drm/i915: add pipe A force quirks to i915 driver
[sfrench/cifs-2.6.git] / fs / ocfs2 / dlm / dlmdomain.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmdomain.c
5  *
6  * defines domain join / leave apis
7  *
8  * Copyright (C) 2004 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation; either
13  * version 2 of the License, or (at your option) any later version.
14  *
15  * This program is distributed in the hope that it will be useful,
16  * but WITHOUT ANY WARRANTY; without even the implied warranty of
17  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
18  * General Public License for more details.
19  *
20  * You should have received a copy of the GNU General Public
21  * License along with this program; if not, write to the
22  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
23  * Boston, MA 021110-1307, USA.
24  *
25  */
26
27 #include <linux/module.h>
28 #include <linux/types.h>
29 #include <linux/slab.h>
30 #include <linux/highmem.h>
31 #include <linux/init.h>
32 #include <linux/spinlock.h>
33 #include <linux/delay.h>
34 #include <linux/err.h>
35 #include <linux/debugfs.h>
36
37 #include "cluster/heartbeat.h"
38 #include "cluster/nodemanager.h"
39 #include "cluster/tcp.h"
40
41 #include "dlmapi.h"
42 #include "dlmcommon.h"
43 #include "dlmdomain.h"
44 #include "dlmdebug.h"
45
46 #include "dlmver.h"
47
48 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
49 #include "cluster/masklog.h"
50
51 /*
52  * ocfs2 node maps are array of long int, which limits to send them freely
53  * across the wire due to endianness issues. To workaround this, we convert
54  * long ints to byte arrays. Following 3 routines are helper functions to
55  * set/test/copy bits within those array of bytes
56  */
57 static inline void byte_set_bit(u8 nr, u8 map[])
58 {
59         map[nr >> 3] |= (1UL << (nr & 7));
60 }
61
62 static inline int byte_test_bit(u8 nr, u8 map[])
63 {
64         return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
65 }
66
67 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
68                         unsigned int sz)
69 {
70         unsigned int nn;
71
72         if (!sz)
73                 return;
74
75         memset(dmap, 0, ((sz + 7) >> 3));
76         for (nn = 0 ; nn < sz; nn++)
77                 if (test_bit(nn, smap))
78                         byte_set_bit(nn, dmap);
79 }
80
81 static void dlm_free_pagevec(void **vec, int pages)
82 {
83         while (pages--)
84                 free_page((unsigned long)vec[pages]);
85         kfree(vec);
86 }
87
88 static void **dlm_alloc_pagevec(int pages)
89 {
90         void **vec = kmalloc(pages * sizeof(void *), GFP_KERNEL);
91         int i;
92
93         if (!vec)
94                 return NULL;
95
96         for (i = 0; i < pages; i++)
97                 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
98                         goto out_free;
99
100         mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
101              pages, (unsigned long)DLM_HASH_PAGES,
102              (unsigned long)DLM_BUCKETS_PER_PAGE);
103         return vec;
104 out_free:
105         dlm_free_pagevec(vec, i);
106         return NULL;
107 }
108
109 /*
110  *
111  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
112  *    dlm_domain_lock
113  *    struct dlm_ctxt->spinlock
114  *    struct dlm_lock_resource->spinlock
115  *    struct dlm_ctxt->master_lock
116  *    struct dlm_ctxt->ast_lock
117  *    dlm_master_list_entry->spinlock
118  *    dlm_lock->spinlock
119  *
120  */
121
122 DEFINE_SPINLOCK(dlm_domain_lock);
123 LIST_HEAD(dlm_domains);
124 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
125
126 /*
127  * The supported protocol version for DLM communication.  Running domains
128  * will have a negotiated version with the same major number and a minor
129  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
130  * be used to determine what a running domain is actually using.
131  */
132 static const struct dlm_protocol_version dlm_protocol = {
133         .pv_major = 1,
134         .pv_minor = 0,
135 };
136
137 #define DLM_DOMAIN_BACKOFF_MS 200
138
139 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
140                                   void **ret_data);
141 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
142                                      void **ret_data);
143 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
144                                    void **ret_data);
145 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
146                                    void **ret_data);
147 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
148                                 struct dlm_protocol_version *request);
149
150 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
151
152 void __dlm_unhash_lockres(struct dlm_lock_resource *lockres)
153 {
154         if (!hlist_unhashed(&lockres->hash_node)) {
155                 hlist_del_init(&lockres->hash_node);
156                 dlm_lockres_put(lockres);
157         }
158 }
159
160 void __dlm_insert_lockres(struct dlm_ctxt *dlm,
161                        struct dlm_lock_resource *res)
162 {
163         struct hlist_head *bucket;
164         struct qstr *q;
165
166         assert_spin_locked(&dlm->spinlock);
167
168         q = &res->lockname;
169         bucket = dlm_lockres_hash(dlm, q->hash);
170
171         /* get a reference for our hashtable */
172         dlm_lockres_get(res);
173
174         hlist_add_head(&res->hash_node, bucket);
175 }
176
177 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
178                                                      const char *name,
179                                                      unsigned int len,
180                                                      unsigned int hash)
181 {
182         struct hlist_head *bucket;
183         struct hlist_node *list;
184
185         mlog_entry("%.*s\n", len, name);
186
187         assert_spin_locked(&dlm->spinlock);
188
189         bucket = dlm_lockres_hash(dlm, hash);
190
191         hlist_for_each(list, bucket) {
192                 struct dlm_lock_resource *res = hlist_entry(list,
193                         struct dlm_lock_resource, hash_node);
194                 if (res->lockname.name[0] != name[0])
195                         continue;
196                 if (unlikely(res->lockname.len != len))
197                         continue;
198                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
199                         continue;
200                 dlm_lockres_get(res);
201                 return res;
202         }
203         return NULL;
204 }
205
206 /* intended to be called by functions which do not care about lock
207  * resources which are being purged (most net _handler functions).
208  * this will return NULL for any lock resource which is found but
209  * currently in the process of dropping its mastery reference.
210  * use __dlm_lookup_lockres_full when you need the lock resource
211  * regardless (e.g. dlm_get_lock_resource) */
212 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
213                                                 const char *name,
214                                                 unsigned int len,
215                                                 unsigned int hash)
216 {
217         struct dlm_lock_resource *res = NULL;
218
219         mlog_entry("%.*s\n", len, name);
220
221         assert_spin_locked(&dlm->spinlock);
222
223         res = __dlm_lookup_lockres_full(dlm, name, len, hash);
224         if (res) {
225                 spin_lock(&res->spinlock);
226                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
227                         spin_unlock(&res->spinlock);
228                         dlm_lockres_put(res);
229                         return NULL;
230                 }
231                 spin_unlock(&res->spinlock);
232         }
233
234         return res;
235 }
236
237 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
238                                     const char *name,
239                                     unsigned int len)
240 {
241         struct dlm_lock_resource *res;
242         unsigned int hash = dlm_lockid_hash(name, len);
243
244         spin_lock(&dlm->spinlock);
245         res = __dlm_lookup_lockres(dlm, name, len, hash);
246         spin_unlock(&dlm->spinlock);
247         return res;
248 }
249
250 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
251 {
252         struct dlm_ctxt *tmp = NULL;
253         struct list_head *iter;
254
255         assert_spin_locked(&dlm_domain_lock);
256
257         /* tmp->name here is always NULL terminated,
258          * but domain may not be! */
259         list_for_each(iter, &dlm_domains) {
260                 tmp = list_entry (iter, struct dlm_ctxt, list);
261                 if (strlen(tmp->name) == len &&
262                     memcmp(tmp->name, domain, len)==0)
263                         break;
264                 tmp = NULL;
265         }
266
267         return tmp;
268 }
269
270 /* For null terminated domain strings ONLY */
271 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
272 {
273         assert_spin_locked(&dlm_domain_lock);
274
275         return __dlm_lookup_domain_full(domain, strlen(domain));
276 }
277
278
279 /* returns true on one of two conditions:
280  * 1) the domain does not exist
281  * 2) the domain exists and it's state is "joined" */
282 static int dlm_wait_on_domain_helper(const char *domain)
283 {
284         int ret = 0;
285         struct dlm_ctxt *tmp = NULL;
286
287         spin_lock(&dlm_domain_lock);
288
289         tmp = __dlm_lookup_domain(domain);
290         if (!tmp)
291                 ret = 1;
292         else if (tmp->dlm_state == DLM_CTXT_JOINED)
293                 ret = 1;
294
295         spin_unlock(&dlm_domain_lock);
296         return ret;
297 }
298
299 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
300 {
301         dlm_destroy_debugfs_subroot(dlm);
302
303         if (dlm->lockres_hash)
304                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
305
306         if (dlm->master_hash)
307                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
308
309         if (dlm->name)
310                 kfree(dlm->name);
311
312         kfree(dlm);
313 }
314
315 /* A little strange - this function will be called while holding
316  * dlm_domain_lock and is expected to be holding it on the way out. We
317  * will however drop and reacquire it multiple times */
318 static void dlm_ctxt_release(struct kref *kref)
319 {
320         struct dlm_ctxt *dlm;
321
322         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
323
324         BUG_ON(dlm->num_joins);
325         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
326
327         /* we may still be in the list if we hit an error during join. */
328         list_del_init(&dlm->list);
329
330         spin_unlock(&dlm_domain_lock);
331
332         mlog(0, "freeing memory from domain %s\n", dlm->name);
333
334         wake_up(&dlm_domain_events);
335
336         dlm_free_ctxt_mem(dlm);
337
338         spin_lock(&dlm_domain_lock);
339 }
340
341 void dlm_put(struct dlm_ctxt *dlm)
342 {
343         spin_lock(&dlm_domain_lock);
344         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
345         spin_unlock(&dlm_domain_lock);
346 }
347
348 static void __dlm_get(struct dlm_ctxt *dlm)
349 {
350         kref_get(&dlm->dlm_refs);
351 }
352
353 /* given a questionable reference to a dlm object, gets a reference if
354  * it can find it in the list, otherwise returns NULL in which case
355  * you shouldn't trust your pointer. */
356 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
357 {
358         struct list_head *iter;
359         struct dlm_ctxt *target = NULL;
360
361         spin_lock(&dlm_domain_lock);
362
363         list_for_each(iter, &dlm_domains) {
364                 target = list_entry (iter, struct dlm_ctxt, list);
365
366                 if (target == dlm) {
367                         __dlm_get(target);
368                         break;
369                 }
370
371                 target = NULL;
372         }
373
374         spin_unlock(&dlm_domain_lock);
375
376         return target;
377 }
378
379 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
380 {
381         int ret;
382
383         spin_lock(&dlm_domain_lock);
384         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
385                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
386         spin_unlock(&dlm_domain_lock);
387
388         return ret;
389 }
390
391 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
392 {
393         if (dlm->dlm_worker) {
394                 flush_workqueue(dlm->dlm_worker);
395                 destroy_workqueue(dlm->dlm_worker);
396                 dlm->dlm_worker = NULL;
397         }
398 }
399
400 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
401 {
402         dlm_unregister_domain_handlers(dlm);
403         dlm_debug_shutdown(dlm);
404         dlm_complete_thread(dlm);
405         dlm_complete_recovery_thread(dlm);
406         dlm_destroy_dlm_worker(dlm);
407
408         /* We've left the domain. Now we can take ourselves out of the
409          * list and allow the kref stuff to help us free the
410          * memory. */
411         spin_lock(&dlm_domain_lock);
412         list_del_init(&dlm->list);
413         spin_unlock(&dlm_domain_lock);
414
415         /* Wake up anyone waiting for us to remove this domain */
416         wake_up(&dlm_domain_events);
417 }
418
419 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
420 {
421         int i, num, n, ret = 0;
422         struct dlm_lock_resource *res;
423         struct hlist_node *iter;
424         struct hlist_head *bucket;
425         int dropped;
426
427         mlog(0, "Migrating locks from domain %s\n", dlm->name);
428
429         num = 0;
430         spin_lock(&dlm->spinlock);
431         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
432 redo_bucket:
433                 n = 0;
434                 bucket = dlm_lockres_hash(dlm, i);
435                 iter = bucket->first;
436                 while (iter) {
437                         n++;
438                         res = hlist_entry(iter, struct dlm_lock_resource,
439                                           hash_node);
440                         dlm_lockres_get(res);
441                         /* migrate, if necessary.  this will drop the dlm
442                          * spinlock and retake it if it does migration. */
443                         dropped = dlm_empty_lockres(dlm, res);
444
445                         spin_lock(&res->spinlock);
446                         __dlm_lockres_calc_usage(dlm, res);
447                         iter = res->hash_node.next;
448                         spin_unlock(&res->spinlock);
449
450                         dlm_lockres_put(res);
451
452                         if (dropped)
453                                 goto redo_bucket;
454                 }
455                 cond_resched_lock(&dlm->spinlock);
456                 num += n;
457                 mlog(0, "%s: touched %d lockreses in bucket %d "
458                      "(tot=%d)\n", dlm->name, n, i, num);
459         }
460         spin_unlock(&dlm->spinlock);
461         wake_up(&dlm->dlm_thread_wq);
462
463         /* let the dlm thread take care of purging, keep scanning until
464          * nothing remains in the hash */
465         if (num) {
466                 mlog(0, "%s: %d lock resources in hash last pass\n",
467                      dlm->name, num);
468                 ret = -EAGAIN;
469         }
470         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
471         return ret;
472 }
473
474 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
475 {
476         int ret;
477
478         spin_lock(&dlm->spinlock);
479         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
480         spin_unlock(&dlm->spinlock);
481
482         return ret;
483 }
484
485 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
486 {
487         /* Yikes, a double spinlock! I need domain_lock for the dlm
488          * state and the dlm spinlock for join state... Sorry! */
489 again:
490         spin_lock(&dlm_domain_lock);
491         spin_lock(&dlm->spinlock);
492
493         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
494                 mlog(0, "Node %d is joining, we wait on it.\n",
495                           dlm->joining_node);
496                 spin_unlock(&dlm->spinlock);
497                 spin_unlock(&dlm_domain_lock);
498
499                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
500                 goto again;
501         }
502
503         dlm->dlm_state = DLM_CTXT_LEAVING;
504         spin_unlock(&dlm->spinlock);
505         spin_unlock(&dlm_domain_lock);
506 }
507
508 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
509 {
510         int node = -1;
511
512         assert_spin_locked(&dlm->spinlock);
513
514         printk(KERN_NOTICE "o2dlm: Nodes in domain %s: ", dlm->name);
515
516         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
517                                      node + 1)) < O2NM_MAX_NODES) {
518                 printk("%d ", node);
519         }
520         printk("\n");
521 }
522
523 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
524                                    void **ret_data)
525 {
526         struct dlm_ctxt *dlm = data;
527         unsigned int node;
528         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
529
530         mlog_entry("%p %u %p", msg, len, data);
531
532         if (!dlm_grab(dlm))
533                 return 0;
534
535         node = exit_msg->node_idx;
536
537         printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s\n", node, dlm->name);
538
539         spin_lock(&dlm->spinlock);
540         clear_bit(node, dlm->domain_map);
541         __dlm_print_nodes(dlm);
542
543         /* notify anything attached to the heartbeat events */
544         dlm_hb_event_notify_attached(dlm, node, 0);
545
546         spin_unlock(&dlm->spinlock);
547
548         dlm_put(dlm);
549
550         return 0;
551 }
552
553 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm,
554                                     unsigned int node)
555 {
556         int status;
557         struct dlm_exit_domain leave_msg;
558
559         mlog(0, "Asking node %u if we can leave the domain %s me = %u\n",
560                   node, dlm->name, dlm->node_num);
561
562         memset(&leave_msg, 0, sizeof(leave_msg));
563         leave_msg.node_idx = dlm->node_num;
564
565         status = o2net_send_message(DLM_EXIT_DOMAIN_MSG, dlm->key,
566                                     &leave_msg, sizeof(leave_msg), node,
567                                     NULL);
568         if (status < 0)
569                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
570                      "node %u\n", status, DLM_EXIT_DOMAIN_MSG, dlm->key, node);
571         mlog(0, "status return %d from o2net_send_message\n", status);
572
573         return status;
574 }
575
576
577 static void dlm_leave_domain(struct dlm_ctxt *dlm)
578 {
579         int node, clear_node, status;
580
581         /* At this point we've migrated away all our locks and won't
582          * accept mastership of new ones. The dlm is responsible for
583          * almost nothing now. We make sure not to confuse any joining
584          * nodes and then commence shutdown procedure. */
585
586         spin_lock(&dlm->spinlock);
587         /* Clear ourselves from the domain map */
588         clear_bit(dlm->node_num, dlm->domain_map);
589         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
590                                      0)) < O2NM_MAX_NODES) {
591                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
592                  * -nodes cannot be added now as the
593                  *   query_join_handlers knows to respond with OK_NO_MAP
594                  * -we catch the right network errors if a node is
595                  *   removed from the map while we're sending him the
596                  *   exit message. */
597                 spin_unlock(&dlm->spinlock);
598
599                 clear_node = 1;
600
601                 status = dlm_send_one_domain_exit(dlm, node);
602                 if (status < 0 &&
603                     status != -ENOPROTOOPT &&
604                     status != -ENOTCONN) {
605                         mlog(ML_NOTICE, "Error %d sending domain exit message "
606                              "to node %d\n", status, node);
607
608                         /* Not sure what to do here but lets sleep for
609                          * a bit in case this was a transient
610                          * error... */
611                         msleep(DLM_DOMAIN_BACKOFF_MS);
612                         clear_node = 0;
613                 }
614
615                 spin_lock(&dlm->spinlock);
616                 /* If we're not clearing the node bit then we intend
617                  * to loop back around to try again. */
618                 if (clear_node)
619                         clear_bit(node, dlm->domain_map);
620         }
621         spin_unlock(&dlm->spinlock);
622 }
623
624 int dlm_joined(struct dlm_ctxt *dlm)
625 {
626         int ret = 0;
627
628         spin_lock(&dlm_domain_lock);
629
630         if (dlm->dlm_state == DLM_CTXT_JOINED)
631                 ret = 1;
632
633         spin_unlock(&dlm_domain_lock);
634
635         return ret;
636 }
637
638 int dlm_shutting_down(struct dlm_ctxt *dlm)
639 {
640         int ret = 0;
641
642         spin_lock(&dlm_domain_lock);
643
644         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
645                 ret = 1;
646
647         spin_unlock(&dlm_domain_lock);
648
649         return ret;
650 }
651
652 void dlm_unregister_domain(struct dlm_ctxt *dlm)
653 {
654         int leave = 0;
655         struct dlm_lock_resource *res;
656
657         spin_lock(&dlm_domain_lock);
658         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
659         BUG_ON(!dlm->num_joins);
660
661         dlm->num_joins--;
662         if (!dlm->num_joins) {
663                 /* We mark it "in shutdown" now so new register
664                  * requests wait until we've completely left the
665                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
666                  * want new domain joins to communicate with us at
667                  * least until we've completed migration of our
668                  * resources. */
669                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
670                 leave = 1;
671         }
672         spin_unlock(&dlm_domain_lock);
673
674         if (leave) {
675                 mlog(0, "shutting down domain %s\n", dlm->name);
676
677                 /* We changed dlm state, notify the thread */
678                 dlm_kick_thread(dlm, NULL);
679
680                 while (dlm_migrate_all_locks(dlm)) {
681                         /* Give dlm_thread time to purge the lockres' */
682                         msleep(500);
683                         mlog(0, "%s: more migration to do\n", dlm->name);
684                 }
685
686                 /* This list should be empty. If not, print remaining lockres */
687                 if (!list_empty(&dlm->tracking_list)) {
688                         mlog(ML_ERROR, "Following lockres' are still on the "
689                              "tracking list:\n");
690                         list_for_each_entry(res, &dlm->tracking_list, tracking)
691                                 dlm_print_one_lock_resource(res);
692                 }
693
694                 dlm_mark_domain_leaving(dlm);
695                 dlm_leave_domain(dlm);
696                 dlm_complete_dlm_shutdown(dlm);
697         }
698         dlm_put(dlm);
699 }
700 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
701
702 static int dlm_query_join_proto_check(char *proto_type, int node,
703                                       struct dlm_protocol_version *ours,
704                                       struct dlm_protocol_version *request)
705 {
706         int rc;
707         struct dlm_protocol_version proto = *request;
708
709         if (!dlm_protocol_compare(ours, &proto)) {
710                 mlog(0,
711                      "node %u wanted to join with %s locking protocol "
712                      "%u.%u, we respond with %u.%u\n",
713                      node, proto_type,
714                      request->pv_major,
715                      request->pv_minor,
716                      proto.pv_major, proto.pv_minor);
717                 request->pv_minor = proto.pv_minor;
718                 rc = 0;
719         } else {
720                 mlog(ML_NOTICE,
721                      "Node %u wanted to join with %s locking "
722                      "protocol %u.%u, but we have %u.%u, disallowing\n",
723                      node, proto_type,
724                      request->pv_major,
725                      request->pv_minor,
726                      ours->pv_major,
727                      ours->pv_minor);
728                 rc = 1;
729         }
730
731         return rc;
732 }
733
734 /*
735  * struct dlm_query_join_packet is made up of four one-byte fields.  They
736  * are effectively in big-endian order already.  However, little-endian
737  * machines swap them before putting the packet on the wire (because
738  * query_join's response is a status, and that status is treated as a u32
739  * on the wire).  Thus, a big-endian and little-endian machines will treat
740  * this structure differently.
741  *
742  * The solution is to have little-endian machines swap the structure when
743  * converting from the structure to the u32 representation.  This will
744  * result in the structure having the correct format on the wire no matter
745  * the host endian format.
746  */
747 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
748                                           u32 *wire)
749 {
750         union dlm_query_join_response response;
751
752         response.packet = *packet;
753         *wire = cpu_to_be32(response.intval);
754 }
755
756 static void dlm_query_join_wire_to_packet(u32 wire,
757                                           struct dlm_query_join_packet *packet)
758 {
759         union dlm_query_join_response response;
760
761         response.intval = cpu_to_be32(wire);
762         *packet = response.packet;
763 }
764
765 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
766                                   void **ret_data)
767 {
768         struct dlm_query_join_request *query;
769         struct dlm_query_join_packet packet = {
770                 .code = JOIN_DISALLOW,
771         };
772         struct dlm_ctxt *dlm = NULL;
773         u32 response;
774         u8 nodenum;
775
776         query = (struct dlm_query_join_request *) msg->buf;
777
778         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
779                   query->domain);
780
781         /*
782          * If heartbeat doesn't consider the node live, tell it
783          * to back off and try again.  This gives heartbeat a chance
784          * to catch up.
785          */
786         if (!o2hb_check_node_heartbeating(query->node_idx)) {
787                 mlog(0, "node %u is not in our live map yet\n",
788                      query->node_idx);
789
790                 packet.code = JOIN_DISALLOW;
791                 goto respond;
792         }
793
794         packet.code = JOIN_OK_NO_MAP;
795
796         spin_lock(&dlm_domain_lock);
797         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
798         if (!dlm)
799                 goto unlock_respond;
800
801         /*
802          * There is a small window where the joining node may not see the
803          * node(s) that just left but still part of the cluster. DISALLOW
804          * join request if joining node has different node map.
805          */
806         nodenum=0;
807         while (nodenum < O2NM_MAX_NODES) {
808                 if (test_bit(nodenum, dlm->domain_map)) {
809                         if (!byte_test_bit(nodenum, query->node_map)) {
810                                 mlog(0, "disallow join as node %u does not "
811                                      "have node %u in its nodemap\n",
812                                      query->node_idx, nodenum);
813                                 packet.code = JOIN_DISALLOW;
814                                 goto unlock_respond;
815                         }
816                 }
817                 nodenum++;
818         }
819
820         /* Once the dlm ctxt is marked as leaving then we don't want
821          * to be put in someone's domain map.
822          * Also, explicitly disallow joining at certain troublesome
823          * times (ie. during recovery). */
824         if (dlm && dlm->dlm_state != DLM_CTXT_LEAVING) {
825                 int bit = query->node_idx;
826                 spin_lock(&dlm->spinlock);
827
828                 if (dlm->dlm_state == DLM_CTXT_NEW &&
829                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
830                         /*If this is a brand new context and we
831                          * haven't started our join process yet, then
832                          * the other node won the race. */
833                         packet.code = JOIN_OK_NO_MAP;
834                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
835                         /* Disallow parallel joins. */
836                         packet.code = JOIN_DISALLOW;
837                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
838                         mlog(0, "node %u trying to join, but recovery "
839                              "is ongoing.\n", bit);
840                         packet.code = JOIN_DISALLOW;
841                 } else if (test_bit(bit, dlm->recovery_map)) {
842                         mlog(0, "node %u trying to join, but it "
843                              "still needs recovery.\n", bit);
844                         packet.code = JOIN_DISALLOW;
845                 } else if (test_bit(bit, dlm->domain_map)) {
846                         mlog(0, "node %u trying to join, but it "
847                              "is still in the domain! needs recovery?\n",
848                              bit);
849                         packet.code = JOIN_DISALLOW;
850                 } else {
851                         /* Alright we're fully a part of this domain
852                          * so we keep some state as to who's joining
853                          * and indicate to him that needs to be fixed
854                          * up. */
855
856                         /* Make sure we speak compatible locking protocols.  */
857                         if (dlm_query_join_proto_check("DLM", bit,
858                                                        &dlm->dlm_locking_proto,
859                                                        &query->dlm_proto)) {
860                                 packet.code = JOIN_PROTOCOL_MISMATCH;
861                         } else if (dlm_query_join_proto_check("fs", bit,
862                                                               &dlm->fs_locking_proto,
863                                                               &query->fs_proto)) {
864                                 packet.code = JOIN_PROTOCOL_MISMATCH;
865                         } else {
866                                 packet.dlm_minor = query->dlm_proto.pv_minor;
867                                 packet.fs_minor = query->fs_proto.pv_minor;
868                                 packet.code = JOIN_OK;
869                                 __dlm_set_joining_node(dlm, query->node_idx);
870                         }
871                 }
872
873                 spin_unlock(&dlm->spinlock);
874         }
875 unlock_respond:
876         spin_unlock(&dlm_domain_lock);
877
878 respond:
879         mlog(0, "We respond with %u\n", packet.code);
880
881         dlm_query_join_packet_to_wire(&packet, &response);
882         return response;
883 }
884
885 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
886                                      void **ret_data)
887 {
888         struct dlm_assert_joined *assert;
889         struct dlm_ctxt *dlm = NULL;
890
891         assert = (struct dlm_assert_joined *) msg->buf;
892
893         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
894                   assert->domain);
895
896         spin_lock(&dlm_domain_lock);
897         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
898         /* XXX should we consider no dlm ctxt an error? */
899         if (dlm) {
900                 spin_lock(&dlm->spinlock);
901
902                 /* Alright, this node has officially joined our
903                  * domain. Set him in the map and clean up our
904                  * leftover join state. */
905                 BUG_ON(dlm->joining_node != assert->node_idx);
906                 set_bit(assert->node_idx, dlm->domain_map);
907                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
908
909                 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s\n",
910                        assert->node_idx, dlm->name);
911                 __dlm_print_nodes(dlm);
912
913                 /* notify anything attached to the heartbeat events */
914                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
915
916                 spin_unlock(&dlm->spinlock);
917         }
918         spin_unlock(&dlm_domain_lock);
919
920         return 0;
921 }
922
923 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
924                                    void **ret_data)
925 {
926         struct dlm_cancel_join *cancel;
927         struct dlm_ctxt *dlm = NULL;
928
929         cancel = (struct dlm_cancel_join *) msg->buf;
930
931         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
932                   cancel->domain);
933
934         spin_lock(&dlm_domain_lock);
935         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
936
937         if (dlm) {
938                 spin_lock(&dlm->spinlock);
939
940                 /* Yikes, this guy wants to cancel his join. No
941                  * problem, we simply cleanup our join state. */
942                 BUG_ON(dlm->joining_node != cancel->node_idx);
943                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
944
945                 spin_unlock(&dlm->spinlock);
946         }
947         spin_unlock(&dlm_domain_lock);
948
949         return 0;
950 }
951
952 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
953                                     unsigned int node)
954 {
955         int status;
956         struct dlm_cancel_join cancel_msg;
957
958         memset(&cancel_msg, 0, sizeof(cancel_msg));
959         cancel_msg.node_idx = dlm->node_num;
960         cancel_msg.name_len = strlen(dlm->name);
961         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
962
963         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
964                                     &cancel_msg, sizeof(cancel_msg), node,
965                                     NULL);
966         if (status < 0) {
967                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
968                      "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
969                      node);
970                 goto bail;
971         }
972
973 bail:
974         return status;
975 }
976
977 /* map_size should be in bytes. */
978 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
979                                  unsigned long *node_map,
980                                  unsigned int map_size)
981 {
982         int status, tmpstat;
983         unsigned int node;
984
985         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
986                          sizeof(unsigned long))) {
987                 mlog(ML_ERROR,
988                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
989                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
990                 return -EINVAL;
991         }
992
993         status = 0;
994         node = -1;
995         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
996                                      node + 1)) < O2NM_MAX_NODES) {
997                 if (node == dlm->node_num)
998                         continue;
999
1000                 tmpstat = dlm_send_one_join_cancel(dlm, node);
1001                 if (tmpstat) {
1002                         mlog(ML_ERROR, "Error return %d cancelling join on "
1003                              "node %d\n", tmpstat, node);
1004                         if (!status)
1005                                 status = tmpstat;
1006                 }
1007         }
1008
1009         if (status)
1010                 mlog_errno(status);
1011         return status;
1012 }
1013
1014 static int dlm_request_join(struct dlm_ctxt *dlm,
1015                             int node,
1016                             enum dlm_query_join_response_code *response)
1017 {
1018         int status;
1019         struct dlm_query_join_request join_msg;
1020         struct dlm_query_join_packet packet;
1021         u32 join_resp;
1022
1023         mlog(0, "querying node %d\n", node);
1024
1025         memset(&join_msg, 0, sizeof(join_msg));
1026         join_msg.node_idx = dlm->node_num;
1027         join_msg.name_len = strlen(dlm->name);
1028         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1029         join_msg.dlm_proto = dlm->dlm_locking_proto;
1030         join_msg.fs_proto = dlm->fs_locking_proto;
1031
1032         /* copy live node map to join message */
1033         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1034
1035         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1036                                     sizeof(join_msg), node, &join_resp);
1037         if (status < 0 && status != -ENOPROTOOPT) {
1038                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1039                      "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1040                      node);
1041                 goto bail;
1042         }
1043         dlm_query_join_wire_to_packet(join_resp, &packet);
1044
1045         /* -ENOPROTOOPT from the net code means the other side isn't
1046             listening for our message type -- that's fine, it means
1047             his dlm isn't up, so we can consider him a 'yes' but not
1048             joined into the domain.  */
1049         if (status == -ENOPROTOOPT) {
1050                 status = 0;
1051                 *response = JOIN_OK_NO_MAP;
1052         } else if (packet.code == JOIN_DISALLOW ||
1053                    packet.code == JOIN_OK_NO_MAP) {
1054                 *response = packet.code;
1055         } else if (packet.code == JOIN_PROTOCOL_MISMATCH) {
1056                 mlog(ML_NOTICE,
1057                      "This node requested DLM locking protocol %u.%u and "
1058                      "filesystem locking protocol %u.%u.  At least one of "
1059                      "the protocol versions on node %d is not compatible, "
1060                      "disconnecting\n",
1061                      dlm->dlm_locking_proto.pv_major,
1062                      dlm->dlm_locking_proto.pv_minor,
1063                      dlm->fs_locking_proto.pv_major,
1064                      dlm->fs_locking_proto.pv_minor,
1065                      node);
1066                 status = -EPROTO;
1067                 *response = packet.code;
1068         } else if (packet.code == JOIN_OK) {
1069                 *response = packet.code;
1070                 /* Use the same locking protocol as the remote node */
1071                 dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1072                 dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1073                 mlog(0,
1074                      "Node %d responds JOIN_OK with DLM locking protocol "
1075                      "%u.%u and fs locking protocol %u.%u\n",
1076                      node,
1077                      dlm->dlm_locking_proto.pv_major,
1078                      dlm->dlm_locking_proto.pv_minor,
1079                      dlm->fs_locking_proto.pv_major,
1080                      dlm->fs_locking_proto.pv_minor);
1081         } else {
1082                 status = -EINVAL;
1083                 mlog(ML_ERROR, "invalid response %d from node %u\n",
1084                      packet.code, node);
1085         }
1086
1087         mlog(0, "status %d, node %d response is %d\n", status, node,
1088              *response);
1089
1090 bail:
1091         return status;
1092 }
1093
1094 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1095                                     unsigned int node)
1096 {
1097         int status;
1098         struct dlm_assert_joined assert_msg;
1099
1100         mlog(0, "Sending join assert to node %u\n", node);
1101
1102         memset(&assert_msg, 0, sizeof(assert_msg));
1103         assert_msg.node_idx = dlm->node_num;
1104         assert_msg.name_len = strlen(dlm->name);
1105         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1106
1107         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1108                                     &assert_msg, sizeof(assert_msg), node,
1109                                     NULL);
1110         if (status < 0)
1111                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1112                      "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1113                      node);
1114
1115         return status;
1116 }
1117
1118 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1119                                   unsigned long *node_map)
1120 {
1121         int status, node, live;
1122
1123         status = 0;
1124         node = -1;
1125         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1126                                      node + 1)) < O2NM_MAX_NODES) {
1127                 if (node == dlm->node_num)
1128                         continue;
1129
1130                 do {
1131                         /* It is very important that this message be
1132                          * received so we spin until either the node
1133                          * has died or it gets the message. */
1134                         status = dlm_send_one_join_assert(dlm, node);
1135
1136                         spin_lock(&dlm->spinlock);
1137                         live = test_bit(node, dlm->live_nodes_map);
1138                         spin_unlock(&dlm->spinlock);
1139
1140                         if (status) {
1141                                 mlog(ML_ERROR, "Error return %d asserting "
1142                                      "join on node %d\n", status, node);
1143
1144                                 /* give us some time between errors... */
1145                                 if (live)
1146                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1147                         }
1148                 } while (status && live);
1149         }
1150 }
1151
1152 struct domain_join_ctxt {
1153         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1154         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1155 };
1156
1157 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1158                                    struct domain_join_ctxt *ctxt,
1159                                    enum dlm_query_join_response_code response)
1160 {
1161         int ret;
1162
1163         if (response == JOIN_DISALLOW) {
1164                 mlog(0, "Latest response of disallow -- should restart\n");
1165                 return 1;
1166         }
1167
1168         spin_lock(&dlm->spinlock);
1169         /* For now, we restart the process if the node maps have
1170          * changed at all */
1171         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1172                      sizeof(dlm->live_nodes_map));
1173         spin_unlock(&dlm->spinlock);
1174
1175         if (ret)
1176                 mlog(0, "Node maps changed -- should restart\n");
1177
1178         return ret;
1179 }
1180
1181 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1182 {
1183         int status = 0, tmpstat, node;
1184         struct domain_join_ctxt *ctxt;
1185         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1186
1187         mlog_entry("%p", dlm);
1188
1189         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1190         if (!ctxt) {
1191                 status = -ENOMEM;
1192                 mlog_errno(status);
1193                 goto bail;
1194         }
1195
1196         /* group sem locking should work for us here -- we're already
1197          * registered for heartbeat events so filling this should be
1198          * atomic wrt getting those handlers called. */
1199         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1200
1201         spin_lock(&dlm->spinlock);
1202         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1203
1204         __dlm_set_joining_node(dlm, dlm->node_num);
1205
1206         spin_unlock(&dlm->spinlock);
1207
1208         node = -1;
1209         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1210                                      node + 1)) < O2NM_MAX_NODES) {
1211                 if (node == dlm->node_num)
1212                         continue;
1213
1214                 status = dlm_request_join(dlm, node, &response);
1215                 if (status < 0) {
1216                         mlog_errno(status);
1217                         goto bail;
1218                 }
1219
1220                 /* Ok, either we got a response or the node doesn't have a
1221                  * dlm up. */
1222                 if (response == JOIN_OK)
1223                         set_bit(node, ctxt->yes_resp_map);
1224
1225                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1226                         status = -EAGAIN;
1227                         goto bail;
1228                 }
1229         }
1230
1231         mlog(0, "Yay, done querying nodes!\n");
1232
1233         /* Yay, everyone agree's we can join the domain. My domain is
1234          * comprised of all nodes who were put in the
1235          * yes_resp_map. Copy that into our domain map and send a join
1236          * assert message to clean up everyone elses state. */
1237         spin_lock(&dlm->spinlock);
1238         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1239                sizeof(ctxt->yes_resp_map));
1240         set_bit(dlm->node_num, dlm->domain_map);
1241         spin_unlock(&dlm->spinlock);
1242
1243         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1244
1245         /* Joined state *must* be set before the joining node
1246          * information, otherwise the query_join handler may read no
1247          * current joiner but a state of NEW and tell joining nodes
1248          * we're not in the domain. */
1249         spin_lock(&dlm_domain_lock);
1250         dlm->dlm_state = DLM_CTXT_JOINED;
1251         dlm->num_joins++;
1252         spin_unlock(&dlm_domain_lock);
1253
1254 bail:
1255         spin_lock(&dlm->spinlock);
1256         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1257         if (!status)
1258                 __dlm_print_nodes(dlm);
1259         spin_unlock(&dlm->spinlock);
1260
1261         if (ctxt) {
1262                 /* Do we need to send a cancel message to any nodes? */
1263                 if (status < 0) {
1264                         tmpstat = dlm_send_join_cancels(dlm,
1265                                                         ctxt->yes_resp_map,
1266                                                         sizeof(ctxt->yes_resp_map));
1267                         if (tmpstat < 0)
1268                                 mlog_errno(tmpstat);
1269                 }
1270                 kfree(ctxt);
1271         }
1272
1273         mlog(0, "returning %d\n", status);
1274         return status;
1275 }
1276
1277 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1278 {
1279         o2hb_unregister_callback(NULL, &dlm->dlm_hb_up);
1280         o2hb_unregister_callback(NULL, &dlm->dlm_hb_down);
1281         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1282 }
1283
1284 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1285 {
1286         int status;
1287
1288         mlog(0, "registering handlers.\n");
1289
1290         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1291                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1292         status = o2hb_register_callback(NULL, &dlm->dlm_hb_down);
1293         if (status)
1294                 goto bail;
1295
1296         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1297                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1298         status = o2hb_register_callback(NULL, &dlm->dlm_hb_up);
1299         if (status)
1300                 goto bail;
1301
1302         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1303                                         sizeof(struct dlm_master_request),
1304                                         dlm_master_request_handler,
1305                                         dlm, NULL, &dlm->dlm_domain_handlers);
1306         if (status)
1307                 goto bail;
1308
1309         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1310                                         sizeof(struct dlm_assert_master),
1311                                         dlm_assert_master_handler,
1312                                         dlm, dlm_assert_master_post_handler,
1313                                         &dlm->dlm_domain_handlers);
1314         if (status)
1315                 goto bail;
1316
1317         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1318                                         sizeof(struct dlm_create_lock),
1319                                         dlm_create_lock_handler,
1320                                         dlm, NULL, &dlm->dlm_domain_handlers);
1321         if (status)
1322                 goto bail;
1323
1324         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1325                                         DLM_CONVERT_LOCK_MAX_LEN,
1326                                         dlm_convert_lock_handler,
1327                                         dlm, NULL, &dlm->dlm_domain_handlers);
1328         if (status)
1329                 goto bail;
1330
1331         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1332                                         DLM_UNLOCK_LOCK_MAX_LEN,
1333                                         dlm_unlock_lock_handler,
1334                                         dlm, NULL, &dlm->dlm_domain_handlers);
1335         if (status)
1336                 goto bail;
1337
1338         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1339                                         DLM_PROXY_AST_MAX_LEN,
1340                                         dlm_proxy_ast_handler,
1341                                         dlm, NULL, &dlm->dlm_domain_handlers);
1342         if (status)
1343                 goto bail;
1344
1345         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1346                                         sizeof(struct dlm_exit_domain),
1347                                         dlm_exit_domain_handler,
1348                                         dlm, NULL, &dlm->dlm_domain_handlers);
1349         if (status)
1350                 goto bail;
1351
1352         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1353                                         sizeof(struct dlm_deref_lockres),
1354                                         dlm_deref_lockres_handler,
1355                                         dlm, NULL, &dlm->dlm_domain_handlers);
1356         if (status)
1357                 goto bail;
1358
1359         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1360                                         sizeof(struct dlm_migrate_request),
1361                                         dlm_migrate_request_handler,
1362                                         dlm, NULL, &dlm->dlm_domain_handlers);
1363         if (status)
1364                 goto bail;
1365
1366         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1367                                         DLM_MIG_LOCKRES_MAX_LEN,
1368                                         dlm_mig_lockres_handler,
1369                                         dlm, NULL, &dlm->dlm_domain_handlers);
1370         if (status)
1371                 goto bail;
1372
1373         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1374                                         sizeof(struct dlm_master_requery),
1375                                         dlm_master_requery_handler,
1376                                         dlm, NULL, &dlm->dlm_domain_handlers);
1377         if (status)
1378                 goto bail;
1379
1380         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1381                                         sizeof(struct dlm_lock_request),
1382                                         dlm_request_all_locks_handler,
1383                                         dlm, NULL, &dlm->dlm_domain_handlers);
1384         if (status)
1385                 goto bail;
1386
1387         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1388                                         sizeof(struct dlm_reco_data_done),
1389                                         dlm_reco_data_done_handler,
1390                                         dlm, NULL, &dlm->dlm_domain_handlers);
1391         if (status)
1392                 goto bail;
1393
1394         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1395                                         sizeof(struct dlm_begin_reco),
1396                                         dlm_begin_reco_handler,
1397                                         dlm, NULL, &dlm->dlm_domain_handlers);
1398         if (status)
1399                 goto bail;
1400
1401         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1402                                         sizeof(struct dlm_finalize_reco),
1403                                         dlm_finalize_reco_handler,
1404                                         dlm, NULL, &dlm->dlm_domain_handlers);
1405         if (status)
1406                 goto bail;
1407
1408 bail:
1409         if (status)
1410                 dlm_unregister_domain_handlers(dlm);
1411
1412         return status;
1413 }
1414
1415 static int dlm_join_domain(struct dlm_ctxt *dlm)
1416 {
1417         int status;
1418         unsigned int backoff;
1419         unsigned int total_backoff = 0;
1420
1421         BUG_ON(!dlm);
1422
1423         mlog(0, "Join domain %s\n", dlm->name);
1424
1425         status = dlm_register_domain_handlers(dlm);
1426         if (status) {
1427                 mlog_errno(status);
1428                 goto bail;
1429         }
1430
1431         status = dlm_debug_init(dlm);
1432         if (status < 0) {
1433                 mlog_errno(status);
1434                 goto bail;
1435         }
1436
1437         status = dlm_launch_thread(dlm);
1438         if (status < 0) {
1439                 mlog_errno(status);
1440                 goto bail;
1441         }
1442
1443         status = dlm_launch_recovery_thread(dlm);
1444         if (status < 0) {
1445                 mlog_errno(status);
1446                 goto bail;
1447         }
1448
1449         dlm->dlm_worker = create_singlethread_workqueue("dlm_wq");
1450         if (!dlm->dlm_worker) {
1451                 status = -ENOMEM;
1452                 mlog_errno(status);
1453                 goto bail;
1454         }
1455
1456         do {
1457                 status = dlm_try_to_join_domain(dlm);
1458
1459                 /* If we're racing another node to the join, then we
1460                  * need to back off temporarily and let them
1461                  * complete. */
1462 #define DLM_JOIN_TIMEOUT_MSECS  90000
1463                 if (status == -EAGAIN) {
1464                         if (signal_pending(current)) {
1465                                 status = -ERESTARTSYS;
1466                                 goto bail;
1467                         }
1468
1469                         if (total_backoff >
1470                             msecs_to_jiffies(DLM_JOIN_TIMEOUT_MSECS)) {
1471                                 status = -ERESTARTSYS;
1472                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1473                                      "%s after %u msecs\n", dlm->name,
1474                                      jiffies_to_msecs(total_backoff));
1475                                 goto bail;
1476                         }
1477
1478                         /*
1479                          * <chip> After you!
1480                          * <dale> No, after you!
1481                          * <chip> I insist!
1482                          * <dale> But you first!
1483                          * ...
1484                          */
1485                         backoff = (unsigned int)(jiffies & 0x3);
1486                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1487                         total_backoff += backoff;
1488                         mlog(0, "backoff %d\n", backoff);
1489                         msleep(backoff);
1490                 }
1491         } while (status == -EAGAIN);
1492
1493         if (status < 0) {
1494                 mlog_errno(status);
1495                 goto bail;
1496         }
1497
1498         status = 0;
1499 bail:
1500         wake_up(&dlm_domain_events);
1501
1502         if (status) {
1503                 dlm_unregister_domain_handlers(dlm);
1504                 dlm_debug_shutdown(dlm);
1505                 dlm_complete_thread(dlm);
1506                 dlm_complete_recovery_thread(dlm);
1507                 dlm_destroy_dlm_worker(dlm);
1508         }
1509
1510         return status;
1511 }
1512
1513 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1514                                 u32 key)
1515 {
1516         int i;
1517         int ret;
1518         struct dlm_ctxt *dlm = NULL;
1519
1520         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1521         if (!dlm) {
1522                 mlog_errno(-ENOMEM);
1523                 goto leave;
1524         }
1525
1526         dlm->name = kstrdup(domain, GFP_KERNEL);
1527         if (dlm->name == NULL) {
1528                 mlog_errno(-ENOMEM);
1529                 kfree(dlm);
1530                 dlm = NULL;
1531                 goto leave;
1532         }
1533
1534         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1535         if (!dlm->lockres_hash) {
1536                 mlog_errno(-ENOMEM);
1537                 kfree(dlm->name);
1538                 kfree(dlm);
1539                 dlm = NULL;
1540                 goto leave;
1541         }
1542
1543         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1544                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1545
1546         dlm->master_hash = (struct hlist_head **)
1547                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
1548         if (!dlm->master_hash) {
1549                 mlog_errno(-ENOMEM);
1550                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1551                 kfree(dlm->name);
1552                 kfree(dlm);
1553                 dlm = NULL;
1554                 goto leave;
1555         }
1556
1557         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1558                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
1559
1560         dlm->key = key;
1561         dlm->node_num = o2nm_this_node();
1562
1563         ret = dlm_create_debugfs_subroot(dlm);
1564         if (ret < 0) {
1565                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
1566                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
1567                 kfree(dlm->name);
1568                 kfree(dlm);
1569                 dlm = NULL;
1570                 goto leave;
1571         }
1572
1573         spin_lock_init(&dlm->spinlock);
1574         spin_lock_init(&dlm->master_lock);
1575         spin_lock_init(&dlm->ast_lock);
1576         spin_lock_init(&dlm->track_lock);
1577         INIT_LIST_HEAD(&dlm->list);
1578         INIT_LIST_HEAD(&dlm->dirty_list);
1579         INIT_LIST_HEAD(&dlm->reco.resources);
1580         INIT_LIST_HEAD(&dlm->reco.received);
1581         INIT_LIST_HEAD(&dlm->reco.node_data);
1582         INIT_LIST_HEAD(&dlm->purge_list);
1583         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
1584         INIT_LIST_HEAD(&dlm->tracking_list);
1585         dlm->reco.state = 0;
1586
1587         INIT_LIST_HEAD(&dlm->pending_asts);
1588         INIT_LIST_HEAD(&dlm->pending_basts);
1589
1590         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
1591                   dlm->recovery_map, &(dlm->recovery_map[0]));
1592
1593         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
1594         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
1595         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
1596
1597         dlm->dlm_thread_task = NULL;
1598         dlm->dlm_reco_thread_task = NULL;
1599         dlm->dlm_worker = NULL;
1600         init_waitqueue_head(&dlm->dlm_thread_wq);
1601         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
1602         init_waitqueue_head(&dlm->reco.event);
1603         init_waitqueue_head(&dlm->ast_wq);
1604         init_waitqueue_head(&dlm->migration_wq);
1605         INIT_LIST_HEAD(&dlm->mle_hb_events);
1606
1607         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
1608         init_waitqueue_head(&dlm->dlm_join_events);
1609
1610         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
1611         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
1612
1613         atomic_set(&dlm->res_tot_count, 0);
1614         atomic_set(&dlm->res_cur_count, 0);
1615         for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
1616                 atomic_set(&dlm->mle_tot_count[i], 0);
1617                 atomic_set(&dlm->mle_cur_count[i], 0);
1618         }
1619
1620         spin_lock_init(&dlm->work_lock);
1621         INIT_LIST_HEAD(&dlm->work_list);
1622         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
1623
1624         kref_init(&dlm->dlm_refs);
1625         dlm->dlm_state = DLM_CTXT_NEW;
1626
1627         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
1628
1629         mlog(0, "context init: refcount %u\n",
1630                   atomic_read(&dlm->dlm_refs.refcount));
1631
1632 leave:
1633         return dlm;
1634 }
1635
1636 /*
1637  * Compare a requested locking protocol version against the current one.
1638  *
1639  * If the major numbers are different, they are incompatible.
1640  * If the current minor is greater than the request, they are incompatible.
1641  * If the current minor is less than or equal to the request, they are
1642  * compatible, and the requester should run at the current minor version.
1643  */
1644 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
1645                                 struct dlm_protocol_version *request)
1646 {
1647         if (existing->pv_major != request->pv_major)
1648                 return 1;
1649
1650         if (existing->pv_minor > request->pv_minor)
1651                 return 1;
1652
1653         if (existing->pv_minor < request->pv_minor)
1654                 request->pv_minor = existing->pv_minor;
1655
1656         return 0;
1657 }
1658
1659 /*
1660  * dlm_register_domain: one-time setup per "domain".
1661  *
1662  * The filesystem passes in the requested locking version via proto.
1663  * If registration was successful, proto will contain the negotiated
1664  * locking protocol.
1665  */
1666 struct dlm_ctxt * dlm_register_domain(const char *domain,
1667                                u32 key,
1668                                struct dlm_protocol_version *fs_proto)
1669 {
1670         int ret;
1671         struct dlm_ctxt *dlm = NULL;
1672         struct dlm_ctxt *new_ctxt = NULL;
1673
1674         if (strlen(domain) > O2NM_MAX_NAME_LEN) {
1675                 ret = -ENAMETOOLONG;
1676                 mlog(ML_ERROR, "domain name length too long\n");
1677                 goto leave;
1678         }
1679
1680         if (!o2hb_check_local_node_heartbeating()) {
1681                 mlog(ML_ERROR, "the local node has not been configured, or is "
1682                      "not heartbeating\n");
1683                 ret = -EPROTO;
1684                 goto leave;
1685         }
1686
1687         mlog(0, "register called for domain \"%s\"\n", domain);
1688
1689 retry:
1690         dlm = NULL;
1691         if (signal_pending(current)) {
1692                 ret = -ERESTARTSYS;
1693                 mlog_errno(ret);
1694                 goto leave;
1695         }
1696
1697         spin_lock(&dlm_domain_lock);
1698
1699         dlm = __dlm_lookup_domain(domain);
1700         if (dlm) {
1701                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
1702                         spin_unlock(&dlm_domain_lock);
1703
1704                         mlog(0, "This ctxt is not joined yet!\n");
1705                         wait_event_interruptible(dlm_domain_events,
1706                                                  dlm_wait_on_domain_helper(
1707                                                          domain));
1708                         goto retry;
1709                 }
1710
1711                 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
1712                         mlog(ML_ERROR,
1713                              "Requested locking protocol version is not "
1714                              "compatible with already registered domain "
1715                              "\"%s\"\n", domain);
1716                         ret = -EPROTO;
1717                         goto leave;
1718                 }
1719
1720                 __dlm_get(dlm);
1721                 dlm->num_joins++;
1722
1723                 spin_unlock(&dlm_domain_lock);
1724
1725                 ret = 0;
1726                 goto leave;
1727         }
1728
1729         /* doesn't exist */
1730         if (!new_ctxt) {
1731                 spin_unlock(&dlm_domain_lock);
1732
1733                 new_ctxt = dlm_alloc_ctxt(domain, key);
1734                 if (new_ctxt)
1735                         goto retry;
1736
1737                 ret = -ENOMEM;
1738                 mlog_errno(ret);
1739                 goto leave;
1740         }
1741
1742         /* a little variable switch-a-roo here... */
1743         dlm = new_ctxt;
1744         new_ctxt = NULL;
1745
1746         /* add the new domain */
1747         list_add_tail(&dlm->list, &dlm_domains);
1748         spin_unlock(&dlm_domain_lock);
1749
1750         /*
1751          * Pass the locking protocol version into the join.  If the join
1752          * succeeds, it will have the negotiated protocol set.
1753          */
1754         dlm->dlm_locking_proto = dlm_protocol;
1755         dlm->fs_locking_proto = *fs_proto;
1756
1757         ret = dlm_join_domain(dlm);
1758         if (ret) {
1759                 mlog_errno(ret);
1760                 dlm_put(dlm);
1761                 goto leave;
1762         }
1763
1764         /* Tell the caller what locking protocol we negotiated */
1765         *fs_proto = dlm->fs_locking_proto;
1766
1767         ret = 0;
1768 leave:
1769         if (new_ctxt)
1770                 dlm_free_ctxt_mem(new_ctxt);
1771
1772         if (ret < 0)
1773                 dlm = ERR_PTR(ret);
1774
1775         return dlm;
1776 }
1777 EXPORT_SYMBOL_GPL(dlm_register_domain);
1778
1779 static LIST_HEAD(dlm_join_handlers);
1780
1781 static void dlm_unregister_net_handlers(void)
1782 {
1783         o2net_unregister_handler_list(&dlm_join_handlers);
1784 }
1785
1786 static int dlm_register_net_handlers(void)
1787 {
1788         int status = 0;
1789
1790         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1791                                         sizeof(struct dlm_query_join_request),
1792                                         dlm_query_join_handler,
1793                                         NULL, NULL, &dlm_join_handlers);
1794         if (status)
1795                 goto bail;
1796
1797         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1798                                         sizeof(struct dlm_assert_joined),
1799                                         dlm_assert_joined_handler,
1800                                         NULL, NULL, &dlm_join_handlers);
1801         if (status)
1802                 goto bail;
1803
1804         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1805                                         sizeof(struct dlm_cancel_join),
1806                                         dlm_cancel_join_handler,
1807                                         NULL, NULL, &dlm_join_handlers);
1808
1809 bail:
1810         if (status < 0)
1811                 dlm_unregister_net_handlers();
1812
1813         return status;
1814 }
1815
1816 /* Domain eviction callback handling.
1817  *
1818  * The file system requires notification of node death *before* the
1819  * dlm completes it's recovery work, otherwise it may be able to
1820  * acquire locks on resources requiring recovery. Since the dlm can
1821  * evict a node from it's domain *before* heartbeat fires, a similar
1822  * mechanism is required. */
1823
1824 /* Eviction is not expected to happen often, so a per-domain lock is
1825  * not necessary. Eviction callbacks are allowed to sleep for short
1826  * periods of time. */
1827 static DECLARE_RWSEM(dlm_callback_sem);
1828
1829 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
1830                                         int node_num)
1831 {
1832         struct list_head *iter;
1833         struct dlm_eviction_cb *cb;
1834
1835         down_read(&dlm_callback_sem);
1836         list_for_each(iter, &dlm->dlm_eviction_callbacks) {
1837                 cb = list_entry(iter, struct dlm_eviction_cb, ec_item);
1838
1839                 cb->ec_func(node_num, cb->ec_data);
1840         }
1841         up_read(&dlm_callback_sem);
1842 }
1843
1844 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
1845                            dlm_eviction_func *f,
1846                            void *data)
1847 {
1848         INIT_LIST_HEAD(&cb->ec_item);
1849         cb->ec_func = f;
1850         cb->ec_data = data;
1851 }
1852 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
1853
1854 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
1855                               struct dlm_eviction_cb *cb)
1856 {
1857         down_write(&dlm_callback_sem);
1858         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
1859         up_write(&dlm_callback_sem);
1860 }
1861 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
1862
1863 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
1864 {
1865         down_write(&dlm_callback_sem);
1866         list_del_init(&cb->ec_item);
1867         up_write(&dlm_callback_sem);
1868 }
1869 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
1870
1871 static int __init dlm_init(void)
1872 {
1873         int status;
1874
1875         dlm_print_version();
1876
1877         status = dlm_init_mle_cache();
1878         if (status) {
1879                 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
1880                 goto error;
1881         }
1882
1883         status = dlm_init_master_caches();
1884         if (status) {
1885                 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
1886                      "o2dlm_lockname slabcaches\n");
1887                 goto error;
1888         }
1889
1890         status = dlm_init_lock_cache();
1891         if (status) {
1892                 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
1893                 goto error;
1894         }
1895
1896         status = dlm_register_net_handlers();
1897         if (status) {
1898                 mlog(ML_ERROR, "Unable to register network handlers\n");
1899                 goto error;
1900         }
1901
1902         status = dlm_create_debugfs_root();
1903         if (status)
1904                 goto error;
1905
1906         return 0;
1907 error:
1908         dlm_unregister_net_handlers();
1909         dlm_destroy_lock_cache();
1910         dlm_destroy_master_caches();
1911         dlm_destroy_mle_cache();
1912         return -1;
1913 }
1914
1915 static void __exit dlm_exit (void)
1916 {
1917         dlm_destroy_debugfs_root();
1918         dlm_unregister_net_handlers();
1919         dlm_destroy_lock_cache();
1920         dlm_destroy_master_caches();
1921         dlm_destroy_mle_cache();
1922 }
1923
1924 MODULE_AUTHOR("Oracle");
1925 MODULE_LICENSE("GPL");
1926
1927 module_init(dlm_init);
1928 module_exit(dlm_exit);