that needs to be fixed
897 			 * up. */
898 
899 			/* Make sure we speak compatible locking protocols.  */
900 			if (dlm_query_join_proto_check("DLM", bit,
901 						       &dlm->dlm_locking_proto,
902 						       &query->dlm_proto)) {
903 				packet.code = JOIN_PROTOCOL_MISMATCH;
904 			} else if (dlm_query_join_proto_check("fs", bit,
905 							      &dlm->fs_locking_proto,
906 							      &query->fs_proto)) {
907 				packet.code = JOIN_PROTOCOL_MISMATCH;
908 			} else {
909 				packet.dlm_minor = query->dlm_proto.pv_minor;
910 				packet.fs_minor = query->fs_proto.pv_minor;
911 				packet.code = JOIN_OK;
912 				__dlm_set_joining_node(dlm, query->node_idx);
913 			}
914 		}
915 
916 		spin_unlock(&dlm->spinlock);
917 	}
918 unlock_respond:
919 	spin_unlock(&dlm_domain_lock);
920 
921 respond:
922 	mlog(0, "We respond with %u\n", packet.code);
923 
924 	dlm_query_join_packet_to_wire(&packet, &response);
925 	return response;
926 }
927 
928 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
929 				     void **ret_data)
930 {
931 	struct dlm_assert_joined *assert;
932 	struct dlm_ctxt *dlm = NULL;
933 
934 	assert = (struct dlm_assert_joined *) msg->buf;
935 
936 	mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
937 		  assert->domain);
938 
939 	spin_lock(&dlm_domain_lock);
940 	dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
941 	/* XXX should we consider no dlm ctxt an error? */
942 	if (dlm) {
943 		spin_lock(&dlm->spinlock);
944 
945 		/* Alright, this node has officially joined our
946 		 * domain. Set him in the map and clean up our
947 		 * leftover join state. */
948 		BUG_ON(dlm->joining_node != assert->node_idx);
949 
950 		if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
951 			mlog(0, "dlm recovery is ongoing, disallow join\n");
952 			spin_unlock(&dlm->spinlock);
953 			spin_unlock(&dlm_domain_lock);
954 			return -EAGAIN;
955 		}
956 
957 		set_bit(assert->node_idx, dlm->domain_map);
958 		clear_bit(assert->node_idx, dlm->exit_domain_map);
959 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
960 
961 		printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
962 		       assert->node_idx, dlm->name);
963 		__dlm_print_nodes(dlm);
964 
965 		/* notify anything attached to the heartbeat events */
966 		dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
967 
968 		spin_unlock(&dlm->spinlock);
969 	}
970 	spin_unlock(&dlm_domain_lock);
971 
972 	return 0;
973 }
974 
975 static int dlm_match_regions(struct dlm_ctxt *dlm,
976 			     struct dlm_query_region *qr,
977 			     char *local, int locallen)
978 {
979 	char *remote = qr->qr_regions;
980 	char *l, *r;
981 	int localnr, i, j, foundit;
982 	int status = 0;
983 
984 	if (!o2hb_global_heartbeat_active()) {
985 		if (qr->qr_numregions) {
986 			mlog(ML_ERROR, "Domain %s: Joining node %d has global "
987 			     "heartbeat enabled but local node %d does not\n",
988 			     qr->qr_domain, qr->qr_node, dlm->node_num);
989 			status = -EINVAL;
990 		}
991 		goto bail;
992 	}
993 
994 	if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
995 		mlog(ML_ERROR, "Domain %s: Local node %d has global "
996 		     "heartbeat enabled but joining node %d does not\n",
997 		     qr->qr_domain, dlm->node_num, qr->qr_node);
998 		status = -EINVAL;
999 		goto bail;
1000 	}
1001 
1002 	r = remote;
1003 	for (i = 0; i < qr->qr_numregions; ++i) {
1004 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1005 		r += O2HB_MAX_REGION_NAME_LEN;
1006 	}
1007 
1008 	localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1009 	localnr = o2hb_get_all_regions(local, (u8)localnr);
1010 
1011 	/* compare local regions with remote */
1012 	l = local;
1013 	for (i = 0; i < localnr; ++i) {
1014 		foundit = 0;
1015 		r = remote;
1016 		for (j = 0; j <= qr->qr_numregions; ++j) {
1017 			if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1018 				foundit = 1;
1019 				break;
1020 			}
1021 			r += O2HB_MAX_REGION_NAME_LEN;
1022 		}
1023 		if (!foundit) {
1024 			status = -EINVAL;
1025 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1026 			     "in local node %d but not in joining node %d\n",
1027 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1028 			     dlm->node_num, qr->qr_node);
1029 			goto bail;
1030 		}
1031 		l += O2HB_MAX_REGION_NAME_LEN;
1032 	}
1033 
1034 	/* compare remote with local regions */
1035 	r = remote;
1036 	for (i = 0; i < qr->qr_numregions; ++i) {
1037 		foundit = 0;
1038 		l = local;
1039 		for (j = 0; j < localnr; ++j) {
1040 			if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1041 				foundit = 1;
1042 				break;
1043 			}
1044 			l += O2HB_MAX_REGION_NAME_LEN;
1045 		}
1046 		if (!foundit) {
1047 			status = -EINVAL;
1048 			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1049 			     "in joining node %d but not in local node %d\n",
1050 			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1051 			     qr->qr_node, dlm->node_num);
1052 			goto bail;
1053 		}
1054 		r += O2HB_MAX_REGION_NAME_LEN;
1055 	}
1056 
1057 bail:
1058 	return status;
1059 }
1060 
1061 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1062 {
1063 	struct dlm_query_region *qr = NULL;
1064 	int status, ret = 0, i;
1065 	char *p;
1066 
1067 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1068 		goto bail;
1069 
1070 	qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1071 	if (!qr) {
1072 		ret = -ENOMEM;
1073 		mlog_errno(ret);
1074 		goto bail;
1075 	}
1076 
1077 	qr->qr_node = dlm->node_num;
1078 	qr->qr_namelen = strlen(dlm->name);
1079 	memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1080 	/* if local hb, the numregions will be zero */
1081 	if (o2hb_global_heartbeat_active())
1082 		qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1083 							 O2NM_MAX_REGIONS);
1084 
1085 	p = qr->qr_regions;
1086 	for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1087 		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1088 
1089 	i = -1;
1090 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1091 				  i + 1)) < O2NM_MAX_NODES) {
1092 		if (i == dlm->node_num)
1093 			continue;
1094 
1095 		mlog(0, "Sending regions to node %d\n", i);
1096 
1097 		ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1098 					 sizeof(struct dlm_query_region),
1099 					 i, &status);
1100 		if (ret >= 0)
1101 			ret = status;
1102 		if (ret) {
1103 			mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1104 			     ret, i);
1105 			break;
1106 		}
1107 	}
1108 
1109 bail:
1110 	kfree(qr);
1111 	return ret;
1112 }
1113 
1114 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1115 				    void *data, void **ret_data)
1116 {
1117 	struct dlm_query_region *qr;
1118 	struct dlm_ctxt *dlm = NULL;
1119 	char *local = NULL;
1120 	int status = 0;
1121 
1122 	qr = (struct dlm_query_region *) msg->buf;
1123 
1124 	mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1125 	     qr->qr_domain);
1126 
1127 	/* buffer used in dlm_mast_regions() */
1128 	local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1129 	if (!local)
1130 		return -ENOMEM;
1131 
1132 	status = -EINVAL;
1133 
1134 	spin_lock(&dlm_domain_lock);
1135 	dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1136 	if (!dlm) {
1137 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1138 		     "before join domain\n", qr->qr_node, qr->qr_domain);
1139 		goto out_domain_lock;
1140 	}
1141 
1142 	spin_lock(&dlm->spinlock);
1143 	if (dlm->joining_node != qr->qr_node) {
1144 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1145 		     "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1146 		     dlm->joining_node);
1147 		goto out_dlm_lock;
1148 	}
1149 
1150 	/* Support for global heartbeat was added in 1.1 */
1151 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1152 	    dlm->dlm_locking_proto.pv_minor == 0) {
1153 		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1154 		     "but active dlm protocol is %d.%d\n", qr->qr_node,
1155 		     qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1156 		     dlm->dlm_locking_proto.pv_minor);
1157 		goto out_dlm_lock;
1158 	}
1159 
1160 	status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1161 
1162 out_dlm_lock:
1163 	spin_unlock(&dlm->spinlock);
1164 
1165 out_domain_lock:
1166 	spin_unlock(&dlm_domain_lock);
1167 
1168 	kfree(local);
1169 
1170 	return status;
1171 }
1172 
1173 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1174 {
1175 	struct o2nm_node *local;
1176 	struct dlm_node_info *remote;
1177 	int i, j;
1178 	int status = 0;
1179 
1180 	for (j = 0; j < qn->qn_numnodes; ++j)
1181 		mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1182 		     &(qn->qn_nodes[j].ni_ipv4_address),
1183 		     ntohs(qn->qn_nodes[j].ni_ipv4_port));
1184 
1185 	for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1186 		local = o2nm_get_node_by_num(i);
1187 		remote = NULL;
1188 		for (j = 0; j < qn->qn_numnodes; ++j) {
1189 			if (qn->qn_nodes[j].ni_nodenum == i) {
1190 				remote = &(qn->qn_nodes[j]);
1191 				break;
1192 			}
1193 		}
1194 
1195 		if (!local && !remote)
1196 			continue;
1197 
1198 		if ((local && !remote) || (!local && remote))
1199 			status = -EINVAL;
1200 
1201 		if (!status &&
1202 		    ((remote->ni_nodenum != local->nd_num) ||
1203 		     (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1204 		     (remote->ni_ipv4_address != local->nd_ipv4_address)))
1205 			status = -EINVAL;
1206 
1207 		if (status) {
1208 			if (remote && !local)
1209 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1210 				     "registered in joining node %d but not in "
1211 				     "local node %d\n", qn->qn_domain,
1212 				     remote->ni_nodenum,
1213 				     &(remote->ni_ipv4_address),
1214 				     ntohs(remote->ni_ipv4_port),
1215 				     qn->qn_nodenum, dlm->node_num);
1216 			if (local && !remote)
1217 				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1218 				     "registered in local node %d but not in "
1219 				     "joining node %d\n", qn->qn_domain,
1220 				     local->nd_num, &(local->nd_ipv4_address),
1221 				     ntohs(local->nd_ipv4_port),
1222 				     dlm->node_num, qn->qn_nodenum);
1223 			BUG_ON((!local && !remote));
1224 		}
1225 
1226 		if (local)
1227 			o2nm_node_put(local);
1228 	}
1229 
1230 	return status;
1231 }
1232 
1233 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1234 {
1235 	struct dlm_query_nodeinfo *qn = NULL;
1236 	struct o2nm_node *node;
1237 	int ret = 0, status, count, i;
1238 
1239 	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1240 		goto bail;
1241 
1242 	qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1243 	if (!qn) {
1244 		ret = -ENOMEM;
1245 		mlog_errno(ret);
1246 		goto bail;
1247 	}
1248 
1249 	for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1250 		node = o2nm_get_node_by_num(i);
1251 		if (!node)
1252 			continue;
1253 		qn->qn_nodes[count].ni_nodenum = node->nd_num;
1254 		qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1255 		qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1256 		mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1257 		     &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1258 		++count;
1259 		o2nm_node_put(node);
1260 	}
1261 
1262 	qn->qn_nodenum = dlm->node_num;
1263 	qn->qn_numnodes = count;
1264 	qn->qn_namelen = strlen(dlm->name);
1265 	memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1266 
1267 	i = -1;
1268 	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1269 				  i + 1)) < O2NM_MAX_NODES) {
1270 		if (i == dlm->node_num)
1271 			continue;
1272 
1273 		mlog(0, "Sending nodeinfo to node %d\n", i);
1274 
1275 		ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1276 					 qn, sizeof(struct dlm_query_nodeinfo),
1277 					 i, &status);
1278 		if (ret >= 0)
1279 			ret = status;
1280 		if (ret) {
1281 			mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1282 			break;
1283 		}
1284 	}
1285 
1286 bail:
1287 	kfree(qn);
1288 	return ret;
1289 }
1290 
1291 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1292 				      void *data, void **ret_data)
1293 {
1294 	struct dlm_query_nodeinfo *qn;
1295 	struct dlm_ctxt *dlm = NULL;
1296 	int locked = 0, status = -EINVAL;
1297 
1298 	qn = (struct dlm_query_nodeinfo *) msg->buf;
1299 
1300 	mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1301 	     qn->qn_domain);
1302 
1303 	spin_lock(&dlm_domain_lock);
1304 	dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1305 	if (!dlm) {
1306 		mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1307 		     "join domain\n", qn->qn_nodenum, qn->qn_domain);
1308 		goto bail;
1309 	}
1310 
1311 	spin_lock(&dlm->spinlock);
1312 	locked = 1;
1313 	if (dlm->joining_node != qn->qn_nodenum) {
1314 		mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1315 		     "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1316 		     dlm->joining_node);
1317 		goto bail;
1318 	}
1319 
1320 	/* Support for node query was added in 1.1 */
1321 	if (dlm->dlm_locking_proto.pv_major == 1 &&
1322 	    dlm->dlm_locking_proto.pv_minor == 0) {
1323 		mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1324 		     "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1325 		     qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1326 		     dlm->dlm_locking_proto.pv_minor);
1327 		goto bail;
1328 	}
1329 
1330 	status = dlm_match_nodes(dlm, qn);
1331 
1332 bail:
1333 	if (locked)
1334 		spin_unlock(&dlm->spinlock);
1335 	spin_unlock(&dlm_domain_lock);
1336 
1337 	return status;
1338 }
1339 
1340 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1341 				   void **ret_data)
1342 {
1343 	struct dlm_cancel_join *cancel;
1344 	struct dlm_ctxt *dlm = NULL;
1345 
1346 	cancel = (struct dlm_cancel_join *) msg->buf;
1347 
1348 	mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1349 		  cancel->domain);
1350 
1351 	spin_lock(&dlm_domain_lock);
1352 	dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1353 
1354 	if (dlm) {
1355 		spin_lock(&dlm->spinlock);
1356 
1357 		/* Yikes, this guy wants to cancel his join. No
1358 		 * problem, we simply cleanup our join state. */
1359 		BUG_ON(dlm->joining_node != cancel->node_idx);
1360 		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1361 
1362 		spin_unlock(&dlm->spinlock);
1363 	}
1364 	spin_unlock(&dlm_domain_lock);
1365 
1366 	return 0;
1367 }
1368 
1369 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1370 				    unsigned int node)
1371 {
1372 	int status;
1373 	struct dlm_cancel_join cancel_msg;
1374 
1375 	memset(&cancel_msg, 0, sizeof(cancel_msg));
1376 	cancel_msg.node_idx = dlm->node_num;
1377 	cancel_msg.name_len = strlen(dlm->name);
1378 	memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1379 
1380 	status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1381 				    &cancel_msg, sizeof(cancel_msg), node,
1382 				    NULL);
1383 	if (status < 0) {
1384 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1385 		     "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1386 		     node);
1387 		goto bail;
1388 	}
1389 
1390 bail:
1391 	return status;
1392 }
1393 
1394 /* map_size should be in bytes. */
1395 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1396 				 unsigned long *node_map,
1397 				 unsigned int map_size)
1398 {
1399 	int status, tmpstat;
1400 	int node;
1401 
1402 	if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1403 			 sizeof(unsigned long))) {
1404 		mlog(ML_ERROR,
1405 		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1406 		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1407 		return -EINVAL;
1408 	}
1409 
1410 	status = 0;
1411 	node = -1;
1412 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1413 				     node + 1)) < O2NM_MAX_NODES) {
1414 		if (node == dlm->node_num)
1415 			continue;
1416 
1417 		tmpstat = dlm_send_one_join_cancel(dlm, node);
1418 		if (tmpstat) {
1419 			mlog(ML_ERROR, "Error return %d cancelling join on "
1420 			     "node %d\n", tmpstat, node);
1421 			if (!status)
1422 				status = tmpstat;
1423 		}
1424 	}
1425 
1426 	if (status)
1427 		mlog_errno(status);
1428 	return status;
1429 }
1430 
1431 static int dlm_request_join(struct dlm_ctxt *dlm,
1432 			    int node,
1433 			    enum dlm_query_join_response_code *response)
1434 {
1435 	int status;
1436 	struct dlm_query_join_request join_msg;
1437 	struct dlm_query_join_packet packet;
1438 	u32 join_resp;
1439 
1440 	mlog(0, "querying node %d\n", node);
1441 
1442 	memset(&join_msg, 0, sizeof(join_msg));
1443 	join_msg.node_idx = dlm->node_num;
1444 	join_msg.name_len = strlen(dlm->name);
1445 	memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1446 	join_msg.dlm_proto = dlm->dlm_locking_proto;
1447 	join_msg.fs_proto = dlm->fs_locking_proto;
1448 
1449 	/* copy live node map to join message */
1450 	byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1451 
1452 	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1453 				    sizeof(join_msg), node, &join_resp);
1454 	if (status < 0 && status != -ENOPROTOOPT) {
1455 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1456 		     "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1457 		     node);
1458 		goto bail;
1459 	}
1460 	dlm_query_join_wire_to_packet(join_resp, &packet);
1461 
1462 	/* -ENOPROTOOPT from the net code means the other side isn't
1463 	    listening for our message type -- that's fine, it means
1464 	    his dlm isn't up, so we can consider him a 'yes' but not
1465 	    joined into the domain.  */
1466 	if (status == -ENOPROTOOPT) {
1467 		status = 0;
1468 		*response = JOIN_OK_NO_MAP;
1469 	} else {
1470 		*response = packet.code;
1471 		switch (packet.code) {
1472 		case JOIN_DISALLOW:
1473 		case JOIN_OK_NO_MAP:
1474 			break;
1475 		case JOIN_PROTOCOL_MISMATCH:
1476 			mlog(ML_NOTICE,
1477 			     "This node requested DLM locking protocol %u.%u and "
1478 			     "filesystem locking protocol %u.%u.  At least one of "
1479 			     "the protocol versions on node %d is not compatible, "
1480 			     "disconnecting\n",
1481 			     dlm->dlm_locking_proto.pv_major,
1482 			     dlm->dlm_locking_proto.pv_minor,
1483 			     dlm->fs_locking_proto.pv_major,
1484 			     dlm->fs_locking_proto.pv_minor,
1485 			     node);
1486 			status = -EPROTO;
1487 			break;
1488 		case JOIN_OK:
1489 			/* Use the same locking protocol as the remote node */
1490 			dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1491 			dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1492 			mlog(0,
1493 			     "Node %d responds JOIN_OK with DLM locking protocol "
1494 			     "%u.%u and fs locking protocol %u.%u\n",
1495 			     node,
1496 			     dlm->dlm_locking_proto.pv_major,
1497 			     dlm->dlm_locking_proto.pv_minor,
1498 			     dlm->fs_locking_proto.pv_major,
1499 			     dlm->fs_locking_proto.pv_minor);
1500 			break;
1501 		default:
1502 			status = -EINVAL;
1503 			mlog(ML_ERROR, "invalid response %d from node %u\n",
1504 			     packet.code, node);
1505 			/* Reset response to JOIN_DISALLOW */
1506 			*response = JOIN_DISALLOW;
1507 			break;
1508 		}
1509 	}
1510 
1511 	mlog(0, "status %d, node %d response is %d\n", status, node,
1512 	     *response);
1513 
1514 bail:
1515 	return status;
1516 }
1517 
1518 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1519 				    unsigned int node)
1520 {
1521 	int status;
1522 	int ret;
1523 	struct dlm_assert_joined assert_msg;
1524 
1525 	mlog(0, "Sending join assert to node %u\n", node);
1526 
1527 	memset(&assert_msg, 0, sizeof(assert_msg));
1528 	assert_msg.node_idx = dlm->node_num;
1529 	assert_msg.name_len = strlen(dlm->name);
1530 	memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1531 
1532 	status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1533 				    &assert_msg, sizeof(assert_msg), node,
1534 				    &ret);
1535 	if (status < 0)
1536 		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1537 		     "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1538 		     node);
1539 	else
1540 		status = ret;
1541 
1542 	return status;
1543 }
1544 
1545 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1546 				  unsigned long *node_map)
1547 {
1548 	int status, node, live;
1549 
1550 	status = 0;
1551 	node = -1;
1552 	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1553 				     node + 1)) < O2NM_MAX_NODES) {
1554 		if (node == dlm->node_num)
1555 			continue;
1556 
1557 		do {
1558 			/* It is very important that this message be
1559 			 * received so we spin until either the node
1560 			 * has died or it gets the message. */
1561 			status = dlm_send_one_join_assert(dlm, node);
1562 
1563 			spin_lock(&dlm->spinlock);
1564 			live = test_bit(node, dlm->live_nodes_map);
1565 			spin_unlock(&dlm->spinlock);
1566 
1567 			if (status) {
1568 				mlog(ML_ERROR, "Error return %d asserting "
1569 				     "join on node %d\n", status, node);
1570 
1571 				/* give us some time between errors... */
1572 				if (live)
1573 					msleep(DLM_DOMAIN_BACKOFF_MS);
1574 			}
1575 		} while (status && live);
1576 	}
1577 }
1578 
1579 struct domain_join_ctxt {
1580 	unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1581 	unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1582 };
1583 
1584 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1585 				   struct domain_join_ctxt *ctxt,
1586 				   enum dlm_query_join_response_code response)
1587 {
1588 	int ret;
1589 
1590 	if (response == JOIN_DISALLOW) {
1591 		mlog(0, "Latest response of disallow -- should restart\n");
1592 		return 1;
1593 	}
1594 
1595 	spin_lock(&dlm->spinlock);
1596 	/* For now, we restart the process if the node maps have
1597 	 * changed at all */
1598 	ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1599 		     sizeof(dlm->live_nodes_map));
1600 	spin_unlock(&dlm->spinlock);
1601 
1602 	if (ret)
1603 		mlog(0, "Node maps changed -- should restart\n");
1604 
1605 	return ret;
1606 }
1607 
1608 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1609 {
1610 	int status = 0, tmpstat, node;
1611 	struct domain_join_ctxt *ctxt;
1612 	enum dlm_query_join_response_code response = JOIN_DISALLOW;
1613 
1614 	mlog(0, "%p", dlm);
1615 
1616 	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1617 	if (!ctxt) {
1618 		status = -ENOMEM;
1619 		mlog_errno(status);
1620 		goto bail;
1621 	}
1622 
1623 	/* group sem locking should work for us here -- we're already
1624 	 * registered for heartbeat events so filling this should be
1625 	 * atomic wrt getting those handlers called. */
1626 	o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1627 
1628 	spin_lock(&dlm->spinlock);
1629 	memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1630 
1631 	__dlm_set_joining_node(dlm, dlm->node_num);
1632 
1633 	spin_unlock(&dlm->spinlock);
1634 
1635 	node = -1;
1636 	while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1637 				     node + 1)) < O2NM_MAX_NODES) {
1638 		if (node == dlm->node_num)
1639 			continue;
1640 
1641 		status = dlm_request_join(dlm, node, &response);
1642 		if (status < 0) {
1643 			mlog_errno(status);
1644 			goto bail;
1645 		}
1646 
1647 		/* Ok, either we got a response or the node doesn't have a
1648 		 * dlm up. */
1649 		if (response == JOIN_OK)
1650 			set_bit(node, ctxt->yes_resp_map);
1651 
1652 		if (dlm_should_restart_join(dlm, ctxt, response)) {
1653 			status = -EAGAIN;
1654 			goto bail;
1655 		}
1656 	}
1657 
1658 	mlog(0, "Yay, done querying nodes!\n");
1659 
1660 	/* Yay, everyone agree's we can join the domain. My domain is
1661 	 * comprised of all nodes who were put in the
1662 	 * yes_resp_map. Copy that into our domain map and send a join
1663 	 * assert message to clean up everyone elses state. */
1664 	spin_lock(&dlm->spinlock);
1665 	memcpy(dlm->domain_map, ctxt->yes_resp_map,
1666 	       sizeof(ctxt->yes_resp_map));
1667 	set_bit(dlm->node_num, dlm->domain_map);
1668 	spin_unlock(&dlm->spinlock);
1669 
1670 	/* Support for global heartbeat and node info was added in 1.1 */
1671 	if (dlm->dlm_locking_proto.pv_major > 1 ||
1672 	    dlm->dlm_locking_proto.pv_minor > 0) {
1673 		status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1674 		if (status) {
1675 			mlog_errno(status);
1676 			goto bail;
1677 		}
1678 		status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1679 		if (status) {
1680 			mlog_errno(status);
1681 			goto bail;
1682 		}
1683 	}
1684 
1685 	dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1686 
1687 	/* Joined state *must* be set before the joining node
1688 	 * information, otherwise the query_join handler may read no
1689 	 * current joiner but a state of NEW and tell joining nodes
1690 	 * we're not in the domain. */
1691 	spin_lock(&dlm_domain_lock);
1692 	dlm->dlm_state = DLM_CTXT_JOINED;
1693 	dlm->num_joins++;
1694 	spin_unlock(&dlm_domain_lock);
1695 
1696 bail:
1697 	spin_lock(&dlm->spinlock);
1698 	__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1699 	if (!status) {
1700 		printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1701 		__dlm_print_nodes(dlm);
1702 	}
1703 	spin_unlock(&dlm->spinlock);
1704 
1705 	if (ctxt) {
1706 		/* Do we need to send a cancel message to any nodes? */
1707 		if (status < 0) {
1708 			tmpstat = dlm_send_join_cancels(dlm,
1709 							ctxt->yes_resp_map,
1710 							sizeof(ctxt->yes_resp_map));
1711 			if (tmpstat < 0)
1712 				mlog_errno(tmpstat);
1713 		}
1714 		kfree(ctxt);
1715 	}
1716 
1717 	mlog(0, "returning %d\n", status);
1718 	return status;
1719 }
1720 
1721 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1722 {
1723 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1724 	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1725 	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1726 }
1727 
1728 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1729 {
1730 	int status;
1731 
1732 	mlog(0, "registering handlers.\n");
1733 
1734 	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1735 			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1736 	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1737 			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1738 
1739 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1740 	if (status)
1741 		goto bail;
1742 
1743 	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1744 	if (status)
1745 		goto bail;
1746 
1747 	status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1748 					sizeof(struct dlm_master_request),
1749 					dlm_master_request_handler,
1750 					dlm, NULL, &dlm->dlm_domain_handlers);
1751 	if (status)
1752 		goto bail;
1753 
1754 	status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1755 					sizeof(struct dlm_assert_master),
1756 					dlm_assert_master_handler,
1757 					dlm, dlm_assert_master_post_handler,
1758 					&dlm->dlm_domain_handlers);
1759 	if (status)
1760 		goto bail;
1761 
1762 	status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1763 					sizeof(struct dlm_create_lock),
1764 					dlm_create_lock_handler,
1765 					dlm, NULL, &dlm->dlm_domain_handlers);
1766 	if (status)
1767 		goto bail;
1768 
1769 	status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1770 					DLM_CONVERT_LOCK_MAX_LEN,
1771 					dlm_convert_lock_handler,
1772 					dlm, NULL, &dlm->dlm_domain_handlers);
1773 	if (status)
1774 		goto bail;
1775 
1776 	status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1777 					DLM_UNLOCK_LOCK_MAX_LEN,
1778 					dlm_unlock_lock_handler,
1779 					dlm, NULL, &dlm->dlm_domain_handlers);
1780 	if (status)
1781 		goto bail;
1782 
1783 	status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1784 					DLM_PROXY_AST_MAX_LEN,
1785 					dlm_proxy_ast_handler,
1786 					dlm, NULL, &dlm->dlm_domain_handlers);
1787 	if (status)
1788 		goto bail;
1789 
1790 	status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1791 					sizeof(struct dlm_exit_domain),
1792 					dlm_exit_domain_handler,
1793 					dlm, NULL, &dlm->dlm_domain_handlers);
1794 	if (status)
1795 		goto bail;
1796 
1797 	status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1798 					sizeof(struct dlm_deref_lockres),
1799 					dlm_deref_lockres_handler,
1800 					dlm, NULL, &dlm->dlm_domain_handlers);
1801 	if (status)
1802 		goto bail;
1803 
1804 	status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1805 					sizeof(struct dlm_migrate_request),
1806 					dlm_migrate_request_handler,
1807 					dlm, NULL, &dlm->dlm_domain_handlers);
1808 	if (status)
1809 		goto bail;
1810 
1811 	status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1812 					DLM_MIG_LOCKRES_MAX_LEN,
1813 					dlm_mig_lockres_handler,
1814 					dlm, NULL, &dlm->dlm_domain_handlers);
1815 	if (status)
1816 		goto bail;
1817 
1818 	status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1819 					sizeof(struct dlm_master_requery),
1820 					dlm_master_requery_handler,
1821 					dlm, NULL, &dlm->dlm_domain_handlers);
1822 	if (status)
1823 		goto bail;
1824 
1825 	status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1826 					sizeof(struct dlm_lock_request),
1827 					dlm_request_all_locks_handler,
1828 					dlm, NULL, &dlm->dlm_domain_handlers);
1829 	if (status)
1830 		goto bail;
1831 
1832 	status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1833 					sizeof(struct dlm_reco_data_done),
1834 					dlm_reco_data_done_handler,
1835 					dlm, NULL, &dlm->dlm_domain_handlers);
1836 	if (status)
1837 		goto bail;
1838 
1839 	status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1840 					sizeof(struct dlm_begin_reco),
1841 					dlm_begin_reco_handler,
1842 					dlm, NULL, &dlm->dlm_domain_handlers);
1843 	if (status)
1844 		goto bail;
1845 
1846 	status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1847 					sizeof(struct dlm_finalize_reco),
1848 					dlm_finalize_reco_handler,
1849 					dlm, NULL, &dlm->dlm_domain_handlers);
1850 	if (status)
1851 		goto bail;
1852 
1853 	status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1854 					sizeof(struct dlm_exit_domain),
1855 					dlm_begin_exit_domain_handler,
1856 					dlm, NULL, &dlm->dlm_domain_handlers);
1857 	if (status)
1858 		goto bail;
1859 
1860 	status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key,
1861 					sizeof(struct dlm_deref_lockres_done),
1862 					dlm_deref_lockres_done_handler,
1863 					dlm, NULL, &dlm->dlm_domain_handlers);
1864 bail:
1865 	if (status)
1866 		dlm_unregister_domain_handlers(dlm);
1867 
1868 	return status;
1869 }
1870 
1871 static int dlm_join_domain(struct dlm_ctxt *dlm)
1872 {
1873 	int status;
1874 	unsigned int backoff;
1875 	unsigned int total_backoff = 0;
1876 	char wq_name[O2NM_MAX_NAME_LEN];
1877 
1878 	BUG_ON(!dlm);
1879 
1880 	mlog(0, "Join domain %s\n", dlm->name);
1881 
1882 	status = dlm_register_domain_handlers(dlm);
1883 	if (status) {
1884 		mlog_errno(status);
1885 		goto bail;
1886 	}
1887 
1888 	status = dlm_launch_thread(dlm);
1889 	if (status < 0) {
1890 		mlog_errno(status);
1891 		goto bail;
1892 	}
1893 
1894 	status = dlm_launch_recovery_thread(dlm);
1895 	if (status < 0) {
1896 		mlog_errno(status);
1897 		goto bail;
1898 	}
1899 
1900 	status = dlm_debug_init(dlm);
1901 	if (status < 0) {
1902 		mlog_errno(status);
1903 		goto bail;
1904 	}
1905 
1906 	snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name);
1907 	dlm->dlm_worker = create_singlethread_workqueue(wq_name);
1908 	if (!dlm->dlm_worker) {
1909 		status = -ENOMEM;
1910 		mlog_errno(status);
1911 		goto bail;
1912 	}
1913 
1914 	do {
1915 		status = dlm_try_to_join_domain(dlm);
1916 
1917 		/* If we're racing another node to the join, then we
1918 		 * need to back off temporarily and let them
1919 		 * complete. */
1920 #define	DLM_JOIN_TIMEOUT_MSECS	90000
1921 		if (status == -EAGAIN) {
1922 			if (signal_pending(current)) {
1923 				status = -ERESTARTSYS;
1924 				goto bail;
1925 			}
1926 
1927 			if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1928 				status = -ERESTARTSYS;
1929 				mlog(ML_NOTICE, "Timed out joining dlm domain "
1930 				     "%s after %u msecs\n", dlm->name,
1931 				     total_backoff);
1932 				goto bail;
1933 			}
1934 
1935 			/*
1936 			 * <chip> After you!
1937 			 * <dale> No, after you!
1938 			 * <chip> I insist!
1939 			 * <dale> But you first!
1940 			 * ...
1941 			 */
1942 			backoff = (unsigned int)(jiffies & 0x3);
1943 			backoff *= DLM_DOMAIN_BACKOFF_MS;
1944 			total_backoff += backoff;
1945 			mlog(0, "backoff %d\n", backoff);
1946 			msleep(backoff);
1947 		}
1948 	} while (status == -EAGAIN);
1949 
1950 	if (status < 0) {
1951 		mlog_errno(status);
1952 		goto bail;
1953 	}
1954 
1955 	status = 0;
1956 bail:
1957 	wake_up(&dlm_domain_events);
1958 
1959 	if (status) {
1960 		dlm_unregister_domain_handlers(dlm);
1961 		dlm_debug_shutdown(dlm);
1962 		dlm_complete_thread(dlm);
1963 		dlm_complete_recovery_thread(dlm);
1964 		dlm_destroy_dlm_worker(dlm);
1965 	}
1966 
1967 	return status;
1968 }
1969 
1970 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1971 				u32 key)
1972 {
1973 	int i;
1974 	int ret;
1975 	struct dlm_ctxt *dlm = NULL;
1976 
1977 	dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1978 	if (!dlm) {
1979 		ret = -ENOMEM;
1980 		mlog_errno(ret);
1981 		goto leave;
1982 	}
1983 
1984 	dlm->name = kstrdup(domain, GFP_KERNEL);
1985 	if (dlm->name == NULL) {
1986 		ret = -ENOMEM;
1987 		mlog_errno(ret);
1988 		goto leave;
1989 	}
1990 
1991 	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1992 	if (!dlm->lockres_hash) {
1993 		ret = -ENOMEM;
1994 		mlog_errno(ret);
1995 		goto leave;
1996 	}
1997 
1998 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
1999 		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
2000 
2001 	dlm->master_hash = (struct hlist_head **)
2002 				dlm_alloc_pagevec(DLM_HASH_PAGES);
2003 	if (!dlm->master_hash) {
2004 		ret = -ENOMEM;
2005 		mlog_errno(ret);
2006 		goto leave;
2007 	}
2008 
2009 	for (i = 0; i < DLM_HASH_BUCKETS; i++)
2010 		INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2011 
2012 	dlm->key = key;
2013 	dlm->node_num = o2nm_this_node();
2014 
2015 	ret = dlm_create_debugfs_subroot(dlm);
2016 	if (ret < 0)
2017 		goto leave;
2018 
2019 	spin_lock_init(&dlm->spinlock);
2020 	spin_lock_init(&dlm->master_lock);
2021 	spin_lock_init(&dlm->ast_lock);
2022 	spin_lock_init(&dlm->track_lock);
2023 	INIT_LIST_HEAD(&dlm->list);
2024 	INIT_LIST_HEAD(&dlm->dirty_list);
2025 	INIT_LIST_HEAD(&dlm->reco.resources);
2026 	INIT_LIST_HEAD(&dlm->reco.node_data);
2027 	INIT_LIST_HEAD(&dlm->purge_list);
2028 	INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2029 	INIT_LIST_HEAD(&dlm->tracking_list);
2030 	dlm->reco.state = 0;
2031 
2032 	INIT_LIST_HEAD(&dlm->pending_asts);
2033 	INIT_LIST_HEAD(&dlm->pending_basts);
2034 
2035 	mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2036 		  dlm->recovery_map, &(dlm->recovery_map[0]));
2037 
2038 	memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2039 	memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2040 	memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2041 
2042 	dlm->dlm_thread_task = NULL;
2043 	dlm->dlm_reco_thread_task = NULL;
2044 	dlm->dlm_worker = NULL;
2045 	init_waitqueue_head(&dlm->dlm_thread_wq);
2046 	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2047 	init_waitqueue_head(&dlm->reco.event);
2048 	init_waitqueue_head(&dlm->ast_wq);
2049 	init_waitqueue_head(&dlm->migration_wq);
2050 	INIT_LIST_HEAD(&dlm->mle_hb_events);
2051 
2052 	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2053 	init_waitqueue_head(&dlm->dlm_join_events);
2054 
2055 	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2056 	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2057 
2058 	atomic_set(&dlm->res_tot_count, 0);
2059 	atomic_set(&dlm->res_cur_count, 0);
2060 	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2061 		atomic_set(&dlm->mle_tot_count[i], 0);
2062 		atomic_set(&dlm->mle_cur_count[i], 0);
2063 	}
2064 
2065 	spin_lock_init(&dlm->work_lock);
2066 	INIT_LIST_HEAD(&dlm->work_list);
2067 	INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2068 
2069 	kref_init(&dlm->dlm_refs);
2070 	dlm->dlm_state = DLM_CTXT_NEW;
2071 
2072 	INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2073 
2074 	mlog(0, "context init: refcount %u\n",
2075 		  atomic_read(&dlm->dlm_refs.refcount));
2076 
2077 leave:
2078 	if (ret < 0 && dlm) {
2079 		if (dlm->master_hash)
2080 			dlm_free_pagevec((void **)dlm->master_hash,
2081 					DLM_HASH_PAGES);
2082 
2083 		if (dlm->lockres_hash)
2084 			dlm_free_pagevec((void **)dlm->lockres_hash,
2085 					DLM_HASH_PAGES);
2086 
2087 		kfree(dlm->name);
2088 		kfree(dlm);
2089 		dlm = NULL;
2090 	}
2091 	return dlm;
2092 }
2093 
2094 /*
2095  * Compare a requested locking protocol version against the current one.
2096  *
2097  * If the major numbers are different, they are incompatible.
2098  * If the current minor is greater than the request, they are incompatible.
2099  * If the current minor is less than or equal to the request, they are
2100  * compatible, and the requester should run at the current minor version.
2101  */
2102 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2103 				struct dlm_protocol_version *request)
2104 {
2105 	if (existing->pv_major != request->pv_major)
2106 		return 1;
2107 
2108 	if (existing->pv_minor > request->pv_minor)
2109 		return 1;
2110 
2111 	if (existing->pv_minor < request->pv_minor)
2112 		request->pv_minor = existing->pv_minor;
2113 
2114 	return 0;
2115 }
2116 
2117 /*
2118  * dlm_register_domain: one-time setup per "domain".
2119  *
2120  * The filesystem passes in the requested locking version via proto.
2121  * If registration was successful, proto will contain the negotiated
2122  * locking protocol.
2123  */
2124 struct dlm_ctxt * dlm_register_domain(const char *domain,
2125 			       u32 key,
2126 			       struct dlm_protocol_version *fs_proto)
2127 {
2128 	int ret;
2129 	struct dlm_ctxt *dlm = NULL;
2130 	struct dlm_ctxt *new_ctxt = NULL;
2131 
2132 	if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2133 		ret = -ENAMETOOLONG;
2134 		mlog(ML_ERROR, "domain name length too long\n");
2135 		goto leave;
2136 	}
2137 
2138 	mlog(0, "register called for domain \"%s\"\n", domain);
2139 
2140 retry:
2141 	dlm = NULL;
2142 	if (signal_pending(current)) {
2143 		ret = -ERESTARTSYS;
2144 		mlog_errno(ret);
2145 		goto leave;
2146 	}
2147 
2148 	spin_lock(&dlm_domain_lock);
2149 
2150 	dlm = __dlm_lookup_domain(domain);
2151 	if (dlm) {
2152 		if (dlm->dlm_state != DLM_CTXT_JOINED) {
2153 			spin_unlock(&dlm_domain_lock);
2154 
2155 			mlog(0, "This ctxt is not joined yet!\n");
2156 			wait_event_interruptible(dlm_domain_events,
2157 						 dlm_wait_on_domain_helper(
2158 							 domain));
2159 			goto retry;
2160 		}
2161 
2162 		if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2163 			spin_unlock(&dlm_domain_lock);
2164 			mlog(ML_ERROR,
2165 			     "Requested locking protocol version is not "
2166 			     "compatible with already registered domain "
2167 			     "\"%s\"\n", domain);
2168 			ret = -EPROTO;
2169 			goto leave;
2170 		}
2171 
2172 		__dlm_get(dlm);
2173 		dlm->num_joins++;
2174 
2175 		spin_unlock(&dlm_domain_lock);
2176 
2177 		ret = 0;
2178 		goto leave;
2179 	}
2180 
2181 	/* doesn't exist */
2182 	if (!new_ctxt) {
2183 		spin_unlock(&dlm_domain_lock);
2184 
2185 		new_ctxt = dlm_alloc_ctxt(domain, key);
2186 		if (new_ctxt)
2187 			goto retry;
2188 
2189 		ret = -ENOMEM;
2190 		mlog_errno(ret);
2191 		goto leave;
2192 	}
2193 
2194 	/* a little variable switch-a-roo here... */
2195 	dlm = new_ctxt;
2196 	new_ctxt = NULL;
2197 
2198 	/* add the new domain */
2199 	list_add_tail(&dlm->list, &dlm_domains);
2200 	spin_unlock(&dlm_domain_lock);
2201 
2202 	/*
2203 	 * Pass the locking protocol version into the join.  If the join
2204 	 * succeeds, it will have the negotiated protocol set.
2205 	 */
2206 	dlm->dlm_locking_proto = dlm_protocol;
2207 	dlm->fs_locking_proto = *fs_proto;
2208 
2209 	ret = dlm_join_domain(dlm);
2210 	if (ret) {
2211 		mlog_errno(ret);
2212 		dlm_put(dlm);
2213 		goto leave;
2214 	}
2215 
2216 	/* Tell the caller what locking protocol we negotiated */
2217 	*fs_proto = dlm->fs_locking_proto;
2218 
2219 	ret = 0;
2220 leave:
2221 	if (new_ctxt)
2222 		dlm_free_ctxt_mem(new_ctxt);
2223 
2224 	if (ret < 0)
2225 		dlm = ERR_PTR(ret);
2226 
2227 	return dlm;
2228 }
2229 EXPORT_SYMBOL_GPL(dlm_register_domain);
2230 
2231 static LIST_HEAD(dlm_join_handlers);
2232 
2233 static void dlm_unregister_net_handlers(void)
2234 {
2235 	o2net_unregister_handler_list(&dlm_join_handlers);
2236 }
2237 
2238 static int dlm_register_net_handlers(void)
2239 {
2240 	int status = 0;
2241 
2242 	status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2243 					sizeof(struct dlm_query_join_request),
2244 					dlm_query_join_handler,
2245 					NULL, NULL, &dlm_join_handlers);
2246 	if (status)
2247 		goto bail;
2248 
2249 	status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2250 					sizeof(struct dlm_assert_joined),
2251 					dlm_assert_joined_handler,
2252 					NULL, NULL, &dlm_join_handlers);
2253 	if (status)
2254 		goto bail;
2255 
2256 	status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2257 					sizeof(struct dlm_cancel_join),
2258 					dlm_cancel_join_handler,
2259 					NULL, NULL, &dlm_join_handlers);
2260 	if (status)
2261 		goto bail;
2262 
2263 	status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2264 					sizeof(struct dlm_query_region),
2265 					dlm_query_region_handler,
2266 					NULL, NULL, &dlm_join_handlers);
2267 
2268 	if (status)
2269 		goto bail;
2270 
2271 	status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2272 					sizeof(struct dlm_query_nodeinfo),
2273 					dlm_query_nodeinfo_handler,
2274 					NULL, NULL, &dlm_join_handlers);
2275 bail:
2276 	if (status < 0)
2277 		dlm_unregister_net_handlers();
2278 
2279 	return status;
2280 }
2281 
2282 /* Domain eviction callback handling.
2283  *
2284  * The file system requires notification of node death *before* the
2285  * dlm completes it's recovery work, otherwise it may be able to
2286  * acquire locks on resources requiring recovery. Since the dlm can
2287  * evict a node from it's domain *before* heartbeat fires, a similar
2288  * mechanism is required. */
2289 
2290 /* Eviction is not expected to happen often, so a per-domain lock is
2291  * not necessary. Eviction callbacks are allowed to sleep for short
2292  * periods of time. */
2293 static DECLARE_RWSEM(dlm_callback_sem);
2294 
2295 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2296 					int node_num)
2297 {
2298 	struct dlm_eviction_cb *cb;
2299 
2300 	down_read(&dlm_callback_sem);
2301 	list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
2302 		cb->ec_func(node_num, cb->ec_data);
2303 	}
2304 	up_read(&dlm_callback_sem);
2305 }
2306 
2307 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2308 			   dlm_eviction_func *f,
2309 			   void *data)
2310 {
2311 	INIT_LIST_HEAD(&cb->ec_item);
2312 	cb->ec_func = f;
2313 	cb->ec_data = data;
2314 }
2315 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2316 
2317 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2318 			      struct dlm_eviction_cb *cb)
2319 {
2320 	down_write(&dlm_callback_sem);
2321 	list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2322 	up_write(&dlm_callback_sem);
2323 }
2324 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2325 
2326 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2327 {
2328 	down_write(&dlm_callback_sem);
2329 	list_del_init(&cb->ec_item);
2330 	up_write(&dlm_callback_sem);
2331 }
2332 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2333 
2334 static int __init dlm_init(void)
2335 {
2336 	int status;
2337 
2338 	status = dlm_init_mle_cache();
2339 	if (status) {
2340 		mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2341 		goto error;
2342 	}
2343 
2344 	status = dlm_init_master_caches();
2345 	if (status) {
2346 		mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2347 		     "o2dlm_lockname slabcaches\n");
2348 		goto error;
2349 	}
2350 
2351 	status = dlm_init_lock_cache();
2352 	if (status) {
2353 		mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2354 		goto error;
2355 	}
2356 
2357 	status = dlm_register_net_handlers();
2358 	if (status) {
2359 		mlog(ML_ERROR, "Unable to register network handlers\n");
2360 		goto error;
2361 	}
2362 
2363 	status = dlm_create_debugfs_root();
2364 	if (status)
2365 		goto error;
2366 
2367 	return 0;
2368 error:
2369 	dlm_unregister_net_handlers();
2370 	dlm_destroy_lock_cache();
2371 	dlm_destroy_master_caches();
2372 	dlm_destroy_mle_cache();
2373 	return -1;
2374 }
2375 
2376 static void __exit dlm_exit (void)
2377 {
2378 	dlm_destroy_debugfs_root();
2379 	dlm_unregister_net_handlers();
2380 	dlm_destroy_lock_cache();
2381 	dlm_destroy_master_caches();
2382 	dlm_destroy_mle_cache();
2383 }
2384 
2385 MODULE_AUTHOR("Oracle");
2386 MODULE_LICENSE("GPL");
2387 MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
2388 
2389 module_init(dlm_init);
2390 module_exit(dlm_exit);
2391