Merge master.kernel.org:/home/rmk/linux-2.6-serial
[linux-2.6] / 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/utsname.h>
34 #include <linux/init.h>
35 #include <linux/sysctl.h>
36 #include <linux/random.h>
37 #include <linux/blkdev.h>
38 #include <linux/socket.h>
39 #include <linux/inet.h>
40 #include <linux/spinlock.h>
41 #include <linux/delay.h>
42
43
44 #include "cluster/heartbeat.h"
45 #include "cluster/nodemanager.h"
46 #include "cluster/tcp.h"
47
48 #include "dlmapi.h"
49 #include "dlmcommon.h"
50
51 #include "dlmconvert.h"
52
53 #define MLOG_MASK_PREFIX ML_DLM
54 #include "cluster/masklog.h"
55
56 static spinlock_t dlm_cookie_lock = SPIN_LOCK_UNLOCKED;
57 static u64 dlm_next_cookie = 1;
58
59 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
60                                                struct dlm_lock_resource *res,
61                                                struct dlm_lock *lock, int flags);
62 static void dlm_init_lock(struct dlm_lock *newlock, int type,
63                           u8 node, u64 cookie);
64 static void dlm_lock_release(struct kref *kref);
65 static void dlm_lock_detach_lockres(struct dlm_lock *lock);
66
67 /* Tell us whether we can grant a new lock request.
68  * locking:
69  *   caller needs:  res->spinlock
70  *   taken:         none
71  *   held on exit:  none
72  * returns: 1 if the lock can be granted, 0 otherwise.
73  */
74 static int dlm_can_grant_new_lock(struct dlm_lock_resource *res,
75                                   struct dlm_lock *lock)
76 {
77         struct list_head *iter;
78         struct dlm_lock *tmplock;
79
80         list_for_each(iter, &res->granted) {
81                 tmplock = list_entry(iter, struct dlm_lock, list);
82
83                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
84                         return 0;
85         }
86
87         list_for_each(iter, &res->converting) {
88                 tmplock = list_entry(iter, struct dlm_lock, list);
89
90                 if (!dlm_lock_compatible(tmplock->ml.type, lock->ml.type))
91                         return 0;
92         }
93
94         return 1;
95 }
96
97 /* performs lock creation at the lockres master site
98  * locking:
99  *   caller needs:  none
100  *   taken:         takes and drops res->spinlock
101  *   held on exit:  none
102  * returns: DLM_NORMAL, DLM_NOTQUEUED
103  */
104 static enum dlm_status dlmlock_master(struct dlm_ctxt *dlm,
105                                       struct dlm_lock_resource *res,
106                                       struct dlm_lock *lock, int flags)
107 {
108         int call_ast = 0, kick_thread = 0;
109         enum dlm_status status = DLM_NORMAL;
110
111         mlog_entry("type=%d\n", lock->ml.type);
112
113         spin_lock(&res->spinlock);
114         /* if called from dlm_create_lock_handler, need to
115          * ensure it will not sleep in dlm_wait_on_lockres */
116         status = __dlm_lockres_state_to_status(res);
117         if (status != DLM_NORMAL &&
118             lock->ml.node != dlm->node_num) {
119                 /* erf.  state changed after lock was dropped. */
120                 spin_unlock(&res->spinlock);
121                 dlm_error(status);
122                 return status;
123         }
124         __dlm_wait_on_lockres(res);
125         __dlm_lockres_reserve_ast(res);
126
127         if (dlm_can_grant_new_lock(res, lock)) {
128                 mlog(0, "I can grant this lock right away\n");
129                 /* got it right away */
130                 lock->lksb->status = DLM_NORMAL;
131                 status = DLM_NORMAL;
132                 dlm_lock_get(lock);
133                 list_add_tail(&lock->list, &res->granted);
134
135                 /* for the recovery lock, we can't allow the ast
136                  * to be queued since the dlmthread is already
137                  * frozen.  but the recovery lock is always locked
138                  * with LKM_NOQUEUE so we do not need the ast in
139                  * this special case */
140                 if (!dlm_is_recovery_lock(res->lockname.name,
141                                           res->lockname.len)) {
142                         kick_thread = 1;
143                         call_ast = 1;
144                 }
145         } else {
146                 /* for NOQUEUE request, unless we get the
147                  * lock right away, return DLM_NOTQUEUED */
148                 if (flags & LKM_NOQUEUE)
149                         status = DLM_NOTQUEUED;
150                 else {
151                         dlm_lock_get(lock);
152                         list_add_tail(&lock->list, &res->blocked);
153                         kick_thread = 1;
154                 }
155         }
156
157         spin_unlock(&res->spinlock);
158         wake_up(&res->wq);
159
160         /* either queue the ast or release it */
161         if (call_ast)
162                 dlm_queue_ast(dlm, lock);
163         else
164                 dlm_lockres_release_ast(dlm, res);
165
166         dlm_lockres_calc_usage(dlm, res);
167         if (kick_thread)
168                 dlm_kick_thread(dlm, res);
169
170         return status;
171 }
172
173 void dlm_revert_pending_lock(struct dlm_lock_resource *res,
174                              struct dlm_lock *lock)
175 {
176         /* remove from local queue if it failed */
177         list_del_init(&lock->list);
178         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
179 }
180
181
182 /*
183  * locking:
184  *   caller needs:  none
185  *   taken:         takes and drops res->spinlock
186  *   held on exit:  none
187  * returns: DLM_DENIED, DLM_RECOVERING, or net status
188  */
189 static enum dlm_status dlmlock_remote(struct dlm_ctxt *dlm,
190                                       struct dlm_lock_resource *res,
191                                       struct dlm_lock *lock, int flags)
192 {
193         enum dlm_status status = DLM_DENIED;
194
195         mlog_entry("type=%d\n", lock->ml.type);
196         mlog(0, "lockres %.*s, flags = 0x%x\n", res->lockname.len,
197              res->lockname.name, flags);
198
199         spin_lock(&res->spinlock);
200
201         /* will exit this call with spinlock held */
202         __dlm_wait_on_lockres(res);
203         res->state |= DLM_LOCK_RES_IN_PROGRESS;
204
205         /* add lock to local (secondary) queue */
206         dlm_lock_get(lock);
207         list_add_tail(&lock->list, &res->blocked);
208         lock->lock_pending = 1;
209         spin_unlock(&res->spinlock);
210
211         /* spec seems to say that you will get DLM_NORMAL when the lock
212          * has been queued, meaning we need to wait for a reply here. */
213         status = dlm_send_remote_lock_request(dlm, res, lock, flags);
214
215         spin_lock(&res->spinlock);
216         res->state &= ~DLM_LOCK_RES_IN_PROGRESS;
217         lock->lock_pending = 0;
218         if (status != DLM_NORMAL) {
219                 if (status != DLM_NOTQUEUED)
220                         dlm_error(status);
221                 dlm_revert_pending_lock(res, lock);
222                 dlm_lock_put(lock);
223         }
224         spin_unlock(&res->spinlock);
225
226         dlm_lockres_calc_usage(dlm, res);
227
228         wake_up(&res->wq);
229         return status;
230 }
231
232
233 /* for remote lock creation.
234  * locking:
235  *   caller needs:  none, but need res->state & DLM_LOCK_RES_IN_PROGRESS
236  *   taken:         none
237  *   held on exit:  none
238  * returns: DLM_NOLOCKMGR, or net status
239  */
240 static enum dlm_status dlm_send_remote_lock_request(struct dlm_ctxt *dlm,
241                                                struct dlm_lock_resource *res,
242                                                struct dlm_lock *lock, int flags)
243 {
244         struct dlm_create_lock create;
245         int tmpret, status = 0;
246         enum dlm_status ret;
247
248         mlog_entry_void();
249
250         memset(&create, 0, sizeof(create));
251         create.node_idx = dlm->node_num;
252         create.requested_type = lock->ml.type;
253         create.cookie = lock->ml.cookie;
254         create.namelen = res->lockname.len;
255         create.flags = cpu_to_be32(flags);
256         memcpy(create.name, res->lockname.name, create.namelen);
257
258         tmpret = o2net_send_message(DLM_CREATE_LOCK_MSG, dlm->key, &create,
259                                     sizeof(create), res->owner, &status);
260         if (tmpret >= 0) {
261                 // successfully sent and received
262                 ret = status;  // this is already a dlm_status
263         } else {
264                 mlog_errno(tmpret);
265                 if (dlm_is_host_down(tmpret)) {
266                         ret = DLM_RECOVERING;
267                         mlog(0, "node %u died so returning DLM_RECOVERING "
268                              "from lock message!\n", res->owner);
269                 } else {
270                         ret = dlm_err_to_dlm_status(tmpret);
271                 }
272         }
273
274         return ret;
275 }
276
277 void dlm_lock_get(struct dlm_lock *lock)
278 {
279         kref_get(&lock->lock_refs);
280 }
281
282 void dlm_lock_put(struct dlm_lock *lock)
283 {
284         kref_put(&lock->lock_refs, dlm_lock_release);
285 }
286
287 static void dlm_lock_release(struct kref *kref)
288 {
289         struct dlm_lock *lock;
290
291         lock = container_of(kref, struct dlm_lock, lock_refs);
292
293         BUG_ON(!list_empty(&lock->list));
294         BUG_ON(!list_empty(&lock->ast_list));
295         BUG_ON(!list_empty(&lock->bast_list));
296         BUG_ON(lock->ast_pending);
297         BUG_ON(lock->bast_pending);
298
299         dlm_lock_detach_lockres(lock);
300
301         if (lock->lksb_kernel_allocated) {
302                 mlog(0, "freeing kernel-allocated lksb\n");
303                 kfree(lock->lksb);
304         }
305         kfree(lock);
306 }
307
308 /* associate a lock with it's lockres, getting a ref on the lockres */
309 void dlm_lock_attach_lockres(struct dlm_lock *lock,
310                              struct dlm_lock_resource *res)
311 {
312         dlm_lockres_get(res);
313         lock->lockres = res;
314 }
315
316 /* drop ref on lockres, if there is still one associated with lock */
317 static void dlm_lock_detach_lockres(struct dlm_lock *lock)
318 {
319         struct dlm_lock_resource *res;
320
321         res = lock->lockres;
322         if (res) {
323                 lock->lockres = NULL;
324                 mlog(0, "removing lock's lockres reference\n");
325                 dlm_lockres_put(res);
326         }
327 }
328
329 static void dlm_init_lock(struct dlm_lock *newlock, int type,
330                           u8 node, u64 cookie)
331 {
332         INIT_LIST_HEAD(&newlock->list);
333         INIT_LIST_HEAD(&newlock->ast_list);
334         INIT_LIST_HEAD(&newlock->bast_list);
335         spin_lock_init(&newlock->spinlock);
336         newlock->ml.type = type;
337         newlock->ml.convert_type = LKM_IVMODE;
338         newlock->ml.highest_blocked = LKM_IVMODE;
339         newlock->ml.node = node;
340         newlock->ml.pad1 = 0;
341         newlock->ml.list = 0;
342         newlock->ml.flags = 0;
343         newlock->ast = NULL;
344         newlock->bast = NULL;
345         newlock->astdata = NULL;
346         newlock->ml.cookie = cpu_to_be64(cookie);
347         newlock->ast_pending = 0;
348         newlock->bast_pending = 0;
349         newlock->convert_pending = 0;
350         newlock->lock_pending = 0;
351         newlock->unlock_pending = 0;
352         newlock->cancel_pending = 0;
353         newlock->lksb_kernel_allocated = 0;
354
355         kref_init(&newlock->lock_refs);
356 }
357
358 struct dlm_lock * dlm_new_lock(int type, u8 node, u64 cookie,
359                                struct dlm_lockstatus *lksb)
360 {
361         struct dlm_lock *lock;
362         int kernel_allocated = 0;
363
364         lock = kcalloc(1, sizeof(*lock), GFP_KERNEL);
365         if (!lock)
366                 return NULL;
367
368         if (!lksb) {
369                 /* zero memory only if kernel-allocated */
370                 lksb = kcalloc(1, sizeof(*lksb), GFP_KERNEL);
371                 if (!lksb) {
372                         kfree(lock);
373                         return NULL;
374                 }
375                 kernel_allocated = 1;
376         }
377
378         dlm_init_lock(lock, type, node, cookie);
379         if (kernel_allocated)
380                 lock->lksb_kernel_allocated = 1;
381         lock->lksb = lksb;
382         lksb->lockid = lock;
383         return lock;
384 }
385
386 /* handler for lock creation net message
387  * locking:
388  *   caller needs:  none
389  *   taken:         takes and drops res->spinlock
390  *   held on exit:  none
391  * returns: DLM_NORMAL, DLM_SYSERR, DLM_IVLOCKID, DLM_NOTQUEUED
392  */
393 int dlm_create_lock_handler(struct o2net_msg *msg, u32 len, void *data)
394 {
395         struct dlm_ctxt *dlm = data;
396         struct dlm_create_lock *create = (struct dlm_create_lock *)msg->buf;
397         struct dlm_lock_resource *res = NULL;
398         struct dlm_lock *newlock = NULL;
399         struct dlm_lockstatus *lksb = NULL;
400         enum dlm_status status = DLM_NORMAL;
401         char *name;
402         unsigned int namelen;
403
404         BUG_ON(!dlm);
405
406         mlog_entry_void();
407
408         if (!dlm_grab(dlm))
409                 return DLM_REJECTED;
410
411         mlog_bug_on_msg(!dlm_domain_fully_joined(dlm),
412                         "Domain %s not fully joined!\n", dlm->name);
413
414         name = create->name;
415         namelen = create->namelen;
416
417         status = DLM_IVBUFLEN;
418         if (namelen > DLM_LOCKID_NAME_MAX) {
419                 dlm_error(status);
420                 goto leave;
421         }
422
423         status = DLM_SYSERR;
424         newlock = dlm_new_lock(create->requested_type,
425                                create->node_idx,
426                                be64_to_cpu(create->cookie), NULL);
427         if (!newlock) {
428                 dlm_error(status);
429                 goto leave;
430         }
431
432         lksb = newlock->lksb;
433
434         if (be32_to_cpu(create->flags) & LKM_GET_LVB) {
435                 lksb->flags |= DLM_LKSB_GET_LVB;
436                 mlog(0, "set DLM_LKSB_GET_LVB flag\n");
437         }
438
439         status = DLM_IVLOCKID;
440         res = dlm_lookup_lockres(dlm, name, namelen);
441         if (!res) {
442                 dlm_error(status);
443                 goto leave;
444         }
445
446         spin_lock(&res->spinlock);
447         status = __dlm_lockres_state_to_status(res);
448         spin_unlock(&res->spinlock);
449
450         if (status != DLM_NORMAL) {
451                 mlog(0, "lockres recovering/migrating/in-progress\n");
452                 goto leave;
453         }
454
455         dlm_lock_attach_lockres(newlock, res);
456
457         status = dlmlock_master(dlm, res, newlock, be32_to_cpu(create->flags));
458 leave:
459         if (status != DLM_NORMAL)
460                 if (newlock)
461                         dlm_lock_put(newlock);
462
463         if (res)
464                 dlm_lockres_put(res);
465
466         dlm_put(dlm);
467
468         return status;
469 }
470
471
472 /* fetch next node-local (u8 nodenum + u56 cookie) into u64 */
473 static inline void dlm_get_next_cookie(u8 node_num, u64 *cookie)
474 {
475         u64 tmpnode = node_num;
476
477         /* shift single byte of node num into top 8 bits */
478         tmpnode <<= 56;
479
480         spin_lock(&dlm_cookie_lock);
481         *cookie = (dlm_next_cookie | tmpnode);
482         if (++dlm_next_cookie & 0xff00000000000000ull) {
483                 mlog(0, "This node's cookie will now wrap!\n");
484                 dlm_next_cookie = 1;
485         }
486         spin_unlock(&dlm_cookie_lock);
487 }
488
489 enum dlm_status dlmlock(struct dlm_ctxt *dlm, int mode,
490                         struct dlm_lockstatus *lksb, int flags,
491                         const char *name, dlm_astlockfunc_t *ast, void *data,
492                         dlm_bastlockfunc_t *bast)
493 {
494         enum dlm_status status;
495         struct dlm_lock_resource *res = NULL;
496         struct dlm_lock *lock = NULL;
497         int convert = 0, recovery = 0;
498
499         /* yes this function is a mess.
500          * TODO: clean this up.  lots of common code in the
501          *       lock and convert paths, especially in the retry blocks */
502         if (!lksb) {
503                 dlm_error(DLM_BADARGS);
504                 return DLM_BADARGS;
505         }
506
507         status = DLM_BADPARAM;
508         if (mode != LKM_EXMODE && mode != LKM_PRMODE && mode != LKM_NLMODE) {
509                 dlm_error(status);
510                 goto error;
511         }
512
513         if (flags & ~LKM_VALID_FLAGS) {
514                 dlm_error(status);
515                 goto error;
516         }
517
518         convert = (flags & LKM_CONVERT);
519         recovery = (flags & LKM_RECOVERY);
520
521         if (recovery &&
522             (!dlm_is_recovery_lock(name, strlen(name)) || convert) ) {
523                 dlm_error(status);
524                 goto error;
525         }
526         if (convert && (flags & LKM_LOCAL)) {
527                 mlog(ML_ERROR, "strange LOCAL convert request!\n");
528                 goto error;
529         }
530
531         if (convert) {
532                 /* CONVERT request */
533
534                 /* if converting, must pass in a valid dlm_lock */
535                 lock = lksb->lockid;
536                 if (!lock) {
537                         mlog(ML_ERROR, "NULL lock pointer in convert "
538                              "request\n");
539                         goto error;
540                 }
541
542                 res = lock->lockres;
543                 if (!res) {
544                         mlog(ML_ERROR, "NULL lockres pointer in convert "
545                              "request\n");
546                         goto error;
547                 }
548                 dlm_lockres_get(res);
549
550                 /* XXX: for ocfs2 purposes, the ast/bast/astdata/lksb are
551                  * static after the original lock call.  convert requests will
552                  * ensure that everything is the same, or return DLM_BADARGS.
553                  * this means that DLM_DENIED_NOASTS will never be returned.
554                  */
555                 if (lock->lksb != lksb || lock->ast != ast ||
556                     lock->bast != bast || lock->astdata != data) {
557                         status = DLM_BADARGS;
558                         mlog(ML_ERROR, "new args:  lksb=%p, ast=%p, bast=%p, "
559                              "astdata=%p\n", lksb, ast, bast, data);
560                         mlog(ML_ERROR, "orig args: lksb=%p, ast=%p, bast=%p, "
561                              "astdata=%p\n", lock->lksb, lock->ast,
562                              lock->bast, lock->astdata);
563                         goto error;
564                 }
565 retry_convert:
566                 dlm_wait_for_recovery(dlm);
567
568                 if (res->owner == dlm->node_num)
569                         status = dlmconvert_master(dlm, res, lock, flags, mode);
570                 else
571                         status = dlmconvert_remote(dlm, res, lock, flags, mode);
572                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
573                     status == DLM_FORWARD) {
574                         /* for now, see how this works without sleeping
575                          * and just retry right away.  I suspect the reco
576                          * or migration will complete fast enough that
577                          * no waiting will be necessary */
578                         mlog(0, "retrying convert with migration/recovery/"
579                              "in-progress\n");
580                         msleep(100);
581                         goto retry_convert;
582                 }
583         } else {
584                 u64 tmpcookie;
585
586                 /* LOCK request */
587                 status = DLM_BADARGS;
588                 if (!name) {
589                         dlm_error(status);
590                         goto error;
591                 }
592
593                 status = DLM_IVBUFLEN;
594                 if (strlen(name) > DLM_LOCKID_NAME_MAX || strlen(name) < 1) {
595                         dlm_error(status);
596                         goto error;
597                 }
598
599                 dlm_get_next_cookie(dlm->node_num, &tmpcookie);
600                 lock = dlm_new_lock(mode, dlm->node_num, tmpcookie, lksb);
601                 if (!lock) {
602                         dlm_error(status);
603                         goto error;
604                 }
605
606                 if (!recovery)
607                         dlm_wait_for_recovery(dlm);
608
609                 /* find or create the lock resource */
610                 res = dlm_get_lock_resource(dlm, name, flags);
611                 if (!res) {
612                         status = DLM_IVLOCKID;
613                         dlm_error(status);
614                         goto error;
615                 }
616
617                 mlog(0, "type=%d, flags = 0x%x\n", mode, flags);
618                 mlog(0, "creating lock: lock=%p res=%p\n", lock, res);
619
620                 dlm_lock_attach_lockres(lock, res);
621                 lock->ast = ast;
622                 lock->bast = bast;
623                 lock->astdata = data;
624
625 retry_lock:
626                 if (flags & LKM_VALBLK) {
627                         mlog(0, "LKM_VALBLK passed by caller\n");
628
629                         /* LVB requests for non PR, PW or EX locks are
630                          * ignored. */
631                         if (mode < LKM_PRMODE)
632                                 flags &= ~LKM_VALBLK;
633                         else {
634                                 flags |= LKM_GET_LVB;
635                                 lock->lksb->flags |= DLM_LKSB_GET_LVB;
636                         }
637                 }
638
639                 if (res->owner == dlm->node_num)
640                         status = dlmlock_master(dlm, res, lock, flags);
641                 else
642                         status = dlmlock_remote(dlm, res, lock, flags);
643
644                 if (status == DLM_RECOVERING || status == DLM_MIGRATING ||
645                     status == DLM_FORWARD) {
646                         mlog(0, "retrying lock with migration/"
647                              "recovery/in progress\n");
648                         msleep(100);
649                         dlm_wait_for_recovery(dlm);
650                         goto retry_lock;
651                 }
652
653                 if (status != DLM_NORMAL) {
654                         lock->lksb->flags &= ~DLM_LKSB_GET_LVB;
655                         if (status != DLM_NOTQUEUED)
656                                 dlm_error(status);
657                         goto error;
658                 }
659         }
660
661 error:
662         if (status != DLM_NORMAL) {
663                 if (lock && !convert)
664                         dlm_lock_put(lock);
665                 // this is kind of unnecessary
666                 lksb->status = status;
667         }
668
669         /* put lockres ref from the convert path
670          * or from dlm_get_lock_resource */
671         if (res)
672                 dlm_lockres_put(res);
673
674         return status;
675 }
676 EXPORT_SYMBOL_GPL(dlmlock);