xref: /linux/fs/dlm/recover.c (revision 3630400697a3d334a391c1dba1b601d852145f2c)
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 (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 
502 static 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 
528 int 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 
567 int 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 
618 static 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 
634 static 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 
661 int 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 
697 void 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 
731 static 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 need to have their lkb_grmode set. */
815 
816 static void recover_conversion(struct dlm_rsb *r)
817 {
818 	struct dlm_ls *ls = r->res_ls;
819 	struct dlm_lkb *lkb;
820 	int grmode = -1;
821 
822 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
823 		if (lkb->lkb_grmode == DLM_LOCK_PR ||
824 		    lkb->lkb_grmode == DLM_LOCK_CW) {
825 			grmode = lkb->lkb_grmode;
826 			break;
827 		}
828 	}
829 
830 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
831 		if (lkb->lkb_grmode != DLM_LOCK_IV)
832 			continue;
833 		if (grmode == -1) {
834 			log_debug(ls, "recover_conversion %x set gr to rq %d",
835 				  lkb->lkb_id, lkb->lkb_rqmode);
836 			lkb->lkb_grmode = lkb->lkb_rqmode;
837 		} else {
838 			log_debug(ls, "recover_conversion %x set gr %d",
839 				  lkb->lkb_id, grmode);
840 			lkb->lkb_grmode = grmode;
841 		}
842 	}
843 }
844 
845 /* We've become the new master for this rsb and waiting/converting locks may
846    need to be granted in dlm_recover_grant() due to locks that may have
847    existed from a removed node. */
848 
849 static void recover_grant(struct dlm_rsb *r)
850 {
851 	if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
852 		rsb_set_flag(r, RSB_RECOVER_GRANT);
853 }
854 
855 void dlm_recover_rsbs(struct dlm_ls *ls, const struct list_head *root_list)
856 {
857 	struct dlm_rsb *r;
858 	unsigned int count = 0;
859 
860 	list_for_each_entry(r, root_list, res_root_list) {
861 		lock_rsb(r);
862 		if (r->res_nodeid != -1 && is_master(r)) {
863 			if (rsb_flag(r, RSB_RECOVER_CONVERT))
864 				recover_conversion(r);
865 
866 			/* recover lvb before granting locks so the updated
867 			   lvb/VALNOTVALID is presented in the completion */
868 			recover_lvb(r);
869 
870 			if (rsb_flag(r, RSB_NEW_MASTER2))
871 				recover_grant(r);
872 			count++;
873 		} else {
874 			rsb_clear_flag(r, RSB_VALNOTVALID);
875 		}
876 		rsb_clear_flag(r, RSB_RECOVER_CONVERT);
877 		rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
878 		rsb_clear_flag(r, RSB_NEW_MASTER2);
879 		unlock_rsb(r);
880 	}
881 
882 	if (count)
883 		log_rinfo(ls, "dlm_recover_rsbs %d done", count);
884 }
885 
886 void dlm_clear_inactive(struct dlm_ls *ls)
887 {
888 	struct dlm_rsb *r, *safe;
889 	unsigned int count = 0;
890 
891 	write_lock_bh(&ls->ls_rsbtbl_lock);
892 	list_for_each_entry_safe(r, safe, &ls->ls_slow_inactive, res_slow_list) {
893 		list_del(&r->res_slow_list);
894 		rhashtable_remove_fast(&ls->ls_rsbtbl, &r->res_node,
895 				       dlm_rhash_rsb_params);
896 
897 		if (!list_empty(&r->res_scan_list))
898 			list_del_init(&r->res_scan_list);
899 
900 		free_inactive_rsb(r);
901 		count++;
902 	}
903 	write_unlock_bh(&ls->ls_rsbtbl_lock);
904 
905 	if (count)
906 		log_rinfo(ls, "dlm_clear_inactive %u done", count);
907 }
908 
909