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

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

Version: ~ [ linux-5.3-rc5 ] ~ [ linux-5.2.9 ] ~ [ linux-5.1.21 ] ~ [ linux-5.0.21 ] ~ [ linux-4.20.17 ] ~ [ linux-4.19.67 ] ~ [ linux-4.18.20 ] ~ [ linux-4.17.19 ] ~ [ linux-4.16.18 ] ~ [ linux-4.15.18 ] ~ [ linux-4.14.139 ] ~ [ linux-4.13.16 ] ~ [ linux-4.12.14 ] ~ [ linux-4.11.12 ] ~ [ linux-4.10.17 ] ~ [ linux-4.9.189 ] ~ [ linux-4.8.17 ] ~ [ linux-4.7.10 ] ~ [ linux-4.6.7 ] ~ [ linux-4.5.7 ] ~ [ linux-4.4.189 ] ~ [ 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-3.16.72 ] ~ [ 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-3.9.11 ] ~ [ linux-3.8.13 ] ~ [ linux-3.7.10 ] ~ [ linux-3.6.11 ] ~ [ linux-3.5.7 ] ~ [ linux-3.4.113 ] ~ [ linux-3.3.8 ] ~ [ linux-3.2.102 ] ~ [ linux-3.1.10 ] ~ [ linux-3.0.101 ] ~ [ 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(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                 flush_workqueue(dlm->dlm_worker);
398                 destroy_workqueue(dlm->dlm_worker);
399                 dlm->dlm_worker = NULL;
400         }
401 }
402 
403 static void dlm_complete_dlm_shutdown(struct dlm_ctxt *dlm)
404 {
405         dlm_unregister_domain_handlers(dlm);
406         dlm_debug_shutdown(dlm);
407         dlm_complete_thread(dlm);
408         dlm_complete_recovery_thread(dlm);
409         dlm_destroy_dlm_worker(dlm);
410 
411         /* We've left the domain. Now we can take ourselves out of the
412          * list and allow the kref stuff to help us free the
413          * memory. */
414         spin_lock(&dlm_domain_lock);
415         list_del_init(&dlm->list);
416         spin_unlock(&dlm_domain_lock);
417 
418         /* Wake up anyone waiting for us to remove this domain */
419         wake_up(&dlm_domain_events);
420 }
421 
422 static int dlm_migrate_all_locks(struct dlm_ctxt *dlm)
423 {
424         int i, num, n, ret = 0;
425         struct dlm_lock_resource *res;
426         struct hlist_node *iter;
427         struct hlist_head *bucket;
428         int dropped;
429 
430         mlog(0, "Migrating locks from domain %s\n", dlm->name);
431 
432         num = 0;
433         spin_lock(&dlm->spinlock);
434         for (i = 0; i < DLM_HASH_BUCKETS; i++) {
435 redo_bucket:
436                 n = 0;
437                 bucket = dlm_lockres_hash(dlm, i);
438                 iter = bucket->first;
439                 while (iter) {
440                         n++;
441                         res = hlist_entry(iter, struct dlm_lock_resource,
442                                           hash_node);
443                         dlm_lockres_get(res);
444                         /* migrate, if necessary.  this will drop the dlm
445                          * spinlock and retake it if it does migration. */
446                         dropped = dlm_empty_lockres(dlm, res);
447 
448                         spin_lock(&res->spinlock);
449                         if (dropped)
450                                 __dlm_lockres_calc_usage(dlm, res);
451                         else
452                                 iter = res->hash_node.next;
453                         spin_unlock(&res->spinlock);
454 
455                         dlm_lockres_put(res);
456 
457                         if (dropped) {
458                                 cond_resched_lock(&dlm->spinlock);
459                                 goto redo_bucket;
460                         }
461                 }
462                 cond_resched_lock(&dlm->spinlock);
463                 num += n;
464         }
465         spin_unlock(&dlm->spinlock);
466         wake_up(&dlm->dlm_thread_wq);
467 
468         /* let the dlm thread take care of purging, keep scanning until
469          * nothing remains in the hash */
470         if (num) {
471                 mlog(0, "%s: %d lock resources in hash last pass\n",
472                      dlm->name, num);
473                 ret = -EAGAIN;
474         }
475         mlog(0, "DONE Migrating locks from domain %s\n", dlm->name);
476         return ret;
477 }
478 
479 static int dlm_no_joining_node(struct dlm_ctxt *dlm)
480 {
481         int ret;
482 
483         spin_lock(&dlm->spinlock);
484         ret = dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN;
485         spin_unlock(&dlm->spinlock);
486 
487         return ret;
488 }
489 
490 static int dlm_begin_exit_domain_handler(struct o2net_msg *msg, u32 len,
491                                          void *data, void **ret_data)
492 {
493         struct dlm_ctxt *dlm = data;
494         unsigned int node;
495         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
496 
497         if (!dlm_grab(dlm))
498                 return 0;
499 
500         node = exit_msg->node_idx;
501         mlog(0, "%s: Node %u sent a begin exit domain message\n", dlm->name, node);
502 
503         spin_lock(&dlm->spinlock);
504         set_bit(node, dlm->exit_domain_map);
505         spin_unlock(&dlm->spinlock);
506 
507         dlm_put(dlm);
508 
509         return 0;
510 }
511 
512 static void dlm_mark_domain_leaving(struct dlm_ctxt *dlm)
513 {
514         /* Yikes, a double spinlock! I need domain_lock for the dlm
515          * state and the dlm spinlock for join state... Sorry! */
516 again:
517         spin_lock(&dlm_domain_lock);
518         spin_lock(&dlm->spinlock);
519 
520         if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
521                 mlog(0, "Node %d is joining, we wait on it.\n",
522                           dlm->joining_node);
523                 spin_unlock(&dlm->spinlock);
524                 spin_unlock(&dlm_domain_lock);
525 
526                 wait_event(dlm->dlm_join_events, dlm_no_joining_node(dlm));
527                 goto again;
528         }
529 
530         dlm->dlm_state = DLM_CTXT_LEAVING;
531         spin_unlock(&dlm->spinlock);
532         spin_unlock(&dlm_domain_lock);
533 }
534 
535 static void __dlm_print_nodes(struct dlm_ctxt *dlm)
536 {
537         int node = -1, num = 0;
538 
539         assert_spin_locked(&dlm->spinlock);
540 
541         printk("( ");
542         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
543                                      node + 1)) < O2NM_MAX_NODES) {
544                 printk("%d ", node);
545                 ++num;
546         }
547         printk(") %u nodes\n", num);
548 }
549 
550 static int dlm_exit_domain_handler(struct o2net_msg *msg, u32 len, void *data,
551                                    void **ret_data)
552 {
553         struct dlm_ctxt *dlm = data;
554         unsigned int node;
555         struct dlm_exit_domain *exit_msg = (struct dlm_exit_domain *) msg->buf;
556 
557         mlog(0, "%p %u %p", msg, len, data);
558 
559         if (!dlm_grab(dlm))
560                 return 0;
561 
562         node = exit_msg->node_idx;
563 
564         spin_lock(&dlm->spinlock);
565         clear_bit(node, dlm->domain_map);
566         clear_bit(node, dlm->exit_domain_map);
567         printk(KERN_NOTICE "o2dlm: Node %u leaves domain %s ", node, dlm->name);
568         __dlm_print_nodes(dlm);
569 
570         /* notify anything attached to the heartbeat events */
571         dlm_hb_event_notify_attached(dlm, node, 0);
572 
573         spin_unlock(&dlm->spinlock);
574 
575         dlm_put(dlm);
576 
577         return 0;
578 }
579 
580 static int dlm_send_one_domain_exit(struct dlm_ctxt *dlm, u32 msg_type,
581                                     unsigned int node)
582 {
583         int status;
584         struct dlm_exit_domain leave_msg;
585 
586         mlog(0, "%s: Sending domain exit message %u to node %u\n", dlm->name,
587              msg_type, node);
588 
589         memset(&leave_msg, 0, sizeof(leave_msg));
590         leave_msg.node_idx = dlm->node_num;
591 
592         status = o2net_send_message(msg_type, dlm->key, &leave_msg,
593                                     sizeof(leave_msg), node, NULL);
594         if (status < 0)
595                 mlog(ML_ERROR, "Error %d sending domain exit message %u "
596                      "to node %u on domain %s\n", status, msg_type, node,
597                      dlm->name);
598 
599         return status;
600 }
601 
602 static void dlm_begin_exit_domain(struct dlm_ctxt *dlm)
603 {
604         int node = -1;
605 
606         /* Support for begin exit domain was added in 1.2 */
607         if (dlm->dlm_locking_proto.pv_major == 1 &&
608             dlm->dlm_locking_proto.pv_minor < 2)
609                 return;
610 
611         /*
612          * Unlike DLM_EXIT_DOMAIN_MSG, DLM_BEGIN_EXIT_DOMAIN_MSG is purely
613          * informational. Meaning if a node does not receive the message,
614          * so be it.
615          */
616         spin_lock(&dlm->spinlock);
617         while (1) {
618                 node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES, node + 1);
619                 if (node >= O2NM_MAX_NODES)
620                         break;
621                 if (node == dlm->node_num)
622                         continue;
623 
624                 spin_unlock(&dlm->spinlock);
625                 dlm_send_one_domain_exit(dlm, DLM_BEGIN_EXIT_DOMAIN_MSG, node);
626                 spin_lock(&dlm->spinlock);
627         }
628         spin_unlock(&dlm->spinlock);
629 }
630 
631 static void dlm_leave_domain(struct dlm_ctxt *dlm)
632 {
633         int node, clear_node, status;
634 
635         /* At this point we've migrated away all our locks and won't
636          * accept mastership of new ones. The dlm is responsible for
637          * almost nothing now. We make sure not to confuse any joining
638          * nodes and then commence shutdown procedure. */
639 
640         spin_lock(&dlm->spinlock);
641         /* Clear ourselves from the domain map */
642         clear_bit(dlm->node_num, dlm->domain_map);
643         while ((node = find_next_bit(dlm->domain_map, O2NM_MAX_NODES,
644                                      0)) < O2NM_MAX_NODES) {
645                 /* Drop the dlm spinlock. This is safe wrt the domain_map.
646                  * -nodes cannot be added now as the
647                  *   query_join_handlers knows to respond with OK_NO_MAP
648                  * -we catch the right network errors if a node is
649                  *   removed from the map while we're sending him the
650                  *   exit message. */
651                 spin_unlock(&dlm->spinlock);
652 
653                 clear_node = 1;
654 
655                 status = dlm_send_one_domain_exit(dlm, DLM_EXIT_DOMAIN_MSG,
656                                                   node);
657                 if (status < 0 &&
658                     status != -ENOPROTOOPT &&
659                     status != -ENOTCONN) {
660                         mlog(ML_NOTICE, "Error %d sending domain exit message "
661                              "to node %d\n", status, node);
662 
663                         /* Not sure what to do here but lets sleep for
664                          * a bit in case this was a transient
665                          * error... */
666                         msleep(DLM_DOMAIN_BACKOFF_MS);
667                         clear_node = 0;
668                 }
669 
670                 spin_lock(&dlm->spinlock);
671                 /* If we're not clearing the node bit then we intend
672                  * to loop back around to try again. */
673                 if (clear_node)
674                         clear_bit(node, dlm->domain_map);
675         }
676         spin_unlock(&dlm->spinlock);
677 }
678 
679 int dlm_shutting_down(struct dlm_ctxt *dlm)
680 {
681         int ret = 0;
682 
683         spin_lock(&dlm_domain_lock);
684 
685         if (dlm->dlm_state == DLM_CTXT_IN_SHUTDOWN)
686                 ret = 1;
687 
688         spin_unlock(&dlm_domain_lock);
689 
690         return ret;
691 }
692 
693 void dlm_unregister_domain(struct dlm_ctxt *dlm)
694 {
695         int leave = 0;
696         struct dlm_lock_resource *res;
697 
698         spin_lock(&dlm_domain_lock);
699         BUG_ON(dlm->dlm_state != DLM_CTXT_JOINED);
700         BUG_ON(!dlm->num_joins);
701 
702         dlm->num_joins--;
703         if (!dlm->num_joins) {
704                 /* We mark it "in shutdown" now so new register
705                  * requests wait until we've completely left the
706                  * domain. Don't use DLM_CTXT_LEAVING yet as we still
707                  * want new domain joins to communicate with us at
708                  * least until we've completed migration of our
709                  * resources. */
710                 dlm->dlm_state = DLM_CTXT_IN_SHUTDOWN;
711                 leave = 1;
712         }
713         spin_unlock(&dlm_domain_lock);
714 
715         if (leave) {
716                 mlog(0, "shutting down domain %s\n", dlm->name);
717                 dlm_begin_exit_domain(dlm);
718 
719                 /* We changed dlm state, notify the thread */
720                 dlm_kick_thread(dlm, NULL);
721 
722                 while (dlm_migrate_all_locks(dlm)) {
723                         /* Give dlm_thread time to purge the lockres' */
724                         msleep(500);
725                         mlog(0, "%s: more migration to do\n", dlm->name);
726                 }
727 
728                 /* This list should be empty. If not, print remaining lockres */
729                 if (!list_empty(&dlm->tracking_list)) {
730                         mlog(ML_ERROR, "Following lockres' are still on the "
731                              "tracking list:\n");
732                         list_for_each_entry(res, &dlm->tracking_list, tracking)
733                                 dlm_print_one_lock_resource(res);
734                 }
735 
736                 dlm_mark_domain_leaving(dlm);
737                 dlm_leave_domain(dlm);
738                 printk(KERN_NOTICE "o2dlm: Leaving domain %s\n", dlm->name);
739                 dlm_force_free_mles(dlm);
740                 dlm_complete_dlm_shutdown(dlm);
741         }
742         dlm_put(dlm);
743 }
744 EXPORT_SYMBOL_GPL(dlm_unregister_domain);
745 
746 static int dlm_query_join_proto_check(char *proto_type, int node,
747                                       struct dlm_protocol_version *ours,
748                                       struct dlm_protocol_version *request)
749 {
750         int rc;
751         struct dlm_protocol_version proto = *request;
752 
753         if (!dlm_protocol_compare(ours, &proto)) {
754                 mlog(0,
755                      "node %u wanted to join with %s locking protocol "
756                      "%u.%u, we respond with %u.%u\n",
757                      node, proto_type,
758                      request->pv_major,
759                      request->pv_minor,
760                      proto.pv_major, proto.pv_minor);
761                 request->pv_minor = proto.pv_minor;
762                 rc = 0;
763         } else {
764                 mlog(ML_NOTICE,
765                      "Node %u wanted to join with %s locking "
766                      "protocol %u.%u, but we have %u.%u, disallowing\n",
767                      node, proto_type,
768                      request->pv_major,
769                      request->pv_minor,
770                      ours->pv_major,
771                      ours->pv_minor);
772                 rc = 1;
773         }
774 
775         return rc;
776 }
777 
778 /*
779  * struct dlm_query_join_packet is made up of four one-byte fields.  They
780  * are effectively in big-endian order already.  However, little-endian
781  * machines swap them before putting the packet on the wire (because
782  * query_join's response is a status, and that status is treated as a u32
783  * on the wire).  Thus, a big-endian and little-endian machines will treat
784  * this structure differently.
785  *
786  * The solution is to have little-endian machines swap the structure when
787  * converting from the structure to the u32 representation.  This will
788  * result in the structure having the correct format on the wire no matter
789  * the host endian format.
790  */
791 static void dlm_query_join_packet_to_wire(struct dlm_query_join_packet *packet,
792                                           u32 *wire)
793 {
794         union dlm_query_join_response response;
795 
796         response.packet = *packet;
797         *wire = be32_to_cpu(response.intval);
798 }
799 
800 static void dlm_query_join_wire_to_packet(u32 wire,
801                                           struct dlm_query_join_packet *packet)
802 {
803         union dlm_query_join_response response;
804 
805         response.intval = cpu_to_be32(wire);
806         *packet = response.packet;
807 }
808 
809 static int dlm_query_join_handler(struct o2net_msg *msg, u32 len, void *data,
810                                   void **ret_data)
811 {
812         struct dlm_query_join_request *query;
813         struct dlm_query_join_packet packet = {
814                 .code = JOIN_DISALLOW,
815         };
816         struct dlm_ctxt *dlm = NULL;
817         u32 response;
818         u8 nodenum;
819 
820         query = (struct dlm_query_join_request *) msg->buf;
821 
822         mlog(0, "node %u wants to join domain %s\n", query->node_idx,
823                   query->domain);
824 
825         /*
826          * If heartbeat doesn't consider the node live, tell it
827          * to back off and try again.  This gives heartbeat a chance
828          * to catch up.
829          */
830         if (!o2hb_check_node_heartbeating_no_sem(query->node_idx)) {
831                 mlog(0, "node %u is not in our live map yet\n",
832                      query->node_idx);
833 
834                 packet.code = JOIN_DISALLOW;
835                 goto respond;
836         }
837 
838         packet.code = JOIN_OK_NO_MAP;
839 
840         spin_lock(&dlm_domain_lock);
841         dlm = __dlm_lookup_domain_full(query->domain, query->name_len);
842         if (!dlm)
843                 goto unlock_respond;
844 
845         /*
846          * There is a small window where the joining node may not see the
847          * node(s) that just left but still part of the cluster. DISALLOW
848          * join request if joining node has different node map.
849          */
850         nodenum=0;
851         while (nodenum < O2NM_MAX_NODES) {
852                 if (test_bit(nodenum, dlm->domain_map)) {
853                         if (!byte_test_bit(nodenum, query->node_map)) {
854                                 mlog(0, "disallow join as node %u does not "
855                                      "have node %u in its nodemap\n",
856                                      query->node_idx, nodenum);
857                                 packet.code = JOIN_DISALLOW;
858                                 goto unlock_respond;
859                         }
860                 }
861                 nodenum++;
862         }
863 
864         /* Once the dlm ctxt is marked as leaving then we don't want
865          * to be put in someone's domain map.
866          * Also, explicitly disallow joining at certain troublesome
867          * times (ie. during recovery). */
868         if (dlm->dlm_state != DLM_CTXT_LEAVING) {
869                 int bit = query->node_idx;
870                 spin_lock(&dlm->spinlock);
871 
872                 if (dlm->dlm_state == DLM_CTXT_NEW &&
873                     dlm->joining_node == DLM_LOCK_RES_OWNER_UNKNOWN) {
874                         /*If this is a brand new context and we
875                          * haven't started our join process yet, then
876                          * the other node won the race. */
877                         packet.code = JOIN_OK_NO_MAP;
878                 } else if (dlm->joining_node != DLM_LOCK_RES_OWNER_UNKNOWN) {
879                         /* Disallow parallel joins. */
880                         packet.code = JOIN_DISALLOW;
881                 } else if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
882                         mlog(0, "node %u trying to join, but recovery "
883                              "is ongoing.\n", bit);
884                         packet.code = JOIN_DISALLOW;
885                 } else if (test_bit(bit, dlm->recovery_map)) {
886                         mlog(0, "node %u trying to join, but it "
887                              "still needs recovery.\n", bit);
888                         packet.code = JOIN_DISALLOW;
889                 } else if (test_bit(bit, dlm->domain_map)) {
890                         mlog(0, "node %u trying to join, but it "
891                              "is still in the domain! needs recovery?\n",
892                              bit);
893                         packet.code = JOIN_DISALLOW;
894                 } else {
895                         /* Alright we're fully a part of this domain
896                          * so we keep some state as to who's joining
897                          * and indicate to him that needs to be fixed
898                          * up. */
899 
900                         /* Make sure we speak compatible locking protocols.  */
901                         if (dlm_query_join_proto_check("DLM", bit,
902                                                        &dlm->dlm_locking_proto,
903                                                        &query->dlm_proto)) {
904                                 packet.code = JOIN_PROTOCOL_MISMATCH;
905                         } else if (dlm_query_join_proto_check("fs", bit,
906                                                               &dlm->fs_locking_proto,
907                                                               &query->fs_proto)) {
908                                 packet.code = JOIN_PROTOCOL_MISMATCH;
909                         } else {
910                                 packet.dlm_minor = query->dlm_proto.pv_minor;
911                                 packet.fs_minor = query->fs_proto.pv_minor;
912                                 packet.code = JOIN_OK;
913                                 __dlm_set_joining_node(dlm, query->node_idx);
914                         }
915                 }
916 
917                 spin_unlock(&dlm->spinlock);
918         }
919 unlock_respond:
920         spin_unlock(&dlm_domain_lock);
921 
922 respond:
923         mlog(0, "We respond with %u\n", packet.code);
924 
925         dlm_query_join_packet_to_wire(&packet, &response);
926         return response;
927 }
928 
929 static int dlm_assert_joined_handler(struct o2net_msg *msg, u32 len, void *data,
930                                      void **ret_data)
931 {
932         struct dlm_assert_joined *assert;
933         struct dlm_ctxt *dlm = NULL;
934 
935         assert = (struct dlm_assert_joined *) msg->buf;
936 
937         mlog(0, "node %u asserts join on domain %s\n", assert->node_idx,
938                   assert->domain);
939 
940         spin_lock(&dlm_domain_lock);
941         dlm = __dlm_lookup_domain_full(assert->domain, assert->name_len);
942         /* XXX should we consider no dlm ctxt an error? */
943         if (dlm) {
944                 spin_lock(&dlm->spinlock);
945 
946                 /* Alright, this node has officially joined our
947                  * domain. Set him in the map and clean up our
948                  * leftover join state. */
949                 BUG_ON(dlm->joining_node != assert->node_idx);
950 
951                 if (dlm->reco.state & DLM_RECO_STATE_ACTIVE) {
952                         mlog(0, "dlm recovery is ongoing, disallow join\n");
953                         spin_unlock(&dlm->spinlock);
954                         spin_unlock(&dlm_domain_lock);
955                         return -EAGAIN;
956                 }
957 
958                 set_bit(assert->node_idx, dlm->domain_map);
959                 clear_bit(assert->node_idx, dlm->exit_domain_map);
960                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
961 
962                 printk(KERN_NOTICE "o2dlm: Node %u joins domain %s ",
963                        assert->node_idx, dlm->name);
964                 __dlm_print_nodes(dlm);
965 
966                 /* notify anything attached to the heartbeat events */
967                 dlm_hb_event_notify_attached(dlm, assert->node_idx, 1);
968 
969                 spin_unlock(&dlm->spinlock);
970         }
971         spin_unlock(&dlm_domain_lock);
972 
973         return 0;
974 }
975 
976 static int dlm_match_regions(struct dlm_ctxt *dlm,
977                              struct dlm_query_region *qr,
978                              char *local, int locallen)
979 {
980         char *remote = qr->qr_regions;
981         char *l, *r;
982         int localnr, i, j, foundit;
983         int status = 0;
984 
985         if (!o2hb_global_heartbeat_active()) {
986                 if (qr->qr_numregions) {
987                         mlog(ML_ERROR, "Domain %s: Joining node %d has global "
988                              "heartbeat enabled but local node %d does not\n",
989                              qr->qr_domain, qr->qr_node, dlm->node_num);
990                         status = -EINVAL;
991                 }
992                 goto bail;
993         }
994 
995         if (o2hb_global_heartbeat_active() && !qr->qr_numregions) {
996                 mlog(ML_ERROR, "Domain %s: Local node %d has global "
997                      "heartbeat enabled but joining node %d does not\n",
998                      qr->qr_domain, dlm->node_num, qr->qr_node);
999                 status = -EINVAL;
1000                 goto bail;
1001         }
1002 
1003         r = remote;
1004         for (i = 0; i < qr->qr_numregions; ++i) {
1005                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, r);
1006                 r += O2HB_MAX_REGION_NAME_LEN;
1007         }
1008 
1009         localnr = min(O2NM_MAX_REGIONS, locallen/O2HB_MAX_REGION_NAME_LEN);
1010         localnr = o2hb_get_all_regions(local, (u8)localnr);
1011 
1012         /* compare local regions with remote */
1013         l = local;
1014         for (i = 0; i < localnr; ++i) {
1015                 foundit = 0;
1016                 r = remote;
1017                 for (j = 0; j <= qr->qr_numregions; ++j) {
1018                         if (!memcmp(l, r, O2HB_MAX_REGION_NAME_LEN)) {
1019                                 foundit = 1;
1020                                 break;
1021                         }
1022                         r += O2HB_MAX_REGION_NAME_LEN;
1023                 }
1024                 if (!foundit) {
1025                         status = -EINVAL;
1026                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1027                              "in local node %d but not in joining node %d\n",
1028                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, l,
1029                              dlm->node_num, qr->qr_node);
1030                         goto bail;
1031                 }
1032                 l += O2HB_MAX_REGION_NAME_LEN;
1033         }
1034 
1035         /* compare remote with local regions */
1036         r = remote;
1037         for (i = 0; i < qr->qr_numregions; ++i) {
1038                 foundit = 0;
1039                 l = local;
1040                 for (j = 0; j < localnr; ++j) {
1041                         if (!memcmp(r, l, O2HB_MAX_REGION_NAME_LEN)) {
1042                                 foundit = 1;
1043                                 break;
1044                         }
1045                         l += O2HB_MAX_REGION_NAME_LEN;
1046                 }
1047                 if (!foundit) {
1048                         status = -EINVAL;
1049                         mlog(ML_ERROR, "Domain %s: Region '%.*s' registered "
1050                              "in joining node %d but not in local node %d\n",
1051                              qr->qr_domain, O2HB_MAX_REGION_NAME_LEN, r,
1052                              qr->qr_node, dlm->node_num);
1053                         goto bail;
1054                 }
1055                 r += O2HB_MAX_REGION_NAME_LEN;
1056         }
1057 
1058 bail:
1059         return status;
1060 }
1061 
1062 static int dlm_send_regions(struct dlm_ctxt *dlm, unsigned long *node_map)
1063 {
1064         struct dlm_query_region *qr = NULL;
1065         int status, ret = 0, i;
1066         char *p;
1067 
1068         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1069                 goto bail;
1070 
1071         qr = kzalloc(sizeof(struct dlm_query_region), GFP_KERNEL);
1072         if (!qr) {
1073                 ret = -ENOMEM;
1074                 mlog_errno(ret);
1075                 goto bail;
1076         }
1077 
1078         qr->qr_node = dlm->node_num;
1079         qr->qr_namelen = strlen(dlm->name);
1080         memcpy(qr->qr_domain, dlm->name, qr->qr_namelen);
1081         /* if local hb, the numregions will be zero */
1082         if (o2hb_global_heartbeat_active())
1083                 qr->qr_numregions = o2hb_get_all_regions(qr->qr_regions,
1084                                                          O2NM_MAX_REGIONS);
1085 
1086         p = qr->qr_regions;
1087         for (i = 0; i < qr->qr_numregions; ++i, p += O2HB_MAX_REGION_NAME_LEN)
1088                 mlog(0, "Region %.*s\n", O2HB_MAX_REGION_NAME_LEN, p);
1089 
1090         i = -1;
1091         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1092                                   i + 1)) < O2NM_MAX_NODES) {
1093                 if (i == dlm->node_num)
1094                         continue;
1095 
1096                 mlog(0, "Sending regions to node %d\n", i);
1097 
1098                 ret = o2net_send_message(DLM_QUERY_REGION, DLM_MOD_KEY, qr,
1099                                          sizeof(struct dlm_query_region),
1100                                          i, &status);
1101                 if (ret >= 0)
1102                         ret = status;
1103                 if (ret) {
1104                         mlog(ML_ERROR, "Region mismatch %d, node %d\n",
1105                              ret, i);
1106                         break;
1107                 }
1108         }
1109 
1110 bail:
1111         kfree(qr);
1112         return ret;
1113 }
1114 
1115 static int dlm_query_region_handler(struct o2net_msg *msg, u32 len,
1116                                     void *data, void **ret_data)
1117 {
1118         struct dlm_query_region *qr;
1119         struct dlm_ctxt *dlm = NULL;
1120         char *local = NULL;
1121         int status = 0;
1122 
1123         qr = (struct dlm_query_region *) msg->buf;
1124 
1125         mlog(0, "Node %u queries hb regions on domain %s\n", qr->qr_node,
1126              qr->qr_domain);
1127 
1128         /* buffer used in dlm_mast_regions() */
1129         local = kmalloc(sizeof(qr->qr_regions), GFP_KERNEL);
1130         if (!local)
1131                 return -ENOMEM;
1132 
1133         status = -EINVAL;
1134 
1135         spin_lock(&dlm_domain_lock);
1136         dlm = __dlm_lookup_domain_full(qr->qr_domain, qr->qr_namelen);
1137         if (!dlm) {
1138                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1139                      "before join domain\n", qr->qr_node, qr->qr_domain);
1140                 goto out_domain_lock;
1141         }
1142 
1143         spin_lock(&dlm->spinlock);
1144         if (dlm->joining_node != qr->qr_node) {
1145                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1146                      "but joining node is %d\n", qr->qr_node, qr->qr_domain,
1147                      dlm->joining_node);
1148                 goto out_dlm_lock;
1149         }
1150 
1151         /* Support for global heartbeat was added in 1.1 */
1152         if (dlm->dlm_locking_proto.pv_major == 1 &&
1153             dlm->dlm_locking_proto.pv_minor == 0) {
1154                 mlog(ML_ERROR, "Node %d queried hb regions on domain %s "
1155                      "but active dlm protocol is %d.%d\n", qr->qr_node,
1156                      qr->qr_domain, dlm->dlm_locking_proto.pv_major,
1157                      dlm->dlm_locking_proto.pv_minor);
1158                 goto out_dlm_lock;
1159         }
1160 
1161         status = dlm_match_regions(dlm, qr, local, sizeof(qr->qr_regions));
1162 
1163 out_dlm_lock:
1164         spin_unlock(&dlm->spinlock);
1165 
1166 out_domain_lock:
1167         spin_unlock(&dlm_domain_lock);
1168 
1169         kfree(local);
1170 
1171         return status;
1172 }
1173 
1174 static int dlm_match_nodes(struct dlm_ctxt *dlm, struct dlm_query_nodeinfo *qn)
1175 {
1176         struct o2nm_node *local;
1177         struct dlm_node_info *remote;
1178         int i, j;
1179         int status = 0;
1180 
1181         for (j = 0; j < qn->qn_numnodes; ++j)
1182                 mlog(0, "Node %3d, %pI4:%u\n", qn->qn_nodes[j].ni_nodenum,
1183                      &(qn->qn_nodes[j].ni_ipv4_address),
1184                      ntohs(qn->qn_nodes[j].ni_ipv4_port));
1185 
1186         for (i = 0; i < O2NM_MAX_NODES && !status; ++i) {
1187                 local = o2nm_get_node_by_num(i);
1188                 remote = NULL;
1189                 for (j = 0; j < qn->qn_numnodes; ++j) {
1190                         if (qn->qn_nodes[j].ni_nodenum == i) {
1191                                 remote = &(qn->qn_nodes[j]);
1192                                 break;
1193                         }
1194                 }
1195 
1196                 if (!local && !remote)
1197                         continue;
1198 
1199                 if ((local && !remote) || (!local && remote))
1200                         status = -EINVAL;
1201 
1202                 if (!status &&
1203                     ((remote->ni_nodenum != local->nd_num) ||
1204                      (remote->ni_ipv4_port != local->nd_ipv4_port) ||
1205                      (remote->ni_ipv4_address != local->nd_ipv4_address)))
1206                         status = -EINVAL;
1207 
1208                 if (status) {
1209                         if (remote && !local)
1210                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1211                                      "registered in joining node %d but not in "
1212                                      "local node %d\n", qn->qn_domain,
1213                                      remote->ni_nodenum,
1214                                      &(remote->ni_ipv4_address),
1215                                      ntohs(remote->ni_ipv4_port),
1216                                      qn->qn_nodenum, dlm->node_num);
1217                         if (local && !remote)
1218                                 mlog(ML_ERROR, "Domain %s: Node %d (%pI4:%u) "
1219                                      "registered in local node %d but not in "
1220                                      "joining node %d\n", qn->qn_domain,
1221                                      local->nd_num, &(local->nd_ipv4_address),
1222                                      ntohs(local->nd_ipv4_port),
1223                                      dlm->node_num, qn->qn_nodenum);
1224                         BUG_ON((!local && !remote));
1225                 }
1226 
1227                 if (local)
1228                         o2nm_node_put(local);
1229         }
1230 
1231         return status;
1232 }
1233 
1234 static int dlm_send_nodeinfo(struct dlm_ctxt *dlm, unsigned long *node_map)
1235 {
1236         struct dlm_query_nodeinfo *qn = NULL;
1237         struct o2nm_node *node;
1238         int ret = 0, status, count, i;
1239 
1240         if (find_next_bit(node_map, O2NM_MAX_NODES, 0) >= O2NM_MAX_NODES)
1241                 goto bail;
1242 
1243         qn = kzalloc(sizeof(struct dlm_query_nodeinfo), GFP_KERNEL);
1244         if (!qn) {
1245                 ret = -ENOMEM;
1246                 mlog_errno(ret);
1247                 goto bail;
1248         }
1249 
1250         for (i = 0, count = 0; i < O2NM_MAX_NODES; ++i) {
1251                 node = o2nm_get_node_by_num(i);
1252                 if (!node)
1253                         continue;
1254                 qn->qn_nodes[count].ni_nodenum = node->nd_num;
1255                 qn->qn_nodes[count].ni_ipv4_port = node->nd_ipv4_port;
1256                 qn->qn_nodes[count].ni_ipv4_address = node->nd_ipv4_address;
1257                 mlog(0, "Node %3d, %pI4:%u\n", node->nd_num,
1258                      &(node->nd_ipv4_address), ntohs(node->nd_ipv4_port));
1259                 ++count;
1260                 o2nm_node_put(node);
1261         }
1262 
1263         qn->qn_nodenum = dlm->node_num;
1264         qn->qn_numnodes = count;
1265         qn->qn_namelen = strlen(dlm->name);
1266         memcpy(qn->qn_domain, dlm->name, qn->qn_namelen);
1267 
1268         i = -1;
1269         while ((i = find_next_bit(node_map, O2NM_MAX_NODES,
1270                                   i + 1)) < O2NM_MAX_NODES) {
1271                 if (i == dlm->node_num)
1272                         continue;
1273 
1274                 mlog(0, "Sending nodeinfo to node %d\n", i);
1275 
1276                 ret = o2net_send_message(DLM_QUERY_NODEINFO, DLM_MOD_KEY,
1277                                          qn, sizeof(struct dlm_query_nodeinfo),
1278                                          i, &status);
1279                 if (ret >= 0)
1280                         ret = status;
1281                 if (ret) {
1282                         mlog(ML_ERROR, "node mismatch %d, node %d\n", ret, i);
1283                         break;
1284                 }
1285         }
1286 
1287 bail:
1288         kfree(qn);
1289         return ret;
1290 }
1291 
1292 static int dlm_query_nodeinfo_handler(struct o2net_msg *msg, u32 len,
1293                                       void *data, void **ret_data)
1294 {
1295         struct dlm_query_nodeinfo *qn;
1296         struct dlm_ctxt *dlm = NULL;
1297         int locked = 0, status = -EINVAL;
1298 
1299         qn = (struct dlm_query_nodeinfo *) msg->buf;
1300 
1301         mlog(0, "Node %u queries nodes on domain %s\n", qn->qn_nodenum,
1302              qn->qn_domain);
1303 
1304         spin_lock(&dlm_domain_lock);
1305         dlm = __dlm_lookup_domain_full(qn->qn_domain, qn->qn_namelen);
1306         if (!dlm) {
1307                 mlog(ML_ERROR, "Node %d queried nodes on domain %s before "
1308                      "join domain\n", qn->qn_nodenum, qn->qn_domain);
1309                 goto bail;
1310         }
1311 
1312         spin_lock(&dlm->spinlock);
1313         locked = 1;
1314         if (dlm->joining_node != qn->qn_nodenum) {
1315                 mlog(ML_ERROR, "Node %d queried nodes on domain %s but "
1316                      "joining node is %d\n", qn->qn_nodenum, qn->qn_domain,
1317                      dlm->joining_node);
1318                 goto bail;
1319         }
1320 
1321         /* Support for node query was added in 1.1 */
1322         if (dlm->dlm_locking_proto.pv_major == 1 &&
1323             dlm->dlm_locking_proto.pv_minor == 0) {
1324                 mlog(ML_ERROR, "Node %d queried nodes on domain %s "
1325                      "but active dlm protocol is %d.%d\n", qn->qn_nodenum,
1326                      qn->qn_domain, dlm->dlm_locking_proto.pv_major,
1327                      dlm->dlm_locking_proto.pv_minor);
1328                 goto bail;
1329         }
1330 
1331         status = dlm_match_nodes(dlm, qn);
1332 
1333 bail:
1334         if (locked)
1335                 spin_unlock(&dlm->spinlock);
1336         spin_unlock(&dlm_domain_lock);
1337 
1338         return status;
1339 }
1340 
1341 static int dlm_cancel_join_handler(struct o2net_msg *msg, u32 len, void *data,
1342                                    void **ret_data)
1343 {
1344         struct dlm_cancel_join *cancel;
1345         struct dlm_ctxt *dlm = NULL;
1346 
1347         cancel = (struct dlm_cancel_join *) msg->buf;
1348 
1349         mlog(0, "node %u cancels join on domain %s\n", cancel->node_idx,
1350                   cancel->domain);
1351 
1352         spin_lock(&dlm_domain_lock);
1353         dlm = __dlm_lookup_domain_full(cancel->domain, cancel->name_len);
1354 
1355         if (dlm) {
1356                 spin_lock(&dlm->spinlock);
1357 
1358                 /* Yikes, this guy wants to cancel his join. No
1359                  * problem, we simply cleanup our join state. */
1360                 BUG_ON(dlm->joining_node != cancel->node_idx);
1361                 __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1362 
1363                 spin_unlock(&dlm->spinlock);
1364         }
1365         spin_unlock(&dlm_domain_lock);
1366 
1367         return 0;
1368 }
1369 
1370 static int dlm_send_one_join_cancel(struct dlm_ctxt *dlm,
1371                                     unsigned int node)
1372 {
1373         int status;
1374         struct dlm_cancel_join cancel_msg;
1375 
1376         memset(&cancel_msg, 0, sizeof(cancel_msg));
1377         cancel_msg.node_idx = dlm->node_num;
1378         cancel_msg.name_len = strlen(dlm->name);
1379         memcpy(cancel_msg.domain, dlm->name, cancel_msg.name_len);
1380 
1381         status = o2net_send_message(DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1382                                     &cancel_msg, sizeof(cancel_msg), node,
1383                                     NULL);
1384         if (status < 0) {
1385                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1386                      "node %u\n", status, DLM_CANCEL_JOIN_MSG, DLM_MOD_KEY,
1387                      node);
1388                 goto bail;
1389         }
1390 
1391 bail:
1392         return status;
1393 }
1394 
1395 /* map_size should be in bytes. */
1396 static int dlm_send_join_cancels(struct dlm_ctxt *dlm,
1397                                  unsigned long *node_map,
1398                                  unsigned int map_size)
1399 {
1400         int status, tmpstat;
1401         int node;
1402 
1403         if (map_size != (BITS_TO_LONGS(O2NM_MAX_NODES) *
1404                          sizeof(unsigned long))) {
1405                 mlog(ML_ERROR,
1406                      "map_size %u != BITS_TO_LONGS(O2NM_MAX_NODES) %u\n",
1407                      map_size, (unsigned)BITS_TO_LONGS(O2NM_MAX_NODES));
1408                 return -EINVAL;
1409         }
1410 
1411         status = 0;
1412         node = -1;
1413         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1414                                      node + 1)) < O2NM_MAX_NODES) {
1415                 if (node == dlm->node_num)
1416                         continue;
1417 
1418                 tmpstat = dlm_send_one_join_cancel(dlm, node);
1419                 if (tmpstat) {
1420                         mlog(ML_ERROR, "Error return %d cancelling join on "
1421                              "node %d\n", tmpstat, node);
1422                         if (!status)
1423                                 status = tmpstat;
1424                 }
1425         }
1426 
1427         if (status)
1428                 mlog_errno(status);
1429         return status;
1430 }
1431 
1432 static int dlm_request_join(struct dlm_ctxt *dlm,
1433                             int node,
1434                             enum dlm_query_join_response_code *response)
1435 {
1436         int status;
1437         struct dlm_query_join_request join_msg;
1438         struct dlm_query_join_packet packet;
1439         u32 join_resp;
1440 
1441         mlog(0, "querying node %d\n", node);
1442 
1443         memset(&join_msg, 0, sizeof(join_msg));
1444         join_msg.node_idx = dlm->node_num;
1445         join_msg.name_len = strlen(dlm->name);
1446         memcpy(join_msg.domain, dlm->name, join_msg.name_len);
1447         join_msg.dlm_proto = dlm->dlm_locking_proto;
1448         join_msg.fs_proto = dlm->fs_locking_proto;
1449 
1450         /* copy live node map to join message */
1451         byte_copymap(join_msg.node_map, dlm->live_nodes_map, O2NM_MAX_NODES);
1452 
1453         status = o2net_send_message(DLM_QUERY_JOIN_MSG, DLM_MOD_KEY, &join_msg,
1454                                     sizeof(join_msg), node, &join_resp);
1455         if (status < 0 && status != -ENOPROTOOPT) {
1456                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1457                      "node %u\n", status, DLM_QUERY_JOIN_MSG, DLM_MOD_KEY,
1458                      node);
1459                 goto bail;
1460         }
1461         dlm_query_join_wire_to_packet(join_resp, &packet);
1462 
1463         /* -ENOPROTOOPT from the net code means the other side isn't
1464             listening for our message type -- that's fine, it means
1465             his dlm isn't up, so we can consider him a 'yes' but not
1466             joined into the domain.  */
1467         if (status == -ENOPROTOOPT) {
1468                 status = 0;
1469                 *response = JOIN_OK_NO_MAP;
1470         } else {
1471                 *response = packet.code;
1472                 switch (packet.code) {
1473                 case JOIN_DISALLOW:
1474                 case JOIN_OK_NO_MAP:
1475                         break;
1476                 case JOIN_PROTOCOL_MISMATCH:
1477                         mlog(ML_NOTICE,
1478                              "This node requested DLM locking protocol %u.%u and "
1479                              "filesystem locking protocol %u.%u.  At least one of "
1480                              "the protocol versions on node %d is not compatible, "
1481                              "disconnecting\n",
1482                              dlm->dlm_locking_proto.pv_major,
1483                              dlm->dlm_locking_proto.pv_minor,
1484                              dlm->fs_locking_proto.pv_major,
1485                              dlm->fs_locking_proto.pv_minor,
1486                              node);
1487                         status = -EPROTO;
1488                         break;
1489                 case JOIN_OK:
1490                         /* Use the same locking protocol as the remote node */
1491                         dlm->dlm_locking_proto.pv_minor = packet.dlm_minor;
1492                         dlm->fs_locking_proto.pv_minor = packet.fs_minor;
1493                         mlog(0,
1494                              "Node %d responds JOIN_OK with DLM locking protocol "
1495                              "%u.%u and fs locking protocol %u.%u\n",
1496                              node,
1497                              dlm->dlm_locking_proto.pv_major,
1498                              dlm->dlm_locking_proto.pv_minor,
1499                              dlm->fs_locking_proto.pv_major,
1500                              dlm->fs_locking_proto.pv_minor);
1501                         break;
1502                 default:
1503                         status = -EINVAL;
1504                         mlog(ML_ERROR, "invalid response %d from node %u\n",
1505                              packet.code, node);
1506                         /* Reset response to JOIN_DISALLOW */
1507                         *response = JOIN_DISALLOW;
1508                         break;
1509                 }
1510         }
1511 
1512         mlog(0, "status %d, node %d response is %d\n", status, node,
1513              *response);
1514 
1515 bail:
1516         return status;
1517 }
1518 
1519 static int dlm_send_one_join_assert(struct dlm_ctxt *dlm,
1520                                     unsigned int node)
1521 {
1522         int status;
1523         int ret;
1524         struct dlm_assert_joined assert_msg;
1525 
1526         mlog(0, "Sending join assert to node %u\n", node);
1527 
1528         memset(&assert_msg, 0, sizeof(assert_msg));
1529         assert_msg.node_idx = dlm->node_num;
1530         assert_msg.name_len = strlen(dlm->name);
1531         memcpy(assert_msg.domain, dlm->name, assert_msg.name_len);
1532 
1533         status = o2net_send_message(DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1534                                     &assert_msg, sizeof(assert_msg), node,
1535                                     &ret);
1536         if (status < 0)
1537                 mlog(ML_ERROR, "Error %d when sending message %u (key 0x%x) to "
1538                      "node %u\n", status, DLM_ASSERT_JOINED_MSG, DLM_MOD_KEY,
1539                      node);
1540         else
1541                 status = ret;
1542 
1543         return status;
1544 }
1545 
1546 static void dlm_send_join_asserts(struct dlm_ctxt *dlm,
1547                                   unsigned long *node_map)
1548 {
1549         int status, node, live;
1550 
1551         status = 0;
1552         node = -1;
1553         while ((node = find_next_bit(node_map, O2NM_MAX_NODES,
1554                                      node + 1)) < O2NM_MAX_NODES) {
1555                 if (node == dlm->node_num)
1556                         continue;
1557 
1558                 do {
1559                         /* It is very important that this message be
1560                          * received so we spin until either the node
1561                          * has died or it gets the message. */
1562                         status = dlm_send_one_join_assert(dlm, node);
1563 
1564                         spin_lock(&dlm->spinlock);
1565                         live = test_bit(node, dlm->live_nodes_map);
1566                         spin_unlock(&dlm->spinlock);
1567 
1568                         if (status) {
1569                                 mlog(ML_ERROR, "Error return %d asserting "
1570                                      "join on node %d\n", status, node);
1571 
1572                                 /* give us some time between errors... */
1573                                 if (live)
1574                                         msleep(DLM_DOMAIN_BACKOFF_MS);
1575                         }
1576                 } while (status && live);
1577         }
1578 }
1579 
1580 struct domain_join_ctxt {
1581         unsigned long live_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1582         unsigned long yes_resp_map[BITS_TO_LONGS(O2NM_MAX_NODES)];
1583 };
1584 
1585 static int dlm_should_restart_join(struct dlm_ctxt *dlm,
1586                                    struct domain_join_ctxt *ctxt,
1587                                    enum dlm_query_join_response_code response)
1588 {
1589         int ret;
1590 
1591         if (response == JOIN_DISALLOW) {
1592                 mlog(0, "Latest response of disallow -- should restart\n");
1593                 return 1;
1594         }
1595 
1596         spin_lock(&dlm->spinlock);
1597         /* For now, we restart the process if the node maps have
1598          * changed at all */
1599         ret = memcmp(ctxt->live_map, dlm->live_nodes_map,
1600                      sizeof(dlm->live_nodes_map));
1601         spin_unlock(&dlm->spinlock);
1602 
1603         if (ret)
1604                 mlog(0, "Node maps changed -- should restart\n");
1605 
1606         return ret;
1607 }
1608 
1609 static int dlm_try_to_join_domain(struct dlm_ctxt *dlm)
1610 {
1611         int status = 0, tmpstat, node;
1612         struct domain_join_ctxt *ctxt;
1613         enum dlm_query_join_response_code response = JOIN_DISALLOW;
1614 
1615         mlog(0, "%p", dlm);
1616 
1617         ctxt = kzalloc(sizeof(*ctxt), GFP_KERNEL);
1618         if (!ctxt) {
1619                 status = -ENOMEM;
1620                 mlog_errno(status);
1621                 goto bail;
1622         }
1623 
1624         /* group sem locking should work for us here -- we're already
1625          * registered for heartbeat events so filling this should be
1626          * atomic wrt getting those handlers called. */
1627         o2hb_fill_node_map(dlm->live_nodes_map, sizeof(dlm->live_nodes_map));
1628 
1629         spin_lock(&dlm->spinlock);
1630         memcpy(ctxt->live_map, dlm->live_nodes_map, sizeof(ctxt->live_map));
1631 
1632         __dlm_set_joining_node(dlm, dlm->node_num);
1633 
1634         spin_unlock(&dlm->spinlock);
1635 
1636         node = -1;
1637         while ((node = find_next_bit(ctxt->live_map, O2NM_MAX_NODES,
1638                                      node + 1)) < O2NM_MAX_NODES) {
1639                 if (node == dlm->node_num)
1640                         continue;
1641 
1642                 status = dlm_request_join(dlm, node, &response);
1643                 if (status < 0) {
1644                         mlog_errno(status);
1645                         goto bail;
1646                 }
1647 
1648                 /* Ok, either we got a response or the node doesn't have a
1649                  * dlm up. */
1650                 if (response == JOIN_OK)
1651                         set_bit(node, ctxt->yes_resp_map);
1652 
1653                 if (dlm_should_restart_join(dlm, ctxt, response)) {
1654                         status = -EAGAIN;
1655                         goto bail;
1656                 }
1657         }
1658 
1659         mlog(0, "Yay, done querying nodes!\n");
1660 
1661         /* Yay, everyone agree's we can join the domain. My domain is
1662          * comprised of all nodes who were put in the
1663          * yes_resp_map. Copy that into our domain map and send a join
1664          * assert message to clean up everyone elses state. */
1665         spin_lock(&dlm->spinlock);
1666         memcpy(dlm->domain_map, ctxt->yes_resp_map,
1667                sizeof(ctxt->yes_resp_map));
1668         set_bit(dlm->node_num, dlm->domain_map);
1669         spin_unlock(&dlm->spinlock);
1670 
1671         /* Support for global heartbeat and node info was added in 1.1 */
1672         if (dlm->dlm_locking_proto.pv_major > 1 ||
1673             dlm->dlm_locking_proto.pv_minor > 0) {
1674                 status = dlm_send_nodeinfo(dlm, ctxt->yes_resp_map);
1675                 if (status) {
1676                         mlog_errno(status);
1677                         goto bail;
1678                 }
1679                 status = dlm_send_regions(dlm, ctxt->yes_resp_map);
1680                 if (status) {
1681                         mlog_errno(status);
1682                         goto bail;
1683                 }
1684         }
1685 
1686         dlm_send_join_asserts(dlm, ctxt->yes_resp_map);
1687 
1688         /* Joined state *must* be set before the joining node
1689          * information, otherwise the query_join handler may read no
1690          * current joiner but a state of NEW and tell joining nodes
1691          * we're not in the domain. */
1692         spin_lock(&dlm_domain_lock);
1693         dlm->dlm_state = DLM_CTXT_JOINED;
1694         dlm->num_joins++;
1695         spin_unlock(&dlm_domain_lock);
1696 
1697 bail:
1698         spin_lock(&dlm->spinlock);
1699         __dlm_set_joining_node(dlm, DLM_LOCK_RES_OWNER_UNKNOWN);
1700         if (!status) {
1701                 printk(KERN_NOTICE "o2dlm: Joining domain %s ", dlm->name);
1702                 __dlm_print_nodes(dlm);
1703         }
1704         spin_unlock(&dlm->spinlock);
1705 
1706         if (ctxt) {
1707                 /* Do we need to send a cancel message to any nodes? */
1708                 if (status < 0) {
1709                         tmpstat = dlm_send_join_cancels(dlm,
1710                                                         ctxt->yes_resp_map,
1711                                                         sizeof(ctxt->yes_resp_map));
1712                         if (tmpstat < 0)
1713                                 mlog_errno(tmpstat);
1714                 }
1715                 kfree(ctxt);
1716         }
1717 
1718         mlog(0, "returning %d\n", status);
1719         return status;
1720 }
1721 
1722 static void dlm_unregister_domain_handlers(struct dlm_ctxt *dlm)
1723 {
1724         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_up);
1725         o2hb_unregister_callback(dlm->name, &dlm->dlm_hb_down);
1726         o2net_unregister_handler_list(&dlm->dlm_domain_handlers);
1727 }
1728 
1729 static int dlm_register_domain_handlers(struct dlm_ctxt *dlm)
1730 {
1731         int status;
1732 
1733         mlog(0, "registering handlers.\n");
1734 
1735         o2hb_setup_callback(&dlm->dlm_hb_down, O2HB_NODE_DOWN_CB,
1736                             dlm_hb_node_down_cb, dlm, DLM_HB_NODE_DOWN_PRI);
1737         o2hb_setup_callback(&dlm->dlm_hb_up, O2HB_NODE_UP_CB,
1738                             dlm_hb_node_up_cb, dlm, DLM_HB_NODE_UP_PRI);
1739 
1740         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_down);
1741         if (status)
1742                 goto bail;
1743 
1744         status = o2hb_register_callback(dlm->name, &dlm->dlm_hb_up);
1745         if (status)
1746                 goto bail;
1747 
1748         status = o2net_register_handler(DLM_MASTER_REQUEST_MSG, dlm->key,
1749                                         sizeof(struct dlm_master_request),
1750                                         dlm_master_request_handler,
1751                                         dlm, NULL, &dlm->dlm_domain_handlers);
1752         if (status)
1753                 goto bail;
1754 
1755         status = o2net_register_handler(DLM_ASSERT_MASTER_MSG, dlm->key,
1756                                         sizeof(struct dlm_assert_master),
1757                                         dlm_assert_master_handler,
1758                                         dlm, dlm_assert_master_post_handler,
1759                                         &dlm->dlm_domain_handlers);
1760         if (status)
1761                 goto bail;
1762 
1763         status = o2net_register_handler(DLM_CREATE_LOCK_MSG, dlm->key,
1764                                         sizeof(struct dlm_create_lock),
1765                                         dlm_create_lock_handler,
1766                                         dlm, NULL, &dlm->dlm_domain_handlers);
1767         if (status)
1768                 goto bail;
1769 
1770         status = o2net_register_handler(DLM_CONVERT_LOCK_MSG, dlm->key,
1771                                         DLM_CONVERT_LOCK_MAX_LEN,
1772                                         dlm_convert_lock_handler,
1773                                         dlm, NULL, &dlm->dlm_domain_handlers);
1774         if (status)
1775                 goto bail;
1776 
1777         status = o2net_register_handler(DLM_UNLOCK_LOCK_MSG, dlm->key,
1778                                         DLM_UNLOCK_LOCK_MAX_LEN,
1779                                         dlm_unlock_lock_handler,
1780                                         dlm, NULL, &dlm->dlm_domain_handlers);
1781         if (status)
1782                 goto bail;
1783 
1784         status = o2net_register_handler(DLM_PROXY_AST_MSG, dlm->key,
1785                                         DLM_PROXY_AST_MAX_LEN,
1786                                         dlm_proxy_ast_handler,
1787                                         dlm, NULL, &dlm->dlm_domain_handlers);
1788         if (status)
1789                 goto bail;
1790 
1791         status = o2net_register_handler(DLM_EXIT_DOMAIN_MSG, dlm->key,
1792                                         sizeof(struct dlm_exit_domain),
1793                                         dlm_exit_domain_handler,
1794                                         dlm, NULL, &dlm->dlm_domain_handlers);
1795         if (status)
1796                 goto bail;
1797 
1798         status = o2net_register_handler(DLM_DEREF_LOCKRES_MSG, dlm->key,
1799                                         sizeof(struct dlm_deref_lockres),
1800                                         dlm_deref_lockres_handler,
1801                                         dlm, NULL, &dlm->dlm_domain_handlers);
1802         if (status)
1803                 goto bail;
1804 
1805         status = o2net_register_handler(DLM_MIGRATE_REQUEST_MSG, dlm->key,
1806                                         sizeof(struct dlm_migrate_request),
1807                                         dlm_migrate_request_handler,
1808                                         dlm, NULL, &dlm->dlm_domain_handlers);
1809         if (status)
1810                 goto bail;
1811 
1812         status = o2net_register_handler(DLM_MIG_LOCKRES_MSG, dlm->key,
1813                                         DLM_MIG_LOCKRES_MAX_LEN,
1814                                         dlm_mig_lockres_handler,
1815                                         dlm, NULL, &dlm->dlm_domain_handlers);
1816         if (status)
1817                 goto bail;
1818 
1819         status = o2net_register_handler(DLM_MASTER_REQUERY_MSG, dlm->key,
1820                                         sizeof(struct dlm_master_requery),
1821                                         dlm_master_requery_handler,
1822                                         dlm, NULL, &dlm->dlm_domain_handlers);
1823         if (status)
1824                 goto bail;
1825 
1826         status = o2net_register_handler(DLM_LOCK_REQUEST_MSG, dlm->key,
1827                                         sizeof(struct dlm_lock_request),
1828                                         dlm_request_all_locks_handler,
1829                                         dlm, NULL, &dlm->dlm_domain_handlers);
1830         if (status)
1831                 goto bail;
1832 
1833         status = o2net_register_handler(DLM_RECO_DATA_DONE_MSG, dlm->key,
1834                                         sizeof(struct dlm_reco_data_done),
1835                                         dlm_reco_data_done_handler,
1836                                         dlm, NULL, &dlm->dlm_domain_handlers);
1837         if (status)
1838                 goto bail;
1839 
1840         status = o2net_register_handler(DLM_BEGIN_RECO_MSG, dlm->key,
1841                                         sizeof(struct dlm_begin_reco),
1842                                         dlm_begin_reco_handler,
1843                                         dlm, NULL, &dlm->dlm_domain_handlers);
1844         if (status)
1845                 goto bail;
1846 
1847         status = o2net_register_handler(DLM_FINALIZE_RECO_MSG, dlm->key,
1848                                         sizeof(struct dlm_finalize_reco),
1849                                         dlm_finalize_reco_handler,
1850                                         dlm, NULL, &dlm->dlm_domain_handlers);
1851         if (status)
1852                 goto bail;
1853 
1854         status = o2net_register_handler(DLM_BEGIN_EXIT_DOMAIN_MSG, dlm->key,
1855                                         sizeof(struct dlm_exit_domain),
1856                                         dlm_begin_exit_domain_handler,
1857                                         dlm, NULL, &dlm->dlm_domain_handlers);
1858         if (status)
1859                 goto bail;
1860 
1861         status = o2net_register_handler(DLM_DEREF_LOCKRES_DONE, dlm->key,
1862                                         sizeof(struct dlm_deref_lockres_done),
1863                                         dlm_deref_lockres_done_handler,
1864                                         dlm, NULL, &dlm->dlm_domain_handlers);
1865 bail:
1866         if (status)
1867                 dlm_unregister_domain_handlers(dlm);
1868 
1869         return status;
1870 }
1871 
1872 static int dlm_join_domain(struct dlm_ctxt *dlm)
1873 {
1874         int status;
1875         unsigned int backoff;
1876         unsigned int total_backoff = 0;
1877         char wq_name[O2NM_MAX_NAME_LEN];
1878 
1879         BUG_ON(!dlm);
1880 
1881         mlog(0, "Join domain %s\n", dlm->name);
1882 
1883         status = dlm_register_domain_handlers(dlm);
1884         if (status) {
1885                 mlog_errno(status);
1886                 goto bail;
1887         }
1888 
1889         status = dlm_launch_thread(dlm);
1890         if (status < 0) {
1891                 mlog_errno(status);
1892                 goto bail;
1893         }
1894 
1895         status = dlm_launch_recovery_thread(dlm);
1896         if (status < 0) {
1897                 mlog_errno(status);
1898                 goto bail;
1899         }
1900 
1901         status = dlm_debug_init(dlm);
1902         if (status < 0) {
1903                 mlog_errno(status);
1904                 goto bail;
1905         }
1906 
1907         snprintf(wq_name, O2NM_MAX_NAME_LEN, "dlm_wq-%s", dlm->name);
1908         dlm->dlm_worker = alloc_workqueue(wq_name, WQ_MEM_RECLAIM, 0);
1909         if (!dlm->dlm_worker) {
1910                 status = -ENOMEM;
1911                 mlog_errno(status);
1912                 goto bail;
1913         }
1914 
1915         do {
1916                 status = dlm_try_to_join_domain(dlm);
1917 
1918                 /* If we're racing another node to the join, then we
1919                  * need to back off temporarily and let them
1920                  * complete. */
1921 #define DLM_JOIN_TIMEOUT_MSECS  90000
1922                 if (status == -EAGAIN) {
1923                         if (signal_pending(current)) {
1924                                 status = -ERESTARTSYS;
1925                                 goto bail;
1926                         }
1927 
1928                         if (total_backoff > DLM_JOIN_TIMEOUT_MSECS) {
1929                                 status = -ERESTARTSYS;
1930                                 mlog(ML_NOTICE, "Timed out joining dlm domain "
1931                                      "%s after %u msecs\n", dlm->name,
1932                                      total_backoff);
1933                                 goto bail;
1934                         }
1935 
1936                         /*
1937                          * <chip> After you!
1938                          * <dale> No, after you!
1939                          * <chip> I insist!
1940                          * <dale> But you first!
1941                          * ...
1942                          */
1943                         backoff = (unsigned int)(jiffies & 0x3);
1944                         backoff *= DLM_DOMAIN_BACKOFF_MS;
1945                         total_backoff += backoff;
1946                         mlog(0, "backoff %d\n", backoff);
1947                         msleep(backoff);
1948                 }
1949         } while (status == -EAGAIN);
1950 
1951         if (status < 0) {
1952                 mlog_errno(status);
1953                 goto bail;
1954         }
1955 
1956         status = 0;
1957 bail:
1958         wake_up(&dlm_domain_events);
1959 
1960         if (status) {
1961                 dlm_unregister_domain_handlers(dlm);
1962                 dlm_debug_shutdown(dlm);
1963                 dlm_complete_thread(dlm);
1964                 dlm_complete_recovery_thread(dlm);
1965                 dlm_destroy_dlm_worker(dlm);
1966         }
1967 
1968         return status;
1969 }
1970 
1971 static struct dlm_ctxt *dlm_alloc_ctxt(const char *domain,
1972                                 u32 key)
1973 {
1974         int i;
1975         int ret;
1976         struct dlm_ctxt *dlm = NULL;
1977 
1978         dlm = kzalloc(sizeof(*dlm), GFP_KERNEL);
1979         if (!dlm) {
1980                 ret = -ENOMEM;
1981                 mlog_errno(ret);
1982                 goto leave;
1983         }
1984 
1985         dlm->name = kstrdup(domain, GFP_KERNEL);
1986         if (dlm->name == NULL) {
1987                 ret = -ENOMEM;
1988                 mlog_errno(ret);
1989                 goto leave;
1990         }
1991 
1992         dlm->lockres_hash = (struct hlist_head **)dlm_alloc_pagevec(DLM_HASH_PAGES);
1993         if (!dlm->lockres_hash) {
1994                 ret = -ENOMEM;
1995                 mlog_errno(ret);
1996                 goto leave;
1997         }
1998 
1999         for (i = 0; i < DLM_HASH_BUCKETS; i++)
2000                 INIT_HLIST_HEAD(dlm_lockres_hash(dlm, i));
2001 
2002         dlm->master_hash = (struct hlist_head **)
2003                                 dlm_alloc_pagevec(DLM_HASH_PAGES);
2004         if (!dlm->master_hash) {
2005                 ret = -ENOMEM;
2006                 mlog_errno(ret);
2007                 goto leave;
2008         }
2009 
2010         for (i = 0; i < DLM_HASH_BUCKETS; i++)
2011                 INIT_HLIST_HEAD(dlm_master_hash(dlm, i));
2012 
2013         dlm->key = key;
2014         dlm->node_num = o2nm_this_node();
2015 
2016         ret = dlm_create_debugfs_subroot(dlm);
2017         if (ret < 0)
2018                 goto leave;
2019 
2020         spin_lock_init(&dlm->spinlock);
2021         spin_lock_init(&dlm->master_lock);
2022         spin_lock_init(&dlm->ast_lock);
2023         spin_lock_init(&dlm->track_lock);
2024         INIT_LIST_HEAD(&dlm->list);
2025         INIT_LIST_HEAD(&dlm->dirty_list);
2026         INIT_LIST_HEAD(&dlm->reco.resources);
2027         INIT_LIST_HEAD(&dlm->reco.node_data);
2028         INIT_LIST_HEAD(&dlm->purge_list);
2029         INIT_LIST_HEAD(&dlm->dlm_domain_handlers);
2030         INIT_LIST_HEAD(&dlm->tracking_list);
2031         dlm->reco.state = 0;
2032 
2033         INIT_LIST_HEAD(&dlm->pending_asts);
2034         INIT_LIST_HEAD(&dlm->pending_basts);
2035 
2036         mlog(0, "dlm->recovery_map=%p, &(dlm->recovery_map[0])=%p\n",
2037                   dlm->recovery_map, &(dlm->recovery_map[0]));
2038 
2039         memset(dlm->recovery_map, 0, sizeof(dlm->recovery_map));
2040         memset(dlm->live_nodes_map, 0, sizeof(dlm->live_nodes_map));
2041         memset(dlm->domain_map, 0, sizeof(dlm->domain_map));
2042 
2043         dlm->dlm_thread_task = NULL;
2044         dlm->dlm_reco_thread_task = NULL;
2045         dlm->dlm_worker = NULL;
2046         init_waitqueue_head(&dlm->dlm_thread_wq);
2047         init_waitqueue_head(&dlm->dlm_reco_thread_wq);
2048         init_waitqueue_head(&dlm->reco.event);
2049         init_waitqueue_head(&dlm->ast_wq);
2050         init_waitqueue_head(&dlm->migration_wq);
2051         INIT_LIST_HEAD(&dlm->mle_hb_events);
2052 
2053         dlm->joining_node = DLM_LOCK_RES_OWNER_UNKNOWN;
2054         init_waitqueue_head(&dlm->dlm_join_events);
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