Loading...
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * dlmthread.c
5 *
6 * standalone DLM module
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/highmem.h>
32#include <linux/init.h>
33#include <linux/sysctl.h>
34#include <linux/random.h>
35#include <linux/blkdev.h>
36#include <linux/socket.h>
37#include <linux/inet.h>
38#include <linux/timer.h>
39#include <linux/kthread.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#include "dlmdomain.h"
50
51#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
52#include "cluster/masklog.h"
53
54static int dlm_thread(void *data);
55static void dlm_flush_asts(struct dlm_ctxt *dlm);
56
57#define dlm_lock_is_remote(dlm, lock) ((lock)->ml.node != (dlm)->node_num)
58
59/* will exit holding res->spinlock, but may drop in function */
60/* waits until flags are cleared on res->state */
61void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
62{
63 DECLARE_WAITQUEUE(wait, current);
64
65 assert_spin_locked(&res->spinlock);
66
67 add_wait_queue(&res->wq, &wait);
68repeat:
69 set_current_state(TASK_UNINTERRUPTIBLE);
70 if (res->state & flags) {
71 spin_unlock(&res->spinlock);
72 schedule();
73 spin_lock(&res->spinlock);
74 goto repeat;
75 }
76 remove_wait_queue(&res->wq, &wait);
77 __set_current_state(TASK_RUNNING);
78}
79
80int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
81{
82 if (list_empty(&res->granted) &&
83 list_empty(&res->converting) &&
84 list_empty(&res->blocked))
85 return 0;
86 return 1;
87}
88
89/* "unused": the lockres has no locks, is not on the dirty list,
90 * has no inflight locks (in the gap between mastery and acquiring
91 * the first lock), and has no bits in its refmap.
92 * truly ready to be freed. */
93int __dlm_lockres_unused(struct dlm_lock_resource *res)
94{
95 int bit;
96
97 assert_spin_locked(&res->spinlock);
98
99 if (__dlm_lockres_has_locks(res))
100 return 0;
101
102 /* Locks are in the process of being created */
103 if (res->inflight_locks)
104 return 0;
105
106 if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
107 return 0;
108
109 if (res->state & DLM_LOCK_RES_RECOVERING)
110 return 0;
111
112 /* Another node has this resource with this node as the master */
113 bit = find_next_bit(res->refmap, O2NM_MAX_NODES, 0);
114 if (bit < O2NM_MAX_NODES)
115 return 0;
116
117 return 1;
118}
119
120
121/* Call whenever you may have added or deleted something from one of
122 * the lockres queue's. This will figure out whether it belongs on the
123 * unused list or not and does the appropriate thing. */
124void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
125 struct dlm_lock_resource *res)
126{
127 assert_spin_locked(&dlm->spinlock);
128 assert_spin_locked(&res->spinlock);
129
130 if (__dlm_lockres_unused(res)){
131 if (list_empty(&res->purge)) {
132 mlog(0, "%s: Adding res %.*s to purge list\n",
133 dlm->name, res->lockname.len, res->lockname.name);
134
135 res->last_used = jiffies;
136 dlm_lockres_get(res);
137 list_add_tail(&res->purge, &dlm->purge_list);
138 dlm->purge_count++;
139 }
140 } else if (!list_empty(&res->purge)) {
141 mlog(0, "%s: Removing res %.*s from purge list\n",
142 dlm->name, res->lockname.len, res->lockname.name);
143
144 list_del_init(&res->purge);
145 dlm_lockres_put(res);
146 dlm->purge_count--;
147 }
148}
149
150void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
151 struct dlm_lock_resource *res)
152{
153 spin_lock(&dlm->spinlock);
154 spin_lock(&res->spinlock);
155
156 __dlm_lockres_calc_usage(dlm, res);
157
158 spin_unlock(&res->spinlock);
159 spin_unlock(&dlm->spinlock);
160}
161
162static void dlm_purge_lockres(struct dlm_ctxt *dlm,
163 struct dlm_lock_resource *res)
164{
165 int master;
166 int ret = 0;
167
168 assert_spin_locked(&dlm->spinlock);
169 assert_spin_locked(&res->spinlock);
170
171 master = (res->owner == dlm->node_num);
172
173 mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
174 res->lockname.len, res->lockname.name, master);
175
176 if (!master) {
177 res->state |= DLM_LOCK_RES_DROPPING_REF;
178 /* drop spinlock... retake below */
179 spin_unlock(&res->spinlock);
180 spin_unlock(&dlm->spinlock);
181
182 spin_lock(&res->spinlock);
183 /* This ensures that clear refmap is sent after the set */
184 __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
185 spin_unlock(&res->spinlock);
186
187 /* clear our bit from the master's refmap, ignore errors */
188 ret = dlm_drop_lockres_ref(dlm, res);
189 if (ret < 0) {
190 if (!dlm_is_host_down(ret))
191 BUG();
192 }
193 spin_lock(&dlm->spinlock);
194 spin_lock(&res->spinlock);
195 }
196
197 if (!list_empty(&res->purge)) {
198 mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
199 dlm->name, res->lockname.len, res->lockname.name, master);
200 list_del_init(&res->purge);
201 dlm_lockres_put(res);
202 dlm->purge_count--;
203 }
204
205 if (!__dlm_lockres_unused(res)) {
206 mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
207 dlm->name, res->lockname.len, res->lockname.name);
208 __dlm_print_one_lock_resource(res);
209 BUG();
210 }
211
212 __dlm_unhash_lockres(dlm, res);
213
214 /* lockres is not in the hash now. drop the flag and wake up
215 * any processes waiting in dlm_get_lock_resource. */
216 if (!master) {
217 res->state &= ~DLM_LOCK_RES_DROPPING_REF;
218 spin_unlock(&res->spinlock);
219 wake_up(&res->wq);
220 } else
221 spin_unlock(&res->spinlock);
222}
223
224static void dlm_run_purge_list(struct dlm_ctxt *dlm,
225 int purge_now)
226{
227 unsigned int run_max, unused;
228 unsigned long purge_jiffies;
229 struct dlm_lock_resource *lockres;
230
231 spin_lock(&dlm->spinlock);
232 run_max = dlm->purge_count;
233
234 while(run_max && !list_empty(&dlm->purge_list)) {
235 run_max--;
236
237 lockres = list_entry(dlm->purge_list.next,
238 struct dlm_lock_resource, purge);
239
240 spin_lock(&lockres->spinlock);
241
242 purge_jiffies = lockres->last_used +
243 msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
244
245 /* Make sure that we want to be processing this guy at
246 * this time. */
247 if (!purge_now && time_after(purge_jiffies, jiffies)) {
248 /* Since resources are added to the purge list
249 * in tail order, we can stop at the first
250 * unpurgable resource -- anyone added after
251 * him will have a greater last_used value */
252 spin_unlock(&lockres->spinlock);
253 break;
254 }
255
256 /* Status of the lockres *might* change so double
257 * check. If the lockres is unused, holding the dlm
258 * spinlock will prevent people from getting and more
259 * refs on it. */
260 unused = __dlm_lockres_unused(lockres);
261 if (!unused ||
262 (lockres->state & DLM_LOCK_RES_MIGRATING)) {
263 mlog(0, "%s: res %.*s is in use or being remastered, "
264 "used %d, state %d\n", dlm->name,
265 lockres->lockname.len, lockres->lockname.name,
266 !unused, lockres->state);
267 list_move_tail(&dlm->purge_list, &lockres->purge);
268 spin_unlock(&lockres->spinlock);
269 continue;
270 }
271
272 dlm_lockres_get(lockres);
273
274 dlm_purge_lockres(dlm, lockres);
275
276 dlm_lockres_put(lockres);
277
278 /* Avoid adding any scheduling latencies */
279 cond_resched_lock(&dlm->spinlock);
280 }
281
282 spin_unlock(&dlm->spinlock);
283}
284
285static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
286 struct dlm_lock_resource *res)
287{
288 struct dlm_lock *lock, *target;
289 int can_grant = 1;
290
291 /*
292 * Because this function is called with the lockres
293 * spinlock, and because we know that it is not migrating/
294 * recovering/in-progress, it is fine to reserve asts and
295 * basts right before queueing them all throughout
296 */
297 assert_spin_locked(&dlm->ast_lock);
298 assert_spin_locked(&res->spinlock);
299 BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
300 DLM_LOCK_RES_RECOVERING|
301 DLM_LOCK_RES_IN_PROGRESS)));
302
303converting:
304 if (list_empty(&res->converting))
305 goto blocked;
306 mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
307 res->lockname.len, res->lockname.name);
308
309 target = list_entry(res->converting.next, struct dlm_lock, list);
310 if (target->ml.convert_type == LKM_IVMODE) {
311 mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
312 dlm->name, res->lockname.len, res->lockname.name);
313 BUG();
314 }
315 list_for_each_entry(lock, &res->granted, list) {
316 if (lock==target)
317 continue;
318 if (!dlm_lock_compatible(lock->ml.type,
319 target->ml.convert_type)) {
320 can_grant = 0;
321 /* queue the BAST if not already */
322 if (lock->ml.highest_blocked == LKM_IVMODE) {
323 __dlm_lockres_reserve_ast(res);
324 __dlm_queue_bast(dlm, lock);
325 }
326 /* update the highest_blocked if needed */
327 if (lock->ml.highest_blocked < target->ml.convert_type)
328 lock->ml.highest_blocked =
329 target->ml.convert_type;
330 }
331 }
332
333 list_for_each_entry(lock, &res->converting, list) {
334 if (lock==target)
335 continue;
336 if (!dlm_lock_compatible(lock->ml.type,
337 target->ml.convert_type)) {
338 can_grant = 0;
339 if (lock->ml.highest_blocked == LKM_IVMODE) {
340 __dlm_lockres_reserve_ast(res);
341 __dlm_queue_bast(dlm, lock);
342 }
343 if (lock->ml.highest_blocked < target->ml.convert_type)
344 lock->ml.highest_blocked =
345 target->ml.convert_type;
346 }
347 }
348
349 /* we can convert the lock */
350 if (can_grant) {
351 spin_lock(&target->spinlock);
352 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
353
354 mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
355 "%d => %d, node %u\n", dlm->name, res->lockname.len,
356 res->lockname.name,
357 dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
358 dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
359 target->ml.type,
360 target->ml.convert_type, target->ml.node);
361
362 target->ml.type = target->ml.convert_type;
363 target->ml.convert_type = LKM_IVMODE;
364 list_move_tail(&target->list, &res->granted);
365
366 BUG_ON(!target->lksb);
367 target->lksb->status = DLM_NORMAL;
368
369 spin_unlock(&target->spinlock);
370
371 __dlm_lockres_reserve_ast(res);
372 __dlm_queue_ast(dlm, target);
373 /* go back and check for more */
374 goto converting;
375 }
376
377blocked:
378 if (list_empty(&res->blocked))
379 goto leave;
380 target = list_entry(res->blocked.next, struct dlm_lock, list);
381
382 list_for_each_entry(lock, &res->granted, list) {
383 if (lock==target)
384 continue;
385 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
386 can_grant = 0;
387 if (lock->ml.highest_blocked == LKM_IVMODE) {
388 __dlm_lockres_reserve_ast(res);
389 __dlm_queue_bast(dlm, lock);
390 }
391 if (lock->ml.highest_blocked < target->ml.type)
392 lock->ml.highest_blocked = target->ml.type;
393 }
394 }
395
396 list_for_each_entry(lock, &res->converting, list) {
397 if (lock==target)
398 continue;
399 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
400 can_grant = 0;
401 if (lock->ml.highest_blocked == LKM_IVMODE) {
402 __dlm_lockres_reserve_ast(res);
403 __dlm_queue_bast(dlm, lock);
404 }
405 if (lock->ml.highest_blocked < target->ml.type)
406 lock->ml.highest_blocked = target->ml.type;
407 }
408 }
409
410 /* we can grant the blocked lock (only
411 * possible if converting list empty) */
412 if (can_grant) {
413 spin_lock(&target->spinlock);
414 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
415
416 mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
417 "node %u\n", dlm->name, res->lockname.len,
418 res->lockname.name,
419 dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
420 dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
421 target->ml.type, target->ml.node);
422
423 /* target->ml.type is already correct */
424 list_move_tail(&target->list, &res->granted);
425
426 BUG_ON(!target->lksb);
427 target->lksb->status = DLM_NORMAL;
428
429 spin_unlock(&target->spinlock);
430
431 __dlm_lockres_reserve_ast(res);
432 __dlm_queue_ast(dlm, target);
433 /* go back and check for more */
434 goto converting;
435 }
436
437leave:
438 return;
439}
440
441/* must have NO locks when calling this with res !=NULL * */
442void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
443{
444 if (res) {
445 spin_lock(&dlm->spinlock);
446 spin_lock(&res->spinlock);
447 __dlm_dirty_lockres(dlm, res);
448 spin_unlock(&res->spinlock);
449 spin_unlock(&dlm->spinlock);
450 }
451 wake_up(&dlm->dlm_thread_wq);
452}
453
454void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
455{
456 assert_spin_locked(&dlm->spinlock);
457 assert_spin_locked(&res->spinlock);
458
459 /* don't shuffle secondary queues */
460 if ((res->owner == dlm->node_num)) {
461 if (res->state & (DLM_LOCK_RES_MIGRATING |
462 DLM_LOCK_RES_BLOCK_DIRTY))
463 return;
464
465 if (list_empty(&res->dirty)) {
466 /* ref for dirty_list */
467 dlm_lockres_get(res);
468 list_add_tail(&res->dirty, &dlm->dirty_list);
469 res->state |= DLM_LOCK_RES_DIRTY;
470 }
471 }
472
473 mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
474 res->lockname.name);
475}
476
477
478/* Launch the NM thread for the mounted volume */
479int dlm_launch_thread(struct dlm_ctxt *dlm)
480{
481 mlog(0, "Starting dlm_thread...\n");
482
483 dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm_thread");
484 if (IS_ERR(dlm->dlm_thread_task)) {
485 mlog_errno(PTR_ERR(dlm->dlm_thread_task));
486 dlm->dlm_thread_task = NULL;
487 return -EINVAL;
488 }
489
490 return 0;
491}
492
493void dlm_complete_thread(struct dlm_ctxt *dlm)
494{
495 if (dlm->dlm_thread_task) {
496 mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
497 kthread_stop(dlm->dlm_thread_task);
498 dlm->dlm_thread_task = NULL;
499 }
500}
501
502static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
503{
504 int empty;
505
506 spin_lock(&dlm->spinlock);
507 empty = list_empty(&dlm->dirty_list);
508 spin_unlock(&dlm->spinlock);
509
510 return empty;
511}
512
513static void dlm_flush_asts(struct dlm_ctxt *dlm)
514{
515 int ret;
516 struct dlm_lock *lock;
517 struct dlm_lock_resource *res;
518 u8 hi;
519
520 spin_lock(&dlm->ast_lock);
521 while (!list_empty(&dlm->pending_asts)) {
522 lock = list_entry(dlm->pending_asts.next,
523 struct dlm_lock, ast_list);
524 /* get an extra ref on lock */
525 dlm_lock_get(lock);
526 res = lock->lockres;
527 mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
528 "node %u\n", dlm->name, res->lockname.len,
529 res->lockname.name,
530 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
531 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
532 lock->ml.type, lock->ml.node);
533
534 BUG_ON(!lock->ast_pending);
535
536 /* remove from list (including ref) */
537 list_del_init(&lock->ast_list);
538 dlm_lock_put(lock);
539 spin_unlock(&dlm->ast_lock);
540
541 if (lock->ml.node != dlm->node_num) {
542 ret = dlm_do_remote_ast(dlm, res, lock);
543 if (ret < 0)
544 mlog_errno(ret);
545 } else
546 dlm_do_local_ast(dlm, res, lock);
547
548 spin_lock(&dlm->ast_lock);
549
550 /* possible that another ast was queued while
551 * we were delivering the last one */
552 if (!list_empty(&lock->ast_list)) {
553 mlog(0, "%s: res %.*s, AST queued while flushing last "
554 "one\n", dlm->name, res->lockname.len,
555 res->lockname.name);
556 } else
557 lock->ast_pending = 0;
558
559 /* drop the extra ref.
560 * this may drop it completely. */
561 dlm_lock_put(lock);
562 dlm_lockres_release_ast(dlm, res);
563 }
564
565 while (!list_empty(&dlm->pending_basts)) {
566 lock = list_entry(dlm->pending_basts.next,
567 struct dlm_lock, bast_list);
568 /* get an extra ref on lock */
569 dlm_lock_get(lock);
570 res = lock->lockres;
571
572 BUG_ON(!lock->bast_pending);
573
574 /* get the highest blocked lock, and reset */
575 spin_lock(&lock->spinlock);
576 BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
577 hi = lock->ml.highest_blocked;
578 lock->ml.highest_blocked = LKM_IVMODE;
579 spin_unlock(&lock->spinlock);
580
581 /* remove from list (including ref) */
582 list_del_init(&lock->bast_list);
583 dlm_lock_put(lock);
584 spin_unlock(&dlm->ast_lock);
585
586 mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
587 "blocked %d, node %u\n",
588 dlm->name, res->lockname.len, res->lockname.name,
589 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
590 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
591 hi, lock->ml.node);
592
593 if (lock->ml.node != dlm->node_num) {
594 ret = dlm_send_proxy_bast(dlm, res, lock, hi);
595 if (ret < 0)
596 mlog_errno(ret);
597 } else
598 dlm_do_local_bast(dlm, res, lock, hi);
599
600 spin_lock(&dlm->ast_lock);
601
602 /* possible that another bast was queued while
603 * we were delivering the last one */
604 if (!list_empty(&lock->bast_list)) {
605 mlog(0, "%s: res %.*s, BAST queued while flushing last "
606 "one\n", dlm->name, res->lockname.len,
607 res->lockname.name);
608 } else
609 lock->bast_pending = 0;
610
611 /* drop the extra ref.
612 * this may drop it completely. */
613 dlm_lock_put(lock);
614 dlm_lockres_release_ast(dlm, res);
615 }
616 wake_up(&dlm->ast_wq);
617 spin_unlock(&dlm->ast_lock);
618}
619
620
621#define DLM_THREAD_TIMEOUT_MS (4 * 1000)
622#define DLM_THREAD_MAX_DIRTY 100
623#define DLM_THREAD_MAX_ASTS 10
624
625static int dlm_thread(void *data)
626{
627 struct dlm_lock_resource *res;
628 struct dlm_ctxt *dlm = data;
629 unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
630
631 mlog(0, "dlm thread running for %s...\n", dlm->name);
632
633 while (!kthread_should_stop()) {
634 int n = DLM_THREAD_MAX_DIRTY;
635
636 /* dlm_shutting_down is very point-in-time, but that
637 * doesn't matter as we'll just loop back around if we
638 * get false on the leading edge of a state
639 * transition. */
640 dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
641
642 /* We really don't want to hold dlm->spinlock while
643 * calling dlm_shuffle_lists on each lockres that
644 * needs to have its queues adjusted and AST/BASTs
645 * run. So let's pull each entry off the dirty_list
646 * and drop dlm->spinlock ASAP. Once off the list,
647 * res->spinlock needs to be taken again to protect
648 * the queues while calling dlm_shuffle_lists. */
649 spin_lock(&dlm->spinlock);
650 while (!list_empty(&dlm->dirty_list)) {
651 int delay = 0;
652 res = list_entry(dlm->dirty_list.next,
653 struct dlm_lock_resource, dirty);
654
655 /* peel a lockres off, remove it from the list,
656 * unset the dirty flag and drop the dlm lock */
657 BUG_ON(!res);
658 dlm_lockres_get(res);
659
660 spin_lock(&res->spinlock);
661 /* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
662 list_del_init(&res->dirty);
663 spin_unlock(&res->spinlock);
664 spin_unlock(&dlm->spinlock);
665 /* Drop dirty_list ref */
666 dlm_lockres_put(res);
667
668 /* lockres can be re-dirtied/re-added to the
669 * dirty_list in this gap, but that is ok */
670
671 spin_lock(&dlm->ast_lock);
672 spin_lock(&res->spinlock);
673 if (res->owner != dlm->node_num) {
674 __dlm_print_one_lock_resource(res);
675 mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
676 " dirty %d\n", dlm->name,
677 !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
678 !!(res->state & DLM_LOCK_RES_MIGRATING),
679 !!(res->state & DLM_LOCK_RES_RECOVERING),
680 !!(res->state & DLM_LOCK_RES_DIRTY));
681 }
682 BUG_ON(res->owner != dlm->node_num);
683
684 /* it is now ok to move lockreses in these states
685 * to the dirty list, assuming that they will only be
686 * dirty for a short while. */
687 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
688 if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
689 DLM_LOCK_RES_RECOVERING)) {
690 /* move it to the tail and keep going */
691 res->state &= ~DLM_LOCK_RES_DIRTY;
692 spin_unlock(&res->spinlock);
693 spin_unlock(&dlm->ast_lock);
694 mlog(0, "%s: res %.*s, inprogress, delay list "
695 "shuffle, state %d\n", dlm->name,
696 res->lockname.len, res->lockname.name,
697 res->state);
698 delay = 1;
699 goto in_progress;
700 }
701
702 /* at this point the lockres is not migrating/
703 * recovering/in-progress. we have the lockres
704 * spinlock and do NOT have the dlm lock.
705 * safe to reserve/queue asts and run the lists. */
706
707 /* called while holding lockres lock */
708 dlm_shuffle_lists(dlm, res);
709 res->state &= ~DLM_LOCK_RES_DIRTY;
710 spin_unlock(&res->spinlock);
711 spin_unlock(&dlm->ast_lock);
712
713 dlm_lockres_calc_usage(dlm, res);
714
715in_progress:
716
717 spin_lock(&dlm->spinlock);
718 /* if the lock was in-progress, stick
719 * it on the back of the list */
720 if (delay) {
721 spin_lock(&res->spinlock);
722 __dlm_dirty_lockres(dlm, res);
723 spin_unlock(&res->spinlock);
724 }
725 dlm_lockres_put(res);
726
727 /* unlikely, but we may need to give time to
728 * other tasks */
729 if (!--n) {
730 mlog(0, "%s: Throttling dlm thread\n",
731 dlm->name);
732 break;
733 }
734 }
735
736 spin_unlock(&dlm->spinlock);
737 dlm_flush_asts(dlm);
738
739 /* yield and continue right away if there is more work to do */
740 if (!n) {
741 cond_resched();
742 continue;
743 }
744
745 wait_event_interruptible_timeout(dlm->dlm_thread_wq,
746 !dlm_dirty_list_empty(dlm) ||
747 kthread_should_stop(),
748 timeout);
749 }
750
751 mlog(0, "quitting DLM thread\n");
752 return 0;
753}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * dlmthread.c
4 *
5 * standalone DLM module
6 *
7 * Copyright (C) 2004 Oracle. All rights reserved.
8 */
9
10
11#include <linux/module.h>
12#include <linux/fs.h>
13#include <linux/types.h>
14#include <linux/highmem.h>
15#include <linux/init.h>
16#include <linux/sysctl.h>
17#include <linux/random.h>
18#include <linux/blkdev.h>
19#include <linux/socket.h>
20#include <linux/inet.h>
21#include <linux/timer.h>
22#include <linux/kthread.h>
23#include <linux/delay.h>
24
25
26#include "../cluster/heartbeat.h"
27#include "../cluster/nodemanager.h"
28#include "../cluster/tcp.h"
29
30#include "dlmapi.h"
31#include "dlmcommon.h"
32#include "dlmdomain.h"
33
34#define MLOG_MASK_PREFIX (ML_DLM|ML_DLM_THREAD)
35#include "../cluster/masklog.h"
36
37static int dlm_thread(void *data);
38static void dlm_flush_asts(struct dlm_ctxt *dlm);
39
40/* will exit holding res->spinlock, but may drop in function */
41/* waits until flags are cleared on res->state */
42void __dlm_wait_on_lockres_flags(struct dlm_lock_resource *res, int flags)
43{
44 DECLARE_WAITQUEUE(wait, current);
45
46 assert_spin_locked(&res->spinlock);
47
48 add_wait_queue(&res->wq, &wait);
49repeat:
50 set_current_state(TASK_UNINTERRUPTIBLE);
51 if (res->state & flags) {
52 spin_unlock(&res->spinlock);
53 schedule();
54 spin_lock(&res->spinlock);
55 goto repeat;
56 }
57 remove_wait_queue(&res->wq, &wait);
58 __set_current_state(TASK_RUNNING);
59}
60
61int __dlm_lockres_has_locks(struct dlm_lock_resource *res)
62{
63 if (list_empty(&res->granted) &&
64 list_empty(&res->converting) &&
65 list_empty(&res->blocked))
66 return 0;
67 return 1;
68}
69
70/* "unused": the lockres has no locks, is not on the dirty list,
71 * has no inflight locks (in the gap between mastery and acquiring
72 * the first lock), and has no bits in its refmap.
73 * truly ready to be freed. */
74int __dlm_lockres_unused(struct dlm_lock_resource *res)
75{
76 int bit;
77
78 assert_spin_locked(&res->spinlock);
79
80 if (__dlm_lockres_has_locks(res))
81 return 0;
82
83 /* Locks are in the process of being created */
84 if (res->inflight_locks)
85 return 0;
86
87 if (!list_empty(&res->dirty) || res->state & DLM_LOCK_RES_DIRTY)
88 return 0;
89
90 if (res->state & (DLM_LOCK_RES_RECOVERING|
91 DLM_LOCK_RES_RECOVERY_WAITING))
92 return 0;
93
94 /* Another node has this resource with this node as the master */
95 bit = find_first_bit(res->refmap, O2NM_MAX_NODES);
96 if (bit < O2NM_MAX_NODES)
97 return 0;
98
99 return 1;
100}
101
102
103/* Call whenever you may have added or deleted something from one of
104 * the lockres queue's. This will figure out whether it belongs on the
105 * unused list or not and does the appropriate thing. */
106void __dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
107 struct dlm_lock_resource *res)
108{
109 assert_spin_locked(&dlm->spinlock);
110 assert_spin_locked(&res->spinlock);
111
112 if (__dlm_lockres_unused(res)){
113 if (list_empty(&res->purge)) {
114 mlog(0, "%s: Adding res %.*s to purge list\n",
115 dlm->name, res->lockname.len, res->lockname.name);
116
117 res->last_used = jiffies;
118 dlm_lockres_get(res);
119 list_add_tail(&res->purge, &dlm->purge_list);
120 dlm->purge_count++;
121 }
122 } else if (!list_empty(&res->purge)) {
123 mlog(0, "%s: Removing res %.*s from purge list\n",
124 dlm->name, res->lockname.len, res->lockname.name);
125
126 list_del_init(&res->purge);
127 dlm_lockres_put(res);
128 dlm->purge_count--;
129 }
130}
131
132void dlm_lockres_calc_usage(struct dlm_ctxt *dlm,
133 struct dlm_lock_resource *res)
134{
135 spin_lock(&dlm->spinlock);
136 spin_lock(&res->spinlock);
137
138 __dlm_lockres_calc_usage(dlm, res);
139
140 spin_unlock(&res->spinlock);
141 spin_unlock(&dlm->spinlock);
142}
143
144/*
145 * Do the real purge work:
146 * unhash the lockres, and
147 * clear flag DLM_LOCK_RES_DROPPING_REF.
148 * It requires dlm and lockres spinlock to be taken.
149 */
150void __dlm_do_purge_lockres(struct dlm_ctxt *dlm,
151 struct dlm_lock_resource *res)
152{
153 assert_spin_locked(&dlm->spinlock);
154 assert_spin_locked(&res->spinlock);
155
156 if (!list_empty(&res->purge)) {
157 mlog(0, "%s: Removing res %.*s from purgelist\n",
158 dlm->name, res->lockname.len, res->lockname.name);
159 list_del_init(&res->purge);
160 dlm_lockres_put(res);
161 dlm->purge_count--;
162 }
163
164 if (!__dlm_lockres_unused(res)) {
165 mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
166 dlm->name, res->lockname.len, res->lockname.name);
167 __dlm_print_one_lock_resource(res);
168 BUG();
169 }
170
171 __dlm_unhash_lockres(dlm, res);
172
173 spin_lock(&dlm->track_lock);
174 if (!list_empty(&res->tracking))
175 list_del_init(&res->tracking);
176 else {
177 mlog(ML_ERROR, "%s: Resource %.*s not on the Tracking list\n",
178 dlm->name, res->lockname.len, res->lockname.name);
179 __dlm_print_one_lock_resource(res);
180 }
181 spin_unlock(&dlm->track_lock);
182
183 /*
184 * lockres is not in the hash now. drop the flag and wake up
185 * any processes waiting in dlm_get_lock_resource.
186 */
187 res->state &= ~DLM_LOCK_RES_DROPPING_REF;
188}
189
190static void dlm_purge_lockres(struct dlm_ctxt *dlm,
191 struct dlm_lock_resource *res)
192{
193 int master;
194 int ret = 0;
195
196 assert_spin_locked(&dlm->spinlock);
197 assert_spin_locked(&res->spinlock);
198
199 master = (res->owner == dlm->node_num);
200
201 mlog(0, "%s: Purging res %.*s, master %d\n", dlm->name,
202 res->lockname.len, res->lockname.name, master);
203
204 if (!master) {
205 if (res->state & DLM_LOCK_RES_DROPPING_REF) {
206 mlog(ML_NOTICE, "%s: res %.*s already in DLM_LOCK_RES_DROPPING_REF state\n",
207 dlm->name, res->lockname.len, res->lockname.name);
208 spin_unlock(&res->spinlock);
209 return;
210 }
211
212 res->state |= DLM_LOCK_RES_DROPPING_REF;
213 /* drop spinlock... retake below */
214 spin_unlock(&res->spinlock);
215 spin_unlock(&dlm->spinlock);
216
217 spin_lock(&res->spinlock);
218 /* This ensures that clear refmap is sent after the set */
219 __dlm_wait_on_lockres_flags(res, DLM_LOCK_RES_SETREF_INPROG);
220 spin_unlock(&res->spinlock);
221
222 /* clear our bit from the master's refmap, ignore errors */
223 ret = dlm_drop_lockres_ref(dlm, res);
224 if (ret < 0) {
225 if (!dlm_is_host_down(ret))
226 BUG();
227 }
228 spin_lock(&dlm->spinlock);
229 spin_lock(&res->spinlock);
230 }
231
232 if (!list_empty(&res->purge)) {
233 mlog(0, "%s: Removing res %.*s from purgelist, master %d\n",
234 dlm->name, res->lockname.len, res->lockname.name, master);
235 list_del_init(&res->purge);
236 dlm_lockres_put(res);
237 dlm->purge_count--;
238 }
239
240 if (!master && ret == DLM_DEREF_RESPONSE_INPROG) {
241 mlog(0, "%s: deref %.*s in progress\n",
242 dlm->name, res->lockname.len, res->lockname.name);
243 spin_unlock(&res->spinlock);
244 return;
245 }
246
247 if (!__dlm_lockres_unused(res)) {
248 mlog(ML_ERROR, "%s: res %.*s in use after deref\n",
249 dlm->name, res->lockname.len, res->lockname.name);
250 __dlm_print_one_lock_resource(res);
251 BUG();
252 }
253
254 __dlm_unhash_lockres(dlm, res);
255
256 spin_lock(&dlm->track_lock);
257 if (!list_empty(&res->tracking))
258 list_del_init(&res->tracking);
259 else {
260 mlog(ML_ERROR, "Resource %.*s not on the Tracking list\n",
261 res->lockname.len, res->lockname.name);
262 __dlm_print_one_lock_resource(res);
263 }
264 spin_unlock(&dlm->track_lock);
265
266 /* lockres is not in the hash now. drop the flag and wake up
267 * any processes waiting in dlm_get_lock_resource. */
268 if (!master) {
269 res->state &= ~DLM_LOCK_RES_DROPPING_REF;
270 spin_unlock(&res->spinlock);
271 wake_up(&res->wq);
272 } else
273 spin_unlock(&res->spinlock);
274}
275
276static void dlm_run_purge_list(struct dlm_ctxt *dlm,
277 int purge_now)
278{
279 unsigned int run_max, unused;
280 unsigned long purge_jiffies;
281 struct dlm_lock_resource *lockres;
282
283 spin_lock(&dlm->spinlock);
284 run_max = dlm->purge_count;
285
286 while(run_max && !list_empty(&dlm->purge_list)) {
287 run_max--;
288
289 lockres = list_entry(dlm->purge_list.next,
290 struct dlm_lock_resource, purge);
291
292 spin_lock(&lockres->spinlock);
293
294 purge_jiffies = lockres->last_used +
295 msecs_to_jiffies(DLM_PURGE_INTERVAL_MS);
296
297 /* Make sure that we want to be processing this guy at
298 * this time. */
299 if (!purge_now && time_after(purge_jiffies, jiffies)) {
300 /* Since resources are added to the purge list
301 * in tail order, we can stop at the first
302 * unpurgable resource -- anyone added after
303 * him will have a greater last_used value */
304 spin_unlock(&lockres->spinlock);
305 break;
306 }
307
308 /* Status of the lockres *might* change so double
309 * check. If the lockres is unused, holding the dlm
310 * spinlock will prevent people from getting and more
311 * refs on it. */
312 unused = __dlm_lockres_unused(lockres);
313 if (!unused ||
314 (lockres->state & DLM_LOCK_RES_MIGRATING) ||
315 (lockres->inflight_assert_workers != 0)) {
316 mlog(0, "%s: res %.*s is in use or being remastered, "
317 "used %d, state %d, assert master workers %u\n",
318 dlm->name, lockres->lockname.len,
319 lockres->lockname.name,
320 !unused, lockres->state,
321 lockres->inflight_assert_workers);
322 list_move_tail(&lockres->purge, &dlm->purge_list);
323 spin_unlock(&lockres->spinlock);
324 continue;
325 }
326
327 dlm_lockres_get(lockres);
328
329 dlm_purge_lockres(dlm, lockres);
330
331 dlm_lockres_put(lockres);
332
333 /* Avoid adding any scheduling latencies */
334 cond_resched_lock(&dlm->spinlock);
335 }
336
337 spin_unlock(&dlm->spinlock);
338}
339
340static void dlm_shuffle_lists(struct dlm_ctxt *dlm,
341 struct dlm_lock_resource *res)
342{
343 struct dlm_lock *lock, *target;
344 int can_grant = 1;
345
346 /*
347 * Because this function is called with the lockres
348 * spinlock, and because we know that it is not migrating/
349 * recovering/in-progress, it is fine to reserve asts and
350 * basts right before queueing them all throughout
351 */
352 assert_spin_locked(&dlm->ast_lock);
353 assert_spin_locked(&res->spinlock);
354 BUG_ON((res->state & (DLM_LOCK_RES_MIGRATING|
355 DLM_LOCK_RES_RECOVERING|
356 DLM_LOCK_RES_IN_PROGRESS)));
357
358converting:
359 if (list_empty(&res->converting))
360 goto blocked;
361 mlog(0, "%s: res %.*s has locks on the convert queue\n", dlm->name,
362 res->lockname.len, res->lockname.name);
363
364 target = list_entry(res->converting.next, struct dlm_lock, list);
365 if (target->ml.convert_type == LKM_IVMODE) {
366 mlog(ML_ERROR, "%s: res %.*s converting lock to invalid mode\n",
367 dlm->name, res->lockname.len, res->lockname.name);
368 BUG();
369 }
370 list_for_each_entry(lock, &res->granted, list) {
371 if (lock==target)
372 continue;
373 if (!dlm_lock_compatible(lock->ml.type,
374 target->ml.convert_type)) {
375 can_grant = 0;
376 /* queue the BAST if not already */
377 if (lock->ml.highest_blocked == LKM_IVMODE) {
378 __dlm_lockres_reserve_ast(res);
379 __dlm_queue_bast(dlm, lock);
380 }
381 /* update the highest_blocked if needed */
382 if (lock->ml.highest_blocked < target->ml.convert_type)
383 lock->ml.highest_blocked =
384 target->ml.convert_type;
385 }
386 }
387
388 list_for_each_entry(lock, &res->converting, list) {
389 if (lock==target)
390 continue;
391 if (!dlm_lock_compatible(lock->ml.type,
392 target->ml.convert_type)) {
393 can_grant = 0;
394 if (lock->ml.highest_blocked == LKM_IVMODE) {
395 __dlm_lockres_reserve_ast(res);
396 __dlm_queue_bast(dlm, lock);
397 }
398 if (lock->ml.highest_blocked < target->ml.convert_type)
399 lock->ml.highest_blocked =
400 target->ml.convert_type;
401 }
402 }
403
404 /* we can convert the lock */
405 if (can_grant) {
406 spin_lock(&target->spinlock);
407 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
408
409 mlog(0, "%s: res %.*s, AST for Converting lock %u:%llu, type "
410 "%d => %d, node %u\n", dlm->name, res->lockname.len,
411 res->lockname.name,
412 dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
413 dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
414 target->ml.type,
415 target->ml.convert_type, target->ml.node);
416
417 target->ml.type = target->ml.convert_type;
418 target->ml.convert_type = LKM_IVMODE;
419 list_move_tail(&target->list, &res->granted);
420
421 BUG_ON(!target->lksb);
422 target->lksb->status = DLM_NORMAL;
423
424 spin_unlock(&target->spinlock);
425
426 __dlm_lockres_reserve_ast(res);
427 __dlm_queue_ast(dlm, target);
428 /* go back and check for more */
429 goto converting;
430 }
431
432blocked:
433 if (list_empty(&res->blocked))
434 goto leave;
435 target = list_entry(res->blocked.next, struct dlm_lock, list);
436
437 list_for_each_entry(lock, &res->granted, list) {
438 if (lock==target)
439 continue;
440 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
441 can_grant = 0;
442 if (lock->ml.highest_blocked == LKM_IVMODE) {
443 __dlm_lockres_reserve_ast(res);
444 __dlm_queue_bast(dlm, lock);
445 }
446 if (lock->ml.highest_blocked < target->ml.type)
447 lock->ml.highest_blocked = target->ml.type;
448 }
449 }
450
451 list_for_each_entry(lock, &res->converting, list) {
452 if (lock==target)
453 continue;
454 if (!dlm_lock_compatible(lock->ml.type, target->ml.type)) {
455 can_grant = 0;
456 if (lock->ml.highest_blocked == LKM_IVMODE) {
457 __dlm_lockres_reserve_ast(res);
458 __dlm_queue_bast(dlm, lock);
459 }
460 if (lock->ml.highest_blocked < target->ml.type)
461 lock->ml.highest_blocked = target->ml.type;
462 }
463 }
464
465 /* we can grant the blocked lock (only
466 * possible if converting list empty) */
467 if (can_grant) {
468 spin_lock(&target->spinlock);
469 BUG_ON(target->ml.highest_blocked != LKM_IVMODE);
470
471 mlog(0, "%s: res %.*s, AST for Blocked lock %u:%llu, type %d, "
472 "node %u\n", dlm->name, res->lockname.len,
473 res->lockname.name,
474 dlm_get_lock_cookie_node(be64_to_cpu(target->ml.cookie)),
475 dlm_get_lock_cookie_seq(be64_to_cpu(target->ml.cookie)),
476 target->ml.type, target->ml.node);
477
478 /* target->ml.type is already correct */
479 list_move_tail(&target->list, &res->granted);
480
481 BUG_ON(!target->lksb);
482 target->lksb->status = DLM_NORMAL;
483
484 spin_unlock(&target->spinlock);
485
486 __dlm_lockres_reserve_ast(res);
487 __dlm_queue_ast(dlm, target);
488 /* go back and check for more */
489 goto converting;
490 }
491
492leave:
493 return;
494}
495
496/* must have NO locks when calling this with res !=NULL * */
497void dlm_kick_thread(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
498{
499 if (res) {
500 spin_lock(&dlm->spinlock);
501 spin_lock(&res->spinlock);
502 __dlm_dirty_lockres(dlm, res);
503 spin_unlock(&res->spinlock);
504 spin_unlock(&dlm->spinlock);
505 }
506 wake_up(&dlm->dlm_thread_wq);
507}
508
509void __dlm_dirty_lockres(struct dlm_ctxt *dlm, struct dlm_lock_resource *res)
510{
511 assert_spin_locked(&dlm->spinlock);
512 assert_spin_locked(&res->spinlock);
513
514 /* don't shuffle secondary queues */
515 if (res->owner == dlm->node_num) {
516 if (res->state & (DLM_LOCK_RES_MIGRATING |
517 DLM_LOCK_RES_BLOCK_DIRTY))
518 return;
519
520 if (list_empty(&res->dirty)) {
521 /* ref for dirty_list */
522 dlm_lockres_get(res);
523 list_add_tail(&res->dirty, &dlm->dirty_list);
524 res->state |= DLM_LOCK_RES_DIRTY;
525 }
526 }
527
528 mlog(0, "%s: res %.*s\n", dlm->name, res->lockname.len,
529 res->lockname.name);
530}
531
532
533/* Launch the NM thread for the mounted volume */
534int dlm_launch_thread(struct dlm_ctxt *dlm)
535{
536 mlog(0, "Starting dlm_thread...\n");
537
538 dlm->dlm_thread_task = kthread_run(dlm_thread, dlm, "dlm-%s",
539 dlm->name);
540 if (IS_ERR(dlm->dlm_thread_task)) {
541 mlog_errno(PTR_ERR(dlm->dlm_thread_task));
542 dlm->dlm_thread_task = NULL;
543 return -EINVAL;
544 }
545
546 return 0;
547}
548
549void dlm_complete_thread(struct dlm_ctxt *dlm)
550{
551 if (dlm->dlm_thread_task) {
552 mlog(ML_KTHREAD, "Waiting for dlm thread to exit\n");
553 kthread_stop(dlm->dlm_thread_task);
554 dlm->dlm_thread_task = NULL;
555 }
556}
557
558static int dlm_dirty_list_empty(struct dlm_ctxt *dlm)
559{
560 int empty;
561
562 spin_lock(&dlm->spinlock);
563 empty = list_empty(&dlm->dirty_list);
564 spin_unlock(&dlm->spinlock);
565
566 return empty;
567}
568
569static void dlm_flush_asts(struct dlm_ctxt *dlm)
570{
571 int ret;
572 struct dlm_lock *lock;
573 struct dlm_lock_resource *res;
574 u8 hi;
575
576 spin_lock(&dlm->ast_lock);
577 while (!list_empty(&dlm->pending_asts)) {
578 lock = list_entry(dlm->pending_asts.next,
579 struct dlm_lock, ast_list);
580 /* get an extra ref on lock */
581 dlm_lock_get(lock);
582 res = lock->lockres;
583 mlog(0, "%s: res %.*s, Flush AST for lock %u:%llu, type %d, "
584 "node %u\n", dlm->name, res->lockname.len,
585 res->lockname.name,
586 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
587 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
588 lock->ml.type, lock->ml.node);
589
590 BUG_ON(!lock->ast_pending);
591
592 /* remove from list (including ref) */
593 list_del_init(&lock->ast_list);
594 dlm_lock_put(lock);
595 spin_unlock(&dlm->ast_lock);
596
597 if (lock->ml.node != dlm->node_num) {
598 ret = dlm_do_remote_ast(dlm, res, lock);
599 if (ret < 0)
600 mlog_errno(ret);
601 } else
602 dlm_do_local_ast(dlm, res, lock);
603
604 spin_lock(&dlm->ast_lock);
605
606 /* possible that another ast was queued while
607 * we were delivering the last one */
608 if (!list_empty(&lock->ast_list)) {
609 mlog(0, "%s: res %.*s, AST queued while flushing last "
610 "one\n", dlm->name, res->lockname.len,
611 res->lockname.name);
612 } else
613 lock->ast_pending = 0;
614
615 /* drop the extra ref.
616 * this may drop it completely. */
617 dlm_lock_put(lock);
618 dlm_lockres_release_ast(dlm, res);
619 }
620
621 while (!list_empty(&dlm->pending_basts)) {
622 lock = list_entry(dlm->pending_basts.next,
623 struct dlm_lock, bast_list);
624 /* get an extra ref on lock */
625 dlm_lock_get(lock);
626 res = lock->lockres;
627
628 BUG_ON(!lock->bast_pending);
629
630 /* get the highest blocked lock, and reset */
631 spin_lock(&lock->spinlock);
632 BUG_ON(lock->ml.highest_blocked <= LKM_IVMODE);
633 hi = lock->ml.highest_blocked;
634 lock->ml.highest_blocked = LKM_IVMODE;
635 spin_unlock(&lock->spinlock);
636
637 /* remove from list (including ref) */
638 list_del_init(&lock->bast_list);
639 dlm_lock_put(lock);
640 spin_unlock(&dlm->ast_lock);
641
642 mlog(0, "%s: res %.*s, Flush BAST for lock %u:%llu, "
643 "blocked %d, node %u\n",
644 dlm->name, res->lockname.len, res->lockname.name,
645 dlm_get_lock_cookie_node(be64_to_cpu(lock->ml.cookie)),
646 dlm_get_lock_cookie_seq(be64_to_cpu(lock->ml.cookie)),
647 hi, lock->ml.node);
648
649 if (lock->ml.node != dlm->node_num) {
650 ret = dlm_send_proxy_bast(dlm, res, lock, hi);
651 if (ret < 0)
652 mlog_errno(ret);
653 } else
654 dlm_do_local_bast(dlm, res, lock, hi);
655
656 spin_lock(&dlm->ast_lock);
657
658 /* possible that another bast was queued while
659 * we were delivering the last one */
660 if (!list_empty(&lock->bast_list)) {
661 mlog(0, "%s: res %.*s, BAST queued while flushing last "
662 "one\n", dlm->name, res->lockname.len,
663 res->lockname.name);
664 } else
665 lock->bast_pending = 0;
666
667 /* drop the extra ref.
668 * this may drop it completely. */
669 dlm_lock_put(lock);
670 dlm_lockres_release_ast(dlm, res);
671 }
672 wake_up(&dlm->ast_wq);
673 spin_unlock(&dlm->ast_lock);
674}
675
676
677#define DLM_THREAD_TIMEOUT_MS (4 * 1000)
678#define DLM_THREAD_MAX_DIRTY 100
679
680static int dlm_thread(void *data)
681{
682 struct dlm_lock_resource *res;
683 struct dlm_ctxt *dlm = data;
684 unsigned long timeout = msecs_to_jiffies(DLM_THREAD_TIMEOUT_MS);
685
686 mlog(0, "dlm thread running for %s...\n", dlm->name);
687
688 while (!kthread_should_stop()) {
689 int n = DLM_THREAD_MAX_DIRTY;
690
691 /* dlm_shutting_down is very point-in-time, but that
692 * doesn't matter as we'll just loop back around if we
693 * get false on the leading edge of a state
694 * transition. */
695 dlm_run_purge_list(dlm, dlm_shutting_down(dlm));
696
697 /* We really don't want to hold dlm->spinlock while
698 * calling dlm_shuffle_lists on each lockres that
699 * needs to have its queues adjusted and AST/BASTs
700 * run. So let's pull each entry off the dirty_list
701 * and drop dlm->spinlock ASAP. Once off the list,
702 * res->spinlock needs to be taken again to protect
703 * the queues while calling dlm_shuffle_lists. */
704 spin_lock(&dlm->spinlock);
705 while (!list_empty(&dlm->dirty_list)) {
706 int delay = 0;
707 res = list_entry(dlm->dirty_list.next,
708 struct dlm_lock_resource, dirty);
709
710 /* peel a lockres off, remove it from the list,
711 * unset the dirty flag and drop the dlm lock */
712 BUG_ON(!res);
713 dlm_lockres_get(res);
714
715 spin_lock(&res->spinlock);
716 /* We clear the DLM_LOCK_RES_DIRTY state once we shuffle lists below */
717 list_del_init(&res->dirty);
718 spin_unlock(&res->spinlock);
719 spin_unlock(&dlm->spinlock);
720 /* Drop dirty_list ref */
721 dlm_lockres_put(res);
722
723 /* lockres can be re-dirtied/re-added to the
724 * dirty_list in this gap, but that is ok */
725
726 spin_lock(&dlm->ast_lock);
727 spin_lock(&res->spinlock);
728 if (res->owner != dlm->node_num) {
729 __dlm_print_one_lock_resource(res);
730 mlog(ML_ERROR, "%s: inprog %d, mig %d, reco %d,"
731 " dirty %d\n", dlm->name,
732 !!(res->state & DLM_LOCK_RES_IN_PROGRESS),
733 !!(res->state & DLM_LOCK_RES_MIGRATING),
734 !!(res->state & DLM_LOCK_RES_RECOVERING),
735 !!(res->state & DLM_LOCK_RES_DIRTY));
736 }
737 BUG_ON(res->owner != dlm->node_num);
738
739 /* it is now ok to move lockreses in these states
740 * to the dirty list, assuming that they will only be
741 * dirty for a short while. */
742 BUG_ON(res->state & DLM_LOCK_RES_MIGRATING);
743 if (res->state & (DLM_LOCK_RES_IN_PROGRESS |
744 DLM_LOCK_RES_RECOVERING |
745 DLM_LOCK_RES_RECOVERY_WAITING)) {
746 /* move it to the tail and keep going */
747 res->state &= ~DLM_LOCK_RES_DIRTY;
748 spin_unlock(&res->spinlock);
749 spin_unlock(&dlm->ast_lock);
750 mlog(0, "%s: res %.*s, inprogress, delay list "
751 "shuffle, state %d\n", dlm->name,
752 res->lockname.len, res->lockname.name,
753 res->state);
754 delay = 1;
755 goto in_progress;
756 }
757
758 /* at this point the lockres is not migrating/
759 * recovering/in-progress. we have the lockres
760 * spinlock and do NOT have the dlm lock.
761 * safe to reserve/queue asts and run the lists. */
762
763 /* called while holding lockres lock */
764 dlm_shuffle_lists(dlm, res);
765 res->state &= ~DLM_LOCK_RES_DIRTY;
766 spin_unlock(&res->spinlock);
767 spin_unlock(&dlm->ast_lock);
768
769 dlm_lockres_calc_usage(dlm, res);
770
771in_progress:
772
773 spin_lock(&dlm->spinlock);
774 /* if the lock was in-progress, stick
775 * it on the back of the list */
776 if (delay) {
777 spin_lock(&res->spinlock);
778 __dlm_dirty_lockres(dlm, res);
779 spin_unlock(&res->spinlock);
780 }
781 dlm_lockres_put(res);
782
783 /* unlikely, but we may need to give time to
784 * other tasks */
785 if (!--n) {
786 mlog(0, "%s: Throttling dlm thread\n",
787 dlm->name);
788 break;
789 }
790 }
791
792 spin_unlock(&dlm->spinlock);
793 dlm_flush_asts(dlm);
794
795 /* yield and continue right away if there is more work to do */
796 if (!n) {
797 cond_resched();
798 continue;
799 }
800
801 wait_event_interruptible_timeout(dlm->dlm_thread_wq,
802 !dlm_dirty_list_empty(dlm) ||
803 kthread_should_stop(),
804 timeout);
805 }
806
807 mlog(0, "quitting DLM thread\n");
808 return 0;
809}