Add the rt linux 4.1.3-rt3 as base
[kvmfornfv.git] / kernel / fs / ocfs2 / dlm / dlmlock.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * dlmlock.c
5  *
6  * underlying calls for lock creation
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
28 #include <linux/module.h>
29 #include <linux/fs.h>
30 #include <linux/types.h>
31 #include <linux/slab.h>
32 #include <linux/highmem.h>
33 #include <linux/init.h>
34 #include <linux/sysctl.h>
35 #include <linux/random.h>
36 #include <linux/blkdev.h>
37 #include <linux/socket.h>
38 #include <linux/inet.h>
39 #include <linux/spinlock.h>
40 #include <linux/delay.h>
41
42
43 #include "cluster/heartbeat.h"
44 #include "cluster/nodemanager.h"
45 #include "cluster/tcp.h"
46
47 #include "dlmapi.h"
48 #include "dlmcommon.h"
49
50 #include "dlmconvert.h"
51
52 #define MLOG_MASK_PREFIX ML_DLM
53 #include "cluster/masklog.h"
54
55 static struct kmem_cache *dlm_lock_cache;
56
57 static DEFINE_SPINLOCK(dlm_cookie_lock);
58 static u64 dlm_next_cookie = 1;
59
60 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
61                                                struct dlm_lock_resource *res,
62                                                struct dlm_lock *lock, int flags);
63 static void dlm_init_lock(struct dlm_lock *newlock, int type,
64                           u8 node, u64 cookie);
65 static void dlm_lock_release(struct kref *kref);
66 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
67
68 int dlm_init_lock_cache(void)
69 {
70         dlm_lock_cache = kmem_cache_create("o2dlm_lock",
71                                            sizeof(struct dlm_lock),
72                                            0, SLAB_HWCACHE_ALIGN, NULL);
73         if (dlm_lock_cache == NULL)
74                 return -ENOMEM;
75         return 0;
76 }
77
78 void dlm_destroy_lock_cache(void)
79 {
80         if (dlm_lock_cache)
81                 kmem_cache_destroy(dlm_lock_cache);
82 }
83
84 /* Tell us whether we can grant a new lock request.
85  * locking:
86  *   caller needs:  res->spinlock
87  *   taken:         none
88  *   held on exit:  none
89  * returns: 1 if the lock can be granted, 0 otherwise.
90  */
91 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
92                                   struct dlm_lock *lock)
93 {
94         struct dlm_lock *tmplock;
95
96         list_for_each_entry(tmplock, &res->granted, list) {
97                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
98                         return 0;
99         }
100
101         list_for_each_entry(tmplock, &res->converting, list) {
102                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
103                         return 0;
104                 if (!dlm_lock_compatible(tmplock->ml.convert_type,
105                                          lock->ml.type))
106                         return 0;
107         }
108
109         return 1;
110 }
111
112 /* performs lock creation at the lockres master site
113  * locking:
114  *   caller needs:  none
115  *   taken:         takes and drops res->spinlock
116  *   held on exit:  none
117  * returns: DLM_NORMAL, DLM_NOTQUEUED
118  */
119 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
120                                       struct dlm_lock_resource *res,
121                                       struct dlm_lock *lock, int flags)
122 {
123         int call_ast = 0, kick_thread = 0;
124         enum dlm_status status = DLM_NORMAL;
125
126         mlog(0, "type=%d\n", lock->ml.type);
127
128         spin_lock(&res->spinlock);
129         /* if called from dlm_create_lock_handler, need to
130          * ensure it will not sleep in dlm_wait_on_lockres */
131         status = __dlm_lockres_state_to_status(res);
132         if (status != DLM_NORMAL &&
133             lock->ml.node != dlm->node_num) {
134                 /* erf.  state changed after lock was dropped. */
135                 spin_unlock(&res->spinlock);
136                 dlm_error(status);
137                 return status;
138         }
139         __dlm_wait_on_lockres(res);
140         __dlm_lockres_reserve_ast(res);
141
142         if (dlm_can_grant_new_lock(res, lock)) {
143                 mlog(0, "I can grant this lock right away\n");
144                 /* got it right away */
145                 lock->lksb->status = DLM_NORMAL;
146                 status = DLM_NORMAL;
147                 dlm_lock_get(lock);
148                 list_add_tail(&lock->list, &res->granted);
149
150                 /* for the recovery lock, we can't allow the ast
151                  * to be queued since the dlmthread is already
152                  * frozen.  but the recovery lock is always locked
153                  * with LKM_NOQUEUE so we do not need the ast in
154                  * this special case */
155                 if (!dlm_is_recovery_lock(res->lockname.name,
156                                           res->lockname.len)) {
157                         kick_thread = 1;
158                         call_ast = 1;
159                 } else {
160                         mlog(0, "%s: returning DLM_NORMAL to "
161                              "node %u for reco lock\n", dlm->name,
162                              lock->ml.node);
163                 }
164         } else {
165                 /* for NOQUEUE request, unless we get the
166                  * lock right away, return DLM_NOTQUEUED */
167                 if (flags & LKM_NOQUEUE) {
168                         status = DLM_NOTQUEUED;
169                         if (dlm_is_recovery_lock(res->lockname.name,
170                                                  res->lockname.len)) {
171                                 mlog(0, "%s: returning NOTQUEUED to "
172                                      "node %u for reco lock\n", dlm->name,
173                                      lock->ml.node);
174                         }
175                 } else {
176                         status = DLM_NORMAL;
177                         dlm_lock_get(lock);
178                         list_add_tail(&lock->list, &res->blocked);
179                         kick_thread = 1;
180                 }
181         }
182
183         spin_unlock(&res->spinlock);
184         wake_up(&res->wq);
185
186         /* either queue the ast or release it */
187         if (call_ast)
188                 dlm_queue_ast(dlm, lock);
189         else
190                 dlm_lockres_release_ast(dlm, res);
191
192         dlm_lockres_calc_usage(dlm, res);
193         if (kick_thread)
194                 dlm_kick_thread(dlm, res);
195
196         return status;
197 }
198
199 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
200                              struct dlm_lock *lock)
201 {
202         /* remove from local queue if it failed */
203         list_del_init(&lock->list);
204         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
205 }
206
207
208 /*
209  * locking:
210  *   caller needs:  none
211  *   taken:         takes and drops res->spinlock
212  *   held on exit:  none
213  * returns: DLM_DENIED, DLM_RECOVERING, or net status
214  */
215 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
216                                       struct dlm_lock_resource *res,
217                                       struct dlm_lock *lock, int flags)
218 {
219         enum dlm_status status = DLM_DENIED;
220         int lockres_changed = 1;
221
222         mlog(0, "type=%d, lockres %.*s, flags = 0x%x\n",
223              lock->ml.type, res->lockname.len,
224              res->lockname.name, flags);
225
226         /*
227          * Wait if resource is getting recovered, remastered, etc.
228          * If the resource was remastered and new owner is self, then exit.
229          */
230         spin_lock(&res->spinlock);
231         __dlm_wait_on_lockres(res);
232         if (res->owner == dlm->node_num) {
233                 spin_unlock(&res->spinlock);
234                 return DLM_RECOVERING;
235         }
236         res->state |= DLM_LOCK_RES_IN_PROGRESS;
237
238         /* add lock to local (secondary) queue */
239         dlm_lock_get(lock);
240         list_add_tail(&lock->list, &res->blocked);
241         lock->lock_pending = 1;
242         spin_unlock(&res->spinlock);
243
244         /* spec seems to say that you will get DLM_NORMAL when the lock
245          * has been queued, meaning we need to wait for a reply here. */
246         status = dlm_send_remote_lock_request(dlm, res, lock, flags);
247
248         spin_lock(&res->spinlock);
249         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
250         lock->lock_pending = 0;
251         if (status != DLM_NORMAL) {
252                 if (status == DLM_RECOVERING &&
253                     dlm_is_recovery_lock(res->lockname.name,
254                                          res->lockname.len)) {
255                         /* recovery lock was mastered by dead node.
256                          * we need to have calc_usage shoot down this
257                          * lockres and completely remaster it. */
258                         mlog(0, "%s: recovery lock was owned by "
259                              "dead node %u, remaster it now.\n",
260                              dlm->name, res->owner);
261                 } else if (status != DLM_NOTQUEUED) {
262                         /*
263                          * DO NOT call calc_usage, as this would unhash
264                          * the remote lockres before we ever get to use
265                          * it.  treat as if we never made any change to
266                          * the lockres.
267                          */
268                         lockres_changed = 0;
269                         dlm_error(status);
270                 }
271                 dlm_revert_pending_lock(res, lock);
272                 dlm_lock_put(lock);
273         } else if (dlm_is_recovery_lock(res->lockname.name,
274                                         res->lockname.len)) {
275                 /* special case for the $RECOVERY lock.
276                  * there will never be an AST delivered to put
277                  * this lock on the proper secondary queue
278                  * (granted), so do it manually. */
279                 mlog(0, "%s: $RECOVERY lock for this node (%u) is "
280                      "mastered by %u; got lock, manually granting (no ast)\n",
281                      dlm->name, dlm->node_num, res->owner);
282                 list_move_tail(&lock->list, &res->granted);
283         }
284         spin_unlock(&res->spinlock);
285
286         if (lockres_changed)
287                 dlm_lockres_calc_usage(dlm, res);
288
289         wake_up(&res->wq);
290         return status;
291 }
292
293
294 /* for remote lock creation.
295  * locking:
296  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
297  *   taken:         none
298  *   held on exit:  none
299  * returns: DLM_NOLOCKMGR, or net status
300  */
301 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
302                                                struct dlm_lock_resource *res,
303                                                struct dlm_lock *lock, int flags)
304 {
305         struct dlm_create_lock create;
306         int tmpret, status = 0;
307         enum dlm_status ret;
308
309         memset(&create, 0, sizeof(create));
310         create.node_idx = dlm->node_num;
311         create.requested_type = lock->ml.type;
312         create.cookie = lock->ml.cookie;
313         create.namelen = res->lockname.len;
314         create.flags = cpu_to_be32(flags);
315         memcpy(create.name, res->lockname.name, create.namelen);
316
317         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
318                                     sizeof(create), res->owner, &status);
319         if (tmpret >= 0) {
320                 ret = status;
321                 if (ret == DLM_REJECTED) {
322                         mlog(ML_ERROR, "%s: res %.*s, Stale lockres no longer "
323                              "owned by node %u. That node is coming back up "
324                              "currently.\n", dlm->name, create.namelen,
325                              create.name, res->owner);
326                         dlm_print_one_lock_resource(res);
327                         BUG();
328                 }
329         } else {
330                 mlog(ML_ERROR, "%s: res %.*s, Error %d send CREATE LOCK to "
331                      "node %u\n", dlm->name, create.namelen, create.name,
332                      tmpret, res->owner);
333                 if (dlm_is_host_down(tmpret))
334                         ret = DLM_RECOVERING;
335                 else
336                         ret = dlm_err_to_dlm_status(tmpret);
337         }
338
339         return ret;
340 }
341
342 void dlm_lock_get(struct dlm_lock *lock)
343 {
344         kref_get(&lock->lock_refs);
345 }
346
347 void dlm_lock_put(struct dlm_lock *lock)
348 {
349         kref_put(&lock->lock_refs, dlm_lock_release);
350 }
351
352 static void dlm_lock_release(struct kref *kref)
353 {
354         struct dlm_lock *lock;
355
356         lock = container_of(kref, struct dlm_lock, lock_refs);
357
358         BUG_ON(!list_empty(&lock->list));
359         BUG_ON(!list_empty(&lock->ast_list));
360         BUG_ON(!list_empty(&lock->bast_list));
361         BUG_ON(lock->ast_pending);
362         BUG_ON(lock->bast_pending);
363
364         dlm_lock_detach_lockres(lock);
365
366         if (lock->lksb_kernel_allocated) {
367                 mlog(0, "freeing kernel-allocated lksb\n");
368                 kfree(lock->lksb);
369         }
370         kmem_cache_free(dlm_lock_cache, lock);
371 }
372
373 /* associate a lock with it's lockres, getting a ref on the lockres */
374 void dlm_lock_attach_lockres(struct dlm_lock *lock,
375                              struct dlm_lock_resource *res)
376 {
377         dlm_lockres_get(res);
378         lock->lockres = res;
379 }
380
381 /* drop ref on lockres, if there is still one associated with lock */
382 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
383 {
384         struct dlm_lock_resource *res;
385
386         res = lock->lockres;
387         if (res) {
388                 lock->lockres = NULL;
389                 mlog(0, "removing lock's lockres reference\n");
390                 dlm_lockres_put(res);
391         }
392 }
393
394 static void dlm_init_lock(struct dlm_lock *newlock, int type,
395                           u8 node, u64 cookie)
396 {
397         INIT_LIST_HEAD(&newlock->list);
398         INIT_LIST_HEAD(&newlock->ast_list);
399         INIT_LIST_HEAD(&newlock->bast_list);
400         spin_lock_init(&newlock->spinlock);
401         newlock->ml.type = type;
402         newlock->ml.convert_type = LKM_IVMODE;
403         newlock->ml.highest_blocked = LKM_IVMODE;
404         newlock->ml.node = node;
405         newlock->ml.pad1 = 0;
406         newlock->ml.list = 0;
407         newlock->ml.flags = 0;
408         newlock->ast = NULL;
409         newlock->bast = NULL;
410         newlock->astdata = NULL;
411         newlock->ml.cookie = cpu_to_be64(cookie);
412         newlock->ast_pending = 0;
413         newlock->bast_pending = 0;
414         newlock->convert_pending = 0;
415         newlock->lock_pending = 0;
416         newlock->unlock_pending = 0;
417         newlock->cancel_pending = 0;
418         newlock->lksb_kernel_allocated = 0;
419
420         kref_init(&newlock->lock_refs);
421 }
422
423 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
424                                struct dlm_lockstatus *lksb)
425 {
426         struct dlm_lock *lock;
427         int kernel_allocated = 0;
428
429         lock = kmem_cache_zalloc(dlm_lock_cache, GFP_NOFS);
430         if (!lock)
431                 return NULL;
432
433         if (!lksb) {
434                 /* zero memory only if kernel-allocated */
435                 lksb = kzalloc(sizeof(*lksb), GFP_NOFS);
436                 if (!lksb) {
437                         kmem_cache_free(dlm_lock_cache, lock);
438                         return NULL;
439                 }
440                 kernel_allocated = 1;
441         }
442
443         dlm_init_lock(lock, type, node, cookie);
444         if (kernel_allocated)
445                 lock->lksb_kernel_allocated = 1;
446         lock->lksb = lksb;
447         lksb->lockid = lock;
448         return lock;
449 }
450
451 /* handler for lock creation net message
452  * locking:
453  *   caller needs:  none
454  *   taken:         takes and drops res->spinlock
455  *   held on exit:  none
456  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
457  */
458 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data,
459                             void **ret_data)
460 {
461         struct dlm_ctxt *dlm = data;
462         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
463         struct dlm_lock_resource *res = NULL;
464         struct dlm_lock *newlock = NULL;
465         struct dlm_lockstatus *lksb = NULL;
466         enum dlm_status status = DLM_NORMAL;
467         char *name;
468         unsigned int namelen;
469
470         BUG_ON(!dlm);
471
472         if (!dlm_grab(dlm))
473                 return DLM_REJECTED;
474
475         name = create->name;
476         namelen = create->namelen;
477         status = DLM_REJECTED;
478         if (!dlm_domain_fully_joined(dlm)) {
479                 mlog(ML_ERROR, "Domain %s not fully joined, but node %u is "
480                      "sending a create_lock message for lock %.*s!\n",
481                      dlm->name, create->node_idx, namelen, name);
482                 dlm_error(status);
483                 goto leave;
484         }
485
486         status = DLM_IVBUFLEN;
487         if (namelen > DLM_LOCKID_NAME_MAX) {
488                 dlm_error(status);
489                 goto leave;
490         }
491
492         status = DLM_SYSERR;
493         newlock = dlm_new_lock(create->requested_type,
494                                create->node_idx,
495                                be64_to_cpu(create->cookie), NULL);
496         if (!newlock) {
497                 dlm_error(status);
498                 goto leave;
499         }
500
501         lksb = newlock->lksb;
502
503         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
504                 lksb->flags |= DLM_LKSB_GET_LVB;
505                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
506         }
507
508         status = DLM_IVLOCKID;
509         res = dlm_lookup_lockres(dlm, name, namelen);
510         if (!res) {
511                 dlm_error(status);
512                 goto leave;
513         }
514
515         spin_lock(&res->spinlock);
516         status = __dlm_lockres_state_to_status(res);
517         spin_unlock(&res->spinlock);
518
519         if (status != DLM_NORMAL) {
520                 mlog(0, "lockres recovering/migrating/in-progress\n");
521                 goto leave;
522         }
523
524         dlm_lock_attach_lockres(newlock, res);
525
526         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
527 leave:
528         if (status != DLM_NORMAL)
529                 if (newlock)
530                         dlm_lock_put(newlock);
531
532         if (res)
533                 dlm_lockres_put(res);
534
535         dlm_put(dlm);
536
537         return status;
538 }
539
540
541 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
542 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
543 {
544         u64 tmpnode = node_num;
545
546         /* shift single byte of node num into top 8 bits */
547         tmpnode <<= 56;
548
549         spin_lock(&dlm_cookie_lock);
550         *cookie = (dlm_next_cookie | tmpnode);
551         if (++dlm_next_cookie & 0xff00000000000000ull) {
552                 mlog(0, "This node's cookie will now wrap!\n");
553                 dlm_next_cookie = 1;
554         }
555         spin_unlock(&dlm_cookie_lock);
556 }
557
558 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
559                         struct dlm_lockstatus *lksb, int flags,
560                         const char *name, int namelen, dlm_astlockfunc_t *ast,
561                         void *data, dlm_bastlockfunc_t *bast)
562 {
563         enum dlm_status status;
564         struct dlm_lock_resource *res = NULL;
565         struct dlm_lock *lock = NULL;
566         int convert = 0, recovery = 0;
567
568         /* yes this function is a mess.
569          * TODO: clean this up.  lots of common code in the
570          *       lock and convert paths, especially in the retry blocks */
571         if (!lksb) {
572                 dlm_error(DLM_BADARGS);
573                 return DLM_BADARGS;
574         }
575
576         status = DLM_BADPARAM;
577         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
578                 dlm_error(status);
579                 goto error;
580         }
581
582         if (flags & ~LKM_VALID_FLAGS) {
583                 dlm_error(status);
584                 goto error;
585         }
586
587         convert = (flags & LKM_CONVERT);
588         recovery = (flags & LKM_RECOVERY);
589
590         if (recovery &&
591             (!dlm_is_recovery_lock(name, namelen) || convert) ) {
592                 dlm_error(status);
593                 goto error;
594         }
595         if (convert && (flags & LKM_LOCAL)) {
596                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
597                 goto error;
598         }
599
600         if (convert) {
601                 /* CONVERT request */
602
603                 /* if converting, must pass in a valid dlm_lock */
604                 lock = lksb->lockid;
605                 if (!lock) {
606                         mlog(ML_ERROR, "NULL lock pointer in convert "
607                              "request\n");
608                         goto error;
609                 }
610
611                 res = lock->lockres;
612                 if (!res) {
613                         mlog(ML_ERROR, "NULL lockres pointer in convert "
614                              "request\n");
615                         goto error;
616                 }
617                 dlm_lockres_get(res);
618
619                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
620                  * static after the original lock call.  convert requests will
621                  * ensure that everything is the same, or return DLM_BADARGS.
622                  * this means that DLM_DENIED_NOASTS will never be returned.
623                  */
624                 if (lock->lksb != lksb || lock->ast != ast ||
625                     lock->bast != bast || lock->astdata != data) {
626                         status = DLM_BADARGS;
627                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
628                              "astdata=%p\n", lksb, ast, bast, data);
629                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
630                              "astdata=%p\n", lock->lksb, lock->ast,
631                              lock->bast, lock->astdata);
632                         goto error;
633                 }
634 retry_convert:
635                 dlm_wait_for_recovery(dlm);
636
637                 if (res->owner == dlm->node_num)
638                         status = dlmconvert_master(dlm, res, lock, flags, mode);
639                 else
640                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
641                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
642                     status == DLM_FORWARD) {
643                         /* for now, see how this works without sleeping
644                          * and just retry right away.  I suspect the reco
645                          * or migration will complete fast enough that
646                          * no waiting will be necessary */
647                         mlog(0, "retrying convert with migration/recovery/"
648                              "in-progress\n");
649                         msleep(100);
650                         goto retry_convert;
651                 }
652         } else {
653                 u64 tmpcookie;
654
655                 /* LOCK request */
656                 status = DLM_BADARGS;
657                 if (!name) {
658                         dlm_error(status);
659                         goto error;
660                 }
661
662                 status = DLM_IVBUFLEN;
663                 if (namelen > DLM_LOCKID_NAME_MAX || namelen < 1) {
664                         dlm_error(status);
665                         goto error;
666                 }
667
668                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
669                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
670                 if (!lock) {
671                         dlm_error(status);
672                         goto error;
673                 }
674
675                 if (!recovery)
676                         dlm_wait_for_recovery(dlm);
677
678                 /* find or create the lock resource */
679                 res = dlm_get_lock_resource(dlm, name, namelen, flags);
680                 if (!res) {
681                         status = DLM_IVLOCKID;
682                         dlm_error(status);
683                         goto error;
684                 }
685
686                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
687                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
688
689                 dlm_lock_attach_lockres(lock, res);
690                 lock->ast = ast;
691                 lock->bast = bast;
692                 lock->astdata = data;
693
694 retry_lock:
695                 if (flags & LKM_VALBLK) {
696                         mlog(0, "LKM_VALBLK passed by caller\n");
697
698                         /* LVB requests for non PR, PW or EX locks are
699                          * ignored. */
700                         if (mode < LKM_PRMODE)
701                                 flags &= ~LKM_VALBLK;
702                         else {
703                                 flags |= LKM_GET_LVB;
704                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
705                         }
706                 }
707
708                 if (res->owner == dlm->node_num)
709                         status = dlmlock_master(dlm, res, lock, flags);
710                 else
711                         status = dlmlock_remote(dlm, res, lock, flags);
712
713                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
714                     status == DLM_FORWARD) {
715                         msleep(100);
716                         if (recovery) {
717                                 if (status != DLM_RECOVERING)
718                                         goto retry_lock;
719                                 /* wait to see the node go down, then
720                                  * drop down and allow the lockres to
721                                  * get cleaned up.  need to remaster. */
722                                 dlm_wait_for_node_death(dlm, res->owner,
723                                                 DLM_NODE_DEATH_WAIT_MAX);
724                         } else {
725                                 dlm_wait_for_recovery(dlm);
726                                 goto retry_lock;
727                         }
728                 }
729
730                 /* Inflight taken in dlm_get_lock_resource() is dropped here */
731                 spin_lock(&res->spinlock);
732                 dlm_lockres_drop_inflight_ref(dlm, res);
733                 spin_unlock(&res->spinlock);
734
735                 dlm_lockres_calc_usage(dlm, res);
736                 dlm_kick_thread(dlm, res);
737
738                 if (status != DLM_NORMAL) {
739                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
740                         if (status != DLM_NOTQUEUED)
741                                 dlm_error(status);
742                         goto error;
743                 }
744         }
745
746 error:
747         if (status != DLM_NORMAL) {
748                 if (lock && !convert)
749                         dlm_lock_put(lock);
750                 // this is kind of unnecessary
751                 lksb->status = status;
752         }
753
754         /* put lockres ref from the convert path
755          * or from dlm_get_lock_resource */
756         if (res)
757                 dlm_lockres_put(res);
758
759         return status;
760 }
761 EXPORT_SYMBOL_GPL(dlmlock);