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 */
55static inline void byte_set_bit(u8 nr, u8 map[])
56{
57	map[nr >> 3] |= (1UL << (nr & 7));
58}
59
60static inline int byte_test_bit(u8 nr, u8 map[])
61{
62	return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
63}
64
65static 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
79static void dlm_free_pagevec(void **vec, int pages)
80{
81	while (pages--)
82		free_page((unsigned long)vec[pages]);
83	kfree(vec);
84}
85
86static 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;
102out_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
120DEFINE_SPINLOCK(dlm_domain_lock);
121LIST_HEAD(dlm_domains);
122static 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 */
136static const struct dlm_protocol_version dlm_protocol = {
137	.pv_major = 1,
138	.pv_minor = 2,
139};
140
141#define DLM_DOMAIN_BACKOFF_MS 200
142
143static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
144				  void **ret_data);
145static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
146				     void **ret_data);
147static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
148				   void **ret_data);
149static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
150				    void *data, void **ret_data);
151static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
152				   void **ret_data);
153static int dlm_protocol_compare(struct dlm_protocol_version *existing,
154				struct dlm_protocol_version *request);
155
156static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
157
158void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
159{
160	if (hlist_unhashed(&res->hash_node))
161		return;
162
163	mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
164	     res->lockname.name);
165	hlist_del_init(&res->hash_node);
166	dlm_lockres_put(res);
167}
168
169void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
170{
171	struct hlist_head *bucket;
172	struct qstr *q;
173
174	assert_spin_locked(&dlm->spinlock);
175
176	q = &res->lockname;
177	bucket = dlm_lockres_hash(dlm, q->hash);
178
179	/* get a reference for our hashtable */
180	dlm_lockres_get(res);
181
182	hlist_add_head(&res->hash_node, bucket);
183
184	mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
185	     res->lockname.name);
186}
187
188struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
189						     const char *name,
190						     unsigned int len,
191						     unsigned int hash)
192{
193	struct hlist_head *bucket;
194	struct dlm_lock_resource *res;
195
196	mlog(0, "%.*s\n", len, name);
197
198	assert_spin_locked(&dlm->spinlock);
199
200	bucket = dlm_lockres_hash(dlm, hash);
201
202	hlist_for_each_entry(res, bucket, hash_node) {
203		if (res->lockname.name[0] != name[0])
204			continue;
205		if (unlikely(res->lockname.len != len))
206			continue;
207		if (memcmp(res->lockname.name + 1, name + 1, len - 1))
208			continue;
209		dlm_lockres_get(res);
210		return res;
211	}
212	return NULL;
213}
214
215/* intended to be called by functions which do not care about lock
216 * resources which are being purged (most net _handler functions).
217 * this will return NULL for any lock resource which is found but
218 * currently in the process of dropping its mastery reference.
219 * use __dlm_lookup_lockres_full when you need the lock resource
220 * regardless (e.g. dlm_get_lock_resource) */
221struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
222						const char *name,
223						unsigned int len,
224						unsigned int hash)
225{
226	struct dlm_lock_resource *res = NULL;
227
228	mlog(0, "%.*s\n", len, name);
229
230	assert_spin_locked(&dlm->spinlock);
231
232	res = __dlm_lookup_lockres_full(dlm, name, len, hash);
233	if (res) {
234		spin_lock(&res->spinlock);
235		if (res->state & DLM_LOCK_RES_DROPPING_REF) {
236			spin_unlock(&res->spinlock);
237			dlm_lockres_put(res);
238			return NULL;
239		}
240		spin_unlock(&res->spinlock);
241	}
242
243	return res;
244}
245
246struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
247				    const char *name,
248				    unsigned int len)
249{
250	struct dlm_lock_resource *res;
251	unsigned int hash = dlm_lockid_hash(name, len);
252
253	spin_lock(&dlm->spinlock);
254	res = __dlm_lookup_lockres(dlm, name, len, hash);
255	spin_unlock(&dlm->spinlock);
256	return res;
257}
258
259static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
260{
261	struct dlm_ctxt *tmp;
262
263	assert_spin_locked(&dlm_domain_lock);
264
265	/* tmp->name here is always NULL terminated,
266	 * but domain may not be! */
267	list_for_each_entry(tmp, &dlm_domains, list) {
268		if (strlen(tmp->name) == len &&
269		    memcmp(tmp->name, domain, len)==0)
270			return tmp;
271	}
272
273	return NULL;
274}
275
276/* For null terminated domain strings ONLY */
277static 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" */
288static 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
305static 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	kfree(dlm->name);
316	kfree(dlm);
317}
318
319/* A little strange - this function will be called while holding
320 * dlm_domain_lock and is expected to be holding it on the way out. We
321 * will however drop and reacquire it multiple times */
322static void dlm_ctxt_release(struct kref *kref)
323{
324	struct dlm_ctxt *dlm;
325
326	dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
327
328	BUG_ON(dlm->num_joins);
329	BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
330
331	/* we may still be in the list if we hit an error during join. */
332	list_del_init(&dlm->list);
333
334	spin_unlock(&dlm_domain_lock);
335
336	mlog(0, "freeing memory from domain %s\n", dlm->name);
337
338	wake_up(&dlm_domain_events);
339
340	dlm_free_ctxt_mem(dlm);
341
342	spin_lock(&dlm_domain_lock);
343}
344
345void dlm_put(struct dlm_ctxt *dlm)
346{
347	spin_lock(&dlm_domain_lock);
348	kref_put(&dlm->dlm_refs, dlm_ctxt_release);
349	spin_unlock(&dlm_domain_lock);
350}
351
352static void __dlm_get(struct dlm_ctxt *dlm)
353{
354	kref_get(&dlm->dlm_refs);
355}
356
357/* given a questionable reference to a dlm object, gets a reference if
358 * it can find it in the list, otherwise returns NULL in which case
359 * you shouldn't trust your pointer. */
360struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
361{
362	struct dlm_ctxt *target;
363	struct dlm_ctxt *ret = NULL;
364
365	spin_lock(&dlm_domain_lock);
366
367	list_for_each_entry(target, &dlm_domains, list) {
368		if (target == dlm) {
369			__dlm_get(target);
370			ret = target;
371			break;
372		}
373	}
374
375	spin_unlock(&dlm_domain_lock);
376
377	return ret;
378}
379
380int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
381{
382	int ret;
383
384	spin_lock(&dlm_domain_lock);
385	ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
386		(dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
387	spin_unlock(&dlm_domain_lock);
388
389	return ret;
390}
391
392static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
393{
394	if (dlm->dlm_worker) {
395		flush_workqueue(dlm->dlm_worker);
396		destroy_workqueue(dlm->dlm_worker);
397		dlm->dlm_worker = NULL;
398	}
399}
400
401static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
402{
403	dlm_unregister_domain_handlers(dlm);
404	dlm_debug_shutdown(dlm);
405	dlm_complete_thread(dlm);
406	dlm_complete_recovery_thread(dlm);
407	dlm_destroy_dlm_worker(dlm);
408
409	/* We've left the domain. Now we can take ourselves out of the
410	 * list and allow the kref stuff to help us free the
411	 * memory. */
412	spin_lock(&dlm_domain_lock);
413	list_del_init(&dlm->list);
414	spin_unlock(&dlm_domain_lock);
415
416	/* Wake up anyone waiting for us to remove this domain */
417	wake_up(&dlm_domain_events);
418}
419
420static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
421{
422	int i, num, n, ret = 0;
423	struct dlm_lock_resource *res;
424	struct hlist_node *iter;
425	struct hlist_head *bucket;
426	int dropped;
427
428	mlog(0, "Migrating locks from domain %s\n", dlm->name);
429
430	num = 0;
431	spin_lock(&dlm->spinlock);
432	for (i = 0; i < DLM_HASH_BUCKETS; i++) {
433redo_bucket:
434		n = 0;
435		bucket = dlm_lockres_hash(dlm, i);
436		iter = bucket->first;
437		while (iter) {
438			n++;
439			res = hlist_entry(iter, struct dlm_lock_resource,
440					  hash_node);
441			dlm_lockres_get(res);
442			/* migrate, if necessary.  this will drop the dlm
443			 * spinlock and retake it if it does migration. */
444			dropped = dlm_empty_lockres(dlm, res);
445
446			spin_lock(&res->spinlock);
447			if (dropped)
448				__dlm_lockres_calc_usage(dlm, res);
449			else
450				iter = res->hash_node.next;
451			spin_unlock(&res->spinlock);
452
453			dlm_lockres_put(res);
454
455			if (dropped) {
456				cond_resched_lock(&dlm->spinlock);
457				goto redo_bucket;
458			}
459		}
460		cond_resched_lock(&dlm->spinlock);
461		num += n;
462	}
463	spin_unlock(&dlm->spinlock);
464	wake_up(&dlm->dlm_thread_wq);
465
466	/* let the dlm thread take care of purging, keep scanning until
467	 * nothing remains in the hash */
468	if (num) {
469		mlog(0, "%s: %d lock resources in hash last pass\n",
470		     dlm->name, num);
471		ret = -EAGAIN;
472	}
473	mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
474	return ret;
475}
476
477static int dlm_no_joining_node(struct dlm_ctxt *dlm)
478{
479	int ret;
480
481	spin_lock(&dlm->spinlock);
482	ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
483	spin_unlock(&dlm->spinlock);
484
485	return ret;
486}
487
488static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
489					 void *data, void **ret_data)
490{
491	struct dlm_ctxt *dlm = data;
492	unsigned int node;
493	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
494
495	if (!dlm_grab(dlm))
496		return 0;
497
498	node = exit_msg->node_idx;
499	mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
500
501	spin_lock(&dlm->spinlock);
502	set_bit(node, dlm->exit_domain_map);
503	spin_unlock(&dlm->spinlock);
504
505	dlm_put(dlm);
506
507	return 0;
508}
509
510static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
511{
512	/* Yikes, a double spinlock! I need domain_lock for the dlm
513	 * state and the dlm spinlock for join state... Sorry! */
514again:
515	spin_lock(&dlm_domain_lock);
516	spin_lock(&dlm->spinlock);
517
518	if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
519		mlog(0, "Node %d is joining, we wait on it.\n",
520			  dlm->joining_node);
521		spin_unlock(&dlm->spinlock);
522		spin_unlock(&dlm_domain_lock);
523
524		wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
525		goto again;
526	}
527
528	dlm->dlm_state = DLM_CTXT_LEAVING;
529	spin_unlock(&dlm->spinlock);
530	spin_unlock(&dlm_domain_lock);
531}
532
533static void __dlm_print_nodes(struct dlm_ctxt *dlm)
534{
535	int node = -1, num = 0;
536
537	assert_spin_locked(&dlm->spinlock);
538
539	printk("( ");
540	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
541				     node + 1)) < O2NM_MAX_NODES) {
542		printk("%d ", node);
543		++num;
544	}
545	printk(") %u nodes\n", num);
546}
547
548static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
549				   void **ret_data)
550{
551	struct dlm_ctxt *dlm = data;
552	unsigned int node;
553	struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
554
555	mlog(0, "%p %u %p", msg, len, data);
556
557	if (!dlm_grab(dlm))
558		return 0;
559
560	node = exit_msg->node_idx;
561
562	spin_lock(&dlm->spinlock);
563	clear_bit(node, dlm->domain_map);
564	clear_bit(node, dlm->exit_domain_map);
565	printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
566	__dlm_print_nodes(dlm);
567
568	/* notify anything attached to the heartbeat events */
569	dlm_hb_event_notify_attached(dlm, node, 0);
570
571	spin_unlock(&dlm->spinlock);
572
573	dlm_put(dlm);
574
575	return 0;
576}
577
578static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
579				    unsigned int node)
580{
581	int status;
582	struct dlm_exit_domain leave_msg;
583
584	mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
585	     msg_type, node);
586
587	memset(&leave_msg, 0, sizeof(leave_msg));
588	leave_msg.node_idx = dlm->node_num;
589
590	status = o2net_send_message(msg_type, dlm->key, &leave_msg,
591				    sizeof(leave_msg), node, NULL);
592	if (status < 0)
593		mlog(ML_ERROR, "Error %d sending domain exit message %u "
594		     "to node %u on domain %s\n", status, msg_type, node,
595		     dlm->name);
596
597	return status;
598}
599
600static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
601{
602	int node = -1;
603
604	/* Support for begin exit domain was added in 1.2 */
605	if (dlm->dlm_locking_proto.pv_major == 1 &&
606	    dlm->dlm_locking_proto.pv_minor < 2)
607		return;
608
609	/*
610	 * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
611	 * informational. Meaning if a node does not receive the message,
612	 * so be it.
613	 */
614	spin_lock(&dlm->spinlock);
615	while (1) {
616		node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
617		if (node >= O2NM_MAX_NODES)
618			break;
619		if (node == dlm->node_num)
620			continue;
621
622		spin_unlock(&dlm->spinlock);
623		dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
624		spin_lock(&dlm->spinlock);
625	}
626	spin_unlock(&dlm->spinlock);
627}
628
629static void dlm_leave_domain(struct dlm_ctxt *dlm)
630{
631	int node, clear_node, status;
632
633	/* At this point we've migrated away all our locks and won't
634	 * accept mastership of new ones. The dlm is responsible for
635	 * almost nothing now. We make sure not to confuse any joining
636	 * nodes and then commence shutdown procedure. */
637
638	spin_lock(&dlm->spinlock);
639	/* Clear ourselves from the domain map */
640	clear_bit(dlm->node_num, dlm->domain_map);
641	while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
642				     0)) < O2NM_MAX_NODES) {
643		/* Drop the dlm spinlock. This is safe wrt the domain_map.
644		 * -nodes cannot be added now as the
645		 *   query_join_handlers knows to respond with OK_NO_MAP
646		 * -we catch the right network errors if a node is
647		 *   removed from the map while we're sending him the
648		 *   exit message. */
649		spin_unlock(&dlm->spinlock);
650
651		clear_node = 1;
652
653		status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
654						  node);
655		if (status < 0 &&
656		    status != -ENOPROTOOPT &&
657		    status != -ENOTCONN) {
658			mlog(ML_NOTICE, "Error %d sending domain exit message "
659			     "to node %d\n", status, node);
660
661			/* Not sure what to do here but lets sleep for
662			 * a bit in case this was a transient
663			 * error... */
664			msleep(DLM_DOMAIN_BACKOFF_MS);
665			clear_node = 0;
666		}
667
668		spin_lock(&dlm->spinlock);
669		/* If we're not clearing the node bit then we intend
670		 * to loop back around to try again. */
671		if (clear_node)
672			clear_bit(node, dlm->domain_map);
673	}
674	spin_unlock(&dlm->spinlock);
675}
676
677int dlm_shutting_down(struct dlm_ctxt *dlm)
678{
679	int ret = 0;
680
681	spin_lock(&dlm_domain_lock);
682
683	if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
684		ret = 1;
685
686	spin_unlock(&dlm_domain_lock);
687
688	return ret;
689}
690
691void dlm_unregister_domain(struct dlm_ctxt *dlm)
692{
693	int leave = 0;
694	struct dlm_lock_resource *res;
695
696	spin_lock(&dlm_domain_lock);
697	BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
698	BUG_ON(!dlm->num_joins);
699
700	dlm->num_joins--;
701	if (!dlm->num_joins) {
702		/* We mark it "in shutdown" now so new register
703		 * requests wait until we've completely left the
704		 * domain. Don't use DLM_CTXT_LEAVING yet as we still
705		 * want new domain joins to communicate with us at
706		 * least until we've completed migration of our
707		 * resources. */
708		dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
709		leave = 1;
710	}
711	spin_unlock(&dlm_domain_lock);
712
713	if (leave) {
714		mlog(0, "shutting down domain %s\n", dlm->name);
715		dlm_begin_exit_domain(dlm);
716
717		/* We changed dlm state, notify the thread */
718		dlm_kick_thread(dlm, NULL);
719
720		while (dlm_migrate_all_locks(dlm)) {
721			/* Give dlm_thread time to purge the lockres' */
722			msleep(500);
723			mlog(0, "%s: more migration to do\n", dlm->name);
724		}
725
726		/* This list should be empty. If not, print remaining lockres */
727		if (!list_empty(&dlm->tracking_list)) {
728			mlog(ML_ERROR, "Following lockres' are still on the "
729			     "tracking list:\n");
730			list_for_each_entry(res, &dlm->tracking_list, tracking)
731				dlm_print_one_lock_resource(res);
732		}
733
734		dlm_mark_domain_leaving(dlm);
735		dlm_leave_domain(dlm);
736		printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
737		dlm_force_free_mles(dlm);
738		dlm_complete_dlm_shutdown(dlm);
739	}
740	dlm_put(dlm);
741}
742EXPORT_SYMBOL_GPL(dlm_unregister_domain);
743
744static int dlm_query_join_proto_check(char *proto_type, int node,
745				      struct dlm_protocol_version *ours,
746				      struct dlm_protocol_version *request)
747{
748	int rc;
749	struct dlm_protocol_version proto = *request;
750
751	if (!dlm_protocol_compare(ours, &proto)) {
752		mlog(0,
753		     "node %u wanted to join with %s locking protocol "
754		     "%u.%u, we respond with %u.%u\n",
755		     node, proto_type,
756		     request->pv_major,
757		     request->pv_minor,
758		     proto.pv_major, proto.pv_minor);
759		request->pv_minor = proto.pv_minor;
760		rc = 0;
761	} else {
762		mlog(ML_NOTICE,
763		     "Node %u wanted to join with %s locking "
764		     "protocol %u.%u, but we have %u.%u, disallowing\n",
765		     node, proto_type,
766		     request->pv_major,
767		     request->pv_minor,
768		     ours->pv_major,
769		     ours->pv_minor);
770		rc = 1;
771	}
772
773	return rc;
774}
775
776/*
777 * struct dlm_query_join_packet is made up of four one-byte fields.  They
778 * are effectively in big-endian order already.  However, little-endian
779 * machines swap them before putting the packet on the wire (because
780 * query_join's response is a status, and that status is treated as a u32
781 * on the wire).  Thus, a big-endian and little-endian machines will treat
782 * this structure differently.
783 *
784 * The solution is to have little-endian machines swap the structure when
785 * converting from the structure to the u32 representation.  This will
786 * result in the structure having the correct format on the wire no matter
787 * the host endian format.
788 */
789static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
790					  u32 *wire)
791{
792	union dlm_query_join_response response;
793
794	response.packet = *packet;
795	*wire = be32_to_cpu(response.intval);
796}
797
798static void dlm_query_join_wire_to_packet(u32 wire,
799					  struct dlm_query_join_packet *packet)
800{
801	union dlm_query_join_response response;
802
803	response.intval = cpu_to_be32(wire);
804	*packet = response.packet;
805}
806
807static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
808				  void **ret_data)
809{
810	struct dlm_query_join_request *query;
811	struct dlm_query_join_packet packet = {
812		.code = JOIN_DISALLOW,
813	};
814	struct dlm_ctxt *dlm = NULL;
815	u32 response;
816	u8 nodenum;
817
818	query = (struct dlm_query_join_request *) msg->buf;
819
820	mlog(0, "node %u wants to join domain %s\n", query->node_idx,
821		  query->domain);
822
823	/*
824	 * If heartbeat doesn't consider the node live, tell it
825	 * to back off and try again.  This gives heartbeat a chance
826	 * to catch up.
827	 */
828	if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
829		mlog(0, "node %u is not in our live map yet\n",
830		     query->node_idx);
831
832		packet.code = JOIN_DISALLOW;
833		goto respond;
834	}
835
836	packet.code = JOIN_OK_NO_MAP;
837
838	spin_lock(&dlm_domain_lock);
839	dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
840	if (!dlm)
841		goto unlock_respond;
842
843	/*
844	 * There is a small window where the joining node may not see the
845	 * node(s) that just left but still part of the cluster. DISALLOW
846	 * join request if joining node has different node map.
847	 */
848	nodenum=0;
849	while (nodenum < O2NM_MAX_NODES) {
850		if (test_bit(nodenum, dlm->domain_map)) {
851			if (!byte_test_bit(nodenum, query->node_map)) {
852				mlog(0, "disallow join as node %u does not "
853				     "have node %u in its nodemap\n",
854				     query->node_idx, nodenum);
855				packet.code = JOIN_DISALLOW;
856				goto unlock_respond;
857			}
858		}
859		nodenum++;
860	}
861
862	/* Once the dlm ctxt is marked as leaving then we don't want
863	 * to be put in someone's domain map.
864	 * Also, explicitly disallow joining at certain troublesome
865	 * times (ie. during recovery). */
866	if (dlm->dlm_state != DLM_CTXT_LEAVING) {
867		int bit = query->node_idx;
868		spin_lock(&dlm->spinlock);
869
870		if (dlm->dlm_state == DLM_CTXT_NEW &&
871		    dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
872			/*If this is a brand new context and we
873			 * haven't started our join process yet, then
874			 * the other node won the race. */
875			packet.code = JOIN_OK_NO_MAP;
876		} else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
877			/* Disallow parallel joins. */
878			packet.code = JOIN_DISALLOW;
879		} else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
880			mlog(0, "node %u trying to join, but recovery "
881			     "is ongoing.\n", bit);
882			packet.code = JOIN_DISALLOW;
883		} else if (test_bit(bit, dlm->recovery_map)) {
884			mlog(0, "node %u trying to join, but it "
885			     "still needs recovery.\n", bit);
886			packet.code = JOIN_DISALLOW;
887		} else if (test_bit(bit, dlm->domain_map)) {
888			mlog(0, "node %u trying to join, but it "
889			     "is still in the domain! needs recovery?\n",
890			     bit);
891			packet.code = JOIN_DISALLOW;
892		} else {
893			/* Alright we're fully a part of this domain
894			 * so we keep some state as to who's joining
895			 * and indicate to him that needs to be fixed
896			 * up. */
897
898			/* Make sure we speak compatible locking protocols.  */
899			if (dlm_query_join_proto_check("DLM", bit,
900						       &dlm->dlm_locking_proto,
901						       &query->dlm_proto)) {
902				packet.code = JOIN_PROTOCOL_MISMATCH;
903			} else if (dlm_query_join_proto_check("fs", bit,
904							      &dlm->fs_locking_proto,
905							      &query->fs_proto)) {
906				packet.code = JOIN_PROTOCOL_MISMATCH;
907			} else {
908				packet.dlm_minor = query->dlm_proto.pv_minor;
909				packet.fs_minor = query->fs_proto.pv_minor;
910				packet.code = JOIN_OK;
911				__dlm_set_joining_node(dlm, query->node_idx);
912			}
913		}
914
915		spin_unlock(&dlm->spinlock);
916	}
917unlock_respond:
918	spin_unlock(&dlm_domain_lock);
919
920respond:
921	mlog(0, "We respond with %u\n", packet.code);
922
923	dlm_query_join_packet_to_wire(&packet, &response);
924	return response;
925}
926
927static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
928				     void **ret_data)
929{
930	struct dlm_assert_joined *assert;
931	struct dlm_ctxt *dlm = NULL;
932
933	assert = (struct dlm_assert_joined *) msg->buf;
934
935	mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
936		  assert->domain);
937
938	spin_lock(&dlm_domain_lock);
939	dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
940	/* XXX should we consider no dlm ctxt an error? */
941	if (dlm) {
942		spin_lock(&dlm->spinlock);
943
944		/* Alright, this node has officially joined our
945		 * domain. Set him in the map and clean up our
946		 * leftover join state. */
947		BUG_ON(dlm->joining_node != assert->node_idx);
948
949		if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
950			mlog(0, "dlm recovery is ongoing, disallow join\n");
951			spin_unlock(&dlm->spinlock);
952			spin_unlock(&dlm_domain_lock);
953			return -EAGAIN;
954		}
955
956		set_bit(assert->node_idx, dlm->domain_map);
957		clear_bit(assert->node_idx, dlm->exit_domain_map);
958		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
959
960		printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
961		       assert->node_idx, dlm->name);
962		__dlm_print_nodes(dlm);
963
964		/* notify anything attached to the heartbeat events */
965		dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
966
967		spin_unlock(&dlm->spinlock);
968	}
969	spin_unlock(&dlm_domain_lock);
970
971	return 0;
972}
973
974static int dlm_match_regions(struct dlm_ctxt *dlm,
975			     struct dlm_query_region *qr,
976			     char *local, int locallen)
977{
978	char *remote = qr->qr_regions;
979	char *l, *r;
980	int localnr, i, j, foundit;
981	int status = 0;
982
983	if (!o2hb_global_heartbeat_active()) {
984		if (qr->qr_numregions) {
985			mlog(ML_ERROR, "Domain %s: Joining node %d has global "
986			     "heartbeat enabled but local node %d does not\n",
987			     qr->qr_domain, qr->qr_node, dlm->node_num);
988			status = -EINVAL;
989		}
990		goto bail;
991	}
992
993	if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
994		mlog(ML_ERROR, "Domain %s: Local node %d has global "
995		     "heartbeat enabled but joining node %d does not\n",
996		     qr->qr_domain, dlm->node_num, qr->qr_node);
997		status = -EINVAL;
998		goto bail;
999	}
1000
1001	r = remote;
1002	for (i = 0; i < qr->qr_numregions; ++i) {
1003		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1004		r += O2HB_MAX_REGION_NAME_LEN;
1005	}
1006
1007	localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1008	localnr = o2hb_get_all_regions(local, (u8)localnr);
1009
1010	/* compare local regions with remote */
1011	l = local;
1012	for (i = 0; i < localnr; ++i) {
1013		foundit = 0;
1014		r = remote;
1015		for (j = 0; j <= qr->qr_numregions; ++j) {
1016			if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1017				foundit = 1;
1018				break;
1019			}
1020			r += O2HB_MAX_REGION_NAME_LEN;
1021		}
1022		if (!foundit) {
1023			status = -EINVAL;
1024			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1025			     "in local node %d but not in joining node %d\n",
1026			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1027			     dlm->node_num, qr->qr_node);
1028			goto bail;
1029		}
1030		l += O2HB_MAX_REGION_NAME_LEN;
1031	}
1032
1033	/* compare remote with local regions */
1034	r = remote;
1035	for (i = 0; i < qr->qr_numregions; ++i) {
1036		foundit = 0;
1037		l = local;
1038		for (j = 0; j < localnr; ++j) {
1039			if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1040				foundit = 1;
1041				break;
1042			}
1043			l += O2HB_MAX_REGION_NAME_LEN;
1044		}
1045		if (!foundit) {
1046			status = -EINVAL;
1047			mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1048			     "in joining node %d but not in local node %d\n",
1049			     qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1050			     qr->qr_node, dlm->node_num);
1051			goto bail;
1052		}
1053		r += O2HB_MAX_REGION_NAME_LEN;
1054	}
1055
1056bail:
1057	return status;
1058}
1059
1060static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1061{
1062	struct dlm_query_region *qr = NULL;
1063	int status, ret = 0, i;
1064	char *p;
1065
1066	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1067		goto bail;
1068
1069	qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1070	if (!qr) {
1071		ret = -ENOMEM;
1072		mlog_errno(ret);
1073		goto bail;
1074	}
1075
1076	qr->qr_node = dlm->node_num;
1077	qr->qr_namelen = strlen(dlm->name);
1078	memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1079	/* if local hb, the numregions will be zero */
1080	if (o2hb_global_heartbeat_active())
1081		qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1082							 O2NM_MAX_REGIONS);
1083
1084	p = qr->qr_regions;
1085	for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1086		mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1087
1088	i = -1;
1089	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1090				  i + 1)) < O2NM_MAX_NODES) {
1091		if (i == dlm->node_num)
1092			continue;
1093
1094		mlog(0, "Sending regions to node %d\n", i);
1095
1096		ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1097					 sizeof(struct dlm_query_region),
1098					 i, &status);
1099		if (ret >= 0)
1100			ret = status;
1101		if (ret) {
1102			mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1103			     ret, i);
1104			break;
1105		}
1106	}
1107
1108bail:
1109	kfree(qr);
1110	return ret;
1111}
1112
1113static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1114				    void *data, void **ret_data)
1115{
1116	struct dlm_query_region *qr;
1117	struct dlm_ctxt *dlm = NULL;
1118	char *local = NULL;
1119	int status = 0;
1120
1121	qr = (struct dlm_query_region *) msg->buf;
1122
1123	mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1124	     qr->qr_domain);
1125
1126	/* buffer used in dlm_mast_regions() */
1127	local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1128	if (!local)
1129		return -ENOMEM;
1130
1131	status = -EINVAL;
1132
1133	spin_lock(&dlm_domain_lock);
1134	dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1135	if (!dlm) {
1136		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1137		     "before join domain\n", qr->qr_node, qr->qr_domain);
1138		goto out_domain_lock;
1139	}
1140
1141	spin_lock(&dlm->spinlock);
1142	if (dlm->joining_node != qr->qr_node) {
1143		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1144		     "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1145		     dlm->joining_node);
1146		goto out_dlm_lock;
1147	}
1148
1149	/* Support for global heartbeat was added in 1.1 */
1150	if (dlm->dlm_locking_proto.pv_major == 1 &&
1151	    dlm->dlm_locking_proto.pv_minor == 0) {
1152		mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1153		     "but active dlm protocol is %d.%d\n", qr->qr_node,
1154		     qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1155		     dlm->dlm_locking_proto.pv_minor);
1156		goto out_dlm_lock;
1157	}
1158
1159	status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1160
1161out_dlm_lock:
1162	spin_unlock(&dlm->spinlock);
1163
1164out_domain_lock:
1165	spin_unlock(&dlm_domain_lock);
1166
1167	kfree(local);
1168
1169	return status;
1170}
1171
1172static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1173{
1174	struct o2nm_node *local;
1175	struct dlm_node_info *remote;
1176	int i, j;
1177	int status = 0;
1178
1179	for (j = 0; j < qn->qn_numnodes; ++j)
1180		mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1181		     &(qn->qn_nodes[j].ni_ipv4_address),
1182		     ntohs(qn->qn_nodes[j].ni_ipv4_port));
1183
1184	for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1185		local = o2nm_get_node_by_num(i);
1186		remote = NULL;
1187		for (j = 0; j < qn->qn_numnodes; ++j) {
1188			if (qn->qn_nodes[j].ni_nodenum == i) {
1189				remote = &(qn->qn_nodes[j]);
1190				break;
1191			}
1192		}
1193
1194		if (!local && !remote)
1195			continue;
1196
1197		if ((local && !remote) || (!local && remote))
1198			status = -EINVAL;
1199
1200		if (!status &&
1201		    ((remote->ni_nodenum != local->nd_num) ||
1202		     (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1203		     (remote->ni_ipv4_address != local->nd_ipv4_address)))
1204			status = -EINVAL;
1205
1206		if (status) {
1207			if (remote && !local)
1208				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1209				     "registered in joining node %d but not in "
1210				     "local node %d\n", qn->qn_domain,
1211				     remote->ni_nodenum,
1212				     &(remote->ni_ipv4_address),
1213				     ntohs(remote->ni_ipv4_port),
1214				     qn->qn_nodenum, dlm->node_num);
1215			if (local && !remote)
1216				mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1217				     "registered in local node %d but not in "
1218				     "joining node %d\n", qn->qn_domain,
1219				     local->nd_num, &(local->nd_ipv4_address),
1220				     ntohs(local->nd_ipv4_port),
1221				     dlm->node_num, qn->qn_nodenum);
1222			BUG_ON((!local && !remote));
1223		}
1224
1225		if (local)
1226			o2nm_node_put(local);
1227	}
1228
1229	return status;
1230}
1231
1232static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1233{
1234	struct dlm_query_nodeinfo *qn = NULL;
1235	struct o2nm_node *node;
1236	int ret = 0, status, count, i;
1237
1238	if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1239		goto bail;
1240
1241	qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1242	if (!qn) {
1243		ret = -ENOMEM;
1244		mlog_errno(ret);
1245		goto bail;
1246	}
1247
1248	for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1249		node = o2nm_get_node_by_num(i);
1250		if (!node)
1251			continue;
1252		qn->qn_nodes[count].ni_nodenum = node->nd_num;
1253		qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1254		qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1255		mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1256		     &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1257		++count;
1258		o2nm_node_put(node);
1259	}
1260
1261	qn->qn_nodenum = dlm->node_num;
1262	qn->qn_numnodes = count;
1263	qn->qn_namelen = strlen(dlm->name);
1264	memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1265
1266	i = -1;
1267	while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1268				  i + 1)) < O2NM_MAX_NODES) {
1269		if (i == dlm->node_num)
1270			continue;
1271
1272		mlog(0, "Sending nodeinfo to node %d\n", i);
1273
1274		ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1275					 qn, sizeof(struct dlm_query_nodeinfo),
1276					 i, &status);
1277		if (ret >= 0)
1278			ret = status;
1279		if (ret) {
1280			mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1281			break;
1282		}
1283	}
1284
1285bail:
1286	kfree(qn);
1287	return ret;
1288}
1289
1290static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1291				      void *data, void **ret_data)
1292{
1293	struct dlm_query_nodeinfo *qn;
1294	struct dlm_ctxt *dlm = NULL;
1295	int locked = 0, status = -EINVAL;
1296
1297	qn = (struct dlm_query_nodeinfo *) msg->buf;
1298
1299	mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1300	     qn->qn_domain);
1301
1302	spin_lock(&dlm_domain_lock);
1303	dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1304	if (!dlm) {
1305		mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1306		     "join domain\n", qn->qn_nodenum, qn->qn_domain);
1307		goto bail;
1308	}
1309
1310	spin_lock(&dlm->spinlock);
1311	locked = 1;
1312	if (dlm->joining_node != qn->qn_nodenum) {
1313		mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1314		     "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1315		     dlm->joining_node);
1316		goto bail;
1317	}
1318
1319	/* Support for node query was added in 1.1 */
1320	if (dlm->dlm_locking_proto.pv_major == 1 &&
1321	    dlm->dlm_locking_proto.pv_minor == 0) {
1322		mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1323		     "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1324		     qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1325		     dlm->dlm_locking_proto.pv_minor);
1326		goto bail;
1327	}
1328
1329	status = dlm_match_nodes(dlm, qn);
1330
1331bail:
1332	if (locked)
1333		spin_unlock(&dlm->spinlock);
1334	spin_unlock(&dlm_domain_lock);
1335
1336	return status;
1337}
1338
1339static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1340				   void **ret_data)
1341{
1342	struct dlm_cancel_join *cancel;
1343	struct dlm_ctxt *dlm = NULL;
1344
1345	cancel = (struct dlm_cancel_join *) msg->buf;
1346
1347	mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1348		  cancel->domain);
1349
1350	spin_lock(&dlm_domain_lock);
1351	dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1352
1353	if (dlm) {
1354		spin_lock(&dlm->spinlock);
1355
1356		/* Yikes, this guy wants to cancel his join. No
1357		 * problem, we simply cleanup our join state. */
1358		BUG_ON(dlm->joining_node != cancel->node_idx);
1359		__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1360
1361		spin_unlock(&dlm->spinlock);
1362	}
1363	spin_unlock(&dlm_domain_lock);
1364
1365	return 0;
1366}
1367
1368static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1369				    unsigned int node)
1370{
1371	int status;
1372	struct dlm_cancel_join cancel_msg;
1373
1374	memset(&cancel_msg, 0, sizeof(cancel_msg));
1375	cancel_msg.node_idx = dlm->node_num;
1376	cancel_msg.name_len = strlen(dlm->name);
1377	memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1378
1379	status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1380				    &cancel_msg, sizeof(cancel_msg), node,
1381				    NULL);
1382	if (status < 0) {
1383		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1384		     "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1385		     node);
1386		goto bail;
1387	}
1388
1389bail:
1390	return status;
1391}
1392
1393/* map_size should be in bytes. */
1394static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1395				 unsigned long *node_map,
1396				 unsigned int map_size)
1397{
1398	int status, tmpstat;
1399	unsigned int node;
1400
1401	if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1402			 sizeof(unsigned long))) {
1403		mlog(ML_ERROR,
1404		     "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1405		     map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1406		return -EINVAL;
1407	}
1408
1409	status = 0;
1410	node = -1;
1411	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1412				     node + 1)) < O2NM_MAX_NODES) {
1413		if (node == dlm->node_num)
1414			continue;
1415
1416		tmpstat = dlm_send_one_join_cancel(dlm, node);
1417		if (tmpstat) {
1418			mlog(ML_ERROR, "Error return %d cancelling join on "
1419			     "node %d\n", tmpstat, node);
1420			if (!status)
1421				status = tmpstat;
1422		}
1423	}
1424
1425	if (status)
1426		mlog_errno(status);
1427	return status;
1428}
1429
1430static int dlm_request_join(struct dlm_ctxt *dlm,
1431			    int node,
1432			    enum dlm_query_join_response_code *response)
1433{
1434	int status;
1435	struct dlm_query_join_request join_msg;
1436	struct dlm_query_join_packet packet;
1437	u32 join_resp;
1438
1439	mlog(0, "querying node %d\n", node);
1440
1441	memset(&join_msg, 0, sizeof(join_msg));
1442	join_msg.node_idx = dlm->node_num;
1443	join_msg.name_len = strlen(dlm->name);
1444	memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1445	join_msg.dlm_proto = dlm->dlm_locking_proto;
1446	join_msg.fs_proto = dlm->fs_locking_proto;
1447
1448	/* copy live node map to join message */
1449	byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1450
1451	status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1452				    sizeof(join_msg), node, &join_resp);
1453	if (status < 0 && status != -ENOPROTOOPT) {
1454		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1455		     "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1456		     node);
1457		goto bail;
1458	}
1459	dlm_query_join_wire_to_packet(join_resp, &packet);
1460
1461	/* -ENOPROTOOPT from the net code means the other side isn't
1462	    listening for our message type -- that's fine, it means
1463	    his dlm isn't up, so we can consider him a 'yes' but not
1464	    joined into the domain.  */
1465	if (status == -ENOPROTOOPT) {
1466		status = 0;
1467		*response = JOIN_OK_NO_MAP;
1468	} else if (packet.code == JOIN_DISALLOW ||
1469		   packet.code == JOIN_OK_NO_MAP) {
1470		*response = packet.code;
1471	} else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1472		mlog(ML_NOTICE,
1473		     "This node requested DLM locking protocol %u.%u and "
1474		     "filesystem locking protocol %u.%u.  At least one of "
1475		     "the protocol versions on node %d is not compatible, "
1476		     "disconnecting\n",
1477		     dlm->dlm_locking_proto.pv_major,
1478		     dlm->dlm_locking_proto.pv_minor,
1479		     dlm->fs_locking_proto.pv_major,
1480		     dlm->fs_locking_proto.pv_minor,
1481		     node);
1482		status = -EPROTO;
1483		*response = packet.code;
1484	} else if (packet.code == JOIN_OK) {
1485		*response = packet.code;
1486		/* Use the same locking protocol as the remote node */
1487		dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1488		dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1489		mlog(0,
1490		     "Node %d responds JOIN_OK with DLM locking protocol "
1491		     "%u.%u and fs locking protocol %u.%u\n",
1492		     node,
1493		     dlm->dlm_locking_proto.pv_major,
1494		     dlm->dlm_locking_proto.pv_minor,
1495		     dlm->fs_locking_proto.pv_major,
1496		     dlm->fs_locking_proto.pv_minor);
1497	} else {
1498		status = -EINVAL;
1499		mlog(ML_ERROR, "invalid response %d from node %u\n",
1500		     packet.code, node);
1501	}
1502
1503	mlog(0, "status %d, node %d response is %d\n", status, node,
1504	     *response);
1505
1506bail:
1507	return status;
1508}
1509
1510static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1511				    unsigned int node)
1512{
1513	int status;
1514	int ret;
1515	struct dlm_assert_joined assert_msg;
1516
1517	mlog(0, "Sending join assert to node %u\n", node);
1518
1519	memset(&assert_msg, 0, sizeof(assert_msg));
1520	assert_msg.node_idx = dlm->node_num;
1521	assert_msg.name_len = strlen(dlm->name);
1522	memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1523
1524	status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1525				    &assert_msg, sizeof(assert_msg), node,
1526				    &ret);
1527	if (status < 0)
1528		mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1529		     "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1530		     node);
1531	else
1532		status = ret;
1533
1534	return status;
1535}
1536
1537static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1538				  unsigned long *node_map)
1539{
1540	int status, node, live;
1541
1542	status = 0;
1543	node = -1;
1544	while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1545				     node + 1)) < O2NM_MAX_NODES) {
1546		if (node == dlm->node_num)
1547			continue;
1548
1549		do {
1550			/* It is very important that this message be
1551			 * received so we spin until either the node
1552			 * has died or it gets the message. */
1553			status = dlm_send_one_join_assert(dlm, node);
1554
1555			spin_lock(&dlm->spinlock);
1556			live = test_bit(node, dlm->live_nodes_map);
1557			spin_unlock(&dlm->spinlock);
1558
1559			if (status) {
1560				mlog(ML_ERROR, "Error return %d asserting "
1561				     "join on node %d\n", status, node);
1562
1563				/* give us some time between errors... */
1564				if (live)
1565					msleep(DLM_DOMAIN_BACKOFF_MS);
1566			}
1567		} while (status && live);
1568	}
1569}
1570
1571struct domain_join_ctxt {
1572	unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1573	unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1574};
1575
1576static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1577				   struct domain_join_ctxt *ctxt,
1578				   enum dlm_query_join_response_code response)
1579{
1580	int ret;
1581
1582	if (response == JOIN_DISALLOW) {
1583		mlog(0, "Latest response of disallow -- should restart\n");
1584		return 1;
1585	}
1586
1587	spin_lock(&dlm->spinlock);
1588	/* For now, we restart the process if the node maps have
1589	 * changed at all */
1590	ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1591		     sizeof(dlm->live_nodes_map));
1592	spin_unlock(&dlm->spinlock);
1593
1594	if (ret)
1595		mlog(0, "Node maps changed -- should restart\n");
1596
1597	return ret;
1598}
1599
1600static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1601{
1602	int status = 0, tmpstat, node;
1603	struct domain_join_ctxt *ctxt;
1604	enum dlm_query_join_response_code response = JOIN_DISALLOW;
1605
1606	mlog(0, "%p", dlm);
1607
1608	ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1609	if (!ctxt) {
1610		status = -ENOMEM;
1611		mlog_errno(status);
1612		goto bail;
1613	}
1614
1615	/* group sem locking should work for us here -- we're already
1616	 * registered for heartbeat events so filling this should be
1617	 * atomic wrt getting those handlers called. */
1618	o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1619
1620	spin_lock(&dlm->spinlock);
1621	memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1622
1623	__dlm_set_joining_node(dlm, dlm->node_num);
1624
1625	spin_unlock(&dlm->spinlock);
1626
1627	node = -1;
1628	while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1629				     node + 1)) < O2NM_MAX_NODES) {
1630		if (node == dlm->node_num)
1631			continue;
1632
1633		status = dlm_request_join(dlm, node, &response);
1634		if (status < 0) {
1635			mlog_errno(status);
1636			goto bail;
1637		}
1638
1639		/* Ok, either we got a response or the node doesn't have a
1640		 * dlm up. */
1641		if (response == JOIN_OK)
1642			set_bit(node, ctxt->yes_resp_map);
1643
1644		if (dlm_should_restart_join(dlm, ctxt, response)) {
1645			status = -EAGAIN;
1646			goto bail;
1647		}
1648	}
1649
1650	mlog(0, "Yay, done querying nodes!\n");
1651
1652	/* Yay, everyone agree's we can join the domain. My domain is
1653	 * comprised of all nodes who were put in the
1654	 * yes_resp_map. Copy that into our domain map and send a join
1655	 * assert message to clean up everyone elses state. */
1656	spin_lock(&dlm->spinlock);
1657	memcpy(dlm->domain_map, ctxt->yes_resp_map,
1658	       sizeof(ctxt->yes_resp_map));
1659	set_bit(dlm->node_num, dlm->domain_map);
1660	spin_unlock(&dlm->spinlock);
1661
1662	/* Support for global heartbeat and node info was added in 1.1 */
1663	if (dlm->dlm_locking_proto.pv_major > 1 ||
1664	    dlm->dlm_locking_proto.pv_minor > 0) {
1665		status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1666		if (status) {
1667			mlog_errno(status);
1668			goto bail;
1669		}
1670		status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1671		if (status) {
1672			mlog_errno(status);
1673			goto bail;
1674		}
1675	}
1676
1677	dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1678
1679	/* Joined state *must* be set before the joining node
1680	 * information, otherwise the query_join handler may read no
1681	 * current joiner but a state of NEW and tell joining nodes
1682	 * we're not in the domain. */
1683	spin_lock(&dlm_domain_lock);
1684	dlm->dlm_state = DLM_CTXT_JOINED;
1685	dlm->num_joins++;
1686	spin_unlock(&dlm_domain_lock);
1687
1688bail:
1689	spin_lock(&dlm->spinlock);
1690	__dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1691	if (!status) {
1692		printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1693		__dlm_print_nodes(dlm);
1694	}
1695	spin_unlock(&dlm->spinlock);
1696
1697	if (ctxt) {
1698		/* Do we need to send a cancel message to any nodes? */
1699		if (status < 0) {
1700			tmpstat = dlm_send_join_cancels(dlm,
1701							ctxt->yes_resp_map,
1702							sizeof(ctxt->yes_resp_map));
1703			if (tmpstat < 0)
1704				mlog_errno(tmpstat);
1705		}
1706		kfree(ctxt);
1707	}
1708
1709	mlog(0, "returning %d\n", status);
1710	return status;
1711}
1712
1713static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1714{
1715	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1716	o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1717	o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1718}
1719
1720static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1721{
1722	int status;
1723
1724	mlog(0, "registering handlers.\n");
1725
1726	o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1727			    dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1728	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1729	if (status)
1730		goto bail;
1731
1732	o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1733			    dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1734	status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1735	if (status)
1736		goto bail;
1737
1738	status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1739					sizeof(struct dlm_master_request),
1740					dlm_master_request_handler,
1741					dlm, NULL, &dlm->dlm_domain_handlers);
1742	if (status)
1743		goto bail;
1744
1745	status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1746					sizeof(struct dlm_assert_master),
1747					dlm_assert_master_handler,
1748					dlm, dlm_assert_master_post_handler,
1749					&dlm->dlm_domain_handlers);
1750	if (status)
1751		goto bail;
1752
1753	status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1754					sizeof(struct dlm_create_lock),
1755					dlm_create_lock_handler,
1756					dlm, NULL, &dlm->dlm_domain_handlers);
1757	if (status)
1758		goto bail;
1759
1760	status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1761					DLM_CONVERT_LOCK_MAX_LEN,
1762					dlm_convert_lock_handler,
1763					dlm, NULL, &dlm->dlm_domain_handlers);
1764	if (status)
1765		goto bail;
1766
1767	status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1768					DLM_UNLOCK_LOCK_MAX_LEN,
1769					dlm_unlock_lock_handler,
1770					dlm, NULL, &dlm->dlm_domain_handlers);
1771	if (status)
1772		goto bail;
1773
1774	status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1775					DLM_PROXY_AST_MAX_LEN,
1776					dlm_proxy_ast_handler,
1777					dlm, NULL, &dlm->dlm_domain_handlers);
1778	if (status)
1779		goto bail;
1780
1781	status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1782					sizeof(struct dlm_exit_domain),
1783					dlm_exit_domain_handler,
1784					dlm, NULL, &dlm->dlm_domain_handlers);
1785	if (status)
1786		goto bail;
1787
1788	status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1789					sizeof(struct dlm_deref_lockres),
1790					dlm_deref_lockres_handler,
1791					dlm, NULL, &dlm->dlm_domain_handlers);
1792	if (status)
1793		goto bail;
1794
1795	status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1796					sizeof(struct dlm_migrate_request),
1797					dlm_migrate_request_handler,
1798					dlm, NULL, &dlm->dlm_domain_handlers);
1799	if (status)
1800		goto bail;
1801
1802	status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1803					DLM_MIG_LOCKRES_MAX_LEN,
1804					dlm_mig_lockres_handler,
1805					dlm, NULL, &dlm->dlm_domain_handlers);
1806	if (status)
1807		goto bail;
1808
1809	status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1810					sizeof(struct dlm_master_requery),
1811					dlm_master_requery_handler,
1812					dlm, NULL, &dlm->dlm_domain_handlers);
1813	if (status)
1814		goto bail;
1815
1816	status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1817					sizeof(struct dlm_lock_request),
1818					dlm_request_all_locks_handler,
1819					dlm, NULL, &dlm->dlm_domain_handlers);
1820	if (status)
1821		goto bail;
1822
1823	status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1824					sizeof(struct dlm_reco_data_done),
1825					dlm_reco_data_done_handler,
1826					dlm, NULL, &dlm->dlm_domain_handlers);
1827	if (status)
1828		goto bail;
1829
1830	status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1831					sizeof(struct dlm_begin_reco),
1832					dlm_begin_reco_handler,
1833					dlm, NULL, &dlm->dlm_domain_handlers);
1834	if (status)
1835		goto bail;
1836
1837	status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1838					sizeof(struct dlm_finalize_reco),
1839					dlm_finalize_reco_handler,
1840					dlm, NULL, &dlm->dlm_domain_handlers);
1841	if (status)
1842		goto bail;
1843
1844	status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1845					sizeof(struct dlm_exit_domain),
1846					dlm_begin_exit_domain_handler,
1847					dlm, NULL, &dlm->dlm_domain_handlers);
1848	if (status)
1849		goto bail;
1850
1851bail:
1852	if (status)
1853		dlm_unregister_domain_handlers(dlm);
1854
1855	return status;
1856}
1857
1858static int dlm_join_domain(struct dlm_ctxt *dlm)
1859{
1860	int status;
1861	unsigned int backoff;
1862	unsigned int total_backoff = 0;
1863
1864	BUG_ON(!dlm);
1865
1866	mlog(0, "Join domain %s\n", dlm->name);
1867
1868	status = dlm_register_domain_handlers(dlm);
1869	if (status) {
1870		mlog_errno(status);
1871		goto bail;
1872	}
1873
1874	status = dlm_launch_thread(dlm);
1875	if (status < 0) {
1876		mlog_errno(status);
1877		goto bail;
1878	}
1879
1880	status = dlm_launch_recovery_thread(dlm);
1881	if (status < 0) {
1882		mlog_errno(status);
1883		goto bail;
1884	}
1885
1886	status = dlm_debug_init(dlm);
1887	if (status < 0) {
1888		mlog_errno(status);
1889		goto bail;
1890	}
1891
1892	dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1893	if (!dlm->dlm_worker) {
1894		status = -ENOMEM;
1895		mlog_errno(status);
1896		goto bail;
1897	}
1898
1899	do {
1900		status = dlm_try_to_join_domain(dlm);
1901
1902		/* If we're racing another node to the join, then we
1903		 * need to back off temporarily and let them
1904		 * complete. */
1905#define	DLM_JOIN_TIMEOUT_MSECS	90000
1906		if (status == -EAGAIN) {
1907			if (signal_pending(current)) {
1908				status = -ERESTARTSYS;
1909				goto bail;
1910			}
1911
1912			if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1913				status = -ERESTARTSYS;
1914				mlog(ML_NOTICE, "Timed out joining dlm domain "
1915				     "%s after %u msecs\n", dlm->name,
1916				     total_backoff);
1917				goto bail;
1918			}
1919
1920			/*
1921			 * <chip> After you!
1922			 * <dale> No, after you!
1923			 * <chip> I insist!
1924			 * <dale> But you first!
1925			 * ...
1926			 */
1927			backoff = (unsigned int)(jiffies & 0x3);
1928			backoff *= DLM_DOMAIN_BACKOFF_MS;
1929			total_backoff += backoff;
1930			mlog(0, "backoff %d\n", backoff);
1931			msleep(backoff);
1932		}
1933	} while (status == -EAGAIN);
1934
1935	if (status < 0) {
1936		mlog_errno(status);
1937		goto bail;
1938	}
1939
1940	status = 0;
1941bail:
1942	wake_up(&dlm_domain_events);
1943
1944	if (status) {
1945		dlm_unregister_domain_handlers(dlm);
1946		dlm_debug_shutdown(dlm);
1947		dlm_complete_thread(dlm);
1948		dlm_complete_recovery_thread(dlm);
1949		dlm_destroy_dlm_worker(dlm);
1950	}
1951
1952	return status;
1953}
1954
1955static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1956				u32 key)
1957{
1958	int i;
1959	int ret;
1960	struct dlm_ctxt *dlm = NULL;
1961
1962	dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1963	if (!dlm) {
1964		ret = -ENOMEM;
1965		mlog_errno(ret);
1966		goto leave;
1967	}
1968
1969	dlm->name = kstrdup(domain, GFP_KERNEL);
1970	if (dlm->name == NULL) {
1971		ret = -ENOMEM;
1972		mlog_errno(ret);
1973		goto leave;
1974	}
1975
1976	dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1977	if (!dlm->lockres_hash) {
1978		ret = -ENOMEM;
1979		mlog_errno(ret);
1980		goto leave;
1981	}
1982
1983	for (i = 0; i < DLM_HASH_BUCKETS; i++)
1984		INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1985
1986	dlm->master_hash = (struct hlist_head **)
1987				dlm_alloc_pagevec(DLM_HASH_PAGES);
1988	if (!dlm->master_hash) {
1989		ret = -ENOMEM;
1990		mlog_errno(ret);
1991		goto leave;
1992	}
1993
1994	for (i = 0; i < DLM_HASH_BUCKETS; i++)
1995		INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1996
1997	dlm->key = key;
1998	dlm->node_num = o2nm_this_node();
1999
2000	ret = dlm_create_debugfs_subroot(dlm);
2001	if (ret < 0)
2002		goto leave;
2003
2004	spin_lock_init(&dlm->spinlock);
2005	spin_lock_init(&dlm->master_lock);
2006	spin_lock_init(&dlm->ast_lock);
2007	spin_lock_init(&dlm->track_lock);
2008	INIT_LIST_HEAD(&dlm->list);
2009	INIT_LIST_HEAD(&dlm->dirty_list);
2010	INIT_LIST_HEAD(&dlm->reco.resources);
2011	INIT_LIST_HEAD(&dlm->reco.node_data);
2012	INIT_LIST_HEAD(&dlm->purge_list);
2013	INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2014	INIT_LIST_HEAD(&dlm->tracking_list);
2015	dlm->reco.state = 0;
2016
2017	INIT_LIST_HEAD(&dlm->pending_asts);
2018	INIT_LIST_HEAD(&dlm->pending_basts);
2019
2020	mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2021		  dlm->recovery_map, &(dlm->recovery_map[0]));
2022
2023	memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2024	memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2025	memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2026
2027	dlm->dlm_thread_task = NULL;
2028	dlm->dlm_reco_thread_task = NULL;
2029	dlm->dlm_worker = NULL;
2030	init_waitqueue_head(&dlm->dlm_thread_wq);
2031	init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2032	init_waitqueue_head(&dlm->reco.event);
2033	init_waitqueue_head(&dlm->ast_wq);
2034	init_waitqueue_head(&dlm->migration_wq);
2035	INIT_LIST_HEAD(&dlm->mle_hb_events);
2036
2037	dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2038	init_waitqueue_head(&dlm->dlm_join_events);
2039
2040	dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2041	dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2042
2043	atomic_set(&dlm->res_tot_count, 0);
2044	atomic_set(&dlm->res_cur_count, 0);
2045	for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2046		atomic_set(&dlm->mle_tot_count[i], 0);
2047		atomic_set(&dlm->mle_cur_count[i], 0);
2048	}
2049
2050	spin_lock_init(&dlm->work_lock);
2051	INIT_LIST_HEAD(&dlm->work_list);
2052	INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2053
2054	kref_init(&dlm->dlm_refs);
2055	dlm->dlm_state = DLM_CTXT_NEW;
2056
2057	INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2058
2059	mlog(0, "context init: refcount %u\n",
2060		  atomic_read(&dlm->dlm_refs.refcount));
2061
2062leave:
2063	if (ret < 0 && dlm) {
2064		if (dlm->master_hash)
2065			dlm_free_pagevec((void **)dlm->master_hash,
2066					DLM_HASH_PAGES);
2067
2068		if (dlm->lockres_hash)
2069			dlm_free_pagevec((void **)dlm->lockres_hash,
2070					DLM_HASH_PAGES);
2071
2072		kfree(dlm->name);
2073		kfree(dlm);
2074		dlm = NULL;
2075	}
2076	return dlm;
2077}
2078
2079/*
2080 * Compare a requested locking protocol version against the current one.
2081 *
2082 * If the major numbers are different, they are incompatible.
2083 * If the current minor is greater than the request, they are incompatible.
2084 * If the current minor is less than or equal to the request, they are
2085 * compatible, and the requester should run at the current minor version.
2086 */
2087static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2088				struct dlm_protocol_version *request)
2089{
2090	if (existing->pv_major != request->pv_major)
2091		return 1;
2092
2093	if (existing->pv_minor > request->pv_minor)
2094		return 1;
2095
2096	if (existing->pv_minor < request->pv_minor)
2097		request->pv_minor = existing->pv_minor;
2098
2099	return 0;
2100}
2101
2102/*
2103 * dlm_register_domain: one-time setup per "domain".
2104 *
2105 * The filesystem passes in the requested locking version via proto.
2106 * If registration was successful, proto will contain the negotiated
2107 * locking protocol.
2108 */
2109struct dlm_ctxt * dlm_register_domain(const char *domain,
2110			       u32 key,
2111			       struct dlm_protocol_version *fs_proto)
2112{
2113	int ret;
2114	struct dlm_ctxt *dlm = NULL;
2115	struct dlm_ctxt *new_ctxt = NULL;
2116
2117	if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2118		ret = -ENAMETOOLONG;
2119		mlog(ML_ERROR, "domain name length too long\n");
2120		goto leave;
2121	}
2122
2123	mlog(0, "register called for domain \"%s\"\n", domain);
2124
2125retry:
2126	dlm = NULL;
2127	if (signal_pending(current)) {
2128		ret = -ERESTARTSYS;
2129		mlog_errno(ret);
2130		goto leave;
2131	}
2132
2133	spin_lock(&dlm_domain_lock);
2134
2135	dlm = __dlm_lookup_domain(domain);
2136	if (dlm) {
2137		if (dlm->dlm_state != DLM_CTXT_JOINED) {
2138			spin_unlock(&dlm_domain_lock);
2139
2140			mlog(0, "This ctxt is not joined yet!\n");
2141			wait_event_interruptible(dlm_domain_events,
2142						 dlm_wait_on_domain_helper(
2143							 domain));
2144			goto retry;
2145		}
2146
2147		if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2148			spin_unlock(&dlm_domain_lock);
2149			mlog(ML_ERROR,
2150			     "Requested locking protocol version is not "
2151			     "compatible with already registered domain "
2152			     "\"%s\"\n", domain);
2153			ret = -EPROTO;
2154			goto leave;
2155		}
2156
2157		__dlm_get(dlm);
2158		dlm->num_joins++;
2159
2160		spin_unlock(&dlm_domain_lock);
2161
2162		ret = 0;
2163		goto leave;
2164	}
2165
2166	/* doesn't exist */
2167	if (!new_ctxt) {
2168		spin_unlock(&dlm_domain_lock);
2169
2170		new_ctxt = dlm_alloc_ctxt(domain, key);
2171		if (new_ctxt)
2172			goto retry;
2173
2174		ret = -ENOMEM;
2175		mlog_errno(ret);
2176		goto leave;
2177	}
2178
2179	/* a little variable switch-a-roo here... */
2180	dlm = new_ctxt;
2181	new_ctxt = NULL;
2182
2183	/* add the new domain */
2184	list_add_tail(&dlm->list, &dlm_domains);
2185	spin_unlock(&dlm_domain_lock);
2186
2187	/*
2188	 * Pass the locking protocol version into the join.  If the join
2189	 * succeeds, it will have the negotiated protocol set.
2190	 */
2191	dlm->dlm_locking_proto = dlm_protocol;
2192	dlm->fs_locking_proto = *fs_proto;
2193
2194	ret = dlm_join_domain(dlm);
2195	if (ret) {
2196		mlog_errno(ret);
2197		dlm_put(dlm);
2198		goto leave;
2199	}
2200
2201	/* Tell the caller what locking protocol we negotiated */
2202	*fs_proto = dlm->fs_locking_proto;
2203
2204	ret = 0;
2205leave:
2206	if (new_ctxt)
2207		dlm_free_ctxt_mem(new_ctxt);
2208
2209	if (ret < 0)
2210		dlm = ERR_PTR(ret);
2211
2212	return dlm;
2213}
2214EXPORT_SYMBOL_GPL(dlm_register_domain);
2215
2216static LIST_HEAD(dlm_join_handlers);
2217
2218static void dlm_unregister_net_handlers(void)
2219{
2220	o2net_unregister_handler_list(&dlm_join_handlers);
2221}
2222
2223static int dlm_register_net_handlers(void)
2224{
2225	int status = 0;
2226
2227	status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2228					sizeof(struct dlm_query_join_request),
2229					dlm_query_join_handler,
2230					NULL, NULL, &dlm_join_handlers);
2231	if (status)
2232		goto bail;
2233
2234	status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2235					sizeof(struct dlm_assert_joined),
2236					dlm_assert_joined_handler,
2237					NULL, NULL, &dlm_join_handlers);
2238	if (status)
2239		goto bail;
2240
2241	status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2242					sizeof(struct dlm_cancel_join),
2243					dlm_cancel_join_handler,
2244					NULL, NULL, &dlm_join_handlers);
2245	if (status)
2246		goto bail;
2247
2248	status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2249					sizeof(struct dlm_query_region),
2250					dlm_query_region_handler,
2251					NULL, NULL, &dlm_join_handlers);
2252
2253	if (status)
2254		goto bail;
2255
2256	status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2257					sizeof(struct dlm_query_nodeinfo),
2258					dlm_query_nodeinfo_handler,
2259					NULL, NULL, &dlm_join_handlers);
2260bail:
2261	if (status < 0)
2262		dlm_unregister_net_handlers();
2263
2264	return status;
2265}
2266
2267/* Domain eviction callback handling.
2268 *
2269 * The file system requires notification of node death *before* the
2270 * dlm completes it's recovery work, otherwise it may be able to
2271 * acquire locks on resources requiring recovery. Since the dlm can
2272 * evict a node from it's domain *before* heartbeat fires, a similar
2273 * mechanism is required. */
2274
2275/* Eviction is not expected to happen often, so a per-domain lock is
2276 * not necessary. Eviction callbacks are allowed to sleep for short
2277 * periods of time. */
2278static DECLARE_RWSEM(dlm_callback_sem);
2279
2280void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2281					int node_num)
2282{
2283	struct dlm_eviction_cb *cb;
2284
2285	down_read(&dlm_callback_sem);
2286	list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
2287		cb->ec_func(node_num, cb->ec_data);
2288	}
2289	up_read(&dlm_callback_sem);
2290}
2291
2292void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2293			   dlm_eviction_func *f,
2294			   void *data)
2295{
2296	INIT_LIST_HEAD(&cb->ec_item);
2297	cb->ec_func = f;
2298	cb->ec_data = data;
2299}
2300EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2301
2302void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2303			      struct dlm_eviction_cb *cb)
2304{
2305	down_write(&dlm_callback_sem);
2306	list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2307	up_write(&dlm_callback_sem);
2308}
2309EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2310
2311void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2312{
2313	down_write(&dlm_callback_sem);
2314	list_del_init(&cb->ec_item);
2315	up_write(&dlm_callback_sem);
2316}
2317EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2318
2319static int __init dlm_init(void)
2320{
2321	int status;
2322
2323	status = dlm_init_mle_cache();
2324	if (status) {
2325		mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2326		goto error;
2327	}
2328
2329	status = dlm_init_master_caches();
2330	if (status) {
2331		mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2332		     "o2dlm_lockname slabcaches\n");
2333		goto error;
2334	}
2335
2336	status = dlm_init_lock_cache();
2337	if (status) {
2338		mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2339		goto error;
2340	}
2341
2342	status = dlm_register_net_handlers();
2343	if (status) {
2344		mlog(ML_ERROR, "Unable to register network handlers\n");
2345		goto error;
2346	}
2347
2348	status = dlm_create_debugfs_root();
2349	if (status)
2350		goto error;
2351
2352	return 0;
2353error:
2354	dlm_unregister_net_handlers();
2355	dlm_destroy_lock_cache();
2356	dlm_destroy_master_caches();
2357	dlm_destroy_mle_cache();
2358	return -1;
2359}
2360
2361static void __exit dlm_exit (void)
2362{
2363	dlm_destroy_debugfs_root();
2364	dlm_unregister_net_handlers();
2365	dlm_destroy_lock_cache();
2366	dlm_destroy_master_caches();
2367	dlm_destroy_mle_cache();
2368}
2369
2370MODULE_AUTHOR("Oracle");
2371MODULE_LICENSE("GPL");
2372MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
2373
2374module_init(dlm_init);
2375module_exit(dlm_exit);
2376