GNU Linux-libre 4.9.337-gnu1
[releases.git] / fs / ocfs2 / dlmfs / userdlm.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * userdlm.c
5  *
6  * Code which implements the kernel side of a minimal userspace
7  * interface to our DLM.
8  *
9  * Many of the functions here are pared down versions of dlmglue.c
10  * functions.
11  *
12  * Copyright (C) 2003, 2004 Oracle.  All rights reserved.
13  *
14  * This program is free software; you can redistribute it and/or
15  * modify it under the terms of the GNU General Public
16  * License as published by the Free Software Foundation; either
17  * version 2 of the License, or (at your option) any later version.
18  *
19  * This program is distributed in the hope that it will be useful,
20  * but WITHOUT ANY WARRANTY; without even the implied warranty of
21  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
22  * General Public License for more details.
23  *
24  * You should have received a copy of the GNU General Public
25  * License along with this program; if not, write to the
26  * Free Software Foundation, Inc., 59 Temple Place - Suite 330,
27  * Boston, MA 021110-1307, USA.
28  */
29
30 #include <linux/signal.h>
31
32 #include <linux/module.h>
33 #include <linux/fs.h>
34 #include <linux/types.h>
35 #include <linux/crc32.h>
36
37 #include "ocfs2_lockingver.h"
38 #include "stackglue.h"
39 #include "userdlm.h"
40
41 #define MLOG_MASK_PREFIX ML_DLMFS
42 #include "cluster/masklog.h"
43
44
45 static inline struct user_lock_res *user_lksb_to_lock_res(struct ocfs2_dlm_lksb *lksb)
46 {
47         return container_of(lksb, struct user_lock_res, l_lksb);
48 }
49
50 static inline int user_check_wait_flag(struct user_lock_res *lockres,
51                                        int flag)
52 {
53         int ret;
54
55         spin_lock(&lockres->l_lock);
56         ret = lockres->l_flags & flag;
57         spin_unlock(&lockres->l_lock);
58
59         return ret;
60 }
61
62 static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
63
64 {
65         wait_event(lockres->l_event,
66                    !user_check_wait_flag(lockres, USER_LOCK_BUSY));
67 }
68
69 static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
70
71 {
72         wait_event(lockres->l_event,
73                    !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
74 }
75
76 /* I heart container_of... */
77 static inline struct ocfs2_cluster_connection *
78 cluster_connection_from_user_lockres(struct user_lock_res *lockres)
79 {
80         struct dlmfs_inode_private *ip;
81
82         ip = container_of(lockres,
83                           struct dlmfs_inode_private,
84                           ip_lockres);
85         return ip->ip_conn;
86 }
87
88 static struct inode *
89 user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
90 {
91         struct dlmfs_inode_private *ip;
92
93         ip = container_of(lockres,
94                           struct dlmfs_inode_private,
95                           ip_lockres);
96         return &ip->ip_vfs_inode;
97 }
98
99 static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
100 {
101         spin_lock(&lockres->l_lock);
102         lockres->l_flags &= ~USER_LOCK_BUSY;
103         spin_unlock(&lockres->l_lock);
104 }
105
106 #define user_log_dlm_error(_func, _stat, _lockres) do {                 \
107         mlog(ML_ERROR, "Dlm error %d while calling %s on "              \
108                 "resource %.*s\n", _stat, _func,                        \
109                 _lockres->l_namelen, _lockres->l_name);                 \
110 } while (0)
111
112 /* WARNING: This function lives in a world where the only three lock
113  * levels are EX, PR, and NL. It *will* have to be adjusted when more
114  * lock types are added. */
115 static inline int user_highest_compat_lock_level(int level)
116 {
117         int new_level = DLM_LOCK_EX;
118
119         if (level == DLM_LOCK_EX)
120                 new_level = DLM_LOCK_NL;
121         else if (level == DLM_LOCK_PR)
122                 new_level = DLM_LOCK_PR;
123         return new_level;
124 }
125
126 static void user_ast(struct ocfs2_dlm_lksb *lksb)
127 {
128         struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
129         int status;
130
131         mlog(ML_BASTS, "AST fired for lockres %.*s, level %d => %d\n",
132              lockres->l_namelen, lockres->l_name, lockres->l_level,
133              lockres->l_requested);
134
135         spin_lock(&lockres->l_lock);
136
137         status = ocfs2_dlm_lock_status(&lockres->l_lksb);
138         if (status) {
139                 mlog(ML_ERROR, "lksb status value of %u on lockres %.*s\n",
140                      status, lockres->l_namelen, lockres->l_name);
141                 spin_unlock(&lockres->l_lock);
142                 return;
143         }
144
145         mlog_bug_on_msg(lockres->l_requested == DLM_LOCK_IV,
146                         "Lockres %.*s, requested ivmode. flags 0x%x\n",
147                         lockres->l_namelen, lockres->l_name, lockres->l_flags);
148
149         /* we're downconverting. */
150         if (lockres->l_requested < lockres->l_level) {
151                 if (lockres->l_requested <=
152                     user_highest_compat_lock_level(lockres->l_blocking)) {
153                         lockres->l_blocking = DLM_LOCK_NL;
154                         lockres->l_flags &= ~USER_LOCK_BLOCKED;
155                 }
156         }
157
158         lockres->l_level = lockres->l_requested;
159         lockres->l_requested = DLM_LOCK_IV;
160         lockres->l_flags |= USER_LOCK_ATTACHED;
161         lockres->l_flags &= ~USER_LOCK_BUSY;
162
163         spin_unlock(&lockres->l_lock);
164
165         wake_up(&lockres->l_event);
166 }
167
168 static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
169 {
170         struct inode *inode;
171         inode = user_dlm_inode_from_user_lockres(lockres);
172         if (!igrab(inode))
173                 BUG();
174 }
175
176 static void user_dlm_unblock_lock(struct work_struct *work);
177
178 static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
179 {
180         if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
181                 user_dlm_grab_inode_ref(lockres);
182
183                 INIT_WORK(&lockres->l_work, user_dlm_unblock_lock);
184
185                 queue_work(user_dlm_worker, &lockres->l_work);
186                 lockres->l_flags |= USER_LOCK_QUEUED;
187         }
188 }
189
190 static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
191 {
192         int queue = 0;
193
194         if (!(lockres->l_flags & USER_LOCK_BLOCKED))
195                 return;
196
197         switch (lockres->l_blocking) {
198         case DLM_LOCK_EX:
199                 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
200                         queue = 1;
201                 break;
202         case DLM_LOCK_PR:
203                 if (!lockres->l_ex_holders)
204                         queue = 1;
205                 break;
206         default:
207                 BUG();
208         }
209
210         if (queue)
211                 __user_dlm_queue_lockres(lockres);
212 }
213
214 static void user_bast(struct ocfs2_dlm_lksb *lksb, int level)
215 {
216         struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
217
218         mlog(ML_BASTS, "BAST fired for lockres %.*s, blocking %d, level %d\n",
219              lockres->l_namelen, lockres->l_name, level, lockres->l_level);
220
221         spin_lock(&lockres->l_lock);
222         lockres->l_flags |= USER_LOCK_BLOCKED;
223         if (level > lockres->l_blocking)
224                 lockres->l_blocking = level;
225
226         __user_dlm_queue_lockres(lockres);
227         spin_unlock(&lockres->l_lock);
228
229         wake_up(&lockres->l_event);
230 }
231
232 static void user_unlock_ast(struct ocfs2_dlm_lksb *lksb, int status)
233 {
234         struct user_lock_res *lockres = user_lksb_to_lock_res(lksb);
235
236         mlog(ML_BASTS, "UNLOCK AST fired for lockres %.*s, flags 0x%x\n",
237              lockres->l_namelen, lockres->l_name, lockres->l_flags);
238
239         if (status)
240                 mlog(ML_ERROR, "dlm returns status %d\n", status);
241
242         spin_lock(&lockres->l_lock);
243         /* The teardown flag gets set early during the unlock process,
244          * so test the cancel flag to make sure that this ast isn't
245          * for a concurrent cancel. */
246         if (lockres->l_flags & USER_LOCK_IN_TEARDOWN
247             && !(lockres->l_flags & USER_LOCK_IN_CANCEL)) {
248                 lockres->l_level = DLM_LOCK_IV;
249         } else if (status == DLM_CANCELGRANT) {
250                 /* We tried to cancel a convert request, but it was
251                  * already granted. Don't clear the busy flag - the
252                  * ast should've done this already. */
253                 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
254                 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
255                 goto out_noclear;
256         } else {
257                 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
258                 /* Cancel succeeded, we want to re-queue */
259                 lockres->l_requested = DLM_LOCK_IV; /* cancel an
260                                                     * upconvert
261                                                     * request. */
262                 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
263                 /* we want the unblock thread to look at it again
264                  * now. */
265                 if (lockres->l_flags & USER_LOCK_BLOCKED)
266                         __user_dlm_queue_lockres(lockres);
267         }
268
269         lockres->l_flags &= ~USER_LOCK_BUSY;
270 out_noclear:
271         spin_unlock(&lockres->l_lock);
272
273         wake_up(&lockres->l_event);
274 }
275
276 /*
277  * This is the userdlmfs locking protocol version.
278  *
279  * See fs/ocfs2/dlmglue.c for more details on locking versions.
280  */
281 static struct ocfs2_locking_protocol user_dlm_lproto = {
282         .lp_max_version = {
283                 .pv_major = OCFS2_LOCKING_PROTOCOL_MAJOR,
284                 .pv_minor = OCFS2_LOCKING_PROTOCOL_MINOR,
285         },
286         .lp_lock_ast            = user_ast,
287         .lp_blocking_ast        = user_bast,
288         .lp_unlock_ast          = user_unlock_ast,
289 };
290
291 static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
292 {
293         struct inode *inode;
294         inode = user_dlm_inode_from_user_lockres(lockres);
295         iput(inode);
296 }
297
298 static void user_dlm_unblock_lock(struct work_struct *work)
299 {
300         int new_level, status;
301         struct user_lock_res *lockres =
302                 container_of(work, struct user_lock_res, l_work);
303         struct ocfs2_cluster_connection *conn =
304                 cluster_connection_from_user_lockres(lockres);
305
306         mlog(0, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
307
308         spin_lock(&lockres->l_lock);
309
310         mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
311                         "Lockres %.*s, flags 0x%x\n",
312                         lockres->l_namelen, lockres->l_name, lockres->l_flags);
313
314         /* notice that we don't clear USER_LOCK_BLOCKED here. If it's
315          * set, we want user_ast clear it. */
316         lockres->l_flags &= ~USER_LOCK_QUEUED;
317
318         /* It's valid to get here and no longer be blocked - if we get
319          * several basts in a row, we might be queued by the first
320          * one, the unblock thread might run and clear the queued
321          * flag, and finally we might get another bast which re-queues
322          * us before our ast for the downconvert is called. */
323         if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
324                 mlog(ML_BASTS, "lockres %.*s USER_LOCK_BLOCKED\n",
325                      lockres->l_namelen, lockres->l_name);
326                 spin_unlock(&lockres->l_lock);
327                 goto drop_ref;
328         }
329
330         if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
331                 mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_TEARDOWN\n",
332                      lockres->l_namelen, lockres->l_name);
333                 spin_unlock(&lockres->l_lock);
334                 goto drop_ref;
335         }
336
337         if (lockres->l_flags & USER_LOCK_BUSY) {
338                 if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
339                         mlog(ML_BASTS, "lockres %.*s USER_LOCK_IN_CANCEL\n",
340                              lockres->l_namelen, lockres->l_name);
341                         spin_unlock(&lockres->l_lock);
342                         goto drop_ref;
343                 }
344
345                 lockres->l_flags |= USER_LOCK_IN_CANCEL;
346                 spin_unlock(&lockres->l_lock);
347
348                 status = ocfs2_dlm_unlock(conn, &lockres->l_lksb,
349                                           DLM_LKF_CANCEL);
350                 if (status)
351                         user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
352                 goto drop_ref;
353         }
354
355         /* If there are still incompat holders, we can exit safely
356          * without worrying about re-queueing this lock as that will
357          * happen on the last call to user_cluster_unlock. */
358         if ((lockres->l_blocking == DLM_LOCK_EX)
359             && (lockres->l_ex_holders || lockres->l_ro_holders)) {
360                 spin_unlock(&lockres->l_lock);
361                 mlog(ML_BASTS, "lockres %.*s, EX/PR Holders %u,%u\n",
362                      lockres->l_namelen, lockres->l_name,
363                      lockres->l_ex_holders, lockres->l_ro_holders);
364                 goto drop_ref;
365         }
366
367         if ((lockres->l_blocking == DLM_LOCK_PR)
368             && lockres->l_ex_holders) {
369                 spin_unlock(&lockres->l_lock);
370                 mlog(ML_BASTS, "lockres %.*s, EX Holders %u\n",
371                      lockres->l_namelen, lockres->l_name,
372                      lockres->l_ex_holders);
373                 goto drop_ref;
374         }
375
376         /* yay, we can downconvert now. */
377         new_level = user_highest_compat_lock_level(lockres->l_blocking);
378         lockres->l_requested = new_level;
379         lockres->l_flags |= USER_LOCK_BUSY;
380         mlog(ML_BASTS, "lockres %.*s, downconvert %d => %d\n",
381              lockres->l_namelen, lockres->l_name, lockres->l_level, new_level);
382         spin_unlock(&lockres->l_lock);
383
384         /* need lock downconvert request now... */
385         status = ocfs2_dlm_lock(conn, new_level, &lockres->l_lksb,
386                                 DLM_LKF_CONVERT|DLM_LKF_VALBLK,
387                                 lockres->l_name,
388                                 lockres->l_namelen);
389         if (status) {
390                 user_log_dlm_error("ocfs2_dlm_lock", status, lockres);
391                 user_recover_from_dlm_error(lockres);
392         }
393
394 drop_ref:
395         user_dlm_drop_inode_ref(lockres);
396 }
397
398 static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
399                                         int level)
400 {
401         switch(level) {
402         case DLM_LOCK_EX:
403                 lockres->l_ex_holders++;
404                 break;
405         case DLM_LOCK_PR:
406                 lockres->l_ro_holders++;
407                 break;
408         default:
409                 BUG();
410         }
411 }
412
413 /* predict what lock level we'll be dropping down to on behalf
414  * of another node, and return true if the currently wanted
415  * level will be compatible with it. */
416 static inline int
417 user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
418                                   int wanted)
419 {
420         BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
421
422         return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
423 }
424
425 int user_dlm_cluster_lock(struct user_lock_res *lockres,
426                           int level,
427                           int lkm_flags)
428 {
429         int status, local_flags;
430         struct ocfs2_cluster_connection *conn =
431                 cluster_connection_from_user_lockres(lockres);
432
433         if (level != DLM_LOCK_EX &&
434             level != DLM_LOCK_PR) {
435                 mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
436                      lockres->l_namelen, lockres->l_name);
437                 status = -EINVAL;
438                 goto bail;
439         }
440
441         mlog(ML_BASTS, "lockres %.*s, level %d, flags = 0x%x\n",
442              lockres->l_namelen, lockres->l_name, level, lkm_flags);
443
444 again:
445         if (signal_pending(current)) {
446                 status = -ERESTARTSYS;
447                 goto bail;
448         }
449
450         spin_lock(&lockres->l_lock);
451         if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
452                 spin_unlock(&lockres->l_lock);
453                 status = -EAGAIN;
454                 goto bail;
455         }
456
457         /* We only compare against the currently granted level
458          * here. If the lock is blocked waiting on a downconvert,
459          * we'll get caught below. */
460         if ((lockres->l_flags & USER_LOCK_BUSY) &&
461             (level > lockres->l_level)) {
462                 /* is someone sitting in dlm_lock? If so, wait on
463                  * them. */
464                 spin_unlock(&lockres->l_lock);
465
466                 user_wait_on_busy_lock(lockres);
467                 goto again;
468         }
469
470         if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
471             (!user_may_continue_on_blocked_lock(lockres, level))) {
472                 /* is the lock is currently blocked on behalf of
473                  * another node */
474                 spin_unlock(&lockres->l_lock);
475
476                 user_wait_on_blocked_lock(lockres);
477                 goto again;
478         }
479
480         if (level > lockres->l_level) {
481                 local_flags = lkm_flags | DLM_LKF_VALBLK;
482                 if (lockres->l_level != DLM_LOCK_IV)
483                         local_flags |= DLM_LKF_CONVERT;
484
485                 lockres->l_requested = level;
486                 lockres->l_flags |= USER_LOCK_BUSY;
487                 spin_unlock(&lockres->l_lock);
488
489                 BUG_ON(level == DLM_LOCK_IV);
490                 BUG_ON(level == DLM_LOCK_NL);
491
492                 /* call dlm_lock to upgrade lock now */
493                 status = ocfs2_dlm_lock(conn, level, &lockres->l_lksb,
494                                         local_flags, lockres->l_name,
495                                         lockres->l_namelen);
496                 if (status) {
497                         if ((lkm_flags & DLM_LKF_NOQUEUE) &&
498                             (status != -EAGAIN))
499                                 user_log_dlm_error("ocfs2_dlm_lock",
500                                                    status, lockres);
501                         user_recover_from_dlm_error(lockres);
502                         goto bail;
503                 }
504
505                 user_wait_on_busy_lock(lockres);
506                 goto again;
507         }
508
509         user_dlm_inc_holders(lockres, level);
510         spin_unlock(&lockres->l_lock);
511
512         status = 0;
513 bail:
514         return status;
515 }
516
517 static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
518                                         int level)
519 {
520         switch(level) {
521         case DLM_LOCK_EX:
522                 BUG_ON(!lockres->l_ex_holders);
523                 lockres->l_ex_holders--;
524                 break;
525         case DLM_LOCK_PR:
526                 BUG_ON(!lockres->l_ro_holders);
527                 lockres->l_ro_holders--;
528                 break;
529         default:
530                 BUG();
531         }
532 }
533
534 void user_dlm_cluster_unlock(struct user_lock_res *lockres,
535                              int level)
536 {
537         if (level != DLM_LOCK_EX &&
538             level != DLM_LOCK_PR) {
539                 mlog(ML_ERROR, "lockres %.*s: invalid request!\n",
540                      lockres->l_namelen, lockres->l_name);
541                 return;
542         }
543
544         spin_lock(&lockres->l_lock);
545         user_dlm_dec_holders(lockres, level);
546         __user_dlm_cond_queue_lockres(lockres);
547         spin_unlock(&lockres->l_lock);
548 }
549
550 void user_dlm_write_lvb(struct inode *inode,
551                         const char *val,
552                         unsigned int len)
553 {
554         struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
555         char *lvb;
556
557         BUG_ON(len > DLM_LVB_LEN);
558
559         spin_lock(&lockres->l_lock);
560
561         BUG_ON(lockres->l_level < DLM_LOCK_EX);
562         lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
563         memcpy(lvb, val, len);
564
565         spin_unlock(&lockres->l_lock);
566 }
567
568 ssize_t user_dlm_read_lvb(struct inode *inode,
569                           char *val,
570                           unsigned int len)
571 {
572         struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
573         char *lvb;
574         ssize_t ret = len;
575
576         BUG_ON(len > DLM_LVB_LEN);
577
578         spin_lock(&lockres->l_lock);
579
580         BUG_ON(lockres->l_level < DLM_LOCK_PR);
581         if (ocfs2_dlm_lvb_valid(&lockres->l_lksb)) {
582                 lvb = ocfs2_dlm_lvb(&lockres->l_lksb);
583                 memcpy(val, lvb, len);
584         } else
585                 ret = 0;
586
587         spin_unlock(&lockres->l_lock);
588         return ret;
589 }
590
591 void user_dlm_lock_res_init(struct user_lock_res *lockres,
592                             struct dentry *dentry)
593 {
594         memset(lockres, 0, sizeof(*lockres));
595
596         spin_lock_init(&lockres->l_lock);
597         init_waitqueue_head(&lockres->l_event);
598         lockres->l_level = DLM_LOCK_IV;
599         lockres->l_requested = DLM_LOCK_IV;
600         lockres->l_blocking = DLM_LOCK_IV;
601
602         /* should have been checked before getting here. */
603         BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
604
605         memcpy(lockres->l_name,
606                dentry->d_name.name,
607                dentry->d_name.len);
608         lockres->l_namelen = dentry->d_name.len;
609 }
610
611 int user_dlm_destroy_lock(struct user_lock_res *lockres)
612 {
613         int status = -EBUSY;
614         struct ocfs2_cluster_connection *conn =
615                 cluster_connection_from_user_lockres(lockres);
616
617         mlog(ML_BASTS, "lockres %.*s\n", lockres->l_namelen, lockres->l_name);
618
619         spin_lock(&lockres->l_lock);
620         if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
621                 spin_unlock(&lockres->l_lock);
622                 goto bail;
623         }
624
625         lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
626
627         while (lockres->l_flags & USER_LOCK_BUSY) {
628                 spin_unlock(&lockres->l_lock);
629
630                 user_wait_on_busy_lock(lockres);
631
632                 spin_lock(&lockres->l_lock);
633         }
634
635         if (lockres->l_ro_holders || lockres->l_ex_holders) {
636                 lockres->l_flags &= ~USER_LOCK_IN_TEARDOWN;
637                 spin_unlock(&lockres->l_lock);
638                 goto bail;
639         }
640
641         status = 0;
642         if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
643                 /*
644                  * lock is never requested, leave USER_LOCK_IN_TEARDOWN set
645                  * to avoid new lock request coming in.
646                  */
647                 spin_unlock(&lockres->l_lock);
648                 goto bail;
649         }
650
651         lockres->l_flags &= ~USER_LOCK_ATTACHED;
652         lockres->l_flags |= USER_LOCK_BUSY;
653         spin_unlock(&lockres->l_lock);
654
655         status = ocfs2_dlm_unlock(conn, &lockres->l_lksb, DLM_LKF_VALBLK);
656         if (status) {
657                 spin_lock(&lockres->l_lock);
658                 lockres->l_flags &= ~USER_LOCK_IN_TEARDOWN;
659                 lockres->l_flags &= ~USER_LOCK_BUSY;
660                 spin_unlock(&lockres->l_lock);
661                 user_log_dlm_error("ocfs2_dlm_unlock", status, lockres);
662                 goto bail;
663         }
664
665         user_wait_on_busy_lock(lockres);
666
667         status = 0;
668 bail:
669         return status;
670 }
671
672 static void user_dlm_recovery_handler_noop(int node_num,
673                                            void *recovery_data)
674 {
675         /* We ignore recovery events */
676         return;
677 }
678
679 void user_dlm_set_locking_protocol(void)
680 {
681         ocfs2_stack_glue_set_max_proto_version(&user_dlm_lproto.lp_max_version);
682 }
683
684 struct ocfs2_cluster_connection *user_dlm_register(const struct qstr *name)
685 {
686         int rc;
687         struct ocfs2_cluster_connection *conn;
688
689         rc = ocfs2_cluster_connect_agnostic(name->name, name->len,
690                                             &user_dlm_lproto,
691                                             user_dlm_recovery_handler_noop,
692                                             NULL, &conn);
693         if (rc)
694                 mlog_errno(rc);
695
696         return rc ? ERR_PTR(rc) : conn;
697 }
698
699 void user_dlm_unregister(struct ocfs2_cluster_connection *conn)
700 {
701         ocfs2_cluster_disconnect(conn, 0);
702 }