ocfs2: Handle the DLM_CANCELGRANT case in user_unlock_ast()
[deliverable/linux.git] / fs / ocfs2 / dlm / 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
38 #include "cluster/nodemanager.h"
39 #include "cluster/heartbeat.h"
40 #include "cluster/tcp.h"
41
42 #include "dlmapi.h"
43
44 #include "userdlm.h"
45
46 #define MLOG_MASK_PREFIX ML_DLMFS
47 #include "cluster/masklog.h"
48
49 static inline int user_check_wait_flag(struct user_lock_res *lockres,
50 int flag)
51 {
52 int ret;
53
54 spin_lock(&lockres->l_lock);
55 ret = lockres->l_flags & flag;
56 spin_unlock(&lockres->l_lock);
57
58 return ret;
59 }
60
61 static inline void user_wait_on_busy_lock(struct user_lock_res *lockres)
62
63 {
64 wait_event(lockres->l_event,
65 !user_check_wait_flag(lockres, USER_LOCK_BUSY));
66 }
67
68 static inline void user_wait_on_blocked_lock(struct user_lock_res *lockres)
69
70 {
71 wait_event(lockres->l_event,
72 !user_check_wait_flag(lockres, USER_LOCK_BLOCKED));
73 }
74
75 /* I heart container_of... */
76 static inline struct dlm_ctxt *
77 dlm_ctxt_from_user_lockres(struct user_lock_res *lockres)
78 {
79 struct dlmfs_inode_private *ip;
80
81 ip = container_of(lockres,
82 struct dlmfs_inode_private,
83 ip_lockres);
84 return ip->ip_dlm;
85 }
86
87 static struct inode *
88 user_dlm_inode_from_user_lockres(struct user_lock_res *lockres)
89 {
90 struct dlmfs_inode_private *ip;
91
92 ip = container_of(lockres,
93 struct dlmfs_inode_private,
94 ip_lockres);
95 return &ip->ip_vfs_inode;
96 }
97
98 static inline void user_recover_from_dlm_error(struct user_lock_res *lockres)
99 {
100 spin_lock(&lockres->l_lock);
101 lockres->l_flags &= ~USER_LOCK_BUSY;
102 spin_unlock(&lockres->l_lock);
103 }
104
105 #define user_log_dlm_error(_func, _stat, _lockres) do { \
106 mlog(ML_ERROR, "Dlm error \"%s\" while calling %s on " \
107 "resource %s: %s\n", dlm_errname(_stat), _func, \
108 _lockres->l_name, dlm_errmsg(_stat)); \
109 } while (0)
110
111 /* WARNING: This function lives in a world where the only three lock
112 * levels are EX, PR, and NL. It *will* have to be adjusted when more
113 * lock types are added. */
114 static inline int user_highest_compat_lock_level(int level)
115 {
116 int new_level = LKM_EXMODE;
117
118 if (level == LKM_EXMODE)
119 new_level = LKM_NLMODE;
120 else if (level == LKM_PRMODE)
121 new_level = LKM_PRMODE;
122 return new_level;
123 }
124
125 static void user_ast(void *opaque)
126 {
127 struct user_lock_res *lockres = opaque;
128 struct dlm_lockstatus *lksb;
129
130 mlog(0, "AST fired for lockres %s\n", lockres->l_name);
131
132 spin_lock(&lockres->l_lock);
133
134 lksb = &(lockres->l_lksb);
135 if (lksb->status != DLM_NORMAL) {
136 mlog(ML_ERROR, "lksb status value of %u on lockres %s\n",
137 lksb->status, lockres->l_name);
138 spin_unlock(&lockres->l_lock);
139 return;
140 }
141
142 mlog_bug_on_msg(lockres->l_requested == LKM_IVMODE,
143 "Lockres %s, requested ivmode. flags 0x%x\n",
144 lockres->l_name, lockres->l_flags);
145
146 /* we're downconverting. */
147 if (lockres->l_requested < lockres->l_level) {
148 if (lockres->l_requested <=
149 user_highest_compat_lock_level(lockres->l_blocking)) {
150 lockres->l_blocking = LKM_NLMODE;
151 lockres->l_flags &= ~USER_LOCK_BLOCKED;
152 }
153 }
154
155 lockres->l_level = lockres->l_requested;
156 lockres->l_requested = LKM_IVMODE;
157 lockres->l_flags |= USER_LOCK_ATTACHED;
158 lockres->l_flags &= ~USER_LOCK_BUSY;
159
160 spin_unlock(&lockres->l_lock);
161
162 wake_up(&lockres->l_event);
163 }
164
165 static inline void user_dlm_grab_inode_ref(struct user_lock_res *lockres)
166 {
167 struct inode *inode;
168 inode = user_dlm_inode_from_user_lockres(lockres);
169 if (!igrab(inode))
170 BUG();
171 }
172
173 static void user_dlm_unblock_lock(void *opaque);
174
175 static void __user_dlm_queue_lockres(struct user_lock_res *lockres)
176 {
177 if (!(lockres->l_flags & USER_LOCK_QUEUED)) {
178 user_dlm_grab_inode_ref(lockres);
179
180 INIT_WORK(&lockres->l_work, user_dlm_unblock_lock,
181 lockres);
182
183 queue_work(user_dlm_worker, &lockres->l_work);
184 lockres->l_flags |= USER_LOCK_QUEUED;
185 }
186 }
187
188 static void __user_dlm_cond_queue_lockres(struct user_lock_res *lockres)
189 {
190 int queue = 0;
191
192 if (!(lockres->l_flags & USER_LOCK_BLOCKED))
193 return;
194
195 switch (lockres->l_blocking) {
196 case LKM_EXMODE:
197 if (!lockres->l_ex_holders && !lockres->l_ro_holders)
198 queue = 1;
199 break;
200 case LKM_PRMODE:
201 if (!lockres->l_ex_holders)
202 queue = 1;
203 break;
204 default:
205 BUG();
206 }
207
208 if (queue)
209 __user_dlm_queue_lockres(lockres);
210 }
211
212 static void user_bast(void *opaque, int level)
213 {
214 struct user_lock_res *lockres = opaque;
215
216 mlog(0, "Blocking AST fired for lockres %s. Blocking level %d\n",
217 lockres->l_name, level);
218
219 spin_lock(&lockres->l_lock);
220 lockres->l_flags |= USER_LOCK_BLOCKED;
221 if (level > lockres->l_blocking)
222 lockres->l_blocking = level;
223
224 __user_dlm_queue_lockres(lockres);
225 spin_unlock(&lockres->l_lock);
226
227 wake_up(&lockres->l_event);
228 }
229
230 static void user_unlock_ast(void *opaque, enum dlm_status status)
231 {
232 struct user_lock_res *lockres = opaque;
233
234 mlog(0, "UNLOCK AST called on lock %s\n", lockres->l_name);
235
236 if (status != DLM_NORMAL && status != DLM_CANCELGRANT)
237 mlog(ML_ERROR, "Dlm returns status %d\n", status);
238
239 spin_lock(&lockres->l_lock);
240 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN)
241 lockres->l_level = LKM_IVMODE;
242 else if (status == DLM_CANCELGRANT) {
243 mlog(0, "Lock %s, cancel fails, flags 0x%x\n",
244 lockres->l_name, lockres->l_flags);
245 /* We tried to cancel a convert request, but it was
246 * already granted. Don't clear the busy flag - the
247 * ast should've done this already. */
248 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
249 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
250 goto out_noclear;
251 } else {
252 BUG_ON(!(lockres->l_flags & USER_LOCK_IN_CANCEL));
253 /* Cancel succeeded, we want to re-queue */
254 mlog(0, "Lock %s, cancel succeeds, flags 0x%x\n",
255 lockres->l_name, lockres->l_flags);
256 lockres->l_requested = LKM_IVMODE; /* cancel an
257 * upconvert
258 * request. */
259 lockres->l_flags &= ~USER_LOCK_IN_CANCEL;
260 /* we want the unblock thread to look at it again
261 * now. */
262 if (lockres->l_flags & USER_LOCK_BLOCKED)
263 __user_dlm_queue_lockres(lockres);
264 }
265
266 lockres->l_flags &= ~USER_LOCK_BUSY;
267 out_noclear:
268 spin_unlock(&lockres->l_lock);
269
270 wake_up(&lockres->l_event);
271 }
272
273 static inline void user_dlm_drop_inode_ref(struct user_lock_res *lockres)
274 {
275 struct inode *inode;
276 inode = user_dlm_inode_from_user_lockres(lockres);
277 iput(inode);
278 }
279
280 static void user_dlm_unblock_lock(void *opaque)
281 {
282 int new_level, status;
283 struct user_lock_res *lockres = (struct user_lock_res *) opaque;
284 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
285
286 mlog(0, "processing lockres %s\n", lockres->l_name);
287
288 spin_lock(&lockres->l_lock);
289
290 mlog_bug_on_msg(!(lockres->l_flags & USER_LOCK_QUEUED),
291 "Lockres %s, flags 0x%x\n",
292 lockres->l_name, lockres->l_flags);
293
294 /* notice that we don't clear USER_LOCK_BLOCKED here. If it's
295 * set, we want user_ast clear it. */
296 lockres->l_flags &= ~USER_LOCK_QUEUED;
297
298 /* It's valid to get here and no longer be blocked - if we get
299 * several basts in a row, we might be queued by the first
300 * one, the unblock thread might run and clear the queued
301 * flag, and finally we might get another bast which re-queues
302 * us before our ast for the downconvert is called. */
303 if (!(lockres->l_flags & USER_LOCK_BLOCKED)) {
304 mlog(0, "Lockres %s, flags 0x%x: queued but not blocking\n",
305 lockres->l_name, lockres->l_flags);
306 spin_unlock(&lockres->l_lock);
307 goto drop_ref;
308 }
309
310 if (lockres->l_flags & USER_LOCK_IN_TEARDOWN) {
311 mlog(0, "lock is in teardown so we do nothing\n");
312 spin_unlock(&lockres->l_lock);
313 goto drop_ref;
314 }
315
316 if (lockres->l_flags & USER_LOCK_BUSY) {
317 mlog(0, "Cancel lock %s, flags 0x%x\n",
318 lockres->l_name, lockres->l_flags);
319
320 if (lockres->l_flags & USER_LOCK_IN_CANCEL) {
321 spin_unlock(&lockres->l_lock);
322 goto drop_ref;
323 }
324
325 lockres->l_flags |= USER_LOCK_IN_CANCEL;
326 spin_unlock(&lockres->l_lock);
327
328 status = dlmunlock(dlm,
329 &lockres->l_lksb,
330 LKM_CANCEL,
331 user_unlock_ast,
332 lockres);
333 if (status != DLM_NORMAL)
334 user_log_dlm_error("dlmunlock", status, lockres);
335 goto drop_ref;
336 }
337
338 /* If there are still incompat holders, we can exit safely
339 * without worrying about re-queueing this lock as that will
340 * happen on the last call to user_cluster_unlock. */
341 if ((lockres->l_blocking == LKM_EXMODE)
342 && (lockres->l_ex_holders || lockres->l_ro_holders)) {
343 spin_unlock(&lockres->l_lock);
344 mlog(0, "can't downconvert for ex: ro = %u, ex = %u\n",
345 lockres->l_ro_holders, lockres->l_ex_holders);
346 goto drop_ref;
347 }
348
349 if ((lockres->l_blocking == LKM_PRMODE)
350 && lockres->l_ex_holders) {
351 spin_unlock(&lockres->l_lock);
352 mlog(0, "can't downconvert for pr: ex = %u\n",
353 lockres->l_ex_holders);
354 goto drop_ref;
355 }
356
357 /* yay, we can downconvert now. */
358 new_level = user_highest_compat_lock_level(lockres->l_blocking);
359 lockres->l_requested = new_level;
360 lockres->l_flags |= USER_LOCK_BUSY;
361 mlog(0, "Downconvert lock from %d to %d\n",
362 lockres->l_level, new_level);
363 spin_unlock(&lockres->l_lock);
364
365 /* need lock downconvert request now... */
366 status = dlmlock(dlm,
367 new_level,
368 &lockres->l_lksb,
369 LKM_CONVERT|LKM_VALBLK,
370 lockres->l_name,
371 user_ast,
372 lockres,
373 user_bast);
374 if (status != DLM_NORMAL) {
375 user_log_dlm_error("dlmlock", status, lockres);
376 user_recover_from_dlm_error(lockres);
377 }
378
379 drop_ref:
380 user_dlm_drop_inode_ref(lockres);
381 }
382
383 static inline void user_dlm_inc_holders(struct user_lock_res *lockres,
384 int level)
385 {
386 switch(level) {
387 case LKM_EXMODE:
388 lockres->l_ex_holders++;
389 break;
390 case LKM_PRMODE:
391 lockres->l_ro_holders++;
392 break;
393 default:
394 BUG();
395 }
396 }
397
398 /* predict what lock level we'll be dropping down to on behalf
399 * of another node, and return true if the currently wanted
400 * level will be compatible with it. */
401 static inline int
402 user_may_continue_on_blocked_lock(struct user_lock_res *lockres,
403 int wanted)
404 {
405 BUG_ON(!(lockres->l_flags & USER_LOCK_BLOCKED));
406
407 return wanted <= user_highest_compat_lock_level(lockres->l_blocking);
408 }
409
410 int user_dlm_cluster_lock(struct user_lock_res *lockres,
411 int level,
412 int lkm_flags)
413 {
414 int status, local_flags;
415 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
416
417 if (level != LKM_EXMODE &&
418 level != LKM_PRMODE) {
419 mlog(ML_ERROR, "lockres %s: invalid request!\n",
420 lockres->l_name);
421 status = -EINVAL;
422 goto bail;
423 }
424
425 mlog(0, "lockres %s: asking for %s lock, passed flags = 0x%x\n",
426 lockres->l_name,
427 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE",
428 lkm_flags);
429
430 again:
431 if (signal_pending(current)) {
432 status = -ERESTARTSYS;
433 goto bail;
434 }
435
436 spin_lock(&lockres->l_lock);
437
438 /* We only compare against the currently granted level
439 * here. If the lock is blocked waiting on a downconvert,
440 * we'll get caught below. */
441 if ((lockres->l_flags & USER_LOCK_BUSY) &&
442 (level > lockres->l_level)) {
443 /* is someone sitting in dlm_lock? If so, wait on
444 * them. */
445 spin_unlock(&lockres->l_lock);
446
447 user_wait_on_busy_lock(lockres);
448 goto again;
449 }
450
451 if ((lockres->l_flags & USER_LOCK_BLOCKED) &&
452 (!user_may_continue_on_blocked_lock(lockres, level))) {
453 /* is the lock is currently blocked on behalf of
454 * another node */
455 spin_unlock(&lockres->l_lock);
456
457 user_wait_on_blocked_lock(lockres);
458 goto again;
459 }
460
461 if (level > lockres->l_level) {
462 local_flags = lkm_flags | LKM_VALBLK;
463 if (lockres->l_level != LKM_IVMODE)
464 local_flags |= LKM_CONVERT;
465
466 lockres->l_requested = level;
467 lockres->l_flags |= USER_LOCK_BUSY;
468 spin_unlock(&lockres->l_lock);
469
470 BUG_ON(level == LKM_IVMODE);
471 BUG_ON(level == LKM_NLMODE);
472
473 mlog(0, "lock %s, get lock from %d to level = %d\n",
474 lockres->l_name, lockres->l_level, level);
475
476 /* call dlm_lock to upgrade lock now */
477 status = dlmlock(dlm,
478 level,
479 &lockres->l_lksb,
480 local_flags,
481 lockres->l_name,
482 user_ast,
483 lockres,
484 user_bast);
485 if (status != DLM_NORMAL) {
486 if ((lkm_flags & LKM_NOQUEUE) &&
487 (status == DLM_NOTQUEUED))
488 status = -EAGAIN;
489 else {
490 user_log_dlm_error("dlmlock", status, lockres);
491 status = -EINVAL;
492 }
493 user_recover_from_dlm_error(lockres);
494 goto bail;
495 }
496
497 mlog(0, "lock %s, successfull return from dlmlock\n",
498 lockres->l_name);
499
500 user_wait_on_busy_lock(lockres);
501 goto again;
502 }
503
504 user_dlm_inc_holders(lockres, level);
505 spin_unlock(&lockres->l_lock);
506
507 mlog(0, "lockres %s: Got %s lock!\n", lockres->l_name,
508 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
509
510 status = 0;
511 bail:
512 return status;
513 }
514
515 static inline void user_dlm_dec_holders(struct user_lock_res *lockres,
516 int level)
517 {
518 switch(level) {
519 case LKM_EXMODE:
520 BUG_ON(!lockres->l_ex_holders);
521 lockres->l_ex_holders--;
522 break;
523 case LKM_PRMODE:
524 BUG_ON(!lockres->l_ro_holders);
525 lockres->l_ro_holders--;
526 break;
527 default:
528 BUG();
529 }
530 }
531
532 void user_dlm_cluster_unlock(struct user_lock_res *lockres,
533 int level)
534 {
535 if (level != LKM_EXMODE &&
536 level != LKM_PRMODE) {
537 mlog(ML_ERROR, "lockres %s: invalid request!\n", lockres->l_name);
538 return;
539 }
540
541 mlog(0, "lockres %s: dropping %s lock\n", lockres->l_name,
542 (level == LKM_EXMODE) ? "LKM_EXMODE" : "LKM_PRMODE");
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 = lockres->l_lksb.lvb;
556
557 BUG_ON(len > DLM_LVB_LEN);
558
559 spin_lock(&lockres->l_lock);
560
561 BUG_ON(lockres->l_level < LKM_EXMODE);
562 memcpy(lvb, val, len);
563
564 spin_unlock(&lockres->l_lock);
565 }
566
567 void user_dlm_read_lvb(struct inode *inode,
568 char *val,
569 unsigned int len)
570 {
571 struct user_lock_res *lockres = &DLMFS_I(inode)->ip_lockres;
572 char *lvb = lockres->l_lksb.lvb;
573
574 BUG_ON(len > DLM_LVB_LEN);
575
576 spin_lock(&lockres->l_lock);
577
578 BUG_ON(lockres->l_level < LKM_PRMODE);
579 memcpy(val, lvb, len);
580
581 spin_unlock(&lockres->l_lock);
582 }
583
584 void user_dlm_lock_res_init(struct user_lock_res *lockres,
585 struct dentry *dentry)
586 {
587 memset(lockres, 0, sizeof(*lockres));
588
589 spin_lock_init(&lockres->l_lock);
590 init_waitqueue_head(&lockres->l_event);
591 lockres->l_level = LKM_IVMODE;
592 lockres->l_requested = LKM_IVMODE;
593 lockres->l_blocking = LKM_IVMODE;
594
595 /* should have been checked before getting here. */
596 BUG_ON(dentry->d_name.len >= USER_DLM_LOCK_ID_MAX_LEN);
597
598 memcpy(lockres->l_name,
599 dentry->d_name.name,
600 dentry->d_name.len);
601 }
602
603 int user_dlm_destroy_lock(struct user_lock_res *lockres)
604 {
605 int status = -EBUSY;
606 struct dlm_ctxt *dlm = dlm_ctxt_from_user_lockres(lockres);
607
608 mlog(0, "asked to destroy %s\n", lockres->l_name);
609
610 spin_lock(&lockres->l_lock);
611 while (lockres->l_flags & USER_LOCK_BUSY) {
612 spin_unlock(&lockres->l_lock);
613
614 mlog(0, "lock %s is busy\n", lockres->l_name);
615
616 user_wait_on_busy_lock(lockres);
617
618 spin_lock(&lockres->l_lock);
619 }
620
621 if (lockres->l_ro_holders || lockres->l_ex_holders) {
622 spin_unlock(&lockres->l_lock);
623 mlog(0, "lock %s has holders\n", lockres->l_name);
624 goto bail;
625 }
626
627 status = 0;
628 if (!(lockres->l_flags & USER_LOCK_ATTACHED)) {
629 spin_unlock(&lockres->l_lock);
630 mlog(0, "lock %s is not attached\n", lockres->l_name);
631 goto bail;
632 }
633
634 lockres->l_flags &= ~USER_LOCK_ATTACHED;
635 lockres->l_flags |= USER_LOCK_BUSY;
636 lockres->l_flags |= USER_LOCK_IN_TEARDOWN;
637 spin_unlock(&lockres->l_lock);
638
639 mlog(0, "unlocking lockres %s\n", lockres->l_name);
640 status = dlmunlock(dlm,
641 &lockres->l_lksb,
642 LKM_VALBLK,
643 user_unlock_ast,
644 lockres);
645 if (status != DLM_NORMAL) {
646 user_log_dlm_error("dlmunlock", status, lockres);
647 status = -EINVAL;
648 goto bail;
649 }
650
651 user_wait_on_busy_lock(lockres);
652
653 status = 0;
654 bail:
655 return status;
656 }
657
658 struct dlm_ctxt *user_dlm_register_context(struct qstr *name)
659 {
660 struct dlm_ctxt *dlm;
661 u32 dlm_key;
662 char *domain;
663
664 domain = kmalloc(name->len + 1, GFP_KERNEL);
665 if (!domain) {
666 mlog_errno(-ENOMEM);
667 return ERR_PTR(-ENOMEM);
668 }
669
670 dlm_key = crc32_le(0, name->name, name->len);
671
672 snprintf(domain, name->len + 1, "%.*s", name->len, name->name);
673
674 dlm = dlm_register_domain(domain, dlm_key);
675 if (IS_ERR(dlm))
676 mlog_errno(PTR_ERR(dlm));
677
678 kfree(domain);
679 return dlm;
680 }
681
682 void user_dlm_unregister_context(struct dlm_ctxt *dlm)
683 {
684 dlm_unregister_domain(dlm);
685 }
This page took 0.058776 seconds and 6 git commands to generate.