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
42 int 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
74 uint32_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
83 static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
84 {
85 ls->ls_recover_status |= status;
86 }
87
88 void 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
95 static 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
128 static 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
154 static 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
169 int 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
213 int dlm_recover_directory_wait(struct dlm_ls *ls, uint64_t seq)
214 {
215 return wait_status(ls, DLM_RS_DIR, seq);
216 }
217
218 int dlm_recover_locks_wait(struct dlm_ls *ls, uint64_t seq)
219 {
220 return wait_status(ls, DLM_RS_LOCKS, seq);
221 }
222
223 int 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
240 static 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
251 static 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
264 static 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
276 static 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
296 static 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
308 static 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;
331 out_unlock:
332 spin_unlock_bh(&ls->ls_recover_xa_lock);
333 return rv;
334 }
335
336 static 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
349 static 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
359 static 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
405 static 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
417 static 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
431 static 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
448 static 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 (is_master(r))
456 return 0;
457
458 is_removed = dlm_is_removed(ls, r->res_nodeid);
459
460 if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
461 return 0;
462
463 our_nodeid = dlm_our_nodeid();
464 dir_nodeid = dlm_dir_nodeid(r);
465
466 if (dir_nodeid == our_nodeid) {
467 if (is_removed) {
468 r->res_master_nodeid = our_nodeid;
469 r->res_nodeid = 0;
470 }
471
472 /* set master of lkbs to ourself when is_removed, or to
473 another new master which we set along with NEW_MASTER
474 in dlm_master_lookup */
475 set_new_master(r);
476 error = 0;
477 } else {
478 recover_xa_add(r);
479 error = dlm_send_rcom_lookup(r, dir_nodeid, seq);
480 }
481
482 (*count)++;
483 return error;
484 }
485
486 /*
487 * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
488 * This is necessary because recovery can be started, aborted and restarted,
489 * causing the master nodeid to briefly change during the aborted recovery, and
490 * change back to the original value in the second recovery. The MSTCPY locks
491 * may or may not have been purged during the aborted recovery. Another node
492 * with an outstanding request in waiters list and a request reply saved in the
493 * requestqueue, cannot know whether it should ignore the reply and resend the
494 * request, or accept the reply and complete the request. It must do the
495 * former if the remote node purged MSTCPY locks, and it must do the later if
496 * the remote node did not. This is solved by always purging MSTCPY locks, in
497 * which case, the request reply would always be ignored and the request
498 * resent.
499 */
500
501 static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
502 {
503 int dir_nodeid = dlm_dir_nodeid(r);
504 int new_master = dir_nodeid;
505
506 if (dir_nodeid == dlm_our_nodeid())
507 new_master = 0;
508
509 dlm_purge_mstcpy_locks(r);
510 r->res_master_nodeid = dir_nodeid;
511 r->res_nodeid = new_master;
512 set_new_master(r);
513 (*count)++;
514 return 0;
515 }
516
517 /*
518 * Go through local root resources and for each rsb which has a master which
519 * has departed, get the new master nodeid from the directory. The dir will
520 * assign mastery to the first node to look up the new master. That means
521 * we'll discover in this lookup if we're the new master of any rsb's.
522 *
523 * We fire off all the dir lookup requests individually and asynchronously to
524 * the correct dir node.
525 */
526
527 int dlm_recover_masters(struct dlm_ls *ls, uint64_t seq,
528 const struct list_head *root_list)
529 {
530 struct dlm_rsb *r;
531 unsigned int total = 0;
532 unsigned int count = 0;
533 int nodir = dlm_no_directory(ls);
534 int error;
535
536 log_rinfo(ls, "dlm_recover_masters");
537
538 list_for_each_entry(r, root_list, res_root_list) {
539 if (dlm_recovery_stopped(ls)) {
540 error = -EINTR;
541 goto out;
542 }
543
544 lock_rsb(r);
545 if (nodir)
546 error = recover_master_static(r, &count);
547 else
548 error = recover_master(r, &count, seq);
549 unlock_rsb(r);
550 cond_resched();
551 total++;
552
553 if (error)
554 goto out;
555 }
556
557 log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
558
559 error = dlm_wait_function(ls, &recover_xa_empty);
560 out:
561 if (error)
562 recover_xa_clear(ls);
563 return error;
564 }
565
566 int dlm_recover_master_reply(struct dlm_ls *ls, const struct dlm_rcom *rc)
567 {
568 struct dlm_rsb *r;
569 int ret_nodeid, new_master;
570
571 r = recover_xa_find(ls, le64_to_cpu(rc->rc_id));
572 if (!r) {
573 log_error(ls, "dlm_recover_master_reply no id %llx",
574 (unsigned long long)le64_to_cpu(rc->rc_id));
575 goto out;
576 }
577
578 ret_nodeid = le32_to_cpu(rc->rc_result);
579
580 if (ret_nodeid == dlm_our_nodeid())
581 new_master = 0;
582 else
583 new_master = ret_nodeid;
584
585 lock_rsb(r);
586 r->res_master_nodeid = ret_nodeid;
587 r->res_nodeid = new_master;
588 set_new_master(r);
589 unlock_rsb(r);
590 recover_xa_del(r);
591
592 if (recover_xa_empty(ls))
593 wake_up(&ls->ls_wait_general);
594 out:
595 return 0;
596 }
597
598
599 /* Lock recovery: rebuild the process-copy locks we hold on a
600 remastered rsb on the new rsb master.
601
602 dlm_recover_locks
603 recover_locks
604 recover_locks_queue
605 dlm_send_rcom_lock -> receive_rcom_lock
606 dlm_recover_master_copy
607 receive_rcom_lock_reply <-
608 dlm_recover_process_copy
609 */
610
611
612 /*
613 * keep a count of the number of lkb's we send to the new master; when we get
614 * an equal number of replies then recovery for the rsb is done
615 */
616
617 static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head,
618 uint64_t seq)
619 {
620 struct dlm_lkb *lkb;
621 int error = 0;
622
623 list_for_each_entry(lkb, head, lkb_statequeue) {
624 error = dlm_send_rcom_lock(r, lkb, seq);
625 if (error)
626 break;
627 r->res_recover_locks_count++;
628 }
629
630 return error;
631 }
632
633 static int recover_locks(struct dlm_rsb *r, uint64_t seq)
634 {
635 int error = 0;
636
637 lock_rsb(r);
638
639 DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
640
641 error = recover_locks_queue(r, &r->res_grantqueue, seq);
642 if (error)
643 goto out;
644 error = recover_locks_queue(r, &r->res_convertqueue, seq);
645 if (error)
646 goto out;
647 error = recover_locks_queue(r, &r->res_waitqueue, seq);
648 if (error)
649 goto out;
650
651 if (r->res_recover_locks_count)
652 recover_list_add(r);
653 else
654 rsb_clear_flag(r, RSB_NEW_MASTER);
655 out:
656 unlock_rsb(r);
657 return error;
658 }
659
660 int dlm_recover_locks(struct dlm_ls *ls, uint64_t seq,
661 const struct list_head *root_list)
662 {
663 struct dlm_rsb *r;
664 int error, count = 0;
665
666 list_for_each_entry(r, root_list, res_root_list) {
667 if (is_master(r)) {
668 rsb_clear_flag(r, RSB_NEW_MASTER);
669 continue;
670 }
671
672 if (!rsb_flag(r, RSB_NEW_MASTER))
673 continue;
674
675 if (dlm_recovery_stopped(ls)) {
676 error = -EINTR;
677 goto out;
678 }
679
680 error = recover_locks(r, seq);
681 if (error)
682 goto out;
683
684 count += r->res_recover_locks_count;
685 }
686
687 log_rinfo(ls, "dlm_recover_locks %d out", count);
688
689 error = dlm_wait_function(ls, &recover_list_empty);
690 out:
691 if (error)
692 recover_list_clear(ls);
693 return error;
694 }
695
696 void dlm_recovered_lock(struct dlm_rsb *r)
697 {
698 DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
699
700 r->res_recover_locks_count--;
701 if (!r->res_recover_locks_count) {
702 rsb_clear_flag(r, RSB_NEW_MASTER);
703 recover_list_del(r);
704 }
705
706 if (recover_list_empty(r->res_ls))
707 wake_up(&r->res_ls->ls_wait_general);
708 }
709
710 /*
711 * The lvb needs to be recovered on all master rsb's. This includes setting
712 * the VALNOTVALID flag if necessary, and determining the correct lvb contents
713 * based on the lvb's of the locks held on the rsb.
714 *
715 * RSB_VALNOTVALID is set in two cases:
716 *
717 * 1. we are master, but not new, and we purged an EX/PW lock held by a
718 * failed node (in dlm_recover_purge which set RSB_RECOVER_LVB_INVAL)
719 *
720 * 2. we are a new master, and there are only NL/CR locks left.
721 * (We could probably improve this by only invaliding in this way when
722 * the previous master left uncleanly. VMS docs mention that.)
723 *
724 * The LVB contents are only considered for changing when this is a new master
725 * of the rsb (NEW_MASTER2). Then, the rsb's lvb is taken from any lkb with
726 * mode > CR. If no lkb's exist with mode above CR, the lvb contents are taken
727 * from the lkb with the largest lvb sequence number.
728 */
729
730 static void recover_lvb(struct dlm_rsb *r)
731 {
732 struct dlm_lkb *big_lkb = NULL, *iter, *high_lkb = NULL;
733 uint32_t high_seq = 0;
734 int lock_lvb_exists = 0;
735 int lvblen = r->res_ls->ls_lvblen;
736
737 if (!rsb_flag(r, RSB_NEW_MASTER2) &&
738 rsb_flag(r, RSB_RECOVER_LVB_INVAL)) {
739 /* case 1 above */
740 rsb_set_flag(r, RSB_VALNOTVALID);
741 return;
742 }
743
744 if (!rsb_flag(r, RSB_NEW_MASTER2))
745 return;
746
747 /* we are the new master, so figure out if VALNOTVALID should
748 be set, and set the rsb lvb from the best lkb available. */
749
750 list_for_each_entry(iter, &r->res_grantqueue, lkb_statequeue) {
751 if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
752 continue;
753
754 lock_lvb_exists = 1;
755
756 if (iter->lkb_grmode > DLM_LOCK_CR) {
757 big_lkb = iter;
758 goto setflag;
759 }
760
761 if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
762 high_lkb = iter;
763 high_seq = iter->lkb_lvbseq;
764 }
765 }
766
767 list_for_each_entry(iter, &r->res_convertqueue, lkb_statequeue) {
768 if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
769 continue;
770
771 lock_lvb_exists = 1;
772
773 if (iter->lkb_grmode > DLM_LOCK_CR) {
774 big_lkb = iter;
775 goto setflag;
776 }
777
778 if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
779 high_lkb = iter;
780 high_seq = iter->lkb_lvbseq;
781 }
782 }
783
784 setflag:
785 if (!lock_lvb_exists)
786 goto out;
787
788 /* lvb is invalidated if only NL/CR locks remain */
789 if (!big_lkb)
790 rsb_set_flag(r, RSB_VALNOTVALID);
791
792 if (!r->res_lvbptr) {
793 r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
794 if (!r->res_lvbptr)
795 goto out;
796 }
797
798 if (big_lkb) {
799 r->res_lvbseq = big_lkb->lkb_lvbseq;
800 memcpy(r->res_lvbptr, big_lkb->lkb_lvbptr, lvblen);
801 } else if (high_lkb) {
802 r->res_lvbseq = high_lkb->lkb_lvbseq;
803 memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
804 } else {
805 r->res_lvbseq = 0;
806 memset(r->res_lvbptr, 0, lvblen);
807 }
808 out:
809 return;
810 }
811
812 /* All master rsb's flagged RECOVER_CONVERT need to be looked at. The locks
813 converting PR->CW or CW->PR need to have their lkb_grmode set. */
814
815 static void recover_conversion(struct dlm_rsb *r)
816 {
817 struct dlm_ls *ls = r->res_ls;
818 struct dlm_lkb *lkb;
819 int grmode = -1;
820
821 list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
822 if (lkb->lkb_grmode == DLM_LOCK_PR ||
823 lkb->lkb_grmode == DLM_LOCK_CW) {
824 grmode = lkb->lkb_grmode;
825 break;
826 }
827 }
828
829 list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
830 if (lkb->lkb_grmode != DLM_LOCK_IV)
831 continue;
832 if (grmode == -1) {
833 log_debug(ls, "recover_conversion %x set gr to rq %d",
834 lkb->lkb_id, lkb->lkb_rqmode);
835 lkb->lkb_grmode = lkb->lkb_rqmode;
836 } else {
837 log_debug(ls, "recover_conversion %x set gr %d",
838 lkb->lkb_id, grmode);
839 lkb->lkb_grmode = grmode;
840 }
841 }
842 }
843
844 /* We've become the new master for this rsb and waiting/converting locks may
845 need to be granted in dlm_recover_grant() due to locks that may have
846 existed from a removed node. */
847
848 static void recover_grant(struct dlm_rsb *r)
849 {
850 if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
851 rsb_set_flag(r, RSB_RECOVER_GRANT);
852 }
853
854 void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list)
855 {
856 struct dlm_rsb *r;
857 unsigned int count = 0;
858
859 list_for_each_entry(r, root_list, res_root_list) {
860 lock_rsb(r);
861 if (is_master(r)) {
862 if (rsb_flag(r, RSB_RECOVER_CONVERT))
863 recover_conversion(r);
864
865 /* recover lvb before granting locks so the updated
866 lvb/VALNOTVALID is presented in the completion */
867 recover_lvb(r);
868
869 if (rsb_flag(r, RSB_NEW_MASTER2))
870 recover_grant(r);
871 count++;
872 } else {
873 rsb_clear_flag(r, RSB_VALNOTVALID);
874 }
875 rsb_clear_flag(r, RSB_RECOVER_CONVERT);
876 rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
877 rsb_clear_flag(r, RSB_NEW_MASTER2);
878 unlock_rsb(r);
879 }
880
881 if (count)
882 log_rinfo(ls, "dlm_recover_rsbs %d done", count);
883 }
884
885 void dlm_clear_inactive(struct dlm_ls *ls)
886 {
887 struct dlm_rsb *r, *safe;
888 unsigned int count = 0;
889
890 write_lock_bh(&ls->ls_rsbtbl_lock);
891 list_for_each_entry_safe(r, safe, &ls->ls_slow_inactive, res_slow_list) {
892 list_del(&r->res_slow_list);
893 rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
894 dlm_rhash_rsb_params);
895
896 if (!list_empty(&r->res_scan_list))
897 list_del_init(&r->res_scan_list);
898
899 free_inactive_rsb(r);
900 count++;
901 }
902 write_unlock_bh(&ls->ls_rsbtbl_lock);
903
904 if (count)
905 log_rinfo(ls, "dlm_clear_inactive %u done", count);
906 }
907
908
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.