?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Merge branch 'fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/evalenti/linux...
[linux/fpc-iii.git] / fs / ocfs2 / dlm / dlmdomain.c
blob12e064b8be9aff94cfe89d8bf4b60cc8eb9d6204
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 struct qstr *q;
176
177 assert_spin_locked(&dlm->spinlock);
178
179 q = &res->lockname;
180 bucket = dlm_lockres_hash(dlm, q->hash);
181
182 /* get a reference for our hashtable */
183 dlm_lockres_get(res);
184
185 hlist_add_head(&res->hash_node, bucket);
186
187 mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
188 res->lockname.name);
189 }
190
191 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
192 const char *name,
193 unsigned int len,
194 unsigned int hash)
195 {
196 struct hlist_head *bucket;
197 struct dlm_lock_resource *res;
198
199 mlog(0, "%.*s\n", len, name);
200
201 assert_spin_locked(&dlm->spinlock);
202
203 bucket = dlm_lockres_hash(dlm, hash);
204
205 hlist_for_each_entry(res, bucket, hash_node) {
206 if (res->lockname.name[0] != name[0])
207 continue;
208 if (unlikely(res->lockname.len != len))
209 continue;
210 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
211 continue;
212 dlm_lockres_get(res);
213 return res;
214 }
215 return NULL;
216 }
217
218 /* intended to be called by functions which do not care about lock
219 * resources which are being purged (most net _handler functions).
220 * this will return NULL for any lock resource which is found but
221 * currently in the process of dropping its mastery reference.
222 * use __dlm_lookup_lockres_full when you need the lock resource
223 * regardless (e.g. dlm_get_lock_resource) */
224 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
225 const char *name,
226 unsigned int len,
227 unsigned int hash)
228 {
229 struct dlm_lock_resource *res = NULL;
230
231 mlog(0, "%.*s\n", len, name);
232
233 assert_spin_locked(&dlm->spinlock);
234
235 res = __dlm_lookup_lockres_full(dlm, name, len, hash);
236 if (res) {
237 spin_lock(&res->spinlock);
238 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
239 spin_unlock(&res->spinlock);
240 dlm_lockres_put(res);
241 return NULL;
242 }
243 spin_unlock(&res->spinlock);
244 }
245
246 return res;
247 }
248
249 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
250 const char *name,
251 unsigned int len)
252 {
253 struct dlm_lock_resource *res;
254 unsigned int hash = dlm_lockid_hash(name, len);
255
256 spin_lock(&dlm->spinlock);
257 res = __dlm_lookup_lockres(dlm, name, len, hash);
258 spin_unlock(&dlm->spinlock);
259 return res;
260 }
261
262 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
263 {
264 struct dlm_ctxt *tmp;
265
266 assert_spin_locked(&dlm_domain_lock);
267
268 /* tmp->name here is always NULL terminated,
269 * but domain may not be! */
270 list_for_each_entry(tmp, &dlm_domains, list) {
271 if (strlen(tmp->name) == len &&
272 memcmp(tmp->name, domain, len)==0)
273 return tmp;
274 }
275
276 return NULL;
277 }
278
279 /* For null terminated domain strings ONLY */
280 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
281 {
282 assert_spin_locked(&dlm_domain_lock);
283
284 return __dlm_lookup_domain_full(domain, strlen(domain));
285 }
286
287
288 /* returns true on one of two conditions:
289 * 1) the domain does not exist
290 * 2) the domain exists and it's state is "joined" */
291 static int dlm_wait_on_domain_helper(const char *domain)
292 {
293 int ret = 0;
294 struct dlm_ctxt *tmp = NULL;
295
296 spin_lock(&dlm_domain_lock);
297
298 tmp = __dlm_lookup_domain(domain);
299 if (!tmp)
300 ret = 1;
301 else if (tmp->dlm_state == DLM_CTXT_JOINED)
302 ret = 1;
303
304 spin_unlock(&dlm_domain_lock);
305 return ret;
306 }
307
308 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
309 {
310 dlm_destroy_debugfs_subroot(dlm);
311
312 if (dlm->lockres_hash)
313 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
314
315 if (dlm->master_hash)
316 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
317
318 kfree(dlm->name);
319 kfree(dlm);
320 }
321
322 /* A little strange - this function will be called while holding
323 * dlm_domain_lock and is expected to be holding it on the way out. We
324 * will however drop and reacquire it multiple times */
325 static void dlm_ctxt_release(struct kref *kref)
326 {
327 struct dlm_ctxt *dlm;
328
329 dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
330
331 BUG_ON(dlm->num_joins);
332 BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
333
334 /* we may still be in the list if we hit an error during join. */
335 list_del_init(&dlm->list);
336
337 spin_unlock(&dlm_domain_lock);
338
339 mlog(0, "freeing memory from domain %s\n", dlm->name);
340
341 wake_up(&dlm_domain_events);
342
343 dlm_free_ctxt_mem(dlm);
344
345 spin_lock(&dlm_domain_lock);
346 }
347
348 void dlm_put(struct dlm_ctxt *dlm)
349 {
350 spin_lock(&dlm_domain_lock);
351 kref_put(&dlm->dlm_refs, dlm_ctxt_release);
352 spin_unlock(&dlm_domain_lock);
353 }
354
355 static void __dlm_get(struct dlm_ctxt *dlm)
356 {
357 kref_get(&dlm->dlm_refs);
358 }
359
360 /* given a questionable reference to a dlm object, gets a reference if
361 * it can find it in the list, otherwise returns NULL in which case
362 * you shouldn't trust your pointer. */
363 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
364 {
365 struct dlm_ctxt *target;
366 struct dlm_ctxt *ret = NULL;
367
368 spin_lock(&dlm_domain_lock);
369
370 list_for_each_entry(target, &dlm_domains, list) {
371 if (target == dlm) {
372 __dlm_get(target);
373 ret = target;
374 break;
375 }
376 }
377
378 spin_unlock(&dlm_domain_lock);
379
380 return ret;
381 }
382
383 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
384 {
385 int ret;
386
387 spin_lock(&dlm_domain_lock);
388 ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
389 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
390 spin_unlock(&dlm_domain_lock);
391
392 return ret;
393 }
394
395 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
396 {
397 if (dlm->dlm_worker) {
398 flush_workqueue(dlm->dlm_worker);
399 destroy_workqueue(dlm->dlm_worker);
400 dlm->dlm_worker = NULL;
401 }
402 }
403
404 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
405 {
406 dlm_unregister_domain_handlers(dlm);
407 dlm_debug_shutdown(dlm);
408 dlm_complete_thread(dlm);
409 dlm_complete_recovery_thread(dlm);
410 dlm_destroy_dlm_worker(dlm);
411
412 /* We've left the domain. Now we can take ourselves out of the
413 * list and allow the kref stuff to help us free the
414 * memory. */
415 spin_lock(&dlm_domain_lock);
416 list_del_init(&dlm->list);
417 spin_unlock(&dlm_domain_lock);
418
419 /* Wake up anyone waiting for us to remove this domain */
420 wake_up(&dlm_domain_events);
421 }
422
423 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
424 {
425 int i, num, n, ret = 0;
426 struct dlm_lock_resource *res;
427 struct hlist_node *iter;
428 struct hlist_head *bucket;
429 int dropped;
430
431 mlog(0, "Migrating locks from domain %s\n", dlm->name);
432
433 num = 0;
434 spin_lock(&dlm->spinlock);
435 for (i = 0; i < DLM_HASH_BUCKETS; i++) {
436 redo_bucket:
437 n = 0;
438 bucket = dlm_lockres_hash(dlm, i);
439 iter = bucket->first;
440 while (iter) {
441 n++;
442 res = hlist_entry(iter, struct dlm_lock_resource,
443 hash_node);
444 dlm_lockres_get(res);
445 /* migrate, if necessary. this will drop the dlm
446 * spinlock and retake it if it does migration. */
447 dropped = dlm_empty_lockres(dlm, res);
448
449 spin_lock(&res->spinlock);
450 if (dropped)
451 __dlm_lockres_calc_usage(dlm, res);
452 else
453 iter = res->hash_node.next;
454 spin_unlock(&res->spinlock);
455
456 dlm_lockres_put(res);
457
458 if (dropped) {
459 cond_resched_lock(&dlm->spinlock);
460 goto redo_bucket;
461 }
462 }
463 cond_resched_lock(&dlm->spinlock);
464 num += n;
465 }
466 spin_unlock(&dlm->spinlock);
467 wake_up(&dlm->dlm_thread_wq);
468
469 /* let the dlm thread take care of purging, keep scanning until
470 * nothing remains in the hash */
471 if (num) {
472 mlog(0, "%s: %d lock resources in hash last pass\n",
473 dlm->name, num);
474 ret = -EAGAIN;
475 }
476 mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
477 return ret;
478 }
479
480 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
481 {
482 int ret;
483
484 spin_lock(&dlm->spinlock);
485 ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
486 spin_unlock(&dlm->spinlock);
487
488 return ret;
489 }
490
491 static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
492 void *data, void **ret_data)
493 {
494 struct dlm_ctxt *dlm = data;
495 unsigned int node;
496 struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
497
498 if (!dlm_grab(dlm))
499 return 0;
500
501 node = exit_msg->node_idx;
502 mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
503
504 spin_lock(&dlm->spinlock);
505 set_bit(node, dlm->exit_domain_map);
506 spin_unlock(&dlm->spinlock);
507
508 dlm_put(dlm);
509
510 return 0;
511 }
512
513 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
514 {
515 /* Yikes, a double spinlock! I need domain_lock for the dlm
516 * state and the dlm spinlock for join state... Sorry! */
517 again:
518 spin_lock(&dlm_domain_lock);
519 spin_lock(&dlm->spinlock);
520
521 if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
522 mlog(0, "Node %d is joining, we wait on it.\n",
523 dlm->joining_node);
524 spin_unlock(&dlm->spinlock);
525 spin_unlock(&dlm_domain_lock);
526
527 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
528 goto again;
529 }
530
531 dlm->dlm_state = DLM_CTXT_LEAVING;
532 spin_unlock(&dlm->spinlock);
533 spin_unlock(&dlm_domain_lock);
534 }
535
536 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
537 {
538 int node = -1, num = 0;
539
540 assert_spin_locked(&dlm->spinlock);
541
542 printk("( ");
543 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
544 node + 1)) < O2NM_MAX_NODES) {
545 printk("%d ", node);
546 ++num;
547 }
548 printk(") %u nodes\n", num);
549 }
550
551 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
552 void **ret_data)
553 {
554 struct dlm_ctxt *dlm = data;
555 unsigned int node;
556 struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
557
558 mlog(0, "%p %u %p", msg, len, data);
559
560 if (!dlm_grab(dlm))
561 return 0;
562
563 node = exit_msg->node_idx;
564
565 spin_lock(&dlm->spinlock);
566 clear_bit(node, dlm->domain_map);
567 clear_bit(node, dlm->exit_domain_map);
568 printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
569 __dlm_print_nodes(dlm);
570
571 /* notify anything attached to the heartbeat events */
572 dlm_hb_event_notify_attached(dlm, node, 0);
573
574 spin_unlock(&dlm->spinlock);
575
576 dlm_put(dlm);
577
578 return 0;
579 }
580
581 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
582 unsigned int node)
583 {
584 int status;
585 struct dlm_exit_domain leave_msg;
586
587 mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
588 msg_type, node);
589
590 memset(&leave_msg, 0, sizeof(leave_msg));
591 leave_msg.node_idx = dlm->node_num;
592
593 status = o2net_send_message(msg_type, dlm->key, &leave_msg,
594 sizeof(leave_msg), node, NULL);
595 if (status < 0)
596 mlog(ML_ERROR, "Error %d sending domain exit message %u "
597 "to node %u on domain %s\n", status, msg_type, node,
598 dlm->name);
599
600 return status;
601 }
602
603 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
604 {
605 int node = -1;
606
607 /* Support for begin exit domain was added in 1.2 */
608 if (dlm->dlm_locking_proto.pv_major == 1 &&
609 dlm->dlm_locking_proto.pv_minor < 2)
610 return;
611
612 /*
613 * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
614 * informational. Meaning if a node does not receive the message,
615 * so be it.
616 */
617 spin_lock(&dlm->spinlock);
618 while (1) {
619 node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
620 if (node >= O2NM_MAX_NODES)
621 break;
622 if (node == dlm->node_num)
623 continue;
624
625 spin_unlock(&dlm->spinlock);
626 dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
627 spin_lock(&dlm->spinlock);
628 }
629 spin_unlock(&dlm->spinlock);
630 }
631
632 static void dlm_leave_domain(struct dlm_ctxt *dlm)
633 {
634 int node, clear_node, status;
635
636 /* At this point we've migrated away all our locks and won't
637 * accept mastership of new ones. The dlm is responsible for
638 * almost nothing now. We make sure not to confuse any joining
639 * nodes and then commence shutdown procedure. */
640
641 spin_lock(&dlm->spinlock);
642 /* Clear ourselves from the domain map */
643 clear_bit(dlm->node_num, dlm->domain_map);
644 while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
645 0)) < O2NM_MAX_NODES) {
646 /* Drop the dlm spinlock. This is safe wrt the domain_map.
647 * -nodes cannot be added now as the
648 * query_join_handlers knows to respond with OK_NO_MAP
649 * -we catch the right network errors if a node is
650 * removed from the map while we're sending him the
651 * exit message. */
652 spin_unlock(&dlm->spinlock);
653
654 clear_node = 1;
655
656 status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
657 node);
658 if (status < 0 &&
659 status != -ENOPROTOOPT &&
660 status != -ENOTCONN) {
661 mlog(ML_NOTICE, "Error %d sending domain exit message "
662 "to node %d\n", status, node);
663
664 /* Not sure what to do here but lets sleep for
665 * a bit in case this was a transient
666 * error... */
667 msleep(DLM_DOMAIN_BACKOFF_MS);
668 clear_node = 0;
669 }
670
671 spin_lock(&dlm->spinlock);
672 /* If we're not clearing the node bit then we intend
673 * to loop back around to try again. */
674 if (clear_node)
675 clear_bit(node, dlm->domain_map);
676 }
677 spin_unlock(&dlm->spinlock);
678 }
679
680 int dlm_shutting_down(struct dlm_ctxt *dlm)
681 {
682 int ret = 0;
683
684 spin_lock(&dlm_domain_lock);
685
686 if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
687 ret = 1;
688
689 spin_unlock(&dlm_domain_lock);
690
691 return ret;
692 }
693
694 void dlm_unregister_domain(struct dlm_ctxt *dlm)
695 {
696 int leave = 0;
697 struct dlm_lock_resource *res;
698
699 spin_lock(&dlm_domain_lock);
700 BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
701 BUG_ON(!dlm->num_joins);
702
703 dlm->num_joins--;
704 if (!dlm->num_joins) {
705 /* We mark it "in shutdown" now so new register
706 * requests wait until we've completely left the
707 * domain. Don't use DLM_CTXT_LEAVING yet as we still
708 * want new domain joins to communicate with us at
709 * least until we've completed migration of our
710 * resources. */
711 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
712 leave = 1;
713 }
714 spin_unlock(&dlm_domain_lock);
715
716 if (leave) {
717 mlog(0, "shutting down domain %s\n", dlm->name);
718 dlm_begin_exit_domain(dlm);
719
720 /* We changed dlm state, notify the thread */
721 dlm_kick_thread(dlm, NULL);
722
723 while (dlm_migrate_all_locks(dlm)) {
724 /* Give dlm_thread time to purge the lockres' */
725 msleep(500);
726 mlog(0, "%s: more migration to do\n", dlm->name);
727 }
728
729 /* This list should be empty. If not, print remaining lockres */
730 if (!list_empty(&dlm->tracking_list)) {
731 mlog(ML_ERROR, "Following lockres' are still on the "
732 "tracking list:\n");
733 list_for_each_entry(res, &dlm->tracking_list, tracking)
734 dlm_print_one_lock_resource(res);
735 }
736
737 dlm_mark_domain_leaving(dlm);
738 dlm_leave_domain(dlm);
739 printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
740 dlm_force_free_mles(dlm);
741 dlm_complete_dlm_shutdown(dlm);
742 }
743 dlm_put(dlm);
744 }
745 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
746
747 static int dlm_query_join_proto_check(char *proto_type, int node,
748 struct dlm_protocol_version *ours,
749 struct dlm_protocol_version *request)
750 {
751 int rc;
752 struct dlm_protocol_version proto = *request;
753
754 if (!dlm_protocol_compare(ours, &proto)) {
755 mlog(0,
756 "node %u wanted to join with %s locking protocol "
757 "%u.%u, we respond with %u.%u\n",
758 node, proto_type,
759 request->pv_major,
760 request->pv_minor,
761 proto.pv_major, proto.pv_minor);
762 request->pv_minor = proto.pv_minor;
763 rc = 0;
764 } else {
765 mlog(ML_NOTICE,
766 "Node %u wanted to join with %s locking "
767 "protocol %u.%u, but we have %u.%u, disallowing\n",
768 node, proto_type,
769 request->pv_major,
770 request->pv_minor,
771 ours->pv_major,
772 ours->pv_minor);
773 rc = 1;
774 }
775
776 return rc;
777 }
778
779 /*
780 * struct dlm_query_join_packet is made up of four one-byte fields. They
781 * are effectively in big-endian order already. However, little-endian
782 * machines swap them before putting the packet on the wire (because
783 * query_join's response is a status, and that status is treated as a u32
784 * on the wire). Thus, a big-endian and little-endian machines will treat
785 * this structure differently.
786 *
787 * The solution is to have little-endian machines swap the structure when
788 * converting from the structure to the u32 representation. This will
789 * result in the structure having the correct format on the wire no matter
790 * the host endian format.
791 */
792 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
793 u32 *wire)
794 {
795 union dlm_query_join_response response;
796
797 response.packet = *packet;
798 *wire = be32_to_cpu(response.intval);
799 }
800
801 static void dlm_query_join_wire_to_packet(u32 wire,
802 struct dlm_query_join_packet *packet)
803 {
804 union dlm_query_join_response response;
805
806 response.intval = cpu_to_be32(wire);
807 *packet = response.packet;
808 }
809
810 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
811 void **ret_data)
812 {
813 struct dlm_query_join_request *query;
814 struct dlm_query_join_packet packet = {
815 .code = JOIN_DISALLOW,
816 };
817 struct dlm_ctxt *dlm = NULL;
818 u32 response;
819 u8 nodenum;
820
821 query = (struct dlm_query_join_request *) msg->buf;
822
823 mlog(0, "node %u wants to join domain %s\n", query->node_idx,
824 query->domain);
825
826 /*
827 * If heartbeat doesn't consider the node live, tell it
828 * to back off and try again. This gives heartbeat a chance
829 * to catch up.
830 */
831 if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
832 mlog(0, "node %u is not in our live map yet\n",
833 query->node_idx);
834
835 packet.code = JOIN_DISALLOW;
836 goto respond;
837 }
838
839 packet.code = JOIN_OK_NO_MAP;
840
841 spin_lock(&dlm_domain_lock);
842 dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
843 if (!dlm)
844 goto unlock_respond;
845
846 /*
847 * There is a small window where the joining node may not see the
848 * node(s) that just left but still part of the cluster. DISALLOW
849 * join request if joining node has different node map.
850 */
851 nodenum=0;
852 while (nodenum < O2NM_MAX_NODES) {
853 if (test_bit(nodenum, dlm->domain_map)) {
854 if (!byte_test_bit(nodenum, query->node_map)) {
855 mlog(0, "disallow join as node %u does not "
856 "have node %u in its nodemap\n",
857 query->node_idx, nodenum);
858 packet.code = JOIN_DISALLOW;
859 goto unlock_respond;
860 }
861 }
862 nodenum++;
863 }
864
865 /* Once the dlm ctxt is marked as leaving then we don't want
866 * to be put in someone's domain map.
867 * Also, explicitly disallow joining at certain troublesome
868 * times (ie. during recovery). */
869 if (dlm->dlm_state != DLM_CTXT_LEAVING) {
870 int bit = query->node_idx;
871 spin_lock(&dlm->spinlock);
872
873 if (dlm->dlm_state == DLM_CTXT_NEW &&
874 dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
875 /*If this is a brand new context and we
876 * haven't started our join process yet, then
877 * the other node won the race. */
878 packet.code = JOIN_OK_NO_MAP;
879 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
880 /* Disallow parallel joins. */
881 packet.code = JOIN_DISALLOW;
882 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
883 mlog(0, "node %u trying to join, but recovery "
884 "is ongoing.\n", bit);
885 packet.code = JOIN_DISALLOW;
886 } else if (test_bit(bit, dlm->recovery_map)) {
887 mlog(0, "node %u trying to join, but it "
888 "still needs recovery.\n", bit);
889 packet.code = JOIN_DISALLOW;
890 } else if (test_bit(bit, dlm->domain_map)) {
891 mlog(0, "node %u trying to join, but it "
892 "is still in the domain! needs recovery?\n",
893 bit);
894 packet.code = JOIN_DISALLOW;
895 } else {
896 /* Alright we're fully a part of this domain
897 * so we keep some state as to who's joining
898 * and indicate to him that needs to be fixed
899 * up. */
900
901 /* Make sure we speak compatible locking protocols. */
902 if (dlm_query_join_proto_check("DLM", bit,
903 &dlm->dlm_locking_proto,
904 &query->dlm_proto)) {
905 packet.code = JOIN_PROTOCOL_MISMATCH;
906 } else if (dlm_query_join_proto_check("fs", bit,
907 &dlm->fs_locking_proto,
908 &query->fs_proto)) {
909 packet.code = JOIN_PROTOCOL_MISMATCH;
910 } else {
911 packet.dlm_minor = query->dlm_proto.pv_minor;
912 packet.fs_minor = query->fs_proto.pv_minor;
913 packet.code = JOIN_OK;
914 __dlm_set_joining_node(dlm, query->node_idx);
915 }
916 }
917
918 spin_unlock(&dlm->spinlock);
919 }
920 unlock_respond:
921 spin_unlock(&dlm_domain_lock);
922
923 respond:
924 mlog(0, "We respond with %u\n", packet.code);
925
926 dlm_query_join_packet_to_wire(&packet, &response);
927 return response;
928 }
929
930 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
931 void **ret_data)
932 {
933 struct dlm_assert_joined *assert;
934 struct dlm_ctxt *dlm = NULL;
935
936 assert = (struct dlm_assert_joined *) msg->buf;
937
938 mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
939 assert->domain);
940
941 spin_lock(&dlm_domain_lock);
942 dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
943 /* XXX should we consider no dlm ctxt an error? */
944 if (dlm) {
945 spin_lock(&dlm->spinlock);
946
947 /* Alright, this node has officially joined our
948 * domain. Set him in the map and clean up our
949 * leftover join state. */
950 BUG_ON(dlm->joining_node != assert->node_idx);
951
952 if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
953 mlog(0, "dlm recovery is ongoing, disallow join\n");
954 spin_unlock(&dlm->spinlock);
955 spin_unlock(&dlm_domain_lock);
956 return -EAGAIN;
957 }
958
959 set_bit(assert->node_idx, dlm->domain_map);
960 clear_bit(assert->node_idx, dlm->exit_domain_map);
961 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
962
963 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
964 assert->node_idx, dlm->name);
965 __dlm_print_nodes(dlm);
966
967 /* notify anything attached to the heartbeat events */
968 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
969
970 spin_unlock(&dlm->spinlock);
971 }
972 spin_unlock(&dlm_domain_lock);
973
974 return 0;
975 }
976
977 static int dlm_match_regions(struct dlm_ctxt *dlm,
978 struct dlm_query_region *qr,
979 char *local, int locallen)
980 {
981 char *remote = qr->qr_regions;
982 char *l, *r;
983 int localnr, i, j, foundit;
984 int status = 0;
985
986 if (!o2hb_global_heartbeat_active()) {
987 if (qr->qr_numregions) {
988 mlog(ML_ERROR, "Domain %s: Joining node %d has global "
989 "heartbeat enabled but local node %d does not\n",
990 qr->qr_domain, qr->qr_node, dlm->node_num);
991 status = -EINVAL;
992 }
993 goto bail;
994 }
995
996 if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
997 mlog(ML_ERROR, "Domain %s: Local node %d has global "
998 "heartbeat enabled but joining node %d does not\n",
999 qr->qr_domain, dlm->node_num, qr->qr_node);
1000 status = -EINVAL;
1001 goto bail;
1002 }
1003
1004 r = remote;
1005 for (i = 0; i < qr->qr_numregions; ++i) {
1006 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1007 r += O2HB_MAX_REGION_NAME_LEN;
1008 }
1009
1010 localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1011 localnr = o2hb_get_all_regions(local, (u8)localnr);
1012
1013 /* compare local regions with remote */
1014 l = local;
1015 for (i = 0; i < localnr; ++i) {
1016 foundit = 0;
1017 r = remote;
1018 for (j = 0; j <= qr->qr_numregions; ++j) {
1019 if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1020 foundit = 1;
1021 break;
1022 }
1023 r += O2HB_MAX_REGION_NAME_LEN;
1024 }
1025 if (!foundit) {
1026 status = -EINVAL;
1027 mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1028 "in local node %d but not in joining node %d\n",
1029 qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1030 dlm->node_num, qr->qr_node);
1031 goto bail;
1032 }
1033 l += O2HB_MAX_REGION_NAME_LEN;
1034 }
1035
1036 /* compare remote with local regions */
1037 r = remote;
1038 for (i = 0; i < qr->qr_numregions; ++i) {
1039 foundit = 0;
1040 l = local;
1041 for (j = 0; j < localnr; ++j) {
1042 if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1043 foundit = 1;
1044 break;
1045 }
1046 l += O2HB_MAX_REGION_NAME_LEN;
1047 }
1048 if (!foundit) {
1049 status = -EINVAL;
1050 mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1051 "in joining node %d but not in local node %d\n",
1052 qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1053 qr->qr_node, dlm->node_num);
1054 goto bail;
1055 }
1056 r += O2HB_MAX_REGION_NAME_LEN;
1057 }
1058
1059 bail:
1060 return status;
1061 }
1062
1063 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1064 {
1065 struct dlm_query_region *qr = NULL;
1066 int status, ret = 0, i;
1067 char *p;
1068
1069 if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1070 goto bail;
1071
1072 qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1073 if (!qr) {
1074 ret = -ENOMEM;
1075 mlog_errno(ret);
1076 goto bail;
1077 }
1078
1079 qr->qr_node = dlm->node_num;
1080 qr->qr_namelen = strlen(dlm->name);
1081 memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1082 /* if local hb, the numregions will be zero */
1083 if (o2hb_global_heartbeat_active())
1084 qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1085 O2NM_MAX_REGIONS);
1086
1087 p = qr->qr_regions;
1088 for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1089 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1090
1091 i = -1;
1092 while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1093 i + 1)) < O2NM_MAX_NODES) {
1094 if (i == dlm->node_num)
1095 continue;
1096
1097 mlog(0, "Sending regions to node %d\n", i);
1098
1099 ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1100 sizeof(struct dlm_query_region),
1101 i, &status);
1102 if (ret >= 0)
1103 ret = status;
1104 if (ret) {
1105 mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1106 ret, i);
1107 break;
1108 }
1109 }
1110
1111 bail:
1112 kfree(qr);
1113 return ret;
1114 }
1115
1116 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1117 void *data, void **ret_data)
1118 {
1119 struct dlm_query_region *qr;
1120 struct dlm_ctxt *dlm = NULL;
1121 char *local = NULL;
1122 int status = 0;
1123
1124 qr = (struct dlm_query_region *) msg->buf;
1125
1126 mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1127 qr->qr_domain);
1128
1129 /* buffer used in dlm_mast_regions() */
1130 local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1131 if (!local)
1132 return -ENOMEM;
1133
1134 status = -EINVAL;
1135
1136 spin_lock(&dlm_domain_lock);
1137 dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1138 if (!dlm) {
1139 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1140 "before join domain\n", qr->qr_node, qr->qr_domain);
1141 goto out_domain_lock;
1142 }
1143
1144 spin_lock(&dlm->spinlock);
1145 if (dlm->joining_node != qr->qr_node) {
1146 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1147 "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1148 dlm->joining_node);
1149 goto out_dlm_lock;
1150 }
1151
1152 /* Support for global heartbeat was added in 1.1 */
1153 if (dlm->dlm_locking_proto.pv_major == 1 &&
1154 dlm->dlm_locking_proto.pv_minor == 0) {
1155 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1156 "but active dlm protocol is %d.%d\n", qr->qr_node,
1157 qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1158 dlm->dlm_locking_proto.pv_minor);
1159 goto out_dlm_lock;
1160 }
1161
1162 status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1163
1164 out_dlm_lock:
1165 spin_unlock(&dlm->spinlock);
1166
1167 out_domain_lock:
1168 spin_unlock(&dlm_domain_lock);
1169
1170 kfree(local);
1171
1172 return status;
1173 }
1174
1175 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1176 {
1177 struct o2nm_node *local;
1178 struct dlm_node_info *remote;
1179 int i, j;
1180 int status = 0;
1181
1182 for (j = 0; j < qn->qn_numnodes; ++j)
1183 mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1184 &(qn->qn_nodes[j].ni_ipv4_address),
1185 ntohs(qn->qn_nodes[j].ni_ipv4_port));
1186
1187 for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1188 local = o2nm_get_node_by_num(i);
1189 remote = NULL;
1190 for (j = 0; j < qn->qn_numnodes; ++j) {
1191 if (qn->qn_nodes[j].ni_nodenum == i) {
1192 remote = &(qn->qn_nodes[j]);
1193 break;
1194 }
1195 }
1196
1197 if (!local && !remote)
1198 continue;
1199
1200 if ((local && !remote) || (!local && remote))
1201 status = -EINVAL;
1202
1203 if (!status &&
1204 ((remote->ni_nodenum != local->nd_num) ||
1205 (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1206 (remote->ni_ipv4_address != local->nd_ipv4_address)))
1207 status = -EINVAL;
1208
1209 if (status) {
1210 if (remote && !local)
1211 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1212 "registered in joining node %d but not in "
1213 "local node %d\n", qn->qn_domain,
1214 remote->ni_nodenum,
1215 &(remote->ni_ipv4_address),
1216 ntohs(remote->ni_ipv4_port),
1217 qn->qn_nodenum, dlm->node_num);
1218 if (local && !remote)
1219 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1220 "registered in local node %d but not in "
1221 "joining node %d\n", qn->qn_domain,
1222 local->nd_num, &(local->nd_ipv4_address),
1223 ntohs(local->nd_ipv4_port),
1224 dlm->node_num, qn->qn_nodenum);
1225 BUG_ON((!local && !remote));
1226 }
1227
1228 if (local)
1229 o2nm_node_put(local);
1230 }
1231
1232 return status;
1233 }
1234
1235 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1236 {
1237 struct dlm_query_nodeinfo *qn = NULL;
1238 struct o2nm_node *node;
1239 int ret = 0, status, count, i;
1240
1241 if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1242 goto bail;
1243
1244 qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1245 if (!qn) {
1246 ret = -ENOMEM;
1247 mlog_errno(ret);
1248 goto bail;
1249 }
1250
1251 for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1252 node = o2nm_get_node_by_num(i);
1253 if (!node)
1254 continue;
1255 qn->qn_nodes[count].ni_nodenum = node->nd_num;
1256 qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1257 qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1258 mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1259 &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1260 ++count;
1261 o2nm_node_put(node);
1262 }
1263
1264 qn->qn_nodenum = dlm->node_num;
1265 qn->qn_numnodes = count;
1266 qn->qn_namelen = strlen(dlm->name);
1267 memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1268
1269 i = -1;
1270 while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1271 i + 1)) < O2NM_MAX_NODES) {
1272 if (i == dlm->node_num)
1273 continue;
1274
1275 mlog(0, "Sending nodeinfo to node %d\n", i);
1276
1277 ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1278 qn, sizeof(struct dlm_query_nodeinfo),
1279 i, &status);
1280 if (ret >= 0)
1281 ret = status;
1282 if (ret) {
1283 mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1284 break;
1285 }
1286 }
1287
1288 bail:
1289 kfree(qn);
1290 return ret;
1291 }
1292
1293 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1294 void *data, void **ret_data)
1295 {
1296 struct dlm_query_nodeinfo *qn;
1297 struct dlm_ctxt *dlm = NULL;
1298 int locked = 0, status = -EINVAL;
1299
1300 qn = (struct dlm_query_nodeinfo *) msg->buf;
1301
1302 mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1303 qn->qn_domain);
1304
1305 spin_lock(&dlm_domain_lock);
1306 dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1307 if (!dlm) {
1308 mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1309 "join domain\n", qn->qn_nodenum, qn->qn_domain);
1310 goto bail;
1311 }
1312
1313 spin_lock(&dlm->spinlock);
1314 locked = 1;
1315 if (dlm->joining_node != qn->qn_nodenum) {
1316 mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1317 "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1318 dlm->joining_node);
1319 goto bail;
1320 }
1321
1322 /* Support for node query was added in 1.1 */
1323 if (dlm->dlm_locking_proto.pv_major == 1 &&
1324 dlm->dlm_locking_proto.pv_minor == 0) {
1325 mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1326 "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1327 qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1328 dlm->dlm_locking_proto.pv_minor);
1329 goto bail;
1330 }
1331
1332 status = dlm_match_nodes(dlm, qn);
1333
1334 bail:
1335 if (locked)
1336 spin_unlock(&dlm->spinlock);
1337 spin_unlock(&dlm_domain_lock);
1338
1339 return status;
1340 }
1341
1342 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1343 void **ret_data)
1344 {
1345 struct dlm_cancel_join *cancel;
1346 struct dlm_ctxt *dlm = NULL;
1347
1348 cancel = (struct dlm_cancel_join *) msg->buf;
1349
1350 mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1351 cancel->domain);
1352
1353 spin_lock(&dlm_domain_lock);
1354 dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1355
1356 if (dlm) {
1357 spin_lock(&dlm->spinlock);
1358
1359 /* Yikes, this guy wants to cancel his join. No
1360 * problem, we simply cleanup our join state. */
1361 BUG_ON(dlm->joining_node != cancel->node_idx);
1362 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1363
1364 spin_unlock(&dlm->spinlock);
1365 }
1366 spin_unlock(&dlm_domain_lock);
1367
1368 return 0;
1369 }
1370
1371 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1372 unsigned int node)
1373 {
1374 int status;
1375 struct dlm_cancel_join cancel_msg;
1376
1377 memset(&cancel_msg, 0, sizeof(cancel_msg));
1378 cancel_msg.node_idx = dlm->node_num;
1379 cancel_msg.name_len = strlen(dlm->name);
1380 memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1381
1382 status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1383 &cancel_msg, sizeof(cancel_msg), node,
1384 NULL);
1385 if (status < 0) {
1386 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1387 "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1388 node);
1389 goto bail;
1390 }
1391
1392 bail:
1393 return status;
1394 }
1395
1396 /* map_size should be in bytes. */
1397 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1398 unsigned long *node_map,
1399 unsigned int map_size)
1400 {
1401 int status, tmpstat;
1402 int node;
1403
1404 if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1405 sizeof(unsigned long))) {
1406 mlog(ML_ERROR,
1407 "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1408 map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1409 return -EINVAL;
1410 }
1411
1412 status = 0;
1413 node = -1;
1414 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1415 node + 1)) < O2NM_MAX_NODES) {
1416 if (node == dlm->node_num)
1417 continue;
1418
1419 tmpstat = dlm_send_one_join_cancel(dlm, node);
1420 if (tmpstat) {
1421 mlog(ML_ERROR, "Error return %d cancelling join on "
1422 "node %d\n", tmpstat, node);
1423 if (!status)
1424 status = tmpstat;
1425 }
1426 }
1427
1428 if (status)
1429 mlog_errno(status);
1430 return status;
1431 }
1432
1433 static int dlm_request_join(struct dlm_ctxt *dlm,
1434 int node,
1435 enum dlm_query_join_response_code *response)
1436 {
1437 int status;
1438 struct dlm_query_join_request join_msg;
1439 struct dlm_query_join_packet packet;
1440 u32 join_resp;
1441
1442 mlog(0, "querying node %d\n", node);
1443
1444 memset(&join_msg, 0, sizeof(join_msg));
1445 join_msg.node_idx = dlm->node_num;
1446 join_msg.name_len = strlen(dlm->name);
1447 memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1448 join_msg.dlm_proto = dlm->dlm_locking_proto;
1449 join_msg.fs_proto = dlm->fs_locking_proto;
1450
1451 /* copy live node map to join message */
1452 byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1453
1454 status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1455 sizeof(join_msg), node, &join_resp);
1456 if (status < 0 && status != -ENOPROTOOPT) {
1457 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1458 "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1459 node);
1460 goto bail;
1461 }
1462 dlm_query_join_wire_to_packet(join_resp, &packet);
1463
1464 /* -ENOPROTOOPT from the net code means the other side isn't
1465 listening for our message type -- that's fine, it means
1466 his dlm isn't up, so we can consider him a 'yes' but not
1467 joined into the domain. */
1468 if (status == -ENOPROTOOPT) {
1469 status = 0;
1470 *response = JOIN_OK_NO_MAP;
1471 } else {
1472 *response = packet.code;
1473 switch (packet.code) {
1474 case JOIN_DISALLOW:
1475 case JOIN_OK_NO_MAP:
1476 break;
1477 case JOIN_PROTOCOL_MISMATCH:
1478 mlog(ML_NOTICE,
1479 "This node requested DLM locking protocol %u.%u and "
1480 "filesystem locking protocol %u.%u. At least one of "
1481 "the protocol versions on node %d is not compatible, "
1482 "disconnecting\n",
1483 dlm->dlm_locking_proto.pv_major,
1484 dlm->dlm_locking_proto.pv_minor,
1485 dlm->fs_locking_proto.pv_major,
1486 dlm->fs_locking_proto.pv_minor,
1487 node);
1488 status = -EPROTO;
1489 break;
1490 case JOIN_OK:
1491 /* Use the same locking protocol as the remote node */
1492 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1493 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1494 mlog(0,
1495 "Node %d responds JOIN_OK with DLM locking protocol "
1496 "%u.%u and fs locking protocol %u.%u\n",
1497 node,
1498 dlm->dlm_locking_proto.pv_major,
1499 dlm->dlm_locking_proto.pv_minor,
1500 dlm->fs_locking_proto.pv_major,
1501 dlm->fs_locking_proto.pv_minor);
1502 break;
1503 default:
1504 status = -EINVAL;
1505 mlog(ML_ERROR, "invalid response %d from node %u\n",
1506 packet.code, node);
1507 /* Reset response to JOIN_DISALLOW */
1508 *response = JOIN_DISALLOW;
1509 break;
1510 }
1511 }
1512
1513 mlog(0, "status %d, node %d response is %d\n", status, node,
1514 *response);
1515
1516 bail:
1517 return status;
1518 }
1519
1520 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1521 unsigned int node)
1522 {
1523 int status;
1524 int ret;
1525 struct dlm_assert_joined assert_msg;
1526
1527 mlog(0, "Sending join assert to node %u\n", node);
1528
1529 memset(&assert_msg, 0, sizeof(assert_msg));
1530 assert_msg.node_idx = dlm->node_num;
1531 assert_msg.name_len = strlen(dlm->name);
1532 memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1533
1534 status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1535 &assert_msg, sizeof(assert_msg), node,
1536 &ret);
1537 if (status < 0)
1538 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1539 "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1540 node);
1541 else
1542 status = ret;
1543
1544 return status;
1545 }
1546
1547 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1548 unsigned long *node_map)
1549 {
1550 int status, node, live;
1551
1552 status = 0;
1553 node = -1;
1554 while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1555 node + 1)) < O2NM_MAX_NODES) {
1556 if (node == dlm->node_num)
1557 continue;
1558
1559 do {
1560 /* It is very important that this message be
1561 * received so we spin until either the node
1562 * has died or it gets the message. */
1563 status = dlm_send_one_join_assert(dlm, node);
1564
1565 spin_lock(&dlm->spinlock);
1566 live = test_bit(node, dlm->live_nodes_map);
1567 spin_unlock(&dlm->spinlock);
1568
1569 if (status) {
1570 mlog(ML_ERROR, "Error return %d asserting "
1571 "join on node %d\n", status, node);
1572
1573 /* give us some time between errors... */
1574 if (live)
1575 msleep(DLM_DOMAIN_BACKOFF_MS);
1576 }
1577 } while (status && live);
1578 }
1579 }
1580
1581 struct domain_join_ctxt {
1582 unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1583 unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1584 };
1585
1586 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1587 struct domain_join_ctxt *ctxt,
1588 enum dlm_query_join_response_code response)
1589 {
1590 int ret;
1591
1592 if (response == JOIN_DISALLOW) {
1593 mlog(0, "Latest response of disallow -- should restart\n");
1594 return 1;
1595 }
1596
1597 spin_lock(&dlm->spinlock);
1598 /* For now, we restart the process if the node maps have
1599 * changed at all */
1600 ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1601 sizeof(dlm->live_nodes_map));
1602 spin_unlock(&dlm->spinlock);
1603
1604 if (ret)
1605 mlog(0, "Node maps changed -- should restart\n");
1606
1607 return ret;
1608 }
1609
1610 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1611 {
1612 int status = 0, tmpstat, node;
1613 struct domain_join_ctxt *ctxt;
1614 enum dlm_query_join_response_code response = JOIN_DISALLOW;
1615
1616 mlog(0, "%p", dlm);
1617
1618 ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1619 if (!ctxt) {
1620 status = -ENOMEM;
1621 mlog_errno(status);
1622 goto bail;
1623 }
1624
1625 /* group sem locking should work for us here -- we're already
1626 * registered for heartbeat events so filling this should be
1627 * atomic wrt getting those handlers called. */
1628 o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1629
1630 spin_lock(&dlm->spinlock);
1631 memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1632
1633 __dlm_set_joining_node(dlm, dlm->node_num);
1634
1635 spin_unlock(&dlm->spinlock);
1636
1637 node = -1;
1638 while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1639 node + 1)) < O2NM_MAX_NODES) {
1640 if (node == dlm->node_num)
1641 continue;
1642
1643 status = dlm_request_join(dlm, node, &response);
1644 if (status < 0) {
1645 mlog_errno(status);
1646 goto bail;
1647 }
1648
1649 /* Ok, either we got a response or the node doesn't have a
1650 * dlm up. */
1651 if (response == JOIN_OK)
1652 set_bit(node, ctxt->yes_resp_map);
1653
1654 if (dlm_should_restart_join(dlm, ctxt, response)) {
1655 status = -EAGAIN;
1656 goto bail;
1657 }
1658 }
1659
1660 mlog(0, "Yay, done querying nodes!\n");
1661
1662 /* Yay, everyone agree's we can join the domain. My domain is
1663 * comprised of all nodes who were put in the
1664 * yes_resp_map. Copy that into our domain map and send a join
1665 * assert message to clean up everyone elses state. */
1666 spin_lock(&dlm->spinlock);
1667 memcpy(dlm->domain_map, ctxt->yes_resp_map,
1668 sizeof(ctxt->yes_resp_map));
1669 set_bit(dlm->node_num, dlm->domain_map);
1670 spin_unlock(&dlm->spinlock);
1671
1672 /* Support for global heartbeat and node info was added in 1.1 */
1673 if (dlm->dlm_locking_proto.pv_major > 1 ||
1674 dlm->dlm_locking_proto.pv_minor > 0) {
1675 status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1676 if (status) {
1677 mlog_errno(status);
1678 goto bail;
1679 }
1680 status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1681 if (status) {
1682 mlog_errno(status);
1683 goto bail;
1684 }
1685 }
1686
1687 dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1688
1689 /* Joined state *must* be set before the joining node
1690 * information, otherwise the query_join handler may read no
1691 * current joiner but a state of NEW and tell joining nodes
1692 * we're not in the domain. */
1693 spin_lock(&dlm_domain_lock);
1694 dlm->dlm_state = DLM_CTXT_JOINED;
1695 dlm->num_joins++;
1696 spin_unlock(&dlm_domain_lock);
1697
1698 bail:
1699 spin_lock(&dlm->spinlock);
1700 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1701 if (!status) {
1702 printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1703 __dlm_print_nodes(dlm);
1704 }
1705 spin_unlock(&dlm->spinlock);
1706
1707 if (ctxt) {
1708 /* Do we need to send a cancel message to any nodes? */
1709 if (status < 0) {
1710 tmpstat = dlm_send_join_cancels(dlm,
1711 ctxt->yes_resp_map,
1712 sizeof(ctxt->yes_resp_map));
1713 if (tmpstat < 0)
1714 mlog_errno(tmpstat);
1715 }
1716 kfree(ctxt);
1717 }
1718
1719 mlog(0, "returning %d\n", status);
1720 return status;
1721 }
1722
1723 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1724 {
1725 o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1726 o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1727 o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1728 }
1729
1730 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1731 {
1732 int status;
1733
1734 mlog(0, "registering handlers.\n");
1735
1736 o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1737 dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1738 o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1739 dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1740
1741 status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1742 if (status)
1743 goto bail;
1744
1745 status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1746 if (status)
1747 goto bail;
1748
1749 status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1750 sizeof(struct dlm_master_request),
1751 dlm_master_request_handler,
1752 dlm, NULL, &dlm->dlm_domain_handlers);
1753 if (status)
1754 goto bail;
1755
1756 status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1757 sizeof(struct dlm_assert_master),
1758 dlm_assert_master_handler,
1759 dlm, dlm_assert_master_post_handler,
1760 &dlm->dlm_domain_handlers);
1761 if (status)
1762 goto bail;
1763
1764 status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1765 sizeof(struct dlm_create_lock),
1766 dlm_create_lock_handler,
1767 dlm, NULL, &dlm->dlm_domain_handlers);
1768 if (status)
1769 goto bail;
1770
1771 status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1772 DLM_CONVERT_LOCK_MAX_LEN,
1773 dlm_convert_lock_handler,
1774 dlm, NULL, &dlm->dlm_domain_handlers);
1775 if (status)
1776 goto bail;
1777
1778 status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1779 DLM_UNLOCK_LOCK_MAX_LEN,
1780 dlm_unlock_lock_handler,
1781 dlm, NULL, &dlm->dlm_domain_handlers);
1782 if (status)
1783 goto bail;
1784
1785 status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1786 DLM_PROXY_AST_MAX_LEN,
1787 dlm_proxy_ast_handler,
1788 dlm, NULL, &dlm->dlm_domain_handlers);
1789 if (status)
1790 goto bail;
1791
1792 status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1793 sizeof(struct dlm_exit_domain),
1794 dlm_exit_domain_handler,
1795 dlm, NULL, &dlm->dlm_domain_handlers);
1796 if (status)
1797 goto bail;
1798
1799 status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1800 sizeof(struct dlm_deref_lockres),
1801 dlm_deref_lockres_handler,
1802 dlm, NULL, &dlm->dlm_domain_handlers);
1803 if (status)
1804 goto bail;
1805
1806 status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1807 sizeof(struct dlm_migrate_request),
1808 dlm_migrate_request_handler,
1809 dlm, NULL, &dlm->dlm_domain_handlers);
1810 if (status)
1811 goto bail;
1812
1813 status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1814 DLM_MIG_LOCKRES_MAX_LEN,
1815 dlm_mig_lockres_handler,
1816 dlm, NULL, &dlm->dlm_domain_handlers);
1817 if (status)
1818 goto bail;
1819
1820 status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1821 sizeof(struct dlm_master_requery),
1822 dlm_master_requery_handler,
1823 dlm, NULL, &dlm->dlm_domain_handlers);
1824 if (status)
1825 goto bail;
1826
1827 status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1828 sizeof(struct dlm_lock_request),
1829 dlm_request_all_locks_handler,
1830 dlm, NULL, &dlm->dlm_domain_handlers);
1831 if (status)
1832 goto bail;
1833
1834 status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1835 sizeof(struct dlm_reco_data_done),
1836 dlm_reco_data_done_handler,
1837 dlm, NULL, &dlm->dlm_domain_handlers);
1838 if (status)
1839 goto bail;
1840
1841 status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1842 sizeof(struct dlm_begin_reco),
1843 dlm_begin_reco_handler,
1844 dlm, NULL, &dlm->dlm_domain_handlers);
1845 if (status)
1846 goto bail;
1847
1848 status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1849 sizeof(struct dlm_finalize_reco),
1850 dlm_finalize_reco_handler,
1851 dlm, NULL, &dlm->dlm_domain_handlers);
1852 if (status)
1853 goto bail;
1854
1855 status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1856 sizeof(struct dlm_exit_domain),
1857 dlm_begin_exit_domain_handler,
1858 dlm, NULL, &dlm->dlm_domain_handlers);
1859 if (status)
1860 goto bail;
1861
1862 status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key,
1863 sizeof(struct dlm_deref_lockres_done),
1864 dlm_deref_lockres_done_handler,
1865 dlm, NULL, &dlm->dlm_domain_handlers);
1866 bail:
1867 if (status)
1868 dlm_unregister_domain_handlers(dlm);
1869
1870 return status;
1871 }
1872
1873 static int dlm_join_domain(struct dlm_ctxt *dlm)
1874 {
1875 int status;
1876 unsigned int backoff;
1877 unsigned int total_backoff = 0;
1878 char wq_name[O2NM_MAX_NAME_LEN];
1879
1880 BUG_ON(!dlm);
1881
1882 mlog(0, "Join domain %s\n", dlm->name);
1883
1884 status = dlm_register_domain_handlers(dlm);
1885 if (status) {
1886 mlog_errno(status);
1887 goto bail;
1888 }
1889
1890 status = dlm_launch_thread(dlm);
1891 if (status < 0) {
1892 mlog_errno(status);
1893 goto bail;
1894 }
1895
1896 status = dlm_launch_recovery_thread(dlm);
1897 if (status < 0) {
1898 mlog_errno(status);
1899 goto bail;
1900 }
1901
1902 status = dlm_debug_init(dlm);
1903 if (status < 0) {
1904 mlog_errno(status);
1905 goto bail;
1906 }
1907
1908 snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name);
1909 dlm->dlm_worker = create_singlethread_workqueue(wq_name);
1910 if (!dlm->dlm_worker) {
1911 status = -ENOMEM;
1912 mlog_errno(status);
1913 goto bail;
1914 }
1915
1916 do {
1917 status = dlm_try_to_join_domain(dlm);
1918
1919 /* If we're racing another node to the join, then we
1920 * need to back off temporarily and let them
1921 * complete. */
1922 #define DLM_JOIN_TIMEOUT_MSECS 90000
1923 if (status == -EAGAIN) {
1924 if (signal_pending(current)) {
1925 status = -ERESTARTSYS;
1926 goto bail;
1927 }
1928
1929 if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1930 status = -ERESTARTSYS;
1931 mlog(ML_NOTICE, "Timed out joining dlm domain "
1932 "%s after %u msecs\n", dlm->name,
1933 total_backoff);
1934 goto bail;
1935 }
1936
1937 /*
1938 * <chip> After you!
1939 * <dale> No, after you!
1940 * <chip> I insist!
1941 * <dale> But you first!
1942 * ...
1943 */
1944 backoff = (unsigned int)(jiffies & 0x3);
1945 backoff *= DLM_DOMAIN_BACKOFF_MS;
1946 total_backoff += backoff;
1947 mlog(0, "backoff %d\n", backoff);
1948 msleep(backoff);
1949 }
1950 } while (status == -EAGAIN);
1951
1952 if (status < 0) {
1953 mlog_errno(status);
1954 goto bail;
1955 }
1956
1957 status = 0;
1958 bail:
1959 wake_up(&dlm_domain_events);
1960
1961 if (status) {
1962 dlm_unregister_domain_handlers(dlm);
1963 dlm_debug_shutdown(dlm);
1964 dlm_complete_thread(dlm);
1965 dlm_complete_recovery_thread(dlm);
1966 dlm_destroy_dlm_worker(dlm);
1967 }
1968
1969 return status;
1970 }
1971
1972 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1973 u32 key)
1974 {
1975 int i;
1976 int ret;
1977 struct dlm_ctxt *dlm = NULL;
1978
1979 dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1980 if (!dlm) {
1981 ret = -ENOMEM;
1982 mlog_errno(ret);
1983 goto leave;
1984 }
1985
1986 dlm->name = kstrdup(domain, GFP_KERNEL);
1987 if (dlm->name == NULL) {
1988 ret = -ENOMEM;
1989 mlog_errno(ret);
1990 goto leave;
1991 }
1992
1993 dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1994 if (!dlm->lockres_hash) {
1995 ret = -ENOMEM;
1996 mlog_errno(ret);
1997 goto leave;
1998 }
1999
2000 for (i = 0; i < DLM_HASH_BUCKETS; i++)
2001 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
2002
2003 dlm->master_hash = (struct hlist_head **)
2004 dlm_alloc_pagevec(DLM_HASH_PAGES);
2005 if (!dlm->master_hash) {
2006 ret = -ENOMEM;
2007 mlog_errno(ret);
2008 goto leave;
2009 }
2010
2011 for (i = 0; i < DLM_HASH_BUCKETS; i++)
2012 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2013
2014 dlm->key = key;
2015 dlm->node_num = o2nm_this_node();
2016
2017 ret = dlm_create_debugfs_subroot(dlm);
2018 if (ret < 0)
2019 goto leave;
2020
2021 spin_lock_init(&dlm->spinlock);
2022 spin_lock_init(&dlm->master_lock);
2023 spin_lock_init(&dlm->ast_lock);
2024 spin_lock_init(&dlm->track_lock);
2025 INIT_LIST_HEAD(&dlm->list);
2026 INIT_LIST_HEAD(&dlm->dirty_list);
2027 INIT_LIST_HEAD(&dlm->reco.resources);
2028 INIT_LIST_HEAD(&dlm->reco.node_data);
2029 INIT_LIST_HEAD(&dlm->purge_list);
2030 INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2031 INIT_LIST_HEAD(&dlm->tracking_list);
2032 dlm->reco.state = 0;
2033
2034 INIT_LIST_HEAD(&dlm->pending_asts);
2035 INIT_LIST_HEAD(&dlm->pending_basts);
2036
2037 mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2038 dlm->recovery_map, &(dlm->recovery_map[0]));
2039
2040 memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2041 memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2042 memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2043
2044 dlm->dlm_thread_task = NULL;
2045 dlm->dlm_reco_thread_task = NULL;
2046 dlm->dlm_worker = NULL;
2047 init_waitqueue_head(&dlm->dlm_thread_wq);
2048 init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2049 init_waitqueue_head(&dlm->reco.event);
2050 init_waitqueue_head(&dlm->ast_wq);
2051 init_waitqueue_head(&dlm->migration_wq);
2052 INIT_LIST_HEAD(&dlm->mle_hb_events);
2053
2054 dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2055 init_waitqueue_head(&dlm->dlm_join_events);
2056
2057 dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2058 dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2059
2060 atomic_set(&dlm->res_tot_count, 0);
2061 atomic_set(&dlm->res_cur_count, 0);
2062 for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2063 atomic_set(&dlm->mle_tot_count[i], 0);
2064 atomic_set(&dlm->mle_cur_count[i], 0);
2065 }
2066
2067 spin_lock_init(&dlm->work_lock);
2068 INIT_LIST_HEAD(&dlm->work_list);
2069 INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2070
2071 kref_init(&dlm->dlm_refs);
2072 dlm->dlm_state = DLM_CTXT_NEW;
2073
2074 INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2075
2076 mlog(0, "context init: refcount %u\n",
2077 atomic_read(&dlm->dlm_refs.refcount));
2078
2079 leave:
2080 if (ret < 0 && dlm) {
2081 if (dlm->master_hash)
2082 dlm_free_pagevec((void **)dlm->master_hash,
2083 DLM_HASH_PAGES);
2084
2085 if (dlm->lockres_hash)
2086 dlm_free_pagevec((void **)dlm->lockres_hash,
2087 DLM_HASH_PAGES);
2088
2089 kfree(dlm->name);
2090 kfree(dlm);
2091 dlm = NULL;
2092 }
2093 return dlm;
2094 }
2095
2096 /*
2097 * Compare a requested locking protocol version against the current one.
2098 *
2099 * If the major numbers are different, they are incompatible.
2100 * If the current minor is greater than the request, they are incompatible.
2101 * If the current minor is less than or equal to the request, they are
2102 * compatible, and the requester should run at the current minor version.
2103 */
2104 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2105 struct dlm_protocol_version *request)
2106 {
2107 if (existing->pv_major != request->pv_major)
2108 return 1;
2109
2110 if (existing->pv_minor > request->pv_minor)
2111 return 1;
2112
2113 if (existing->pv_minor < request->pv_minor)
2114 request->pv_minor = existing->pv_minor;
2115
2116 return 0;
2117 }
2118
2119 /*
2120 * dlm_register_domain: one-time setup per "domain".
2121 *
2122 * The filesystem passes in the requested locking version via proto.
2123 * If registration was successful, proto will contain the negotiated
2124 * locking protocol.
2125 */
2126 struct dlm_ctxt * dlm_register_domain(const char *domain,
2127 u32 key,
2128 struct dlm_protocol_version *fs_proto)
2129 {
2130 int ret;
2131 struct dlm_ctxt *dlm = NULL;
2132 struct dlm_ctxt *new_ctxt = NULL;
2133
2134 if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2135 ret = -ENAMETOOLONG;
2136 mlog(ML_ERROR, "domain name length too long\n");
2137 goto leave;
2138 }
2139
2140 mlog(0, "register called for domain \"%s\"\n", domain);
2141
2142 retry:
2143 dlm = NULL;
2144 if (signal_pending(current)) {
2145 ret = -ERESTARTSYS;
2146 mlog_errno(ret);
2147 goto leave;
2148 }
2149
2150 spin_lock(&dlm_domain_lock);
2151
2152 dlm = __dlm_lookup_domain(domain);
2153 if (dlm) {
2154 if (dlm->dlm_state != DLM_CTXT_JOINED) {
2155 spin_unlock(&dlm_domain_lock);
2156
2157 mlog(0, "This ctxt is not joined yet!\n");
2158 wait_event_interruptible(dlm_domain_events,
2159 dlm_wait_on_domain_helper(
2160 domain));
2161 goto retry;
2162 }
2163
2164 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2165 spin_unlock(&dlm_domain_lock);
2166 mlog(ML_ERROR,
2167 "Requested locking protocol version is not "
2168 "compatible with already registered domain "
2169 "\"%s\"\n", domain);
2170 ret = -EPROTO;
2171 goto leave;
2172 }
2173
2174 __dlm_get(dlm);
2175 dlm->num_joins++;
2176
2177 spin_unlock(&dlm_domain_lock);
2178
2179 ret = 0;
2180 goto leave;
2181 }
2182
2183 /* doesn't exist */
2184 if (!new_ctxt) {
2185 spin_unlock(&dlm_domain_lock);
2186
2187 new_ctxt = dlm_alloc_ctxt(domain, key);
2188 if (new_ctxt)
2189 goto retry;
2190
2191 ret = -ENOMEM;
2192 mlog_errno(ret);
2193 goto leave;
2194 }
2195
2196 /* a little variable switch-a-roo here... */
2197 dlm = new_ctxt;
2198 new_ctxt = NULL;
2199
2200 /* add the new domain */
2201 list_add_tail(&dlm->list, &dlm_domains);
2202 spin_unlock(&dlm_domain_lock);
2203
2204 /*
2205 * Pass the locking protocol version into the join. If the join
2206 * succeeds, it will have the negotiated protocol set.
2207 */
2208 dlm->dlm_locking_proto = dlm_protocol;
2209 dlm->fs_locking_proto = *fs_proto;
2210
2211 ret = dlm_join_domain(dlm);
2212 if (ret) {
2213 mlog_errno(ret);
2214 dlm_put(dlm);
2215 goto leave;
2216 }
2217
2218 /* Tell the caller what locking protocol we negotiated */
2219 *fs_proto = dlm->fs_locking_proto;
2220
2221 ret = 0;
2222 leave:
2223 if (new_ctxt)
2224 dlm_free_ctxt_mem(new_ctxt);
2225
2226 if (ret < 0)
2227 dlm = ERR_PTR(ret);
2228
2229 return dlm;
2230 }
2231 EXPORT_SYMBOL_GPL(dlm_register_domain);
2232
2233 static LIST_HEAD(dlm_join_handlers);
2234
2235 static void dlm_unregister_net_handlers(void)
2236 {
2237 o2net_unregister_handler_list(&dlm_join_handlers);
2238 }
2239
2240 static int dlm_register_net_handlers(void)
2241 {
2242 int status = 0;
2243
2244 status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2245 sizeof(struct dlm_query_join_request),
2246 dlm_query_join_handler,
2247 NULL, NULL, &dlm_join_handlers);
2248 if (status)
2249 goto bail;
2250
2251 status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2252 sizeof(struct dlm_assert_joined),
2253 dlm_assert_joined_handler,
2254 NULL, NULL, &dlm_join_handlers);
2255 if (status)
2256 goto bail;
2257
2258 status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2259 sizeof(struct dlm_cancel_join),
2260 dlm_cancel_join_handler,
2261 NULL, NULL, &dlm_join_handlers);
2262 if (status)
2263 goto bail;
2264
2265 status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2266 sizeof(struct dlm_query_region),
2267 dlm_query_region_handler,
2268 NULL, NULL, &dlm_join_handlers);
2269
2270 if (status)
2271 goto bail;
2272
2273 status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2274 sizeof(struct dlm_query_nodeinfo),
2275 dlm_query_nodeinfo_handler,
2276 NULL, NULL, &dlm_join_handlers);
2277 bail:
2278 if (status < 0)
2279 dlm_unregister_net_handlers();
2280
2281 return status;
2282 }
2283
2284 /* Domain eviction callback handling.
2285 *
2286 * The file system requires notification of node death *before* the
2287 * dlm completes it's recovery work, otherwise it may be able to
2288 * acquire locks on resources requiring recovery. Since the dlm can
2289 * evict a node from it's domain *before* heartbeat fires, a similar
2290 * mechanism is required. */
2291
2292 /* Eviction is not expected to happen often, so a per-domain lock is
2293 * not necessary. Eviction callbacks are allowed to sleep for short
2294 * periods of time. */
2295 static DECLARE_RWSEM(dlm_callback_sem);
2296
2297 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2298 int node_num)
2299 {
2300 struct dlm_eviction_cb *cb;
2301
2302 down_read(&dlm_callback_sem);
2303 list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
2304 cb->ec_func(node_num, cb->ec_data);
2305 }
2306 up_read(&dlm_callback_sem);
2307 }
2308
2309 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2310 dlm_eviction_func *f,
2311 void *data)
2312 {
2313 INIT_LIST_HEAD(&cb->ec_item);
2314 cb->ec_func = f;
2315 cb->ec_data = data;
2316 }
2317 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2318
2319 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2320 struct dlm_eviction_cb *cb)
2321 {
2322 down_write(&dlm_callback_sem);
2323 list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2324 up_write(&dlm_callback_sem);
2325 }
2326 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2327
2328 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2329 {
2330 down_write(&dlm_callback_sem);
2331 list_del_init(&cb->ec_item);
2332 up_write(&dlm_callback_sem);
2333 }
2334 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2335
2336 static int __init dlm_init(void)
2337 {
2338 int status;
2339
2340 status = dlm_init_mle_cache();
2341 if (status) {
2342 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2343 goto error;
2344 }
2345
2346 status = dlm_init_master_caches();
2347 if (status) {
2348 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2349 "o2dlm_lockname slabcaches\n");
2350 goto error;
2351 }
2352
2353 status = dlm_init_lock_cache();
2354 if (status) {
2355 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2356 goto error;
2357 }
2358
2359 status = dlm_register_net_handlers();
2360 if (status) {
2361 mlog(ML_ERROR, "Unable to register network handlers\n");
2362 goto error;
2363 }
2364
2365 status = dlm_create_debugfs_root();
2366 if (status)
2367 goto error;
2368
2369 return 0;
2370 error:
2371 dlm_unregister_net_handlers();
2372 dlm_destroy_lock_cache();
2373 dlm_destroy_master_caches();
2374 dlm_destroy_mle_cache();
2375 return -1;
2376 }
2377
2378 static void __exit dlm_exit (void)
2379 {
2380 dlm_destroy_debugfs_root();
2381 dlm_unregister_net_handlers();
2382 dlm_destroy_lock_cache();
2383 dlm_destroy_master_caches();
2384 dlm_destroy_mle_cache();
2385 }
2386
2387 MODULE_AUTHOR("Oracle");
2388 MODULE_LICENSE("GPL");
2389 MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
2390
2391 module_init(dlm_init);
2392 module_exit(dlm_exit);
Linux with added FPC-III support