xref: /openbmc/linux/fs/dlm/recover.c (revision dea54fba)
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 
44 int 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 
72 uint32_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 
81 static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
82 {
83 	ls->ls_recover_status |= status;
84 }
85 
86 void 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 
93 static 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 
126 static 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 
152 static 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 
167 int 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 
210 int dlm_recover_directory_wait(struct dlm_ls *ls)
211 {
212 	return wait_status(ls, DLM_RS_DIR);
213 }
214 
215 int dlm_recover_locks_wait(struct dlm_ls *ls)
216 {
217 	return wait_status(ls, DLM_RS_LOCKS);
218 }
219 
220 int 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 
237 static 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 
248 static 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 
261 static 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 
273 static 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 
293 static 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 
305 static 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;
324 out_unlock:
325 	spin_unlock(&ls->ls_recover_idr_lock);
326 	idr_preload_end();
327 	return rv;
328 }
329 
330 static 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 
343 static 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 
353 static 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 
399 static 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 
411 static 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 
425 static 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 
442 static 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 
495 static 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 
521 int 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 
564 int 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 
615 static 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 
630 static 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 
657 int 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 
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 *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 
817 static 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 
850 static 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 
856 void 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 
891 int 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 
921 void 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 
933 void 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 
956