Loading...
1/******************************************************************************
2*******************************************************************************
3**
4** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
6**
7** This copyrighted material is made available to anyone wishing to use,
8** modify, copy, or redistribute it subject to the terms and conditions
9** of the GNU General Public License v.2.
10**
11*******************************************************************************
12******************************************************************************/
13
14#include "dlm_internal.h"
15#include "lockspace.h"
16#include "dir.h"
17#include "config.h"
18#include "ast.h"
19#include "memory.h"
20#include "rcom.h"
21#include "lock.h"
22#include "lowcomms.h"
23#include "member.h"
24#include "recover.h"
25
26
27/*
28 * Recovery waiting routines: these functions wait for a particular reply from
29 * a remote node, or for the remote node to report a certain status. They need
30 * to abort if the lockspace is stopped indicating a node has failed (perhaps
31 * the one being waited for).
32 */
33
34/*
35 * Wait until given function returns non-zero or lockspace is stopped
36 * (LS_RECOVERY_STOP set due to failure of a node in ls_nodes). When another
37 * function thinks it could have completed the waited-on task, they should wake
38 * up ls_wait_general to get an immediate response rather than waiting for the
39 * timeout. This uses a timeout so it can check periodically if the wait
40 * should abort due to node failure (which doesn't cause a wake_up).
41 * This should only be called by the dlm_recoverd thread.
42 */
43
44int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls))
45{
46 int error = 0;
47 int rv;
48
49 while (1) {
50 rv = wait_event_timeout(ls->ls_wait_general,
51 testfn(ls) || dlm_recovery_stopped(ls),
52 dlm_config.ci_recover_timer * HZ);
53 if (rv)
54 break;
55 }
56
57 if (dlm_recovery_stopped(ls)) {
58 log_debug(ls, "dlm_wait_function aborted");
59 error = -EINTR;
60 }
61 return error;
62}
63
64/*
65 * An efficient way for all nodes to wait for all others to have a certain
66 * status. The node with the lowest nodeid polls all the others for their
67 * status (wait_status_all) and all the others poll the node with the low id
68 * for its accumulated result (wait_status_low). When all nodes have set
69 * status flag X, then status flag X_ALL will be set on the low nodeid.
70 */
71
72uint32_t dlm_recover_status(struct dlm_ls *ls)
73{
74 uint32_t status;
75 spin_lock(&ls->ls_recover_lock);
76 status = ls->ls_recover_status;
77 spin_unlock(&ls->ls_recover_lock);
78 return status;
79}
80
81static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
82{
83 ls->ls_recover_status |= status;
84}
85
86void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
87{
88 spin_lock(&ls->ls_recover_lock);
89 _set_recover_status(ls, status);
90 spin_unlock(&ls->ls_recover_lock);
91}
92
93static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
94 int save_slots)
95{
96 struct dlm_rcom *rc = ls->ls_recover_buf;
97 struct dlm_member *memb;
98 int error = 0, delay;
99
100 list_for_each_entry(memb, &ls->ls_nodes, list) {
101 delay = 0;
102 for (;;) {
103 if (dlm_recovery_stopped(ls)) {
104 error = -EINTR;
105 goto out;
106 }
107
108 error = dlm_rcom_status(ls, memb->nodeid, 0);
109 if (error)
110 goto out;
111
112 if (save_slots)
113 dlm_slot_save(ls, rc, memb);
114
115 if (rc->rc_result & wait_status)
116 break;
117 if (delay < 1000)
118 delay += 20;
119 msleep(delay);
120 }
121 }
122 out:
123 return error;
124}
125
126static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status,
127 uint32_t status_flags)
128{
129 struct dlm_rcom *rc = ls->ls_recover_buf;
130 int error = 0, delay = 0, nodeid = ls->ls_low_nodeid;
131
132 for (;;) {
133 if (dlm_recovery_stopped(ls)) {
134 error = -EINTR;
135 goto out;
136 }
137
138 error = dlm_rcom_status(ls, nodeid, status_flags);
139 if (error)
140 break;
141
142 if (rc->rc_result & wait_status)
143 break;
144 if (delay < 1000)
145 delay += 20;
146 msleep(delay);
147 }
148 out:
149 return error;
150}
151
152static int wait_status(struct dlm_ls *ls, uint32_t status)
153{
154 uint32_t status_all = status << 1;
155 int error;
156
157 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
158 error = wait_status_all(ls, status, 0);
159 if (!error)
160 dlm_set_recover_status(ls, status_all);
161 } else
162 error = wait_status_low(ls, status_all, 0);
163
164 return error;
165}
166
167int dlm_recover_members_wait(struct dlm_ls *ls)
168{
169 struct dlm_member *memb;
170 struct dlm_slot *slots;
171 int num_slots, slots_size;
172 int error, rv;
173 uint32_t gen;
174
175 list_for_each_entry(memb, &ls->ls_nodes, list) {
176 memb->slot = -1;
177 memb->generation = 0;
178 }
179
180 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
181 error = wait_status_all(ls, DLM_RS_NODES, 1);
182 if (error)
183 goto out;
184
185 /* slots array is sparse, slots_size may be > num_slots */
186
187 rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen);
188 if (!rv) {
189 spin_lock(&ls->ls_recover_lock);
190 _set_recover_status(ls, DLM_RS_NODES_ALL);
191 ls->ls_num_slots = num_slots;
192 ls->ls_slots_size = slots_size;
193 ls->ls_slots = slots;
194 ls->ls_generation = gen;
195 spin_unlock(&ls->ls_recover_lock);
196 } else {
197 dlm_set_recover_status(ls, DLM_RS_NODES_ALL);
198 }
199 } else {
200 error = wait_status_low(ls, DLM_RS_NODES_ALL, DLM_RSF_NEED_SLOTS);
201 if (error)
202 goto out;
203
204 dlm_slots_copy_in(ls);
205 }
206 out:
207 return error;
208}
209
210int dlm_recover_directory_wait(struct dlm_ls *ls)
211{
212 return wait_status(ls, DLM_RS_DIR);
213}
214
215int dlm_recover_locks_wait(struct dlm_ls *ls)
216{
217 return wait_status(ls, DLM_RS_LOCKS);
218}
219
220int dlm_recover_done_wait(struct dlm_ls *ls)
221{
222 return wait_status(ls, DLM_RS_DONE);
223}
224
225/*
226 * The recover_list contains all the rsb's for which we've requested the new
227 * master nodeid. As replies are returned from the resource directories the
228 * rsb's are removed from the list. When the list is empty we're done.
229 *
230 * The recover_list is later similarly used for all rsb's for which we've sent
231 * new lkb's and need to receive new corresponding lkid's.
232 *
233 * We use the address of the rsb struct as a simple local identifier for the
234 * rsb so we can match an rcom reply with the rsb it was sent for.
235 */
236
237static int recover_list_empty(struct dlm_ls *ls)
238{
239 int empty;
240
241 spin_lock(&ls->ls_recover_list_lock);
242 empty = list_empty(&ls->ls_recover_list);
243 spin_unlock(&ls->ls_recover_list_lock);
244
245 return empty;
246}
247
248static void recover_list_add(struct dlm_rsb *r)
249{
250 struct dlm_ls *ls = r->res_ls;
251
252 spin_lock(&ls->ls_recover_list_lock);
253 if (list_empty(&r->res_recover_list)) {
254 list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
255 ls->ls_recover_list_count++;
256 dlm_hold_rsb(r);
257 }
258 spin_unlock(&ls->ls_recover_list_lock);
259}
260
261static void recover_list_del(struct dlm_rsb *r)
262{
263 struct dlm_ls *ls = r->res_ls;
264
265 spin_lock(&ls->ls_recover_list_lock);
266 list_del_init(&r->res_recover_list);
267 ls->ls_recover_list_count--;
268 spin_unlock(&ls->ls_recover_list_lock);
269
270 dlm_put_rsb(r);
271}
272
273static void recover_list_clear(struct dlm_ls *ls)
274{
275 struct dlm_rsb *r, *s;
276
277 spin_lock(&ls->ls_recover_list_lock);
278 list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
279 list_del_init(&r->res_recover_list);
280 r->res_recover_locks_count = 0;
281 dlm_put_rsb(r);
282 ls->ls_recover_list_count--;
283 }
284
285 if (ls->ls_recover_list_count != 0) {
286 log_error(ls, "warning: recover_list_count %d",
287 ls->ls_recover_list_count);
288 ls->ls_recover_list_count = 0;
289 }
290 spin_unlock(&ls->ls_recover_list_lock);
291}
292
293static int recover_idr_empty(struct dlm_ls *ls)
294{
295 int empty = 1;
296
297 spin_lock(&ls->ls_recover_idr_lock);
298 if (ls->ls_recover_list_count)
299 empty = 0;
300 spin_unlock(&ls->ls_recover_idr_lock);
301
302 return empty;
303}
304
305static int recover_idr_add(struct dlm_rsb *r)
306{
307 struct dlm_ls *ls = r->res_ls;
308 int rv;
309
310 idr_preload(GFP_NOFS);
311 spin_lock(&ls->ls_recover_idr_lock);
312 if (r->res_id) {
313 rv = -1;
314 goto out_unlock;
315 }
316 rv = idr_alloc(&ls->ls_recover_idr, r, 1, 0, GFP_NOWAIT);
317 if (rv < 0)
318 goto out_unlock;
319
320 r->res_id = rv;
321 ls->ls_recover_list_count++;
322 dlm_hold_rsb(r);
323 rv = 0;
324out_unlock:
325 spin_unlock(&ls->ls_recover_idr_lock);
326 idr_preload_end();
327 return rv;
328}
329
330static void recover_idr_del(struct dlm_rsb *r)
331{
332 struct dlm_ls *ls = r->res_ls;
333
334 spin_lock(&ls->ls_recover_idr_lock);
335 idr_remove(&ls->ls_recover_idr, r->res_id);
336 r->res_id = 0;
337 ls->ls_recover_list_count--;
338 spin_unlock(&ls->ls_recover_idr_lock);
339
340 dlm_put_rsb(r);
341}
342
343static struct dlm_rsb *recover_idr_find(struct dlm_ls *ls, uint64_t id)
344{
345 struct dlm_rsb *r;
346
347 spin_lock(&ls->ls_recover_idr_lock);
348 r = idr_find(&ls->ls_recover_idr, (int)id);
349 spin_unlock(&ls->ls_recover_idr_lock);
350 return r;
351}
352
353static void recover_idr_clear(struct dlm_ls *ls)
354{
355 struct dlm_rsb *r;
356 int id;
357
358 spin_lock(&ls->ls_recover_idr_lock);
359
360 idr_for_each_entry(&ls->ls_recover_idr, r, id) {
361 idr_remove(&ls->ls_recover_idr, id);
362 r->res_id = 0;
363 r->res_recover_locks_count = 0;
364 ls->ls_recover_list_count--;
365
366 dlm_put_rsb(r);
367 }
368
369 if (ls->ls_recover_list_count != 0) {
370 log_error(ls, "warning: recover_list_count %d",
371 ls->ls_recover_list_count);
372 ls->ls_recover_list_count = 0;
373 }
374 spin_unlock(&ls->ls_recover_idr_lock);
375}
376
377
378/* Master recovery: find new master node for rsb's that were
379 mastered on nodes that have been removed.
380
381 dlm_recover_masters
382 recover_master
383 dlm_send_rcom_lookup -> receive_rcom_lookup
384 dlm_dir_lookup
385 receive_rcom_lookup_reply <-
386 dlm_recover_master_reply
387 set_new_master
388 set_master_lkbs
389 set_lock_master
390*/
391
392/*
393 * Set the lock master for all LKBs in a lock queue
394 * If we are the new master of the rsb, we may have received new
395 * MSTCPY locks from other nodes already which we need to ignore
396 * when setting the new nodeid.
397 */
398
399static void set_lock_master(struct list_head *queue, int nodeid)
400{
401 struct dlm_lkb *lkb;
402
403 list_for_each_entry(lkb, queue, lkb_statequeue) {
404 if (!(lkb->lkb_flags & DLM_IFL_MSTCPY)) {
405 lkb->lkb_nodeid = nodeid;
406 lkb->lkb_remid = 0;
407 }
408 }
409}
410
411static void set_master_lkbs(struct dlm_rsb *r)
412{
413 set_lock_master(&r->res_grantqueue, r->res_nodeid);
414 set_lock_master(&r->res_convertqueue, r->res_nodeid);
415 set_lock_master(&r->res_waitqueue, r->res_nodeid);
416}
417
418/*
419 * Propagate the new master nodeid to locks
420 * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
421 * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which
422 * rsb's to consider.
423 */
424
425static void set_new_master(struct dlm_rsb *r)
426{
427 set_master_lkbs(r);
428 rsb_set_flag(r, RSB_NEW_MASTER);
429 rsb_set_flag(r, RSB_NEW_MASTER2);
430}
431
432/*
433 * We do async lookups on rsb's that need new masters. The rsb's
434 * waiting for a lookup reply are kept on the recover_list.
435 *
436 * Another node recovering the master may have sent us a rcom lookup,
437 * and our dlm_master_lookup() set it as the new master, along with
438 * NEW_MASTER so that we'll recover it here (this implies dir_nodeid
439 * equals our_nodeid below).
440 */
441
442static int recover_master(struct dlm_rsb *r, unsigned int *count)
443{
444 struct dlm_ls *ls = r->res_ls;
445 int our_nodeid, dir_nodeid;
446 int is_removed = 0;
447 int error;
448
449 if (is_master(r))
450 return 0;
451
452 is_removed = dlm_is_removed(ls, r->res_nodeid);
453
454 if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
455 return 0;
456
457 our_nodeid = dlm_our_nodeid();
458 dir_nodeid = dlm_dir_nodeid(r);
459
460 if (dir_nodeid == our_nodeid) {
461 if (is_removed) {
462 r->res_master_nodeid = our_nodeid;
463 r->res_nodeid = 0;
464 }
465
466 /* set master of lkbs to ourself when is_removed, or to
467 another new master which we set along with NEW_MASTER
468 in dlm_master_lookup */
469 set_new_master(r);
470 error = 0;
471 } else {
472 recover_idr_add(r);
473 error = dlm_send_rcom_lookup(r, dir_nodeid);
474 }
475
476 (*count)++;
477 return error;
478}
479
480/*
481 * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
482 * This is necessary because recovery can be started, aborted and restarted,
483 * causing the master nodeid to briefly change during the aborted recovery, and
484 * change back to the original value in the second recovery. The MSTCPY locks
485 * may or may not have been purged during the aborted recovery. Another node
486 * with an outstanding request in waiters list and a request reply saved in the
487 * requestqueue, cannot know whether it should ignore the reply and resend the
488 * request, or accept the reply and complete the request. It must do the
489 * former if the remote node purged MSTCPY locks, and it must do the later if
490 * the remote node did not. This is solved by always purging MSTCPY locks, in
491 * which case, the request reply would always be ignored and the request
492 * resent.
493 */
494
495static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
496{
497 int dir_nodeid = dlm_dir_nodeid(r);
498 int new_master = dir_nodeid;
499
500 if (dir_nodeid == dlm_our_nodeid())
501 new_master = 0;
502
503 dlm_purge_mstcpy_locks(r);
504 r->res_master_nodeid = dir_nodeid;
505 r->res_nodeid = new_master;
506 set_new_master(r);
507 (*count)++;
508 return 0;
509}
510
511/*
512 * Go through local root resources and for each rsb which has a master which
513 * has departed, get the new master nodeid from the directory. The dir will
514 * assign mastery to the first node to look up the new master. That means
515 * we'll discover in this lookup if we're the new master of any rsb's.
516 *
517 * We fire off all the dir lookup requests individually and asynchronously to
518 * the correct dir node.
519 */
520
521int dlm_recover_masters(struct dlm_ls *ls)
522{
523 struct dlm_rsb *r;
524 unsigned int total = 0;
525 unsigned int count = 0;
526 int nodir = dlm_no_directory(ls);
527 int error;
528
529 log_rinfo(ls, "dlm_recover_masters");
530
531 down_read(&ls->ls_root_sem);
532 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
533 if (dlm_recovery_stopped(ls)) {
534 up_read(&ls->ls_root_sem);
535 error = -EINTR;
536 goto out;
537 }
538
539 lock_rsb(r);
540 if (nodir)
541 error = recover_master_static(r, &count);
542 else
543 error = recover_master(r, &count);
544 unlock_rsb(r);
545 cond_resched();
546 total++;
547
548 if (error) {
549 up_read(&ls->ls_root_sem);
550 goto out;
551 }
552 }
553 up_read(&ls->ls_root_sem);
554
555 log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
556
557 error = dlm_wait_function(ls, &recover_idr_empty);
558 out:
559 if (error)
560 recover_idr_clear(ls);
561 return error;
562}
563
564int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
565{
566 struct dlm_rsb *r;
567 int ret_nodeid, new_master;
568
569 r = recover_idr_find(ls, rc->rc_id);
570 if (!r) {
571 log_error(ls, "dlm_recover_master_reply no id %llx",
572 (unsigned long long)rc->rc_id);
573 goto out;
574 }
575
576 ret_nodeid = rc->rc_result;
577
578 if (ret_nodeid == dlm_our_nodeid())
579 new_master = 0;
580 else
581 new_master = ret_nodeid;
582
583 lock_rsb(r);
584 r->res_master_nodeid = ret_nodeid;
585 r->res_nodeid = new_master;
586 set_new_master(r);
587 unlock_rsb(r);
588 recover_idr_del(r);
589
590 if (recover_idr_empty(ls))
591 wake_up(&ls->ls_wait_general);
592 out:
593 return 0;
594}
595
596
597/* Lock recovery: rebuild the process-copy locks we hold on a
598 remastered rsb on the new rsb master.
599
600 dlm_recover_locks
601 recover_locks
602 recover_locks_queue
603 dlm_send_rcom_lock -> receive_rcom_lock
604 dlm_recover_master_copy
605 receive_rcom_lock_reply <-
606 dlm_recover_process_copy
607*/
608
609
610/*
611 * keep a count of the number of lkb's we send to the new master; when we get
612 * an equal number of replies then recovery for the rsb is done
613 */
614
615static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head)
616{
617 struct dlm_lkb *lkb;
618 int error = 0;
619
620 list_for_each_entry(lkb, head, lkb_statequeue) {
621 error = dlm_send_rcom_lock(r, lkb);
622 if (error)
623 break;
624 r->res_recover_locks_count++;
625 }
626
627 return error;
628}
629
630static int recover_locks(struct dlm_rsb *r)
631{
632 int error = 0;
633
634 lock_rsb(r);
635
636 DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
637
638 error = recover_locks_queue(r, &r->res_grantqueue);
639 if (error)
640 goto out;
641 error = recover_locks_queue(r, &r->res_convertqueue);
642 if (error)
643 goto out;
644 error = recover_locks_queue(r, &r->res_waitqueue);
645 if (error)
646 goto out;
647
648 if (r->res_recover_locks_count)
649 recover_list_add(r);
650 else
651 rsb_clear_flag(r, RSB_NEW_MASTER);
652 out:
653 unlock_rsb(r);
654 return error;
655}
656
657int dlm_recover_locks(struct dlm_ls *ls)
658{
659 struct dlm_rsb *r;
660 int error, count = 0;
661
662 down_read(&ls->ls_root_sem);
663 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
664 if (is_master(r)) {
665 rsb_clear_flag(r, RSB_NEW_MASTER);
666 continue;
667 }
668
669 if (!rsb_flag(r, RSB_NEW_MASTER))
670 continue;
671
672 if (dlm_recovery_stopped(ls)) {
673 error = -EINTR;
674 up_read(&ls->ls_root_sem);
675 goto out;
676 }
677
678 error = recover_locks(r);
679 if (error) {
680 up_read(&ls->ls_root_sem);
681 goto out;
682 }
683
684 count += r->res_recover_locks_count;
685 }
686 up_read(&ls->ls_root_sem);
687
688 log_rinfo(ls, "dlm_recover_locks %d out", count);
689
690 error = dlm_wait_function(ls, &recover_list_empty);
691 out:
692 if (error)
693 recover_list_clear(ls);
694 return error;
695}
696
697void dlm_recovered_lock(struct dlm_rsb *r)
698{
699 DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
700
701 r->res_recover_locks_count--;
702 if (!r->res_recover_locks_count) {
703 rsb_clear_flag(r, RSB_NEW_MASTER);
704 recover_list_del(r);
705 }
706
707 if (recover_list_empty(r->res_ls))
708 wake_up(&r->res_ls->ls_wait_general);
709}
710
711/*
712 * The lvb needs to be recovered on all master rsb's. This includes setting
713 * the VALNOTVALID flag if necessary, and determining the correct lvb contents
714 * based on the lvb's of the locks held on the rsb.
715 *
716 * RSB_VALNOTVALID is set in two cases:
717 *
718 * 1. we are master, but not new, and we purged an EX/PW lock held by a
719 * failed node (in dlm_recover_purge which set RSB_RECOVER_LVB_INVAL)
720 *
721 * 2. we are a new master, and there are only NL/CR locks left.
722 * (We could probably improve this by only invaliding in this way when
723 * the previous master left uncleanly. VMS docs mention that.)
724 *
725 * The LVB contents are only considered for changing when this is a new master
726 * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with
727 * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken
728 * from the lkb with the largest lvb sequence number.
729 */
730
731static void recover_lvb(struct dlm_rsb *r)
732{
733 struct dlm_lkb *lkb, *high_lkb = NULL;
734 uint32_t high_seq = 0;
735 int lock_lvb_exists = 0;
736 int big_lock_exists = 0;
737 int lvblen = r->res_ls->ls_lvblen;
738
739 if (!rsb_flag(r, RSB_NEW_MASTER2) &&
740 rsb_flag(r, RSB_RECOVER_LVB_INVAL)) {
741 /* case 1 above */
742 rsb_set_flag(r, RSB_VALNOTVALID);
743 return;
744 }
745
746 if (!rsb_flag(r, RSB_NEW_MASTER2))
747 return;
748
749 /* we are the new master, so figure out if VALNOTVALID should
750 be set, and set the rsb lvb from the best lkb available. */
751
752 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
753 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
754 continue;
755
756 lock_lvb_exists = 1;
757
758 if (lkb->lkb_grmode > DLM_LOCK_CR) {
759 big_lock_exists = 1;
760 goto setflag;
761 }
762
763 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
764 high_lkb = lkb;
765 high_seq = lkb->lkb_lvbseq;
766 }
767 }
768
769 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
770 if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
771 continue;
772
773 lock_lvb_exists = 1;
774
775 if (lkb->lkb_grmode > DLM_LOCK_CR) {
776 big_lock_exists = 1;
777 goto setflag;
778 }
779
780 if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
781 high_lkb = lkb;
782 high_seq = lkb->lkb_lvbseq;
783 }
784 }
785
786 setflag:
787 if (!lock_lvb_exists)
788 goto out;
789
790 /* lvb is invalidated if only NL/CR locks remain */
791 if (!big_lock_exists)
792 rsb_set_flag(r, RSB_VALNOTVALID);
793
794 if (!r->res_lvbptr) {
795 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
796 if (!r->res_lvbptr)
797 goto out;
798 }
799
800 if (big_lock_exists) {
801 r->res_lvbseq = lkb->lkb_lvbseq;
802 memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen);
803 } else if (high_lkb) {
804 r->res_lvbseq = high_lkb->lkb_lvbseq;
805 memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
806 } else {
807 r->res_lvbseq = 0;
808 memset(r->res_lvbptr, 0, lvblen);
809 }
810 out:
811 return;
812}
813
814/* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks
815 converting PR->CW or CW->PR need to have their lkb_grmode set. */
816
817static void recover_conversion(struct dlm_rsb *r)
818{
819 struct dlm_ls *ls = r->res_ls;
820 struct dlm_lkb *lkb;
821 int grmode = -1;
822
823 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
824 if (lkb->lkb_grmode == DLM_LOCK_PR ||
825 lkb->lkb_grmode == DLM_LOCK_CW) {
826 grmode = lkb->lkb_grmode;
827 break;
828 }
829 }
830
831 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
832 if (lkb->lkb_grmode != DLM_LOCK_IV)
833 continue;
834 if (grmode == -1) {
835 log_debug(ls, "recover_conversion %x set gr to rq %d",
836 lkb->lkb_id, lkb->lkb_rqmode);
837 lkb->lkb_grmode = lkb->lkb_rqmode;
838 } else {
839 log_debug(ls, "recover_conversion %x set gr %d",
840 lkb->lkb_id, grmode);
841 lkb->lkb_grmode = grmode;
842 }
843 }
844}
845
846/* We've become the new master for this rsb and waiting/converting locks may
847 need to be granted in dlm_recover_grant() due to locks that may have
848 existed from a removed node. */
849
850static void recover_grant(struct dlm_rsb *r)
851{
852 if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
853 rsb_set_flag(r, RSB_RECOVER_GRANT);
854}
855
856void dlm_recover_rsbs(struct dlm_ls *ls)
857{
858 struct dlm_rsb *r;
859 unsigned int count = 0;
860
861 down_read(&ls->ls_root_sem);
862 list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
863 lock_rsb(r);
864 if (is_master(r)) {
865 if (rsb_flag(r, RSB_RECOVER_CONVERT))
866 recover_conversion(r);
867
868 /* recover lvb before granting locks so the updated
869 lvb/VALNOTVALID is presented in the completion */
870 recover_lvb(r);
871
872 if (rsb_flag(r, RSB_NEW_MASTER2))
873 recover_grant(r);
874 count++;
875 } else {
876 rsb_clear_flag(r, RSB_VALNOTVALID);
877 }
878 rsb_clear_flag(r, RSB_RECOVER_CONVERT);
879 rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
880 rsb_clear_flag(r, RSB_NEW_MASTER2);
881 unlock_rsb(r);
882 }
883 up_read(&ls->ls_root_sem);
884
885 if (count)
886 log_rinfo(ls, "dlm_recover_rsbs %d done", count);
887}
888
889/* Create a single list of all root rsb's to be used during recovery */
890
891int dlm_create_root_list(struct dlm_ls *ls)
892{
893 struct rb_node *n;
894 struct dlm_rsb *r;
895 int i, error = 0;
896
897 down_write(&ls->ls_root_sem);
898 if (!list_empty(&ls->ls_root_list)) {
899 log_error(ls, "root list not empty");
900 error = -EINVAL;
901 goto out;
902 }
903
904 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
905 spin_lock(&ls->ls_rsbtbl[i].lock);
906 for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
907 r = rb_entry(n, struct dlm_rsb, res_hashnode);
908 list_add(&r->res_root_list, &ls->ls_root_list);
909 dlm_hold_rsb(r);
910 }
911
912 if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss))
913 log_error(ls, "dlm_create_root_list toss not empty");
914 spin_unlock(&ls->ls_rsbtbl[i].lock);
915 }
916 out:
917 up_write(&ls->ls_root_sem);
918 return error;
919}
920
921void dlm_release_root_list(struct dlm_ls *ls)
922{
923 struct dlm_rsb *r, *safe;
924
925 down_write(&ls->ls_root_sem);
926 list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
927 list_del_init(&r->res_root_list);
928 dlm_put_rsb(r);
929 }
930 up_write(&ls->ls_root_sem);
931}
932
933void dlm_clear_toss(struct dlm_ls *ls)
934{
935 struct rb_node *n, *next;
936 struct dlm_rsb *r;
937 unsigned int count = 0;
938 int i;
939
940 for (i = 0; i < ls->ls_rsbtbl_size; i++) {
941 spin_lock(&ls->ls_rsbtbl[i].lock);
942 for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) {
943 next = rb_next(n);
944 r = rb_entry(n, struct dlm_rsb, res_hashnode);
945 rb_erase(n, &ls->ls_rsbtbl[i].toss);
946 dlm_free_rsb(r);
947 count++;
948 }
949 spin_unlock(&ls->ls_rsbtbl[i].lock);
950 }
951
952 if (count)
953 log_rinfo(ls, "dlm_clear_toss %u done", count);
954}
955
1// SPDX-License-Identifier: GPL-2.0-only
2/******************************************************************************
3*******************************************************************************
4**
5** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
6** Copyright (C) 2004-2005 Red Hat, Inc. All rights reserved.
7**
8**
9*******************************************************************************
10******************************************************************************/
11
12#include "dlm_internal.h"
13#include "lockspace.h"
14#include "dir.h"
15#include "config.h"
16#include "ast.h"
17#include "memory.h"
18#include "rcom.h"
19#include "lock.h"
20#include "lowcomms.h"
21#include "member.h"
22#include "recover.h"
23
24
25/*
26 * Recovery waiting routines: these functions wait for a particular reply from
27 * a remote node, or for the remote node to report a certain status. They need
28 * to abort if the lockspace is stopped indicating a node has failed (perhaps
29 * the one being waited for).
30 */
31
32/*
33 * Wait until given function returns non-zero or lockspace is stopped
34 * (LS_RECOVERY_STOP set due to failure of a node in ls_nodes). When another
35 * function thinks it could have completed the waited-on task, they should wake
36 * up ls_wait_general to get an immediate response rather than waiting for the
37 * timeout. This uses a timeout so it can check periodically if the wait
38 * should abort due to node failure (which doesn't cause a wake_up).
39 * This should only be called by the dlm_recoverd thread.
40 */
41
42int dlm_wait_function(struct dlm_ls *ls, int (*testfn) (struct dlm_ls *ls))
43{
44 int error = 0;
45 int rv;
46
47 while (1) {
48 rv = wait_event_timeout(ls->ls_wait_general,
49 testfn(ls) || dlm_recovery_stopped(ls),
50 dlm_config.ci_recover_timer * HZ);
51 if (rv)
52 break;
53 if (test_bit(LSFL_RCOM_WAIT, &ls->ls_flags)) {
54 log_debug(ls, "dlm_wait_function timed out");
55 return -ETIMEDOUT;
56 }
57 }
58
59 if (dlm_recovery_stopped(ls)) {
60 log_debug(ls, "dlm_wait_function aborted");
61 error = -EINTR;
62 }
63 return error;
64}
65
66/*
67 * An efficient way for all nodes to wait for all others to have a certain
68 * status. The node with the lowest nodeid polls all the others for their
69 * status (wait_status_all) and all the others poll the node with the low id
70 * for its accumulated result (wait_status_low). When all nodes have set
71 * status flag X, then status flag X_ALL will be set on the low nodeid.
72 */
73
74uint32_t dlm_recover_status(struct dlm_ls *ls)
75{
76 uint32_t status;
77 spin_lock_bh(&ls->ls_recover_lock);
78 status = ls->ls_recover_status;
79 spin_unlock_bh(&ls->ls_recover_lock);
80 return status;
81}
82
83static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
84{
85 ls->ls_recover_status |= status;
86}
87
88void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
89{
90 spin_lock_bh(&ls->ls_recover_lock);
91 _set_recover_status(ls, status);
92 spin_unlock_bh(&ls->ls_recover_lock);
93}
94
95static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
96 int save_slots, uint64_t seq)
97{
98 struct dlm_rcom *rc = ls->ls_recover_buf;
99 struct dlm_member *memb;
100 int error = 0, delay;
101
102 list_for_each_entry(memb, &ls->ls_nodes, list) {
103 delay = 0;
104 for (;;) {
105 if (dlm_recovery_stopped(ls)) {
106 error = -EINTR;
107 goto out;
108 }
109
110 error = dlm_rcom_status(ls, memb->nodeid, 0, seq);
111 if (error)
112 goto out;
113
114 if (save_slots)
115 dlm_slot_save(ls, rc, memb);
116
117 if (le32_to_cpu(rc->rc_result) & wait_status)
118 break;
119 if (delay < 1000)
120 delay += 20;
121 msleep(delay);
122 }
123 }
124 out:
125 return error;
126}
127
128static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status,
129 uint32_t status_flags, uint64_t seq)
130{
131 struct dlm_rcom *rc = ls->ls_recover_buf;
132 int error = 0, delay = 0, nodeid = ls->ls_low_nodeid;
133
134 for (;;) {
135 if (dlm_recovery_stopped(ls)) {
136 error = -EINTR;
137 goto out;
138 }
139
140 error = dlm_rcom_status(ls, nodeid, status_flags, seq);
141 if (error)
142 break;
143
144 if (le32_to_cpu(rc->rc_result) & wait_status)
145 break;
146 if (delay < 1000)
147 delay += 20;
148 msleep(delay);
149 }
150 out:
151 return error;
152}
153
154static int wait_status(struct dlm_ls *ls, uint32_t status, uint64_t seq)
155{
156 uint32_t status_all = status << 1;
157 int error;
158
159 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
160 error = wait_status_all(ls, status, 0, seq);
161 if (!error)
162 dlm_set_recover_status(ls, status_all);
163 } else
164 error = wait_status_low(ls, status_all, 0, seq);
165
166 return error;
167}
168
169int dlm_recover_members_wait(struct dlm_ls *ls, uint64_t seq)
170{
171 struct dlm_member *memb;
172 struct dlm_slot *slots;
173 int num_slots, slots_size;
174 int error, rv;
175 uint32_t gen;
176
177 list_for_each_entry(memb, &ls->ls_nodes, list) {
178 memb->slot = -1;
179 memb->generation = 0;
180 }
181
182 if (ls->ls_low_nodeid == dlm_our_nodeid()) {
183 error = wait_status_all(ls, DLM_RS_NODES, 1, seq);
184 if (error)
185 goto out;
186
187 /* slots array is sparse, slots_size may be > num_slots */
188
189 rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen);
190 if (!rv) {
191 spin_lock_bh(&ls->ls_recover_lock);
192 _set_recover_status(ls, DLM_RS_NODES_ALL);
193 ls->ls_num_slots = num_slots;
194 ls->ls_slots_size = slots_size;
195 ls->ls_slots = slots;
196 ls->ls_generation = gen;
197 spin_unlock_bh(&ls->ls_recover_lock);
198 } else {
199 dlm_set_recover_status(ls, DLM_RS_NODES_ALL);
200 }
201 } else {
202 error = wait_status_low(ls, DLM_RS_NODES_ALL,
203 DLM_RSF_NEED_SLOTS, seq);
204 if (error)
205 goto out;
206
207 dlm_slots_copy_in(ls);
208 }
209 out:
210 return error;
211}
212
213int dlm_recover_directory_wait(struct dlm_ls *ls, uint64_t seq)
214{
215 return wait_status(ls, DLM_RS_DIR, seq);
216}
217
218int dlm_recover_locks_wait(struct dlm_ls *ls, uint64_t seq)
219{
220 return wait_status(ls, DLM_RS_LOCKS, seq);
221}
222
223int dlm_recover_done_wait(struct dlm_ls *ls, uint64_t seq)
224{
225 return wait_status(ls, DLM_RS_DONE, seq);
226}
227
228/*
229 * The recover_list contains all the rsb's for which we've requested the new
230 * master nodeid. As replies are returned from the resource directories the
231 * rsb's are removed from the list. When the list is empty we're done.
232 *
233 * The recover_list is later similarly used for all rsb's for which we've sent
234 * new lkb's and need to receive new corresponding lkid's.
235 *
236 * We use the address of the rsb struct as a simple local identifier for the
237 * rsb so we can match an rcom reply with the rsb it was sent for.
238 */
239
240static int recover_list_empty(struct dlm_ls *ls)
241{
242 int empty;
243
244 spin_lock_bh(&ls->ls_recover_list_lock);
245 empty = list_empty(&ls->ls_recover_list);
246 spin_unlock_bh(&ls->ls_recover_list_lock);
247
248 return empty;
249}
250
251static void recover_list_add(struct dlm_rsb *r)
252{
253 struct dlm_ls *ls = r->res_ls;
254
255 spin_lock_bh(&ls->ls_recover_list_lock);
256 if (list_empty(&r->res_recover_list)) {
257 list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
258 ls->ls_recover_list_count++;
259 dlm_hold_rsb(r);
260 }
261 spin_unlock_bh(&ls->ls_recover_list_lock);
262}
263
264static void recover_list_del(struct dlm_rsb *r)
265{
266 struct dlm_ls *ls = r->res_ls;
267
268 spin_lock_bh(&ls->ls_recover_list_lock);
269 list_del_init(&r->res_recover_list);
270 ls->ls_recover_list_count--;
271 spin_unlock_bh(&ls->ls_recover_list_lock);
272
273 dlm_put_rsb(r);
274}
275
276static void recover_list_clear(struct dlm_ls *ls)
277{
278 struct dlm_rsb *r, *s;
279
280 spin_lock_bh(&ls->ls_recover_list_lock);
281 list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
282 list_del_init(&r->res_recover_list);
283 r->res_recover_locks_count = 0;
284 dlm_put_rsb(r);
285 ls->ls_recover_list_count--;
286 }
287
288 if (ls->ls_recover_list_count != 0) {
289 log_error(ls, "warning: recover_list_count %d",
290 ls->ls_recover_list_count);
291 ls->ls_recover_list_count = 0;
292 }
293 spin_unlock_bh(&ls->ls_recover_list_lock);
294}
295
296static int recover_xa_empty(struct dlm_ls *ls)
297{
298 int empty = 1;
299
300 spin_lock_bh(&ls->ls_recover_xa_lock);
301 if (ls->ls_recover_list_count)
302 empty = 0;
303 spin_unlock_bh(&ls->ls_recover_xa_lock);
304
305 return empty;
306}
307
308static int recover_xa_add(struct dlm_rsb *r)
309{
310 struct dlm_ls *ls = r->res_ls;
311 struct xa_limit limit = {
312 .min = 1,
313 .max = UINT_MAX,
314 };
315 uint32_t id;
316 int rv;
317
318 spin_lock_bh(&ls->ls_recover_xa_lock);
319 if (r->res_id) {
320 rv = -1;
321 goto out_unlock;
322 }
323 rv = xa_alloc(&ls->ls_recover_xa, &id, r, limit, GFP_ATOMIC);
324 if (rv < 0)
325 goto out_unlock;
326
327 r->res_id = id;
328 ls->ls_recover_list_count++;
329 dlm_hold_rsb(r);
330 rv = 0;
331out_unlock:
332 spin_unlock_bh(&ls->ls_recover_xa_lock);
333 return rv;
334}
335
336static void recover_xa_del(struct dlm_rsb *r)
337{
338 struct dlm_ls *ls = r->res_ls;
339
340 spin_lock_bh(&ls->ls_recover_xa_lock);
341 xa_erase_bh(&ls->ls_recover_xa, r->res_id);
342 r->res_id = 0;
343 ls->ls_recover_list_count--;
344 spin_unlock_bh(&ls->ls_recover_xa_lock);
345
346 dlm_put_rsb(r);
347}
348
349static struct dlm_rsb *recover_xa_find(struct dlm_ls *ls, uint64_t id)
350{
351 struct dlm_rsb *r;
352
353 spin_lock_bh(&ls->ls_recover_xa_lock);
354 r = xa_load(&ls->ls_recover_xa, (int)id);
355 spin_unlock_bh(&ls->ls_recover_xa_lock);
356 return r;
357}
358
359static void recover_xa_clear(struct dlm_ls *ls)
360{
361 struct dlm_rsb *r;
362 unsigned long id;
363
364 spin_lock_bh(&ls->ls_recover_xa_lock);
365
366 xa_for_each(&ls->ls_recover_xa, id, r) {
367 xa_erase_bh(&ls->ls_recover_xa, id);
368 r->res_id = 0;
369 r->res_recover_locks_count = 0;
370 ls->ls_recover_list_count--;
371
372 dlm_put_rsb(r);
373 }
374
375 if (ls->ls_recover_list_count != 0) {
376 log_error(ls, "warning: recover_list_count %d",
377 ls->ls_recover_list_count);
378 ls->ls_recover_list_count = 0;
379 }
380 spin_unlock_bh(&ls->ls_recover_xa_lock);
381}
382
383
384/* Master recovery: find new master node for rsb's that were
385 mastered on nodes that have been removed.
386
387 dlm_recover_masters
388 recover_master
389 dlm_send_rcom_lookup -> receive_rcom_lookup
390 dlm_dir_lookup
391 receive_rcom_lookup_reply <-
392 dlm_recover_master_reply
393 set_new_master
394 set_master_lkbs
395 set_lock_master
396*/
397
398/*
399 * Set the lock master for all LKBs in a lock queue
400 * If we are the new master of the rsb, we may have received new
401 * MSTCPY locks from other nodes already which we need to ignore
402 * when setting the new nodeid.
403 */
404
405static void set_lock_master(struct list_head *queue, int nodeid)
406{
407 struct dlm_lkb *lkb;
408
409 list_for_each_entry(lkb, queue, lkb_statequeue) {
410 if (!test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags)) {
411 lkb->lkb_nodeid = nodeid;
412 lkb->lkb_remid = 0;
413 }
414 }
415}
416
417static void set_master_lkbs(struct dlm_rsb *r)
418{
419 set_lock_master(&r->res_grantqueue, r->res_nodeid);
420 set_lock_master(&r->res_convertqueue, r->res_nodeid);
421 set_lock_master(&r->res_waitqueue, r->res_nodeid);
422}
423
424/*
425 * Propagate the new master nodeid to locks
426 * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
427 * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which
428 * rsb's to consider.
429 */
430
431static void set_new_master(struct dlm_rsb *r)
432{
433 set_master_lkbs(r);
434 rsb_set_flag(r, RSB_NEW_MASTER);
435 rsb_set_flag(r, RSB_NEW_MASTER2);
436}
437
438/*
439 * We do async lookups on rsb's that need new masters. The rsb's
440 * waiting for a lookup reply are kept on the recover_list.
441 *
442 * Another node recovering the master may have sent us a rcom lookup,
443 * and our dlm_master_lookup() set it as the new master, along with
444 * NEW_MASTER so that we'll recover it here (this implies dir_nodeid
445 * equals our_nodeid below).
446 */
447
448static int recover_master(struct dlm_rsb *r, unsigned int *count, uint64_t seq)
449{
450 struct dlm_ls *ls = r->res_ls;
451 int our_nodeid, dir_nodeid;
452 int is_removed = 0;
453 int error;
454
455 if (r->res_nodeid != -1 && is_master(r))
456 return 0;
457
458 if (r->res_nodeid != -1)
459 is_removed = dlm_is_removed(ls, r->res_nodeid);
460
461 if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
462 return 0;
463
464 our_nodeid = dlm_our_nodeid();
465 dir_nodeid = dlm_dir_nodeid(r);
466
467 if (dir_nodeid == our_nodeid) {
468 if (is_removed) {
469 r->res_master_nodeid = our_nodeid;
470 r->res_nodeid = 0;
471 }
472
473 /* set master of lkbs to ourself when is_removed, or to
474 another new master which we set along with NEW_MASTER
475 in dlm_master_lookup */
476 set_new_master(r);
477 error = 0;
478 } else {
479 recover_xa_add(r);
480 error = dlm_send_rcom_lookup(r, dir_nodeid, seq);
481 }
482
483 (*count)++;
484 return error;
485}
486
487/*
488 * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
489 * This is necessary because recovery can be started, aborted and restarted,
490 * causing the master nodeid to briefly change during the aborted recovery, and
491 * change back to the original value in the second recovery. The MSTCPY locks
492 * may or may not have been purged during the aborted recovery. Another node
493 * with an outstanding request in waiters list and a request reply saved in the
494 * requestqueue, cannot know whether it should ignore the reply and resend the
495 * request, or accept the reply and complete the request. It must do the
496 * former if the remote node purged MSTCPY locks, and it must do the later if
497 * the remote node did not. This is solved by always purging MSTCPY locks, in
498 * which case, the request reply would always be ignored and the request
499 * resent.
500 */
501
502static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
503{
504 int dir_nodeid = dlm_dir_nodeid(r);
505 int new_master = dir_nodeid;
506
507 if (dir_nodeid == dlm_our_nodeid())
508 new_master = 0;
509
510 dlm_purge_mstcpy_locks(r);
511 r->res_master_nodeid = dir_nodeid;
512 r->res_nodeid = new_master;
513 set_new_master(r);
514 (*count)++;
515 return 0;
516}
517
518/*
519 * Go through local root resources and for each rsb which has a master which
520 * has departed, get the new master nodeid from the directory. The dir will
521 * assign mastery to the first node to look up the new master. That means
522 * we'll discover in this lookup if we're the new master of any rsb's.
523 *
524 * We fire off all the dir lookup requests individually and asynchronously to
525 * the correct dir node.
526 */
527
528int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
529 const struct list_head *root_list)
530{
531 struct dlm_rsb *r;
532 unsigned int total = 0;
533 unsigned int count = 0;
534 int nodir = dlm_no_directory(ls);
535 int error;
536
537 log_rinfo(ls, "dlm_recover_masters");
538
539 list_for_each_entry(r, root_list, res_root_list) {
540 if (dlm_recovery_stopped(ls)) {
541 error = -EINTR;
542 goto out;
543 }
544
545 lock_rsb(r);
546 if (nodir)
547 error = recover_master_static(r, &count);
548 else
549 error = recover_master(r, &count, seq);
550 unlock_rsb(r);
551 cond_resched();
552 total++;
553
554 if (error)
555 goto out;
556 }
557
558 log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
559
560 error = dlm_wait_function(ls, &recover_xa_empty);
561 out:
562 if (error)
563 recover_xa_clear(ls);
564 return error;
565}
566
567int dlm_recover_master_reply(struct dlm_ls *ls, const struct dlm_rcom *rc)
568{
569 struct dlm_rsb *r;
570 int ret_nodeid, new_master;
571
572 r = recover_xa_find(ls, le64_to_cpu(rc->rc_id));
573 if (!r) {
574 log_error(ls, "dlm_recover_master_reply no id %llx",
575 (unsigned long long)le64_to_cpu(rc->rc_id));
576 goto out;
577 }
578
579 ret_nodeid = le32_to_cpu(rc->rc_result);
580
581 if (ret_nodeid == dlm_our_nodeid())
582 new_master = 0;
583 else
584 new_master = ret_nodeid;
585
586 lock_rsb(r);
587 r->res_master_nodeid = ret_nodeid;
588 r->res_nodeid = new_master;
589 set_new_master(r);
590 unlock_rsb(r);
591 recover_xa_del(r);
592
593 if (recover_xa_empty(ls))
594 wake_up(&ls->ls_wait_general);
595 out:
596 return 0;
597}
598
599
600/* Lock recovery: rebuild the process-copy locks we hold on a
601 remastered rsb on the new rsb master.
602
603 dlm_recover_locks
604 recover_locks
605 recover_locks_queue
606 dlm_send_rcom_lock -> receive_rcom_lock
607 dlm_recover_master_copy
608 receive_rcom_lock_reply <-
609 dlm_recover_process_copy
610*/
611
612
613/*
614 * keep a count of the number of lkb's we send to the new master; when we get
615 * an equal number of replies then recovery for the rsb is done
616 */
617
618static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head,
619 uint64_t seq)
620{
621 struct dlm_lkb *lkb;
622 int error = 0;
623
624 list_for_each_entry(lkb, head, lkb_statequeue) {
625 error = dlm_send_rcom_lock(r, lkb, seq);
626 if (error)
627 break;
628 r->res_recover_locks_count++;
629 }
630
631 return error;
632}
633
634static int recover_locks(struct dlm_rsb *r, uint64_t seq)
635{
636 int error = 0;
637
638 lock_rsb(r);
639
640 DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
641
642 error = recover_locks_queue(r, &r->res_grantqueue, seq);
643 if (error)
644 goto out;
645 error = recover_locks_queue(r, &r->res_convertqueue, seq);
646 if (error)
647 goto out;
648 error = recover_locks_queue(r, &r->res_waitqueue, seq);
649 if (error)
650 goto out;
651
652 if (r->res_recover_locks_count)
653 recover_list_add(r);
654 else
655 rsb_clear_flag(r, RSB_NEW_MASTER);
656 out:
657 unlock_rsb(r);
658 return error;
659}
660
661int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
662 const struct list_head *root_list)
663{
664 struct dlm_rsb *r;
665 int error, count = 0;
666
667 list_for_each_entry(r, root_list, res_root_list) {
668 if (r->res_nodeid != -1 && is_master(r)) {
669 rsb_clear_flag(r, RSB_NEW_MASTER);
670 continue;
671 }
672
673 if (!rsb_flag(r, RSB_NEW_MASTER))
674 continue;
675
676 if (dlm_recovery_stopped(ls)) {
677 error = -EINTR;
678 goto out;
679 }
680
681 error = recover_locks(r, seq);
682 if (error)
683 goto out;
684
685 count += r->res_recover_locks_count;
686 }
687
688 log_rinfo(ls, "dlm_recover_locks %d out", count);
689
690 error = dlm_wait_function(ls, &recover_list_empty);
691 out:
692 if (error)
693 recover_list_clear(ls);
694 return error;
695}
696
697void dlm_recovered_lock(struct dlm_rsb *r)
698{
699 DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
700
701 r->res_recover_locks_count--;
702 if (!r->res_recover_locks_count) {
703 rsb_clear_flag(r, RSB_NEW_MASTER);
704 recover_list_del(r);
705 }
706
707 if (recover_list_empty(r->res_ls))
708 wake_up(&r->res_ls->ls_wait_general);
709}
710
711/*
712 * The lvb needs to be recovered on all master rsb's. This includes setting
713 * the VALNOTVALID flag if necessary, and determining the correct lvb contents
714 * based on the lvb's of the locks held on the rsb.
715 *
716 * RSB_VALNOTVALID is set in two cases:
717 *
718 * 1. we are master, but not new, and we purged an EX/PW lock held by a
719 * failed node (in dlm_recover_purge which set RSB_RECOVER_LVB_INVAL)
720 *
721 * 2. we are a new master, and there are only NL/CR locks left.
722 * (We could probably improve this by only invaliding in this way when
723 * the previous master left uncleanly. VMS docs mention that.)
724 *
725 * The LVB contents are only considered for changing when this is a new master
726 * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with
727 * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken
728 * from the lkb with the largest lvb sequence number.
729 */
730
731static void recover_lvb(struct dlm_rsb *r)
732{
733 struct dlm_lkb *big_lkb = NULL, *iter, *high_lkb = NULL;
734 uint32_t high_seq = 0;
735 int lock_lvb_exists = 0;
736 int lvblen = r->res_ls->ls_lvblen;
737
738 if (!rsb_flag(r, RSB_NEW_MASTER2) &&
739 rsb_flag(r, RSB_RECOVER_LVB_INVAL)) {
740 /* case 1 above */
741 rsb_set_flag(r, RSB_VALNOTVALID);
742 return;
743 }
744
745 if (!rsb_flag(r, RSB_NEW_MASTER2))
746 return;
747
748 /* we are the new master, so figure out if VALNOTVALID should
749 be set, and set the rsb lvb from the best lkb available. */
750
751 list_for_each_entry(iter, &r->res_grantqueue, lkb_statequeue) {
752 if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
753 continue;
754
755 lock_lvb_exists = 1;
756
757 if (iter->lkb_grmode > DLM_LOCK_CR) {
758 big_lkb = iter;
759 goto setflag;
760 }
761
762 if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
763 high_lkb = iter;
764 high_seq = iter->lkb_lvbseq;
765 }
766 }
767
768 list_for_each_entry(iter, &r->res_convertqueue, lkb_statequeue) {
769 if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
770 continue;
771
772 lock_lvb_exists = 1;
773
774 if (iter->lkb_grmode > DLM_LOCK_CR) {
775 big_lkb = iter;
776 goto setflag;
777 }
778
779 if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
780 high_lkb = iter;
781 high_seq = iter->lkb_lvbseq;
782 }
783 }
784
785 setflag:
786 if (!lock_lvb_exists)
787 goto out;
788
789 /* lvb is invalidated if only NL/CR locks remain */
790 if (!big_lkb)
791 rsb_set_flag(r, RSB_VALNOTVALID);
792
793 if (!r->res_lvbptr) {
794 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
795 if (!r->res_lvbptr)
796 goto out;
797 }
798
799 if (big_lkb) {
800 r->res_lvbseq = big_lkb->lkb_lvbseq;
801 memcpy(r->res_lvbptr, big_lkb->lkb_lvbptr, lvblen);
802 } else if (high_lkb) {
803 r->res_lvbseq = high_lkb->lkb_lvbseq;
804 memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
805 } else {
806 r->res_lvbseq = 0;
807 memset(r->res_lvbptr, 0, lvblen);
808 }
809 out:
810 return;
811}
812
813/* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks
814 * converting PR->CW or CW->PR may need to have their lkb_grmode changed.
815 */
816
817static void recover_conversion(struct dlm_rsb *r)
818{
819 struct dlm_ls *ls = r->res_ls;
820 uint32_t other_lkid = 0;
821 int other_grmode = -1;
822 struct dlm_lkb *lkb;
823
824 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
825 if (lkb->lkb_grmode == DLM_LOCK_PR ||
826 lkb->lkb_grmode == DLM_LOCK_CW) {
827 other_grmode = lkb->lkb_grmode;
828 other_lkid = lkb->lkb_id;
829 break;
830 }
831 }
832
833 if (other_grmode == -1)
834 return;
835
836 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
837 /* Lock recovery created incompatible granted modes, so
838 * change the granted mode of the converting lock to
839 * NL. The rqmode of the converting lock should be CW,
840 * which means the converting lock should be granted at
841 * the end of recovery.
842 */
843 if (((lkb->lkb_grmode == DLM_LOCK_PR) && (other_grmode == DLM_LOCK_CW)) ||
844 ((lkb->lkb_grmode == DLM_LOCK_CW) && (other_grmode == DLM_LOCK_PR))) {
845 log_limit(ls, "%s %x gr %d rq %d, remote %d %x, other_lkid %u, other gr %d, set gr=NL",
846 __func__, lkb->lkb_id, lkb->lkb_grmode,
847 lkb->lkb_rqmode, lkb->lkb_nodeid,
848 lkb->lkb_remid, other_lkid, other_grmode);
849 lkb->lkb_grmode = DLM_LOCK_NL;
850 }
851 }
852}
853
854/* We've become the new master for this rsb and waiting/converting locks may
855 need to be granted in dlm_recover_grant() due to locks that may have
856 existed from a removed node. */
857
858static void recover_grant(struct dlm_rsb *r)
859{
860 if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
861 rsb_set_flag(r, RSB_RECOVER_GRANT);
862}
863
864void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list)
865{
866 struct dlm_rsb *r;
867 unsigned int count = 0;
868
869 list_for_each_entry(r, root_list, res_root_list) {
870 lock_rsb(r);
871 if (r->res_nodeid != -1 && is_master(r)) {
872 if (rsb_flag(r, RSB_RECOVER_CONVERT))
873 recover_conversion(r);
874
875 /* recover lvb before granting locks so the updated
876 lvb/VALNOTVALID is presented in the completion */
877 recover_lvb(r);
878
879 if (rsb_flag(r, RSB_NEW_MASTER2))
880 recover_grant(r);
881 count++;
882 } else {
883 rsb_clear_flag(r, RSB_VALNOTVALID);
884 }
885 rsb_clear_flag(r, RSB_RECOVER_CONVERT);
886 rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
887 rsb_clear_flag(r, RSB_NEW_MASTER2);
888 unlock_rsb(r);
889 }
890
891 if (count)
892 log_rinfo(ls, "dlm_recover_rsbs %d done", count);
893}
894
895void dlm_clear_inactive(struct dlm_ls *ls)
896{
897 struct dlm_rsb *r, *safe;
898 unsigned int count = 0;
899
900 write_lock_bh(&ls->ls_rsbtbl_lock);
901 list_for_each_entry_safe(r, safe, &ls->ls_slow_inactive, res_slow_list) {
902 list_del(&r->res_slow_list);
903 rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
904 dlm_rhash_rsb_params);
905
906 if (!list_empty(&r->res_scan_list))
907 list_del_init(&r->res_scan_list);
908
909 free_inactive_rsb(r);
910 count++;
911 }
912 write_unlock_bh(&ls->ls_rsbtbl_lock);
913
914 if (count)
915 log_rinfo(ls, "dlm_clear_inactive %u done", count);
916}
917