~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

TOMOYO Linux Cross Reference
Linux/fs/ocfs2/dlm/dlmdomain.c

Version: ~ [ linux-5.11-rc3 ] ~ [ linux-5.10.7 ] ~ [ linux-5.9.16 ] ~ [ linux-5.8.18 ] ~ [ linux-5.7.19 ] ~ [ linux-5.6.19 ] ~ [ linux-5.5.19 ] ~ [ linux-5.4.89 ] ~ [ linux-5.3.18 ] ~ [ linux-5.2.21 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.167 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.215 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.251 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.251 ] ~ [ linux-4.3.6 ] ~ [ linux-4.2.8 ] ~ [ linux-4.1.52 ] ~ [ linux-4.0.9 ] ~ [ linux-3.19.8 ] ~ [ linux-3.18.140 ] ~ [ linux-3.17.8 ] ~ [ linux-3.16.85 ] ~ [ linux-3.15.10 ] ~ [ linux-3.14.79 ] ~ [ linux-3.13.11 ] ~ [ linux-3.12.74 ] ~ [ linux-3.11.10 ] ~ [ linux-3.10.108 ] ~ [ linux-2.6.32.71 ] ~ [ linux-2.6.0 ] ~ [ linux-2.4.37.11 ] ~ [ unix-v6-master ] ~ [ ccs-tools-1.8.5 ] ~ [ policy-sample ] ~
Architecture: ~ [ i386 ] ~ [ alpha ] ~ [ m68k ] ~ [ mips ] ~ [ ppc ] ~ [ sparc ] ~ [ sparc64 ] ~

  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 #include <linux/sched/signal.h>
 37 
 38 #include "cluster/heartbeat.h"
 39 #include "cluster/nodemanager.h"
 40 #include "cluster/tcp.h"
 41 
 42 #include "dlmapi.h"
 43 #include "dlmcommon.h"
 44 #include "dlmdomain.h"
 45 #include "dlmdebug.h"
 46 
 47 #define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_DOMAIN)
 48 #include "cluster/masklog.h"
 49 
 50 /*
 51  * ocfs2 node maps are array of long int, which limits to send them freely
 52  * across the wire due to endianness issues. To workaround this, we convert
 53  * long ints to byte arrays. Following 3 routines are helper functions to
 54  * set/test/copy bits within those array of bytes
 55  */
 56 static inline void byte_set_bit(u8 nr, u8 map[])
 57 {
 58         map[nr >> 3] |= (1UL << (nr & 7));
 59 }
 60 
 61 static inline int byte_test_bit(u8 nr, u8 map[])
 62 {
 63         return ((1UL << (nr & 7)) & (map[nr >> 3])) != 0;
 64 }
 65 
 66 static inline void byte_copymap(u8 dmap[], unsigned long smap[],
 67                         unsigned int sz)
 68 {
 69         unsigned int nn;
 70 
 71         if (!sz)
 72                 return;
 73 
 74         memset(dmap, 0, ((sz + 7) >> 3));
 75         for (nn = 0 ; nn < sz; nn++)
 76                 if (test_bit(nn, smap))
 77                         byte_set_bit(nn, dmap);
 78 }
 79 
 80 static void dlm_free_pagevec(void **vec, int pages)
 81 {
 82         while (pages--)
 83                 free_page((unsigned long)vec[pages]);
 84         kfree(vec);
 85 }
 86 
 87 static void **dlm_alloc_pagevec(int pages)
 88 {
 89         void **vec = kmalloc_array(pages, sizeof(void *), GFP_KERNEL);
 90         int i;
 91 
 92         if (!vec)
 93                 return NULL;
 94 
 95         for (i = 0; i < pages; i++)
 96                 if (!(vec[i] = (void *)__get_free_page(GFP_KERNEL)))
 97                         goto out_free;
 98 
 99         mlog(0, "Allocated DLM hash pagevec; %d pages (%lu expected), %lu buckets per page\n",
100              pages, (unsigned long)DLM_HASH_PAGES,
101              (unsigned long)DLM_BUCKETS_PER_PAGE);
102         return vec;
103 out_free:
104         dlm_free_pagevec(vec, i);
105         return NULL;
106 }
107 
108 /*
109  *
110  * spinlock lock ordering: if multiple locks are needed, obey this ordering:
111  *    dlm_domain_lock
112  *    struct dlm_ctxt->spinlock
113  *    struct dlm_lock_resource->spinlock
114  *    struct dlm_ctxt->master_lock
115  *    struct dlm_ctxt->ast_lock
116  *    dlm_master_list_entry->spinlock
117  *    dlm_lock->spinlock
118  *
119  */
120 
121 DEFINE_SPINLOCK(dlm_domain_lock);
122 LIST_HEAD(dlm_domains);
123 static DECLARE_WAIT_QUEUE_HEAD(dlm_domain_events);
124 
125 /*
126  * The supported protocol version for DLM communication.  Running domains
127  * will have a negotiated version with the same major number and a minor
128  * number equal or smaller.  The dlm_ctxt->dlm_locking_proto field should
129  * be used to determine what a running domain is actually using.
130  *
131  * New in version 1.1:
132  *      - Message DLM_QUERY_REGION added to support global heartbeat
133  *      - Message DLM_QUERY_NODEINFO added to allow online node removes
134  * New in version 1.2:
135  *      - Message DLM_BEGIN_EXIT_DOMAIN_MSG added to mark start of exit domain
136  * New in version 1.3:
137  *      - Message DLM_DEREF_LOCKRES_DONE added to inform non-master that the
138  *        refmap is cleared
139  */
140 static const struct dlm_protocol_version dlm_protocol = {
141         .pv_major = 1,
142         .pv_minor = 3,
143 };
144 
145 #define DLM_DOMAIN_BACKOFF_MS 200
146 
147 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
148                                   void **ret_data);
149 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
150                                      void **ret_data);
151 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
152                                    void **ret_data);
153 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
154                                     void *data, void **ret_data);
155 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
156                                    void **ret_data);
157 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
158                                 struct dlm_protocol_version *request);
159 
160 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm);
161 
162 void __dlm_unhash_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
163 {
164         if (hlist_unhashed(&res->hash_node))
165                 return;
166 
167         mlog(0, "%s: Unhash res %.*s\n", dlm->name, res->lockname.len,
168              res->lockname.name);
169         hlist_del_init(&res->hash_node);
170         dlm_lockres_put(res);
171 }
172 
173 void __dlm_insert_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
174 {
175         struct hlist_head *bucket;
176 
177         assert_spin_locked(&dlm->spinlock);
178 
179         bucket = dlm_lockres_hash(dlm, res->lockname.hash);
180 
181         /* get a reference for our hashtable */
182         dlm_lockres_get(res);
183 
184         hlist_add_head(&res->hash_node, bucket);
185 
186         mlog(0, "%s: Hash res %.*s\n", dlm->name, res->lockname.len,
187              res->lockname.name);
188 }
189 
190 struct dlm_lock_resource * __dlm_lookup_lockres_full(struct dlm_ctxt *dlm,
191                                                      const char *name,
192                                                      unsigned int len,
193                                                      unsigned int hash)
194 {
195         struct hlist_head *bucket;
196         struct dlm_lock_resource *res;
197 
198         mlog(0, "%.*s\n", len, name);
199 
200         assert_spin_locked(&dlm->spinlock);
201 
202         bucket = dlm_lockres_hash(dlm, hash);
203 
204         hlist_for_each_entry(res, bucket, hash_node) {
205                 if (res->lockname.name[0] != name[0])
206                         continue;
207                 if (unlikely(res->lockname.len != len))
208                         continue;
209                 if (memcmp(res->lockname.name + 1, name + 1, len - 1))
210                         continue;
211                 dlm_lockres_get(res);
212                 return res;
213         }
214         return NULL;
215 }
216 
217 /* intended to be called by functions which do not care about lock
218  * resources which are being purged (most net _handler functions).
219  * this will return NULL for any lock resource which is found but
220  * currently in the process of dropping its mastery reference.
221  * use __dlm_lookup_lockres_full when you need the lock resource
222  * regardless (e.g. dlm_get_lock_resource) */
223 struct dlm_lock_resource * __dlm_lookup_lockres(struct dlm_ctxt *dlm,
224                                                 const char *name,
225                                                 unsigned int len,
226                                                 unsigned int hash)
227 {
228         struct dlm_lock_resource *res = NULL;
229 
230         mlog(0, "%.*s\n", len, name);
231 
232         assert_spin_locked(&dlm->spinlock);
233 
234         res = __dlm_lookup_lockres_full(dlm, name, len, hash);
235         if (res) {
236                 spin_lock(&res->spinlock);
237                 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
238                         spin_unlock(&res->spinlock);
239                         dlm_lockres_put(res);
240                         return NULL;
241                 }
242                 spin_unlock(&res->spinlock);
243         }
244 
245         return res;
246 }
247 
248 struct dlm_lock_resource * dlm_lookup_lockres(struct dlm_ctxt *dlm,
249                                     const char *name,
250                                     unsigned int len)
251 {
252         struct dlm_lock_resource *res;
253         unsigned int hash = dlm_lockid_hash(name, len);
254 
255         spin_lock(&dlm->spinlock);
256         res = __dlm_lookup_lockres(dlm, name, len, hash);
257         spin_unlock(&dlm->spinlock);
258         return res;
259 }
260 
261 static struct dlm_ctxt * __dlm_lookup_domain_full(const char *domain, int len)
262 {
263         struct dlm_ctxt *tmp;
264 
265         assert_spin_locked(&dlm_domain_lock);
266 
267         /* tmp->name here is always NULL terminated,
268          * but domain may not be! */
269         list_for_each_entry(tmp, &dlm_domains, list) {
270                 if (strlen(tmp->name) == len &&
271                     memcmp(tmp->name, domain, len)==0)
272                         return tmp;
273         }
274 
275         return NULL;
276 }
277 
278 /* For null terminated domain strings ONLY */
279 static struct dlm_ctxt * __dlm_lookup_domain(const char *domain)
280 {
281         assert_spin_locked(&dlm_domain_lock);
282 
283         return __dlm_lookup_domain_full(domain, strlen(domain));
284 }
285 
286 
287 /* returns true on one of two conditions:
288  * 1) the domain does not exist
289  * 2) the domain exists and it's state is "joined" */
290 static int dlm_wait_on_domain_helper(const char *domain)
291 {
292         int ret = 0;
293         struct dlm_ctxt *tmp = NULL;
294 
295         spin_lock(&dlm_domain_lock);
296 
297         tmp = __dlm_lookup_domain(domain);
298         if (!tmp)
299                 ret = 1;
300         else if (tmp->dlm_state == DLM_CTXT_JOINED)
301                 ret = 1;
302 
303         spin_unlock(&dlm_domain_lock);
304         return ret;
305 }
306 
307 static void dlm_free_ctxt_mem(struct dlm_ctxt *dlm)
308 {
309         dlm_destroy_debugfs_subroot(dlm);
310 
311         if (dlm->lockres_hash)
312                 dlm_free_pagevec((void **)dlm->lockres_hash, DLM_HASH_PAGES);
313 
314         if (dlm->master_hash)
315                 dlm_free_pagevec((void **)dlm->master_hash, DLM_HASH_PAGES);
316 
317         kfree(dlm->name);
318         kfree(dlm);
319 }
320 
321 /* A little strange - this function will be called while holding
322  * dlm_domain_lock and is expected to be holding it on the way out. We
323  * will however drop and reacquire it multiple times */
324 static void dlm_ctxt_release(struct kref *kref)
325 {
326         struct dlm_ctxt *dlm;
327 
328         dlm = container_of(kref, struct dlm_ctxt, dlm_refs);
329 
330         BUG_ON(dlm->num_joins);
331         BUG_ON(dlm->dlm_state == DLM_CTXT_JOINED);
332 
333         /* we may still be in the list if we hit an error during join. */
334         list_del_init(&dlm->list);
335 
336         spin_unlock(&dlm_domain_lock);
337 
338         mlog(0, "freeing memory from domain %s\n", dlm->name);
339 
340         wake_up(&dlm_domain_events);
341 
342         dlm_free_ctxt_mem(dlm);
343 
344         spin_lock(&dlm_domain_lock);
345 }
346 
347 void dlm_put(struct dlm_ctxt *dlm)
348 {
349         spin_lock(&dlm_domain_lock);
350         kref_put(&dlm->dlm_refs, dlm_ctxt_release);
351         spin_unlock(&dlm_domain_lock);
352 }
353 
354 static void __dlm_get(struct dlm_ctxt *dlm)
355 {
356         kref_get(&dlm->dlm_refs);
357 }
358 
359 /* given a questionable reference to a dlm object, gets a reference if
360  * it can find it in the list, otherwise returns NULL in which case
361  * you shouldn't trust your pointer. */
362 struct dlm_ctxt *dlm_grab(struct dlm_ctxt *dlm)
363 {
364         struct dlm_ctxt *target;
365         struct dlm_ctxt *ret = NULL;
366 
367         spin_lock(&dlm_domain_lock);
368 
369         list_for_each_entry(target, &dlm_domains, list) {
370                 if (target == dlm) {
371                         __dlm_get(target);
372                         ret = target;
373                         break;
374                 }
375         }
376 
377         spin_unlock(&dlm_domain_lock);
378 
379         return ret;
380 }
381 
382 int dlm_domain_fully_joined(struct dlm_ctxt *dlm)
383 {
384         int ret;
385 
386         spin_lock(&dlm_domain_lock);
387         ret = (dlm->dlm_state == DLM_CTXT_JOINED) ||
388                 (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN);
389         spin_unlock(&dlm_domain_lock);
390 
391         return ret;
392 }
393 
394 static void dlm_destroy_dlm_worker(struct dlm_ctxt *dlm)
395 {
396         if (dlm->dlm_worker) {
397                 destroy_workqueue(dlm->dlm_worker);
398                 dlm->dlm_worker = NULL;
399         }
400 }
401 
402 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
403 {
404         dlm_unregister_domain_handlers(dlm);
405         dlm_debug_shutdown(dlm);
406         dlm_complete_thread(dlm);
407         dlm_complete_recovery_thread(dlm);
408         dlm_destroy_dlm_worker(dlm);
409 
410         /* We've left the domain. Now we can take ourselves out of the
411          * list and allow the kref stuff to help us free the
412          * memory. */
413         spin_lock(&dlm_domain_lock);
414         list_del_init(&dlm->list);
415         spin_unlock(&dlm_domain_lock);
416 
417         /* Wake up anyone waiting for us to remove this domain */
418         wake_up(&dlm_domain_events);
419 }
420 
421 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
422 {
423         int i, num, n, ret = 0;
424         struct dlm_lock_resource *res;
425         struct hlist_node *iter;
426         struct hlist_head *bucket;
427         int dropped;
428 
429         mlog(0, "Migrating locks from domain %s\n", dlm->name);
430 
431         num = 0;
432         spin_lock(&dlm->spinlock);
433         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
434 redo_bucket:
435                 n = 0;
436                 bucket = dlm_lockres_hash(dlm, i);
437                 iter = bucket->first;
438                 while (iter) {
439                         n++;
440                         res = hlist_entry(iter, struct dlm_lock_resource,
441                                           hash_node);
442                         dlm_lockres_get(res);
443                         /* migrate, if necessary.  this will drop the dlm
444                          * spinlock and retake it if it does migration. */
445                         dropped = dlm_empty_lockres(dlm, res);
446 
447                         spin_lock(&res->spinlock);
448                         if (dropped)
449                                 __dlm_lockres_calc_usage(dlm, res);
450                         else
451                                 iter = res->hash_node.next;
452                         spin_unlock(&res->spinlock);
453 
454                         dlm_lockres_put(res);
455 
456                         if (dropped) {
457                                 cond_resched_lock(&dlm->spinlock);
458                                 goto redo_bucket;
459                         }
460                 }
461                 cond_resched_lock(&dlm->spinlock);
462                 num += n;
463         }
464 
465         if (!num) {
466                 if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
467                         mlog(0, "%s: perhaps there are more lock resources "
468                              "need to be migrated after dlm recovery\n", dlm->name);
469                         ret = -EAGAIN;
470                 } else {
471                         mlog(0, "%s: we won't do dlm recovery after migrating "
472                              "all lock resources\n", dlm->name);
473                         dlm->migrate_done = 1;
474                 }
475         }
476 
477         spin_unlock(&dlm->spinlock);
478         wake_up(&dlm->dlm_thread_wq);
479 
480         /* let the dlm thread take care of purging, keep scanning until
481          * nothing remains in the hash */
482         if (num) {
483                 mlog(0, "%s: %d lock resources in hash last pass\n",
484                      dlm->name, num);
485                 ret = -EAGAIN;
486         }
487         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
488         return ret;
489 }
490 
491 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
492 {
493         int ret;
494 
495         spin_lock(&dlm->spinlock);
496         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
497         spin_unlock(&dlm->spinlock);
498 
499         return ret;
500 }
501 
502 static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
503                                          void *data, void **ret_data)
504 {
505         struct dlm_ctxt *dlm = data;
506         unsigned int node;
507         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
508 
509         if (!dlm_grab(dlm))
510                 return 0;
511 
512         node = exit_msg->node_idx;
513         mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
514 
515         spin_lock(&dlm->spinlock);
516         set_bit(node, dlm->exit_domain_map);
517         spin_unlock(&dlm->spinlock);
518 
519         dlm_put(dlm);
520 
521         return 0;
522 }
523 
524 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
525 {
526         /* Yikes, a double spinlock! I need domain_lock for the dlm
527          * state and the dlm spinlock for join state... Sorry! */
528 again:
529         spin_lock(&dlm_domain_lock);
530         spin_lock(&dlm->spinlock);
531 
532         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
533                 mlog(0, "Node %d is joining, we wait on it.\n",
534                           dlm->joining_node);
535                 spin_unlock(&dlm->spinlock);
536                 spin_unlock(&dlm_domain_lock);
537 
538                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
539                 goto again;
540         }
541 
542         dlm->dlm_state = DLM_CTXT_LEAVING;
543         spin_unlock(&dlm->spinlock);
544         spin_unlock(&dlm_domain_lock);
545 }
546 
547 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
548 {
549         int node = -1, num = 0;
550 
551         assert_spin_locked(&dlm->spinlock);
552 
553         printk("( ");
554         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
555                                      node + 1)) < O2NM_MAX_NODES) {
556                 printk("%d ", node);
557                 ++num;
558         }
559         printk(") %u nodes\n", num);
560 }
561 
562 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
563                                    void **ret_data)
564 {
565         struct dlm_ctxt *dlm = data;
566         unsigned int node;
567         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
568 
569         mlog(0, "%p %u %p", msg, len, data);
570 
571         if (!dlm_grab(dlm))
572                 return 0;
573 
574         node = exit_msg->node_idx;
575 
576         spin_lock(&dlm->spinlock);
577         clear_bit(node, dlm->domain_map);
578         clear_bit(node, dlm->exit_domain_map);
579         printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
580         __dlm_print_nodes(dlm);
581 
582         /* notify anything attached to the heartbeat events */
583         dlm_hb_event_notify_attached(dlm, node, 0);
584 
585         spin_unlock(&dlm->spinlock);
586 
587         dlm_put(dlm);
588 
589         return 0;
590 }
591 
592 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
593                                     unsigned int node)
594 {
595         int status;
596         struct dlm_exit_domain leave_msg;
597 
598         mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
599              msg_type, node);
600 
601         memset(&leave_msg, 0, sizeof(leave_msg));
602         leave_msg.node_idx = dlm->node_num;
603 
604         status = o2net_send_message(msg_type, dlm->key, &leave_msg,
605                                     sizeof(leave_msg), node, NULL);
606         if (status < 0)
607                 mlog(ML_ERROR, "Error %d sending domain exit message %u "
608                      "to node %u on domain %s\n", status, msg_type, node,
609                      dlm->name);
610 
611         return status;
612 }
613 
614 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
615 {
616         int node = -1;
617 
618         /* Support for begin exit domain was added in 1.2 */
619         if (dlm->dlm_locking_proto.pv_major == 1 &&
620             dlm->dlm_locking_proto.pv_minor < 2)
621                 return;
622 
623         /*
624          * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
625          * informational. Meaning if a node does not receive the message,
626          * so be it.
627          */
628         spin_lock(&dlm->spinlock);
629         while (1) {
630                 node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
631                 if (node >= O2NM_MAX_NODES)
632                         break;
633                 if (node == dlm->node_num)
634                         continue;
635 
636                 spin_unlock(&dlm->spinlock);
637                 dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
638                 spin_lock(&dlm->spinlock);
639         }
640         spin_unlock(&dlm->spinlock);
641 }
642 
643 static void dlm_leave_domain(struct dlm_ctxt *dlm)
644 {
645         int node, clear_node, status;
646 
647         /* At this point we've migrated away all our locks and won't
648          * accept mastership of new ones. The dlm is responsible for
649          * almost nothing now. We make sure not to confuse any joining
650          * nodes and then commence shutdown procedure. */
651 
652         spin_lock(&dlm->spinlock);
653         /* Clear ourselves from the domain map */
654         clear_bit(dlm->node_num, dlm->domain_map);
655         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
656                                      0)) < O2NM_MAX_NODES) {
657                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
658                  * -nodes cannot be added now as the
659                  *   query_join_handlers knows to respond with OK_NO_MAP
660                  * -we catch the right network errors if a node is
661                  *   removed from the map while we're sending him the
662                  *   exit message. */
663                 spin_unlock(&dlm->spinlock);
664 
665                 clear_node = 1;
666 
667                 status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
668                                                   node);
669                 if (status < 0 &&
670                     status != -ENOPROTOOPT &&
671                     status != -ENOTCONN) {
672                         mlog(ML_NOTICE, "Error %d sending domain exit message "
673                              "to node %d\n", status, node);
674 
675                         /* Not sure what to do here but lets sleep for
676                          * a bit in case this was a transient
677                          * error... */
678                         msleep(DLM_DOMAIN_BACKOFF_MS);
679                         clear_node = 0;
680                 }
681 
682                 spin_lock(&dlm->spinlock);
683                 /* If we're not clearing the node bit then we intend
684                  * to loop back around to try again. */
685                 if (clear_node)
686                         clear_bit(node, dlm->domain_map);
687         }
688         spin_unlock(&dlm->spinlock);
689 }
690 
691 void dlm_unregister_domain(struct dlm_ctxt *dlm)
692 {
693         int leave = 0;
694         struct dlm_lock_resource *res;
695 
696         spin_lock(&dlm_domain_lock);
697         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
698         BUG_ON(!dlm->num_joins);
699 
700         dlm->num_joins--;
701         if (!dlm->num_joins) {
702                 /* We mark it "in shutdown" now so new register
703                  * requests wait until we've completely left the
704                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
705                  * want new domain joins to communicate with us at
706                  * least until we've completed migration of our
707                  * resources. */
708                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
709                 leave = 1;
710         }
711         spin_unlock(&dlm_domain_lock);
712 
713         if (leave) {
714                 mlog(0, "shutting down domain %s\n", dlm->name);
715                 dlm_begin_exit_domain(dlm);
716 
717                 /* We changed dlm state, notify the thread */
718                 dlm_kick_thread(dlm, NULL);
719 
720                 while (dlm_migrate_all_locks(dlm)) {
721                         /* Give dlm_thread time to purge the lockres' */
722                         msleep(500);
723                         mlog(0, "%s: more migration to do\n", dlm->name);
724                 }
725 
726                 /* This list should be empty. If not, print remaining lockres */
727                 if (!list_empty(&dlm->tracking_list)) {
728                         mlog(ML_ERROR, "Following lockres' are still on the "
729                              "tracking list:\n");
730                         list_for_each_entry(res, &dlm->tracking_list, tracking)
731                                 dlm_print_one_lock_resource(res);
732                 }
733 
734                 dlm_mark_domain_leaving(dlm);
735                 dlm_leave_domain(dlm);
736                 printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
737                 dlm_force_free_mles(dlm);
738                 dlm_complete_dlm_shutdown(dlm);
739         }
740         dlm_put(dlm);
741 }
742 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
743 
744 static int dlm_query_join_proto_check(char *proto_type, int node,
745                                       struct dlm_protocol_version *ours,
746                                       struct dlm_protocol_version *request)
747 {
748         int rc;
749         struct dlm_protocol_version proto = *request;
750 
751         if (!dlm_protocol_compare(ours, &proto)) {
752                 mlog(0,
753                      "node %u wanted to join with %s locking protocol "
754                      "%u.%u, we respond with %u.%u\n",
755                      node, proto_type,
756                      request->pv_major,
757                      request->pv_minor,
758                      proto.pv_major, proto.pv_minor);
759                 request->pv_minor = proto.pv_minor;
760                 rc = 0;
761         } else {
762                 mlog(ML_NOTICE,
763                      "Node %u wanted to join with %s locking "
764                      "protocol %u.%u, but we have %u.%u, disallowing\n",
765                      node, proto_type,
766                      request->pv_major,
767                      request->pv_minor,
768                      ours->pv_major,
769                      ours->pv_minor);
770                 rc = 1;
771         }
772 
773         return rc;
774 }
775 
776 /*
777  * struct dlm_query_join_packet is made up of four one-byte fields.  They
778  * are effectively in big-endian order already.  However, little-endian
779  * machines swap them before putting the packet on the wire (because
780  * query_join's response is a status, and that status is treated as a u32
781  * on the wire).  Thus, a big-endian and little-endian machines will treat
782  * this structure differently.
783  *
784  * The solution is to have little-endian machines swap the structure when
785  * converting from the structure to the u32 representation.  This will
786  * result in the structure having the correct format on the wire no matter
787  * the host endian format.
788  */
789 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
790                                           u32 *wire)
791 {
792         union dlm_query_join_response response;
793 
794         response.packet = *packet;
795         *wire = be32_to_cpu(response.intval);
796 }
797 
798 static void dlm_query_join_wire_to_packet(u32 wire,
799                                           struct dlm_query_join_packet *packet)
800 {
801         union dlm_query_join_response response;
802 
803         response.intval = cpu_to_be32(wire);
804         *packet = response.packet;
805 }
806 
807 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
808                                   void **ret_data)
809 {
810         struct dlm_query_join_request *query;
811         struct dlm_query_join_packet packet = {
812                 .code = JOIN_DISALLOW,
813         };
814         struct dlm_ctxt *dlm = NULL;
815         u32 response;
816         u8 nodenum;
817 
818         query = (struct dlm_query_join_request *) msg->buf;
819 
820         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
821                   query->domain);
822 
823         /*
824          * If heartbeat doesn't consider the node live, tell it
825          * to back off and try again.  This gives heartbeat a chance
826          * to catch up.
827          */
828         if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
829                 mlog(0, "node %u is not in our live map yet\n",
830                      query->node_idx);
831 
832                 packet.code = JOIN_DISALLOW;
833                 goto respond;
834         }
835 
836         packet.code = JOIN_OK_NO_MAP;
837 
838         spin_lock(&dlm_domain_lock);
839         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
840         if (!dlm)
841                 goto unlock_respond;
842 
843         /*
844          * There is a small window where the joining node may not see the
845          * node(s) that just left but still part of the cluster. DISALLOW
846          * join request if joining node has different node map.
847          */
848         nodenum=0;
849         while (nodenum < O2NM_MAX_NODES) {
850                 if (test_bit(nodenum, dlm->domain_map)) {
851                         if (!byte_test_bit(nodenum, query->node_map)) {
852                                 mlog(0, "disallow join as node %u does not "
853                                      "have node %u in its nodemap\n",
854                                      query->node_idx, nodenum);
855                                 packet.code = JOIN_DISALLOW;
856                                 goto unlock_respond;
857                         }
858                 }
859                 nodenum++;
860         }
861 
862         /* Once the dlm ctxt is marked as leaving then we don't want
863          * to be put in someone's domain map.
864          * Also, explicitly disallow joining at certain troublesome
865          * times (ie. during recovery). */
866         if (dlm->dlm_state != DLM_CTXT_LEAVING) {
867                 int bit = query->node_idx;
868                 spin_lock(&dlm->spinlock);
869 
870                 if (dlm->dlm_state == DLM_CTXT_NEW &&
871                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
872                         /*If this is a brand new context and we
873                          * haven't started our join process yet, then
874                          * the other node won the race. */
875                         packet.code = JOIN_OK_NO_MAP;
876                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
877                         /* Disallow parallel joins. */
878                         packet.code = JOIN_DISALLOW;
879                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
880                         mlog(0, "node %u trying to join, but recovery "
881                              "is ongoing.\n", bit);
882                         packet.code = JOIN_DISALLOW;
883                 } else if (test_bit(bit, dlm->recovery_map)) {
884                         mlog(0, "node %u trying to join, but it "
885                              "still needs recovery.\n", bit);
886                         packet.code = JOIN_DISALLOW;
887                 } else if (test_bit(bit, dlm->domain_map)) {
888                         mlog(0, "node %u trying to join, but it "
889                              "is still in the domain! needs recovery?\n",
890                              bit);
891                         packet.code = JOIN_DISALLOW;
892                 } else {
893                         /* Alright we're fully a part of this domain
894                          * so we keep some state as to who's joining
895                          * and indicate to him that needs to be fixed
896                          * up. */
897 
898                         /* Make sure we speak compatible locking protocols.  */
899                         if (dlm_query_join_proto_check("DLM", bit,
900                                                        &dlm->dlm_locking_proto,
901                                                        &query->dlm_proto)) {
902                                 packet.code = JOIN_PROTOCOL_MISMATCH;
903                         } else if (dlm_query_join_proto_check("fs", bit,
904                                                               &dlm->fs_locking_proto,
905                                                               &query->fs_proto)) {
906                                 packet.code = JOIN_PROTOCOL_MISMATCH;
907                         } else {
908                                 packet.dlm_minor = query->dlm_proto.pv_minor;
909                                 packet.fs_minor = query->fs_proto.pv_minor;
910                                 packet.code = JOIN_OK;
911                                 __dlm_set_joining_node(dlm, query->node_idx);
912                         }
913                 }
914 
915                 spin_unlock(&dlm->spinlock);
916         }
917 unlock_respond:
918         spin_unlock(&dlm_domain_lock);
919 
920 respond:
921         mlog(0, "We respond with %u\n", packet.code);
922 
923         dlm_query_join_packet_to_wire(&packet, &response);
924         return response;
925 }
926 
927 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
928                                      void **ret_data)
929 {
930         struct dlm_assert_joined *assert;
931         struct dlm_ctxt *dlm = NULL;
932 
933         assert = (struct dlm_assert_joined *) msg->buf;
934 
935         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
936                   assert->domain);
937 
938         spin_lock(&dlm_domain_lock);
939         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
940         /* XXX should we consider no dlm ctxt an error? */
941         if (dlm) {
942                 spin_lock(&dlm->spinlock);
943 
944                 /* Alright, this node has officially joined our
945                  * domain. Set him in the map and clean up our
946                  * leftover join state. */
947                 BUG_ON(dlm->joining_node != assert->node_idx);
948 
949                 if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
950                         mlog(0, "dlm recovery is ongoing, disallow join\n");
951                         spin_unlock(&dlm->spinlock);
952                         spin_unlock(&dlm_domain_lock);
953                         return -EAGAIN;
954                 }
955 
956                 set_bit(assert->node_idx, dlm->domain_map);
957                 clear_bit(assert->node_idx, dlm->exit_domain_map);
958                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
959 
960                 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
961                        assert->node_idx, dlm->name);
962                 __dlm_print_nodes(dlm);
963 
964                 /* notify anything attached to the heartbeat events */
965                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
966 
967                 spin_unlock(&dlm->spinlock);
968         }
969         spin_unlock(&dlm_domain_lock);
970 
971         return 0;
972 }
973 
974 static int dlm_match_regions(struct dlm_ctxt *dlm,
975                              struct dlm_query_region *qr,
976                              char *local, int locallen)
977 {
978         char *remote = qr->qr_regions;
979         char *l, *r;
980         int localnr, i, j, foundit;
981         int status = 0;
982 
983         if (!o2hb_global_heartbeat_active()) {
984                 if (qr->qr_numregions) {
985                         mlog(ML_ERROR, "Domain %s: Joining node %d has global "
986                              "heartbeat enabled but local node %d does not\n",
987                              qr->qr_domain, qr->qr_node, dlm->node_num);
988                         status = -EINVAL;
989                 }
990                 goto bail;
991         }
992 
993         if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
994                 mlog(ML_ERROR, "Domain %s: Local node %d has global "
995                      "heartbeat enabled but joining node %d does not\n",
996                      qr->qr_domain, dlm->node_num, qr->qr_node);
997                 status = -EINVAL;
998                 goto bail;
999         }
1000 
1001         r = remote;
1002         for (i = 0; i < qr->qr_numregions; ++i) {
1003                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1004                 r += O2HB_MAX_REGION_NAME_LEN;
1005         }
1006 
1007         localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1008         localnr = o2hb_get_all_regions(local, (u8)localnr);
1009 
1010         /* compare local regions with remote */
1011         l = local;
1012         for (i = 0; i < localnr; ++i) {
1013                 foundit = 0;
1014                 r = remote;
1015                 for (j = 0; j <= qr->qr_numregions; ++j) {
1016                         if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1017                                 foundit = 1;
1018                                 break;
1019                         }
1020                         r += O2HB_MAX_REGION_NAME_LEN;
1021                 }
1022                 if (!foundit) {
1023                         status = -EINVAL;
1024                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1025                              "in local node %d but not in joining node %d\n",
1026                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1027                              dlm->node_num, qr->qr_node);
1028                         goto bail;
1029                 }
1030                 l += O2HB_MAX_REGION_NAME_LEN;
1031         }
1032 
1033         /* compare remote with local regions */
1034         r = remote;
1035         for (i = 0; i < qr->qr_numregions; ++i) {
1036                 foundit = 0;
1037                 l = local;
1038                 for (j = 0; j < localnr; ++j) {
1039                         if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1040                                 foundit = 1;
1041                                 break;
1042                         }
1043                         l += O2HB_MAX_REGION_NAME_LEN;
1044                 }
1045                 if (!foundit) {
1046                         status = -EINVAL;
1047                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1048                              "in joining node %d but not in local node %d\n",
1049                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1050                              qr->qr_node, dlm->node_num);
1051                         goto bail;
1052                 }
1053                 r += O2HB_MAX_REGION_NAME_LEN;
1054         }
1055 
1056 bail:
1057         return status;
1058 }
1059 
1060 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1061 {
1062         struct dlm_query_region *qr = NULL;
1063         int status, ret = 0, i;
1064         char *p;
1065 
1066         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1067                 goto bail;
1068 
1069         qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1070         if (!qr) {
1071                 ret = -ENOMEM;
1072                 mlog_errno(ret);
1073                 goto bail;
1074         }
1075 
1076         qr->qr_node = dlm->node_num;
1077         qr->qr_namelen = strlen(dlm->name);
1078         memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1079         /* if local hb, the numregions will be zero */
1080         if (o2hb_global_heartbeat_active())
1081                 qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1082                                                          O2NM_MAX_REGIONS);
1083 
1084         p = qr->qr_regions;
1085         for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1086                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1087 
1088         i = -1;
1089         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1090                                   i + 1)) < O2NM_MAX_NODES) {
1091                 if (i == dlm->node_num)
1092                         continue;
1093 
1094                 mlog(0, "Sending regions to node %d\n", i);
1095 
1096                 ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1097                                          sizeof(struct dlm_query_region),
1098                                          i, &status);
1099                 if (ret >= 0)
1100                         ret = status;
1101                 if (ret) {
1102                         mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1103                              ret, i);
1104                         break;
1105                 }
1106         }
1107 
1108 bail:
1109         kfree(qr);
1110         return ret;
1111 }
1112 
1113 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1114                                     void *data, void **ret_data)
1115 {
1116         struct dlm_query_region *qr;
1117         struct dlm_ctxt *dlm = NULL;
1118         char *local = NULL;
1119         int status = 0;
1120 
1121         qr = (struct dlm_query_region *) msg->buf;
1122 
1123         mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1124              qr->qr_domain);
1125 
1126         /* buffer used in dlm_mast_regions() */
1127         local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1128         if (!local)
1129                 return -ENOMEM;
1130 
1131         status = -EINVAL;
1132 
1133         spin_lock(&dlm_domain_lock);
1134         dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1135         if (!dlm) {
1136                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1137                      "before join domain\n", qr->qr_node, qr->qr_domain);
1138                 goto out_domain_lock;
1139         }
1140 
1141         spin_lock(&dlm->spinlock);
1142         if (dlm->joining_node != qr->qr_node) {
1143                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1144                      "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1145                      dlm->joining_node);
1146                 goto out_dlm_lock;
1147         }
1148 
1149         /* Support for global heartbeat was added in 1.1 */
1150         if (dlm->dlm_locking_proto.pv_major == 1 &&
1151             dlm->dlm_locking_proto.pv_minor == 0) {
1152                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1153                      "but active dlm protocol is %d.%d\n", qr->qr_node,
1154                      qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1155                      dlm->dlm_locking_proto.pv_minor);
1156                 goto out_dlm_lock;
1157         }
1158 
1159         status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1160 
1161 out_dlm_lock:
1162         spin_unlock(&dlm->spinlock);
1163 
1164 out_domain_lock:
1165         spin_unlock(&dlm_domain_lock);
1166 
1167         kfree(local);
1168 
1169         return status;
1170 }
1171 
1172 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1173 {
1174         struct o2nm_node *local;
1175         struct dlm_node_info *remote;
1176         int i, j;
1177         int status = 0;
1178 
1179         for (j = 0; j < qn->qn_numnodes; ++j)
1180                 mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1181                      &(qn->qn_nodes[j].ni_ipv4_address),
1182                      ntohs(qn->qn_nodes[j].ni_ipv4_port));
1183 
1184         for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1185                 local = o2nm_get_node_by_num(i);
1186                 remote = NULL;
1187                 for (j = 0; j < qn->qn_numnodes; ++j) {
1188                         if (qn->qn_nodes[j].ni_nodenum == i) {
1189                                 remote = &(qn->qn_nodes[j]);
1190                                 break;
1191                         }
1192                 }
1193 
1194                 if (!local && !remote)
1195                         continue;
1196 
1197                 if ((local && !remote) || (!local && remote))
1198                         status = -EINVAL;
1199 
1200                 if (!status &&
1201                     ((remote->ni_nodenum != local->nd_num) ||
1202                      (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1203                      (remote->ni_ipv4_address != local->nd_ipv4_address)))
1204                         status = -EINVAL;
1205 
1206                 if (status) {
1207                         if (remote && !local)
1208                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1209                                      "registered in joining node %d but not in "
1210                                      "local node %d\n", qn->qn_domain,
1211                                      remote->ni_nodenum,
1212                                      &(remote->ni_ipv4_address),
1213                                      ntohs(remote->ni_ipv4_port),
1214                                      qn->qn_nodenum, dlm->node_num);
1215                         if (local && !remote)
1216                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1217                                      "registered in local node %d but not in "
1218                                      "joining node %d\n", qn->qn_domain,
1219                                      local->nd_num, &(local->nd_ipv4_address),
1220                                      ntohs(local->nd_ipv4_port),
1221                                      dlm->node_num, qn->qn_nodenum);
1222                         BUG_ON((!local && !remote));
1223                 }
1224 
1225                 if (local)
1226                         o2nm_node_put(local);
1227         }
1228 
1229         return status;
1230 }
1231 
1232 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1233 {
1234         struct dlm_query_nodeinfo *qn = NULL;
1235         struct o2nm_node *node;
1236         int ret = 0, status, count, i;
1237 
1238         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1239                 goto bail;
1240 
1241         qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1242         if (!qn) {
1243                 ret = -ENOMEM;
1244                 mlog_errno(ret);
1245                 goto bail;
1246         }
1247 
1248         for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1249                 node = o2nm_get_node_by_num(i);
1250                 if (!node)
1251                         continue;
1252                 qn->qn_nodes[count].ni_nodenum = node->nd_num;
1253                 qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1254                 qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1255                 mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1256                      &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1257                 ++count;
1258                 o2nm_node_put(node);
1259         }
1260 
1261         qn->qn_nodenum = dlm->node_num;
1262         qn->qn_numnodes = count;
1263         qn->qn_namelen = strlen(dlm->name);
1264         memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1265 
1266         i = -1;
1267         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1268                                   i + 1)) < O2NM_MAX_NODES) {
1269                 if (i == dlm->node_num)
1270                         continue;
1271 
1272                 mlog(0, "Sending nodeinfo to node %d\n", i);
1273 
1274                 ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1275                                          qn, sizeof(struct dlm_query_nodeinfo),
1276                                          i, &status);
1277                 if (ret >= 0)
1278                         ret = status;
1279                 if (ret) {
1280                         mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1281                         break;
1282                 }
1283         }
1284 
1285 bail:
1286         kfree(qn);
1287         return ret;
1288 }
1289 
1290 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1291                                       void *data, void **ret_data)
1292 {
1293         struct dlm_query_nodeinfo *qn;
1294         struct dlm_ctxt *dlm = NULL;
1295         int locked = 0, status = -EINVAL;
1296 
1297         qn = (struct dlm_query_nodeinfo *) msg->buf;
1298 
1299         mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1300              qn->qn_domain);
1301 
1302         spin_lock(&dlm_domain_lock);
1303         dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1304         if (!dlm) {
1305                 mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1306                      "join domain\n", qn->qn_nodenum, qn->qn_domain);
1307                 goto bail;
1308         }
1309 
1310         spin_lock(&dlm->spinlock);
1311         locked = 1;
1312         if (dlm->joining_node != qn->qn_nodenum) {
1313                 mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1314                      "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1315                      dlm->joining_node);
1316                 goto bail;
1317         }
1318 
1319         /* Support for node query was added in 1.1 */
1320         if (dlm->dlm_locking_proto.pv_major == 1 &&
1321             dlm->dlm_locking_proto.pv_minor == 0) {
1322                 mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1323                      "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1324                      qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1325                      dlm->dlm_locking_proto.pv_minor);
1326                 goto bail;
1327         }
1328 
1329         status = dlm_match_nodes(dlm, qn);
1330 
1331 bail:
1332         if (locked)
1333                 spin_unlock(&dlm->spinlock);
1334         spin_unlock(&dlm_domain_lock);
1335 
1336         return status;
1337 }
1338 
1339 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1340                                    void **ret_data)
1341 {
1342         struct dlm_cancel_join *cancel;
1343         struct dlm_ctxt *dlm = NULL;
1344 
1345         cancel = (struct dlm_cancel_join *) msg->buf;
1346 
1347         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1348                   cancel->domain);
1349 
1350         spin_lock(&dlm_domain_lock);
1351         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1352 
1353         if (dlm) {
1354                 spin_lock(&dlm->spinlock);
1355 
1356                 /* Yikes, this guy wants to cancel his join. No
1357                  * problem, we simply cleanup our join state. */
1358                 BUG_ON(dlm->joining_node != cancel->node_idx);
1359                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1360 
1361                 spin_unlock(&dlm->spinlock);
1362         }
1363         spin_unlock(&dlm_domain_lock);
1364 
1365         return 0;
1366 }
1367 
1368 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1369                                     unsigned int node)
1370 {
1371         int status;
1372         struct dlm_cancel_join cancel_msg;
1373 
1374         memset(&cancel_msg, 0, sizeof(cancel_msg));
1375         cancel_msg.node_idx = dlm->node_num;
1376         cancel_msg.name_len = strlen(dlm->name);
1377         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1378 
1379         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1380                                     &cancel_msg, sizeof(cancel_msg), node,
1381                                     NULL);
1382         if (status < 0) {
1383                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1384                      "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1385                      node);
1386                 goto bail;
1387         }
1388 
1389 bail:
1390         return status;
1391 }
1392 
1393 /* map_size should be in bytes. */
1394 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1395                                  unsigned long *node_map,
1396                                  unsigned int map_size)
1397 {
1398         int status, tmpstat;
1399         int node;
1400 
1401         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1402                          sizeof(unsigned long))) {
1403                 mlog(ML_ERROR,
1404                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1405                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1406                 return -EINVAL;
1407         }
1408 
1409         status = 0;
1410         node = -1;
1411         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1412                                      node + 1)) < O2NM_MAX_NODES) {
1413                 if (node == dlm->node_num)
1414                         continue;
1415 
1416                 tmpstat = dlm_send_one_join_cancel(dlm, node);
1417                 if (tmpstat) {
1418                         mlog(ML_ERROR, "Error return %d cancelling join on "
1419                              "node %d\n", tmpstat, node);
1420                         if (!status)
1421                                 status = tmpstat;
1422                 }
1423         }
1424 
1425         if (status)
1426                 mlog_errno(status);
1427         return status;
1428 }
1429 
1430 static int dlm_request_join(struct dlm_ctxt *dlm,
1431                             int node,
1432                             enum dlm_query_join_response_code *response)
1433 {
1434         int status;
1435         struct dlm_query_join_request join_msg;
1436         struct dlm_query_join_packet packet;
1437         u32 join_resp;
1438 
1439         mlog(0, "querying node %d\n", node);
1440 
1441         memset(&join_msg, 0, sizeof(join_msg));
1442         join_msg.node_idx = dlm->node_num;
1443         join_msg.name_len = strlen(dlm->name);
1444         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1445         join_msg.dlm_proto = dlm->dlm_locking_proto;
1446         join_msg.fs_proto = dlm->fs_locking_proto;
1447 
1448         /* copy live node map to join message */
1449         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1450 
1451         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1452                                     sizeof(join_msg), node, &join_resp);
1453         if (status < 0 && status != -ENOPROTOOPT) {
1454                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1455                      "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1456                      node);
1457                 goto bail;
1458         }
1459         dlm_query_join_wire_to_packet(join_resp, &packet);
1460 
1461         /* -ENOPROTOOPT from the net code means the other side isn't
1462             listening for our message type -- that's fine, it means
1463             his dlm isn't up, so we can consider him a 'yes' but not
1464             joined into the domain.  */
1465         if (status == -ENOPROTOOPT) {
1466                 status = 0;
1467                 *response = JOIN_OK_NO_MAP;
1468         } else {
1469                 *response = packet.code;
1470                 switch (packet.code) {
1471                 case JOIN_DISALLOW:
1472                 case JOIN_OK_NO_MAP:
1473                         break;
1474                 case JOIN_PROTOCOL_MISMATCH:
1475                         mlog(ML_NOTICE,
1476                              "This node requested DLM locking protocol %u.%u and "
1477                              "filesystem locking protocol %u.%u.  At least one of "
1478                              "the protocol versions on node %d is not compatible, "
1479                              "disconnecting\n",
1480                              dlm->dlm_locking_proto.pv_major,
1481                              dlm->dlm_locking_proto.pv_minor,
1482                              dlm->fs_locking_proto.pv_major,
1483                              dlm->fs_locking_proto.pv_minor,
1484                              node);
1485                         status = -EPROTO;
1486                         break;
1487                 case JOIN_OK:
1488                         /* Use the same locking protocol as the remote node */
1489                         dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1490                         dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1491                         mlog(0,
1492                              "Node %d responds JOIN_OK with DLM locking protocol "
1493                              "%u.%u and fs locking protocol %u.%u\n",
1494                              node,
1495                              dlm->dlm_locking_proto.pv_major,
1496                              dlm->dlm_locking_proto.pv_minor,
1497                              dlm->fs_locking_proto.pv_major,
1498                              dlm->fs_locking_proto.pv_minor);
1499                         break;
1500                 default:
1501                         status = -EINVAL;
1502                         mlog(ML_ERROR, "invalid response %d from node %u\n",
1503                              packet.code, node);
1504                         /* Reset response to JOIN_DISALLOW */
1505                         *response = JOIN_DISALLOW;
1506                         break;
1507                 }
1508         }
1509 
1510         mlog(0, "status %d, node %d response is %d\n", status, node,
1511              *response);
1512 
1513 bail:
1514         return status;
1515 }
1516 
1517 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1518                                     unsigned int node)
1519 {
1520         int status;
1521         int ret;
1522         struct dlm_assert_joined assert_msg;
1523 
1524         mlog(0, "Sending join assert to node %u\n", node);
1525 
1526         memset(&assert_msg, 0, sizeof(assert_msg));
1527         assert_msg.node_idx = dlm->node_num;
1528         assert_msg.name_len = strlen(dlm->name);
1529         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1530 
1531         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1532                                     &assert_msg, sizeof(assert_msg), node,
1533                                     &ret);
1534         if (status < 0)
1535                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1536                      "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1537                      node);
1538         else
1539                 status = ret;
1540 
1541         return status;
1542 }
1543 
1544 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1545                                   unsigned long *node_map)
1546 {
1547         int status, node, live;
1548 
1549         status = 0;
1550         node = -1;
1551         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1552                                      node + 1)) < O2NM_MAX_NODES) {
1553                 if (node == dlm->node_num)
1554                         continue;
1555 
1556                 do {
1557                         /* It is very important that this message be
1558                          * received so we spin until either the node
1559                          * has died or it gets the message. */
1560                         status = dlm_send_one_join_assert(dlm, node);
1561 
1562                         spin_lock(&dlm->spinlock);
1563                         live = test_bit(node, dlm->live_nodes_map);
1564                         spin_unlock(&dlm->spinlock);
1565 
1566                         if (status) {
1567                                 mlog(ML_ERROR, "Error return %d asserting "
1568                                      "join on node %d\n", status, node);
1569 
1570                                 /* give us some time between errors... */
1571                                 if (live)
1572                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1573                         }
1574                 } while (status && live);
1575         }
1576 }
1577 
1578 struct domain_join_ctxt {
1579         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1580         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1581 };
1582 
1583 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1584                                    struct domain_join_ctxt *ctxt,
1585                                    enum dlm_query_join_response_code response)
1586 {
1587         int ret;
1588 
1589         if (response == JOIN_DISALLOW) {
1590                 mlog(0, "Latest response of disallow -- should restart\n");
1591                 return 1;
1592         }
1593 
1594         spin_lock(&dlm->spinlock);
1595         /* For now, we restart the process if the node maps have
1596          * changed at all */
1597         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1598                      sizeof(dlm->live_nodes_map));
1599         spin_unlock(&dlm->spinlock);
1600 
1601         if (ret)
1602                 mlog(0, "Node maps changed -- should restart\n");
1603 
1604         return ret;
1605 }
1606 
1607 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1608 {
1609         int status = 0, tmpstat, node;
1610         struct domain_join_ctxt *ctxt;
1611         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1612 
1613         mlog(0, "%p", dlm);
1614 
1615         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1616         if (!ctxt) {
1617                 status = -ENOMEM;
1618                 mlog_errno(status);
1619                 goto bail;
1620         }
1621 
1622         /* group sem locking should work for us here -- we're already
1623          * registered for heartbeat events so filling this should be
1624          * atomic wrt getting those handlers called. */
1625         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1626 
1627         spin_lock(&dlm->spinlock);
1628         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1629 
1630         __dlm_set_joining_node(dlm, dlm->node_num);
1631 
1632         spin_unlock(&dlm->spinlock);
1633 
1634         node = -1;
1635         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1636                                      node + 1)) < O2NM_MAX_NODES) {
1637                 if (node == dlm->node_num)
1638                         continue;
1639 
1640                 status = dlm_request_join(dlm, node, &response);
1641                 if (status < 0) {
1642                         mlog_errno(status);
1643                         goto bail;
1644                 }
1645 
1646                 /* Ok, either we got a response or the node doesn't have a
1647                  * dlm up. */
1648                 if (response == JOIN_OK)
1649                         set_bit(node, ctxt->yes_resp_map);
1650 
1651                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1652                         status = -EAGAIN;
1653                         goto bail;
1654                 }
1655         }
1656 
1657         mlog(0, "Yay, done querying nodes!\n");
1658 
1659         /* Yay, everyone agree's we can join the domain. My domain is
1660          * comprised of all nodes who were put in the
1661          * yes_resp_map. Copy that into our domain map and send a join
1662          * assert message to clean up everyone elses state. */
1663         spin_lock(&dlm->spinlock);
1664         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1665                sizeof(ctxt->yes_resp_map));
1666         set_bit(dlm->node_num, dlm->domain_map);
1667         spin_unlock(&dlm->spinlock);
1668 
1669         /* Support for global heartbeat and node info was added in 1.1 */
1670         if (dlm->dlm_locking_proto.pv_major > 1 ||
1671             dlm->dlm_locking_proto.pv_minor > 0) {
1672                 status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1673                 if (status) {
1674                         mlog_errno(status);
1675                         goto bail;
1676                 }
1677                 status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1678                 if (status) {
1679                         mlog_errno(status);
1680                         goto bail;
1681                 }
1682         }
1683 
1684         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1685 
1686         /* Joined state *must* be set before the joining node
1687          * information, otherwise the query_join handler may read no
1688          * current joiner but a state of NEW and tell joining nodes
1689          * we're not in the domain. */
1690         spin_lock(&dlm_domain_lock);
1691         dlm->dlm_state = DLM_CTXT_JOINED;
1692         dlm->num_joins++;
1693         spin_unlock(&dlm_domain_lock);
1694 
1695 bail:
1696         spin_lock(&dlm->spinlock);
1697         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1698         if (!status) {
1699                 printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1700                 __dlm_print_nodes(dlm);
1701         }
1702         spin_unlock(&dlm->spinlock);
1703 
1704         if (ctxt) {
1705                 /* Do we need to send a cancel message to any nodes? */
1706                 if (status < 0) {
1707                         tmpstat = dlm_send_join_cancels(dlm,
1708                                                         ctxt->yes_resp_map,
1709                                                         sizeof(ctxt->yes_resp_map));
1710                         if (tmpstat < 0)
1711                                 mlog_errno(tmpstat);
1712                 }
1713                 kfree(ctxt);
1714         }
1715 
1716         mlog(0, "returning %d\n", status);
1717         return status;
1718 }
1719 
1720 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1721 {
1722         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1723         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1724         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1725 }
1726 
1727 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1728 {
1729         int status;
1730 
1731         mlog(0, "registering handlers.\n");
1732 
1733         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1734                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1735         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1736                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1737 
1738         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1739         if (status)
1740                 goto bail;
1741 
1742         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1743         if (status)
1744                 goto bail;
1745 
1746         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1747                                         sizeof(struct dlm_master_request),
1748                                         dlm_master_request_handler,
1749                                         dlm, NULL, &dlm->dlm_domain_handlers);
1750         if (status)
1751                 goto bail;
1752 
1753         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1754                                         sizeof(struct dlm_assert_master),
1755                                         dlm_assert_master_handler,
1756                                         dlm, dlm_assert_master_post_handler,
1757                                         &dlm->dlm_domain_handlers);
1758         if (status)
1759                 goto bail;
1760 
1761         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1762                                         sizeof(struct dlm_create_lock),
1763                                         dlm_create_lock_handler,
1764                                         dlm, NULL, &dlm->dlm_domain_handlers);
1765         if (status)
1766                 goto bail;
1767 
1768         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1769                                         DLM_CONVERT_LOCK_MAX_LEN,
1770                                         dlm_convert_lock_handler,
1771                                         dlm, NULL, &dlm->dlm_domain_handlers);
1772         if (status)
1773                 goto bail;
1774 
1775         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1776                                         DLM_UNLOCK_LOCK_MAX_LEN,
1777                                         dlm_unlock_lock_handler,
1778                                         dlm, NULL, &dlm->dlm_domain_handlers);
1779         if (status)
1780                 goto bail;
1781 
1782         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1783                                         DLM_PROXY_AST_MAX_LEN,
1784                                         dlm_proxy_ast_handler,
1785                                         dlm, NULL, &dlm->dlm_domain_handlers);
1786         if (status)
1787                 goto bail;
1788 
1789         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1790                                         sizeof(struct dlm_exit_domain),
1791                                         dlm_exit_domain_handler,
1792                                         dlm, NULL, &dlm->dlm_domain_handlers);
1793         if (status)
1794                 goto bail;
1795 
1796         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1797                                         sizeof(struct dlm_deref_lockres),
1798                                         dlm_deref_lockres_handler,
1799                                         dlm, NULL, &dlm->dlm_domain_handlers);
1800         if (status)
1801                 goto bail;
1802 
1803         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1804                                         sizeof(struct dlm_migrate_request),
1805                                         dlm_migrate_request_handler,
1806                                         dlm, NULL, &dlm->dlm_domain_handlers);
1807         if (status)
1808                 goto bail;
1809 
1810         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1811                                         DLM_MIG_LOCKRES_MAX_LEN,
1812                                         dlm_mig_lockres_handler,
1813                                         dlm, NULL, &dlm->dlm_domain_handlers);
1814         if (status)
1815                 goto bail;
1816 
1817         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1818                                         sizeof(struct dlm_master_requery),
1819                                         dlm_master_requery_handler,
1820                                         dlm, NULL, &dlm->dlm_domain_handlers);
1821         if (status)
1822                 goto bail;
1823 
1824         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1825                                         sizeof(struct dlm_lock_request),
1826                                         dlm_request_all_locks_handler,
1827                                         dlm, NULL, &dlm->dlm_domain_handlers);
1828         if (status)
1829                 goto bail;
1830 
1831         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1832                                         sizeof(struct dlm_reco_data_done),
1833                                         dlm_reco_data_done_handler,
1834                                         dlm, NULL, &dlm->dlm_domain_handlers);
1835         if (status)
1836                 goto bail;
1837 
1838         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1839                                         sizeof(struct dlm_begin_reco),
1840                                         dlm_begin_reco_handler,
1841                                         dlm, NULL, &dlm->dlm_domain_handlers);
1842         if (status)
1843                 goto bail;
1844 
1845         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1846                                         sizeof(struct dlm_finalize_reco),
1847                                         dlm_finalize_reco_handler,
1848                                         dlm, NULL, &dlm->dlm_domain_handlers);
1849         if (status)
1850                 goto bail;
1851 
1852         status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1853                                         sizeof(struct dlm_exit_domain),
1854                                         dlm_begin_exit_domain_handler,
1855                                         dlm, NULL, &dlm->dlm_domain_handlers);
1856         if (status)
1857                 goto bail;
1858 
1859         status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key,
1860                                         sizeof(struct dlm_deref_lockres_done),
1861                                         dlm_deref_lockres_done_handler,
1862                                         dlm, NULL, &dlm->dlm_domain_handlers);
1863 bail:
1864         if (status)
1865                 dlm_unregister_domain_handlers(dlm);
1866 
1867         return status;
1868 }
1869 
1870 static int dlm_join_domain(struct dlm_ctxt *dlm)
1871 {
1872         int status;
1873         unsigned int backoff;
1874         unsigned int total_backoff = 0;
1875         char wq_name[O2NM_MAX_NAME_LEN];
1876 
1877         BUG_ON(!dlm);
1878 
1879         mlog(0, "Join domain %s\n", dlm->name);
1880 
1881         status = dlm_register_domain_handlers(dlm);
1882         if (status) {
1883                 mlog_errno(status);
1884                 goto bail;
1885         }
1886 
1887         status = dlm_launch_thread(dlm);
1888         if (status < 0) {
1889                 mlog_errno(status);
1890                 goto bail;
1891         }
1892 
1893         status = dlm_launch_recovery_thread(dlm);
1894         if (status < 0) {
1895                 mlog_errno(status);
1896                 goto bail;
1897         }
1898 
1899         status = dlm_debug_init(dlm);
1900         if (status < 0) {
1901                 mlog_errno(status);
1902                 goto bail;
1903         }
1904 
1905         snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name);
1906         dlm->dlm_worker = alloc_workqueue(wq_name, WQ_MEM_RECLAIM, 0);
1907         if (!dlm->dlm_worker) {
1908                 status = -ENOMEM;
1909                 mlog_errno(status);
1910                 goto bail;
1911         }
1912 
1913         do {
1914                 status = dlm_try_to_join_domain(dlm);
1915 
1916                 /* If we're racing another node to the join, then we
1917                  * need to back off temporarily and let them
1918                  * complete. */
1919 #define DLM_JOIN_TIMEOUT_MSECS  90000
1920                 if (status == -EAGAIN) {
1921                         if (signal_pending(current)) {
1922                                 status = -ERESTARTSYS;
1923                                 goto bail;
1924                         }
1925 
1926                         if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1927                                 status = -ERESTARTSYS;
1928                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1929                                      "%s after %u msecs\n", dlm->name,
1930                                      total_backoff);
1931                                 goto bail;
1932                         }
1933 
1934                         /*
1935                          * <chip> After you!
1936                          * <dale> No, after you!
1937                          * <chip> I insist!
1938                          * <dale> But you first!
1939                          * ...
1940                          */
1941                         backoff = (unsigned int)(jiffies & 0x3);
1942                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1943                         total_backoff += backoff;
1944                         mlog(0, "backoff %d\n", backoff);
1945                         msleep(backoff);
1946                 }
1947         } while (status == -EAGAIN);
1948 
1949         if (status < 0) {
1950                 mlog_errno(status);
1951                 goto bail;
1952         }
1953 
1954         status = 0;
1955 bail:
1956         wake_up(&dlm_domain_events);
1957 
1958         if (status) {
1959                 dlm_unregister_domain_handlers(dlm);
1960                 dlm_debug_shutdown(dlm);
1961                 dlm_complete_thread(dlm);
1962                 dlm_complete_recovery_thread(dlm);
1963                 dlm_destroy_dlm_worker(dlm);
1964         }
1965 
1966         return status;
1967 }
1968 
1969 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1970                                 u32 key)
1971 {
1972         int i;
1973         int ret;
1974         struct dlm_ctxt *dlm = NULL;
1975 
1976         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1977         if (!dlm) {
1978                 ret = -ENOMEM;
1979                 mlog_errno(ret);
1980                 goto leave;
1981         }
1982 
1983         dlm->name = kstrdup(domain, GFP_KERNEL);
1984         if (dlm->name == NULL) {
1985                 ret = -ENOMEM;
1986                 mlog_errno(ret);
1987                 goto leave;
1988         }
1989 
1990         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1991         if (!dlm->lockres_hash) {
1992                 ret = -ENOMEM;
1993                 mlog_errno(ret);
1994                 goto leave;
1995         }
1996 
1997         for (i = 0; i < DLM_HASH_BUCKETS; i++)
1998                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
1999 
2000         dlm->master_hash = (struct hlist_head **)
2001                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
2002         if (!dlm->master_hash) {
2003                 ret = -ENOMEM;
2004                 mlog_errno(ret);
2005                 goto leave;
2006         }
2007 
2008         for (i = 0; i < DLM_HASH_BUCKETS; i++)
2009                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2010 
2011         dlm->key = key;
2012         dlm->node_num = o2nm_this_node();
2013 
2014         ret = dlm_create_debugfs_subroot(dlm);
2015         if (ret < 0)
2016                 goto leave;
2017 
2018         spin_lock_init(&dlm->spinlock);
2019         spin_lock_init(&dlm->master_lock);
2020         spin_lock_init(&dlm->ast_lock);
2021         spin_lock_init(&dlm->track_lock);
2022         INIT_LIST_HEAD(&dlm->list);
2023         INIT_LIST_HEAD(&dlm->dirty_list);
2024         INIT_LIST_HEAD(&dlm->reco.resources);
2025         INIT_LIST_HEAD(&dlm->reco.node_data);
2026         INIT_LIST_HEAD(&dlm->purge_list);
2027         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2028         INIT_LIST_HEAD(&dlm->tracking_list);
2029         dlm->reco.state = 0;
2030 
2031         INIT_LIST_HEAD(&dlm->pending_asts);
2032         INIT_LIST_HEAD(&dlm->pending_basts);
2033 
2034         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2035                   dlm->recovery_map, &(dlm->recovery_map[0]));
2036 
2037         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2038         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2039         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2040 
2041         dlm->dlm_thread_task = NULL;
2042         dlm->dlm_reco_thread_task = NULL;
2043         dlm->dlm_worker = NULL;
2044         init_waitqueue_head(&dlm->dlm_thread_wq);
2045         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2046         init_waitqueue_head(&dlm->reco.event);
2047         init_waitqueue_head(&dlm->ast_wq);
2048         init_waitqueue_head(&dlm->migration_wq);
2049         INIT_LIST_HEAD(&dlm->mle_hb_events);
2050 
2051         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2052         init_waitqueue_head(&dlm->dlm_join_events);
2053 
2054         dlm->migrate_done = 0;
2055 
2056         dlm->reco.new_master = O2NM_INVALID_NODE_NUM;
2057         dlm->reco.dead_node = O2NM_INVALID_NODE_NUM;
2058 
2059         atomic_set(&dlm->res_tot_count, 0);
2060         atomic_set(&dlm->res_cur_count, 0);
2061         for (i = 0; i < DLM_MLE_NUM_TYPES; ++i) {
2062                 atomic_set(&dlm->mle_tot_count[i], 0);
2063                 atomic_set(&dlm->mle_cur_count[i], 0);
2064         }
2065 
2066         spin_lock_init(&dlm->work_lock);
2067         INIT_LIST_HEAD(&dlm->work_list);
2068         INIT_WORK(&dlm->dispatched_work, dlm_dispatch_work);
2069 
2070         kref_init(&dlm->dlm_refs);
2071         dlm->dlm_state = DLM_CTXT_NEW;
2072 
2073         INIT_LIST_HEAD(&dlm->dlm_eviction_callbacks);
2074 
2075         mlog(0, "context init: refcount %u\n",
2076                   kref_read(&dlm->dlm_refs));
2077 
2078 leave:
2079         if (ret < 0 && dlm) {
2080                 if (dlm->master_hash)
2081                         dlm_free_pagevec((void **)dlm->master_hash,
2082                                         DLM_HASH_PAGES);
2083 
2084                 if (dlm->lockres_hash)
2085                         dlm_free_pagevec((void **)dlm->lockres_hash,
2086                                         DLM_HASH_PAGES);
2087 
2088                 kfree(dlm->name);
2089                 kfree(dlm);
2090                 dlm = NULL;
2091         }
2092         return dlm;
2093 }
2094 
2095 /*
2096  * Compare a requested locking protocol version against the current one.
2097  *
2098  * If the major numbers are different, they are incompatible.
2099  * If the current minor is greater than the request, they are incompatible.
2100  * If the current minor is less than or equal to the request, they are
2101  * compatible, and the requester should run at the current minor version.
2102  */
2103 static int dlm_protocol_compare(struct dlm_protocol_version *existing,
2104                                 struct dlm_protocol_version *request)
2105 {
2106         if (existing->pv_major != request->pv_major)
2107                 return 1;
2108 
2109         if (existing->pv_minor > request->pv_minor)
2110                 return 1;
2111 
2112         if (existing->pv_minor < request->pv_minor)
2113                 request->pv_minor = existing->pv_minor;
2114 
2115         return 0;
2116 }
2117 
2118 /*
2119  * dlm_register_domain: one-time setup per "domain".
2120  *
2121  * The filesystem passes in the requested locking version via proto.
2122  * If registration was successful, proto will contain the negotiated
2123  * locking protocol.
2124  */
2125 struct dlm_ctxt * dlm_register_domain(const char *domain,
2126                                u32 key,
2127                                struct dlm_protocol_version *fs_proto)
2128 {
2129         int ret;
2130         struct dlm_ctxt *dlm = NULL;
2131         struct dlm_ctxt *new_ctxt = NULL;
2132 
2133         if (strlen(domain) >= O2NM_MAX_NAME_LEN) {
2134                 ret = -ENAMETOOLONG;
2135                 mlog(ML_ERROR, "domain name length too long\n");
2136                 goto leave;
2137         }
2138 
2139         mlog(0, "register called for domain \"%s\"\n", domain);
2140 
2141 retry:
2142         dlm = NULL;
2143         if (signal_pending(current)) {
2144                 ret = -ERESTARTSYS;
2145                 mlog_errno(ret);
2146                 goto leave;
2147         }
2148 
2149         spin_lock(&dlm_domain_lock);
2150 
2151         dlm = __dlm_lookup_domain(domain);
2152         if (dlm) {
2153                 if (dlm->dlm_state != DLM_CTXT_JOINED) {
2154                         spin_unlock(&dlm_domain_lock);
2155 
2156                         mlog(0, "This ctxt is not joined yet!\n");
2157                         wait_event_interruptible(dlm_domain_events,
2158                                                  dlm_wait_on_domain_helper(
2159                                                          domain));
2160                         goto retry;
2161                 }
2162 
2163                 if (dlm_protocol_compare(&dlm->fs_locking_proto, fs_proto)) {
2164                         spin_unlock(&dlm_domain_lock);
2165                         mlog(ML_ERROR,
2166                              "Requested locking protocol version is not "
2167                              "compatible with already registered domain "
2168                              "\"%s\"\n", domain);
2169                         ret = -EPROTO;
2170                         goto leave;
2171                 }
2172 
2173                 __dlm_get(dlm);
2174                 dlm->num_joins++;
2175 
2176                 spin_unlock(&dlm_domain_lock);
2177 
2178                 ret = 0;
2179                 goto leave;
2180         }
2181 
2182         /* doesn't exist */
2183         if (!new_ctxt) {
2184                 spin_unlock(&dlm_domain_lock);
2185 
2186                 new_ctxt = dlm_alloc_ctxt(domain, key);
2187                 if (new_ctxt)
2188                         goto retry;
2189 
2190                 ret = -ENOMEM;
2191                 mlog_errno(ret);
2192                 goto leave;
2193         }
2194 
2195         /* a little variable switch-a-roo here... */
2196         dlm = new_ctxt;
2197         new_ctxt = NULL;
2198 
2199         /* add the new domain */
2200         list_add_tail(&dlm->list, &dlm_domains);
2201         spin_unlock(&dlm_domain_lock);
2202 
2203         /*
2204          * Pass the locking protocol version into the join.  If the join
2205          * succeeds, it will have the negotiated protocol set.
2206          */
2207         dlm->dlm_locking_proto = dlm_protocol;
2208         dlm->fs_locking_proto = *fs_proto;
2209 
2210         ret = dlm_join_domain(dlm);
2211         if (ret) {
2212                 mlog_errno(ret);
2213                 dlm_put(dlm);
2214                 goto leave;
2215         }
2216 
2217         /* Tell the caller what locking protocol we negotiated */
2218         *fs_proto = dlm->fs_locking_proto;
2219 
2220         ret = 0;
2221 leave:
2222         if (new_ctxt)
2223                 dlm_free_ctxt_mem(new_ctxt);
2224 
2225         if (ret < 0)
2226                 dlm = ERR_PTR(ret);
2227 
2228         return dlm;
2229 }
2230 EXPORT_SYMBOL_GPL(dlm_register_domain);
2231 
2232 static LIST_HEAD(dlm_join_handlers);
2233 
2234 static void dlm_unregister_net_handlers(void)
2235 {
2236         o2net_unregister_handler_list(&dlm_join_handlers);
2237 }
2238 
2239 static int dlm_register_net_handlers(void)
2240 {
2241         int status = 0;
2242 
2243         status = o2net_register_handler(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
2244                                         sizeof(struct dlm_query_join_request),
2245                                         dlm_query_join_handler,
2246                                         NULL, NULL, &dlm_join_handlers);
2247         if (status)
2248                 goto bail;
2249 
2250         status = o2net_register_handler(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
2251                                         sizeof(struct dlm_assert_joined),
2252                                         dlm_assert_joined_handler,
2253                                         NULL, NULL, &dlm_join_handlers);
2254         if (status)
2255                 goto bail;
2256 
2257         status = o2net_register_handler(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
2258                                         sizeof(struct dlm_cancel_join),
2259                                         dlm_cancel_join_handler,
2260                                         NULL, NULL, &dlm_join_handlers);
2261         if (status)
2262                 goto bail;
2263 
2264         status = o2net_register_handler(DLM_QUERY_REGION, DLM_MOD_KEY,
2265                                         sizeof(struct dlm_query_region),
2266                                         dlm_query_region_handler,
2267                                         NULL, NULL, &dlm_join_handlers);
2268 
2269         if (status)
2270                 goto bail;
2271 
2272         status = o2net_register_handler(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
2273                                         sizeof(struct dlm_query_nodeinfo),
2274                                         dlm_query_nodeinfo_handler,
2275                                         NULL, NULL, &dlm_join_handlers);
2276 bail:
2277         if (status < 0)
2278                 dlm_unregister_net_handlers();
2279 
2280         return status;
2281 }
2282 
2283 /* Domain eviction callback handling.
2284  *
2285  * The file system requires notification of node death *before* the
2286  * dlm completes it's recovery work, otherwise it may be able to
2287  * acquire locks on resources requiring recovery. Since the dlm can
2288  * evict a node from it's domain *before* heartbeat fires, a similar
2289  * mechanism is required. */
2290 
2291 /* Eviction is not expected to happen often, so a per-domain lock is
2292  * not necessary. Eviction callbacks are allowed to sleep for short
2293  * periods of time. */
2294 static DECLARE_RWSEM(dlm_callback_sem);
2295 
2296 void dlm_fire_domain_eviction_callbacks(struct dlm_ctxt *dlm,
2297                                         int node_num)
2298 {
2299         struct dlm_eviction_cb *cb;
2300 
2301         down_read(&dlm_callback_sem);
2302         list_for_each_entry(cb, &dlm->dlm_eviction_callbacks, ec_item) {
2303                 cb->ec_func(node_num, cb->ec_data);
2304         }
2305         up_read(&dlm_callback_sem);
2306 }
2307 
2308 void dlm_setup_eviction_cb(struct dlm_eviction_cb *cb,
2309                            dlm_eviction_func *f,
2310                            void *data)
2311 {
2312         INIT_LIST_HEAD(&cb->ec_item);
2313         cb->ec_func = f;
2314         cb->ec_data = data;
2315 }
2316 EXPORT_SYMBOL_GPL(dlm_setup_eviction_cb);
2317 
2318 void dlm_register_eviction_cb(struct dlm_ctxt *dlm,
2319                               struct dlm_eviction_cb *cb)
2320 {
2321         down_write(&dlm_callback_sem);
2322         list_add_tail(&cb->ec_item, &dlm->dlm_eviction_callbacks);
2323         up_write(&dlm_callback_sem);
2324 }
2325 EXPORT_SYMBOL_GPL(dlm_register_eviction_cb);
2326 
2327 void dlm_unregister_eviction_cb(struct dlm_eviction_cb *cb)
2328 {
2329         down_write(&dlm_callback_sem);
2330         list_del_init(&cb->ec_item);
2331         up_write(&dlm_callback_sem);
2332 }
2333 EXPORT_SYMBOL_GPL(dlm_unregister_eviction_cb);
2334 
2335 static int __init dlm_init(void)
2336 {
2337         int status;
2338 
2339         status = dlm_init_mle_cache();
2340         if (status) {
2341                 mlog(ML_ERROR, "Could not create o2dlm_mle slabcache\n");
2342                 goto error;
2343         }
2344 
2345         status = dlm_init_master_caches();
2346         if (status) {
2347                 mlog(ML_ERROR, "Could not create o2dlm_lockres and "
2348                      "o2dlm_lockname slabcaches\n");
2349                 goto error;
2350         }
2351 
2352         status = dlm_init_lock_cache();
2353         if (status) {
2354                 mlog(ML_ERROR, "Count not create o2dlm_lock slabcache\n");
2355                 goto error;
2356         }
2357 
2358         status = dlm_register_net_handlers();
2359         if (status) {
2360                 mlog(ML_ERROR, "Unable to register network handlers\n");
2361                 goto error;
2362         }
2363 
2364         status = dlm_create_debugfs_root();
2365         if (status)
2366                 goto error;
2367 
2368         return 0;
2369 error:
2370         dlm_unregister_net_handlers();
2371         dlm_destroy_lock_cache();
2372         dlm_destroy_master_caches();
2373         dlm_destroy_mle_cache();
2374         return -1;
2375 }
2376 
2377 static void __exit dlm_exit (void)
2378 {
2379         dlm_destroy_debugfs_root();
2380         dlm_unregister_net_handlers();
2381         dlm_destroy_lock_cache();
2382         dlm_destroy_master_caches();
2383         dlm_destroy_mle_cache();
2384 }
2385 
2386 MODULE_AUTHOR("Oracle");
2387 MODULE_LICENSE("GPL");
2388 MODULE_DESCRIPTION("OCFS2 Distributed Lock Management");
2389 
2390 module_init(dlm_init);
2391 module_exit(dlm_exit);
2392 

~ [ source navigation ] ~ [ diff markup ] ~ [ identifier search ] ~

kernel.org | git.kernel.org | LWN.net | Project Home | Wiki (Japanese) | Wiki (English) | SVN repository | Mail admin

Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.

osdn.jp