xref: /linux/fs/dlm/recover.c (revision be709d48329a500621d2a05835283150ae137b45)
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 		if (test_bit(LSFL_RCOM_WAIT, &ls->ls_flags)) {
56 			log_debug(ls, "dlm_wait_function timed out");
57 			return -ETIMEDOUT;
58 		}
59 	}
60 
61 	if (dlm_recovery_stopped(ls)) {
62 		log_debug(ls, "dlm_wait_function aborted");
63 		error = -EINTR;
64 	}
65 	return error;
66 }
67 
68 /*
69  * An efficient way for all nodes to wait for all others to have a certain
70  * status.  The node with the lowest nodeid polls all the others for their
71  * status (wait_status_all) and all the others poll the node with the low id
72  * for its accumulated result (wait_status_low).  When all nodes have set
73  * status flag X, then status flag X_ALL will be set on the low nodeid.
74  */
75 
76 uint32_t dlm_recover_status(struct dlm_ls *ls)
77 {
78 	uint32_t status;
79 	spin_lock(&ls->ls_recover_lock);
80 	status = ls->ls_recover_status;
81 	spin_unlock(&ls->ls_recover_lock);
82 	return status;
83 }
84 
85 static void _set_recover_status(struct dlm_ls *ls, uint32_t status)
86 {
87 	ls->ls_recover_status |= status;
88 }
89 
90 void dlm_set_recover_status(struct dlm_ls *ls, uint32_t status)
91 {
92 	spin_lock(&ls->ls_recover_lock);
93 	_set_recover_status(ls, status);
94 	spin_unlock(&ls->ls_recover_lock);
95 }
96 
97 static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
98 			   int save_slots)
99 {
100 	struct dlm_rcom *rc = ls->ls_recover_buf;
101 	struct dlm_member *memb;
102 	int error = 0, delay;
103 
104 	list_for_each_entry(memb, &ls->ls_nodes, list) {
105 		delay = 0;
106 		for (;;) {
107 			if (dlm_recovery_stopped(ls)) {
108 				error = -EINTR;
109 				goto out;
110 			}
111 
112 			error = dlm_rcom_status(ls, memb->nodeid, 0);
113 			if (error)
114 				goto out;
115 
116 			if (save_slots)
117 				dlm_slot_save(ls, rc, memb);
118 
119 			if (rc->rc_result & wait_status)
120 				break;
121 			if (delay < 1000)
122 				delay += 20;
123 			msleep(delay);
124 		}
125 	}
126  out:
127 	return error;
128 }
129 
130 static int wait_status_low(struct dlm_ls *ls, uint32_t wait_status,
131 			   uint32_t status_flags)
132 {
133 	struct dlm_rcom *rc = ls->ls_recover_buf;
134 	int error = 0, delay = 0, nodeid = ls->ls_low_nodeid;
135 
136 	for (;;) {
137 		if (dlm_recovery_stopped(ls)) {
138 			error = -EINTR;
139 			goto out;
140 		}
141 
142 		error = dlm_rcom_status(ls, nodeid, status_flags);
143 		if (error)
144 			break;
145 
146 		if (rc->rc_result & wait_status)
147 			break;
148 		if (delay < 1000)
149 			delay += 20;
150 		msleep(delay);
151 	}
152  out:
153 	return error;
154 }
155 
156 static int wait_status(struct dlm_ls *ls, uint32_t status)
157 {
158 	uint32_t status_all = status << 1;
159 	int error;
160 
161 	if (ls->ls_low_nodeid == dlm_our_nodeid()) {
162 		error = wait_status_all(ls, status, 0);
163 		if (!error)
164 			dlm_set_recover_status(ls, status_all);
165 	} else
166 		error = wait_status_low(ls, status_all, 0);
167 
168 	return error;
169 }
170 
171 int dlm_recover_members_wait(struct dlm_ls *ls)
172 {
173 	struct dlm_member *memb;
174 	struct dlm_slot *slots;
175 	int num_slots, slots_size;
176 	int error, rv;
177 	uint32_t gen;
178 
179 	list_for_each_entry(memb, &ls->ls_nodes, list) {
180 		memb->slot = -1;
181 		memb->generation = 0;
182 	}
183 
184 	if (ls->ls_low_nodeid == dlm_our_nodeid()) {
185 		error = wait_status_all(ls, DLM_RS_NODES, 1);
186 		if (error)
187 			goto out;
188 
189 		/* slots array is sparse, slots_size may be > num_slots */
190 
191 		rv = dlm_slots_assign(ls, &num_slots, &slots_size, &slots, &gen);
192 		if (!rv) {
193 			spin_lock(&ls->ls_recover_lock);
194 			_set_recover_status(ls, DLM_RS_NODES_ALL);
195 			ls->ls_num_slots = num_slots;
196 			ls->ls_slots_size = slots_size;
197 			ls->ls_slots = slots;
198 			ls->ls_generation = gen;
199 			spin_unlock(&ls->ls_recover_lock);
200 		} else {
201 			dlm_set_recover_status(ls, DLM_RS_NODES_ALL);
202 		}
203 	} else {
204 		error = wait_status_low(ls, DLM_RS_NODES_ALL, DLM_RSF_NEED_SLOTS);
205 		if (error)
206 			goto out;
207 
208 		dlm_slots_copy_in(ls);
209 	}
210  out:
211 	return error;
212 }
213 
214 int dlm_recover_directory_wait(struct dlm_ls *ls)
215 {
216 	return wait_status(ls, DLM_RS_DIR);
217 }
218 
219 int dlm_recover_locks_wait(struct dlm_ls *ls)
220 {
221 	return wait_status(ls, DLM_RS_LOCKS);
222 }
223 
224 int dlm_recover_done_wait(struct dlm_ls *ls)
225 {
226 	return wait_status(ls, DLM_RS_DONE);
227 }
228 
229 /*
230  * The recover_list contains all the rsb's for which we've requested the new
231  * master nodeid.  As replies are returned from the resource directories the
232  * rsb's are removed from the list.  When the list is empty we're done.
233  *
234  * The recover_list is later similarly used for all rsb's for which we've sent
235  * new lkb's and need to receive new corresponding lkid's.
236  *
237  * We use the address of the rsb struct as a simple local identifier for the
238  * rsb so we can match an rcom reply with the rsb it was sent for.
239  */
240 
241 static int recover_list_empty(struct dlm_ls *ls)
242 {
243 	int empty;
244 
245 	spin_lock(&ls->ls_recover_list_lock);
246 	empty = list_empty(&ls->ls_recover_list);
247 	spin_unlock(&ls->ls_recover_list_lock);
248 
249 	return empty;
250 }
251 
252 static void recover_list_add(struct dlm_rsb *r)
253 {
254 	struct dlm_ls *ls = r->res_ls;
255 
256 	spin_lock(&ls->ls_recover_list_lock);
257 	if (list_empty(&r->res_recover_list)) {
258 		list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
259 		ls->ls_recover_list_count++;
260 		dlm_hold_rsb(r);
261 	}
262 	spin_unlock(&ls->ls_recover_list_lock);
263 }
264 
265 static void recover_list_del(struct dlm_rsb *r)
266 {
267 	struct dlm_ls *ls = r->res_ls;
268 
269 	spin_lock(&ls->ls_recover_list_lock);
270 	list_del_init(&r->res_recover_list);
271 	ls->ls_recover_list_count--;
272 	spin_unlock(&ls->ls_recover_list_lock);
273 
274 	dlm_put_rsb(r);
275 }
276 
277 static void recover_list_clear(struct dlm_ls *ls)
278 {
279 	struct dlm_rsb *r, *s;
280 
281 	spin_lock(&ls->ls_recover_list_lock);
282 	list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
283 		list_del_init(&r->res_recover_list);
284 		r->res_recover_locks_count = 0;
285 		dlm_put_rsb(r);
286 		ls->ls_recover_list_count--;
287 	}
288 
289 	if (ls->ls_recover_list_count != 0) {
290 		log_error(ls, "warning: recover_list_count %d",
291 			  ls->ls_recover_list_count);
292 		ls->ls_recover_list_count = 0;
293 	}
294 	spin_unlock(&ls->ls_recover_list_lock);
295 }
296 
297 static int recover_idr_empty(struct dlm_ls *ls)
298 {
299 	int empty = 1;
300 
301 	spin_lock(&ls->ls_recover_idr_lock);
302 	if (ls->ls_recover_list_count)
303 		empty = 0;
304 	spin_unlock(&ls->ls_recover_idr_lock);
305 
306 	return empty;
307 }
308 
309 static int recover_idr_add(struct dlm_rsb *r)
310 {
311 	struct dlm_ls *ls = r->res_ls;
312 	int rv;
313 
314 	idr_preload(GFP_NOFS);
315 	spin_lock(&ls->ls_recover_idr_lock);
316 	if (r->res_id) {
317 		rv = -1;
318 		goto out_unlock;
319 	}
320 	rv = idr_alloc(&ls->ls_recover_idr, r, 1, 0, GFP_NOWAIT);
321 	if (rv < 0)
322 		goto out_unlock;
323 
324 	r->res_id = rv;
325 	ls->ls_recover_list_count++;
326 	dlm_hold_rsb(r);
327 	rv = 0;
328 out_unlock:
329 	spin_unlock(&ls->ls_recover_idr_lock);
330 	idr_preload_end();
331 	return rv;
332 }
333 
334 static void recover_idr_del(struct dlm_rsb *r)
335 {
336 	struct dlm_ls *ls = r->res_ls;
337 
338 	spin_lock(&ls->ls_recover_idr_lock);
339 	idr_remove(&ls->ls_recover_idr, r->res_id);
340 	r->res_id = 0;
341 	ls->ls_recover_list_count--;
342 	spin_unlock(&ls->ls_recover_idr_lock);
343 
344 	dlm_put_rsb(r);
345 }
346 
347 static struct dlm_rsb *recover_idr_find(struct dlm_ls *ls, uint64_t id)
348 {
349 	struct dlm_rsb *r;
350 
351 	spin_lock(&ls->ls_recover_idr_lock);
352 	r = idr_find(&ls->ls_recover_idr, (int)id);
353 	spin_unlock(&ls->ls_recover_idr_lock);
354 	return r;
355 }
356 
357 static void recover_idr_clear(struct dlm_ls *ls)
358 {
359 	struct dlm_rsb *r;
360 	int id;
361 
362 	spin_lock(&ls->ls_recover_idr_lock);
363 
364 	idr_for_each_entry(&ls->ls_recover_idr, r, id) {
365 		idr_remove(&ls->ls_recover_idr, id);
366 		r->res_id = 0;
367 		r->res_recover_locks_count = 0;
368 		ls->ls_recover_list_count--;
369 
370 		dlm_put_rsb(r);
371 	}
372 
373 	if (ls->ls_recover_list_count != 0) {
374 		log_error(ls, "warning: recover_list_count %d",
375 			  ls->ls_recover_list_count);
376 		ls->ls_recover_list_count = 0;
377 	}
378 	spin_unlock(&ls->ls_recover_idr_lock);
379 }
380 
381 
382 /* Master recovery: find new master node for rsb's that were
383    mastered on nodes that have been removed.
384 
385    dlm_recover_masters
386    recover_master
387    dlm_send_rcom_lookup            ->  receive_rcom_lookup
388                                        dlm_dir_lookup
389    receive_rcom_lookup_reply       <-
390    dlm_recover_master_reply
391    set_new_master
392    set_master_lkbs
393    set_lock_master
394 */
395 
396 /*
397  * Set the lock master for all LKBs in a lock queue
398  * If we are the new master of the rsb, we may have received new
399  * MSTCPY locks from other nodes already which we need to ignore
400  * when setting the new nodeid.
401  */
402 
403 static void set_lock_master(struct list_head *queue, int nodeid)
404 {
405 	struct dlm_lkb *lkb;
406 
407 	list_for_each_entry(lkb, queue, lkb_statequeue) {
408 		if (!(lkb->lkb_flags & DLM_IFL_MSTCPY)) {
409 			lkb->lkb_nodeid = nodeid;
410 			lkb->lkb_remid = 0;
411 		}
412 	}
413 }
414 
415 static void set_master_lkbs(struct dlm_rsb *r)
416 {
417 	set_lock_master(&r->res_grantqueue, r->res_nodeid);
418 	set_lock_master(&r->res_convertqueue, r->res_nodeid);
419 	set_lock_master(&r->res_waitqueue, r->res_nodeid);
420 }
421 
422 /*
423  * Propagate the new master nodeid to locks
424  * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
425  * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which
426  * rsb's to consider.
427  */
428 
429 static void set_new_master(struct dlm_rsb *r)
430 {
431 	set_master_lkbs(r);
432 	rsb_set_flag(r, RSB_NEW_MASTER);
433 	rsb_set_flag(r, RSB_NEW_MASTER2);
434 }
435 
436 /*
437  * We do async lookups on rsb's that need new masters.  The rsb's
438  * waiting for a lookup reply are kept on the recover_list.
439  *
440  * Another node recovering the master may have sent us a rcom lookup,
441  * and our dlm_master_lookup() set it as the new master, along with
442  * NEW_MASTER so that we'll recover it here (this implies dir_nodeid
443  * equals our_nodeid below).
444  */
445 
446 static int recover_master(struct dlm_rsb *r, unsigned int *count)
447 {
448 	struct dlm_ls *ls = r->res_ls;
449 	int our_nodeid, dir_nodeid;
450 	int is_removed = 0;
451 	int error;
452 
453 	if (is_master(r))
454 		return 0;
455 
456 	is_removed = dlm_is_removed(ls, r->res_nodeid);
457 
458 	if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
459 		return 0;
460 
461 	our_nodeid = dlm_our_nodeid();
462 	dir_nodeid = dlm_dir_nodeid(r);
463 
464 	if (dir_nodeid == our_nodeid) {
465 		if (is_removed) {
466 			r->res_master_nodeid = our_nodeid;
467 			r->res_nodeid = 0;
468 		}
469 
470 		/* set master of lkbs to ourself when is_removed, or to
471 		   another new master which we set along with NEW_MASTER
472 		   in dlm_master_lookup */
473 		set_new_master(r);
474 		error = 0;
475 	} else {
476 		recover_idr_add(r);
477 		error = dlm_send_rcom_lookup(r, dir_nodeid);
478 	}
479 
480 	(*count)++;
481 	return error;
482 }
483 
484 /*
485  * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
486  * This is necessary because recovery can be started, aborted and restarted,
487  * causing the master nodeid to briefly change during the aborted recovery, and
488  * change back to the original value in the second recovery.  The MSTCPY locks
489  * may or may not have been purged during the aborted recovery.  Another node
490  * with an outstanding request in waiters list and a request reply saved in the
491  * requestqueue, cannot know whether it should ignore the reply and resend the
492  * request, or accept the reply and complete the request.  It must do the
493  * former if the remote node purged MSTCPY locks, and it must do the later if
494  * the remote node did not.  This is solved by always purging MSTCPY locks, in
495  * which case, the request reply would always be ignored and the request
496  * resent.
497  */
498 
499 static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
500 {
501 	int dir_nodeid = dlm_dir_nodeid(r);
502 	int new_master = dir_nodeid;
503 
504 	if (dir_nodeid == dlm_our_nodeid())
505 		new_master = 0;
506 
507 	dlm_purge_mstcpy_locks(r);
508 	r->res_master_nodeid = dir_nodeid;
509 	r->res_nodeid = new_master;
510 	set_new_master(r);
511 	(*count)++;
512 	return 0;
513 }
514 
515 /*
516  * Go through local root resources and for each rsb which has a master which
517  * has departed, get the new master nodeid from the directory.  The dir will
518  * assign mastery to the first node to look up the new master.  That means
519  * we'll discover in this lookup if we're the new master of any rsb's.
520  *
521  * We fire off all the dir lookup requests individually and asynchronously to
522  * the correct dir node.
523  */
524 
525 int dlm_recover_masters(struct dlm_ls *ls)
526 {
527 	struct dlm_rsb *r;
528 	unsigned int total = 0;
529 	unsigned int count = 0;
530 	int nodir = dlm_no_directory(ls);
531 	int error;
532 
533 	log_rinfo(ls, "dlm_recover_masters");
534 
535 	down_read(&ls->ls_root_sem);
536 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
537 		if (dlm_recovery_stopped(ls)) {
538 			up_read(&ls->ls_root_sem);
539 			error = -EINTR;
540 			goto out;
541 		}
542 
543 		lock_rsb(r);
544 		if (nodir)
545 			error = recover_master_static(r, &count);
546 		else
547 			error = recover_master(r, &count);
548 		unlock_rsb(r);
549 		cond_resched();
550 		total++;
551 
552 		if (error) {
553 			up_read(&ls->ls_root_sem);
554 			goto out;
555 		}
556 	}
557 	up_read(&ls->ls_root_sem);
558 
559 	log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
560 
561 	error = dlm_wait_function(ls, &recover_idr_empty);
562  out:
563 	if (error)
564 		recover_idr_clear(ls);
565 	return error;
566 }
567 
568 int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
569 {
570 	struct dlm_rsb *r;
571 	int ret_nodeid, new_master;
572 
573 	r = recover_idr_find(ls, rc->rc_id);
574 	if (!r) {
575 		log_error(ls, "dlm_recover_master_reply no id %llx",
576 			  (unsigned long long)rc->rc_id);
577 		goto out;
578 	}
579 
580 	ret_nodeid = rc->rc_result;
581 
582 	if (ret_nodeid == dlm_our_nodeid())
583 		new_master = 0;
584 	else
585 		new_master = ret_nodeid;
586 
587 	lock_rsb(r);
588 	r->res_master_nodeid = ret_nodeid;
589 	r->res_nodeid = new_master;
590 	set_new_master(r);
591 	unlock_rsb(r);
592 	recover_idr_del(r);
593 
594 	if (recover_idr_empty(ls))
595 		wake_up(&ls->ls_wait_general);
596  out:
597 	return 0;
598 }
599 
600 
601 /* Lock recovery: rebuild the process-copy locks we hold on a
602    remastered rsb on the new rsb master.
603 
604    dlm_recover_locks
605    recover_locks
606    recover_locks_queue
607    dlm_send_rcom_lock              ->  receive_rcom_lock
608                                        dlm_recover_master_copy
609    receive_rcom_lock_reply         <-
610    dlm_recover_process_copy
611 */
612 
613 
614 /*
615  * keep a count of the number of lkb's we send to the new master; when we get
616  * an equal number of replies then recovery for the rsb is done
617  */
618 
619 static int recover_locks_queue(struct dlm_rsb *r, struct list_head *head)
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);
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)
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);
643 	if (error)
644 		goto out;
645 	error = recover_locks_queue(r, &r->res_convertqueue);
646 	if (error)
647 		goto out;
648 	error = recover_locks_queue(r, &r->res_waitqueue);
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)
662 {
663 	struct dlm_rsb *r;
664 	int error, count = 0;
665 
666 	down_read(&ls->ls_root_sem);
667 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
668 		if (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 			up_read(&ls->ls_root_sem);
679 			goto out;
680 		}
681 
682 		error = recover_locks(r);
683 		if (error) {
684 			up_read(&ls->ls_root_sem);
685 			goto out;
686 		}
687 
688 		count += r->res_recover_locks_count;
689 	}
690 	up_read(&ls->ls_root_sem);
691 
692 	log_rinfo(ls, "dlm_recover_locks %d out", count);
693 
694 	error = dlm_wait_function(ls, &recover_list_empty);
695  out:
696 	if (error)
697 		recover_list_clear(ls);
698 	return error;
699 }
700 
701 void dlm_recovered_lock(struct dlm_rsb *r)
702 {
703 	DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
704 
705 	r->res_recover_locks_count--;
706 	if (!r->res_recover_locks_count) {
707 		rsb_clear_flag(r, RSB_NEW_MASTER);
708 		recover_list_del(r);
709 	}
710 
711 	if (recover_list_empty(r->res_ls))
712 		wake_up(&r->res_ls->ls_wait_general);
713 }
714 
715 /*
716  * The lvb needs to be recovered on all master rsb's.  This includes setting
717  * the VALNOTVALID flag if necessary, and determining the correct lvb contents
718  * based on the lvb's of the locks held on the rsb.
719  *
720  * RSB_VALNOTVALID is set in two cases:
721  *
722  * 1. we are master, but not new, and we purged an EX/PW lock held by a
723  * failed node (in dlm_recover_purge which set RSB_RECOVER_LVB_INVAL)
724  *
725  * 2. we are a new master, and there are only NL/CR locks left.
726  * (We could probably improve this by only invaliding in this way when
727  * the previous master left uncleanly.  VMS docs mention that.)
728  *
729  * The LVB contents are only considered for changing when this is a new master
730  * of the rsb (NEW_MASTER2).  Then, the rsb's lvb is taken from any lkb with
731  * mode > CR.  If no lkb's exist with mode above CR, the lvb contents are taken
732  * from the lkb with the largest lvb sequence number.
733  */
734 
735 static void recover_lvb(struct dlm_rsb *r)
736 {
737 	struct dlm_lkb *lkb, *high_lkb = NULL;
738 	uint32_t high_seq = 0;
739 	int lock_lvb_exists = 0;
740 	int big_lock_exists = 0;
741 	int lvblen = r->res_ls->ls_lvblen;
742 
743 	if (!rsb_flag(r, RSB_NEW_MASTER2) &&
744 	    rsb_flag(r, RSB_RECOVER_LVB_INVAL)) {
745 		/* case 1 above */
746 		rsb_set_flag(r, RSB_VALNOTVALID);
747 		return;
748 	}
749 
750 	if (!rsb_flag(r, RSB_NEW_MASTER2))
751 		return;
752 
753 	/* we are the new master, so figure out if VALNOTVALID should
754 	   be set, and set the rsb lvb from the best lkb available. */
755 
756 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
757 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
758 			continue;
759 
760 		lock_lvb_exists = 1;
761 
762 		if (lkb->lkb_grmode > DLM_LOCK_CR) {
763 			big_lock_exists = 1;
764 			goto setflag;
765 		}
766 
767 		if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
768 			high_lkb = lkb;
769 			high_seq = lkb->lkb_lvbseq;
770 		}
771 	}
772 
773 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
774 		if (!(lkb->lkb_exflags & DLM_LKF_VALBLK))
775 			continue;
776 
777 		lock_lvb_exists = 1;
778 
779 		if (lkb->lkb_grmode > DLM_LOCK_CR) {
780 			big_lock_exists = 1;
781 			goto setflag;
782 		}
783 
784 		if (((int)lkb->lkb_lvbseq - (int)high_seq) >= 0) {
785 			high_lkb = lkb;
786 			high_seq = lkb->lkb_lvbseq;
787 		}
788 	}
789 
790  setflag:
791 	if (!lock_lvb_exists)
792 		goto out;
793 
794 	/* lvb is invalidated if only NL/CR locks remain */
795 	if (!big_lock_exists)
796 		rsb_set_flag(r, RSB_VALNOTVALID);
797 
798 	if (!r->res_lvbptr) {
799 		r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
800 		if (!r->res_lvbptr)
801 			goto out;
802 	}
803 
804 	if (big_lock_exists) {
805 		r->res_lvbseq = lkb->lkb_lvbseq;
806 		memcpy(r->res_lvbptr, lkb->lkb_lvbptr, lvblen);
807 	} else if (high_lkb) {
808 		r->res_lvbseq = high_lkb->lkb_lvbseq;
809 		memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
810 	} else {
811 		r->res_lvbseq = 0;
812 		memset(r->res_lvbptr, 0, lvblen);
813 	}
814  out:
815 	return;
816 }
817 
818 /* All master rsb's flagged RECOVER_CONVERT need to be looked at.  The locks
819    converting PR->CW or CW->PR need to have their lkb_grmode set. */
820 
821 static void recover_conversion(struct dlm_rsb *r)
822 {
823 	struct dlm_ls *ls = r->res_ls;
824 	struct dlm_lkb *lkb;
825 	int grmode = -1;
826 
827 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
828 		if (lkb->lkb_grmode == DLM_LOCK_PR ||
829 		    lkb->lkb_grmode == DLM_LOCK_CW) {
830 			grmode = lkb->lkb_grmode;
831 			break;
832 		}
833 	}
834 
835 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
836 		if (lkb->lkb_grmode != DLM_LOCK_IV)
837 			continue;
838 		if (grmode == -1) {
839 			log_debug(ls, "recover_conversion %x set gr to rq %d",
840 				  lkb->lkb_id, lkb->lkb_rqmode);
841 			lkb->lkb_grmode = lkb->lkb_rqmode;
842 		} else {
843 			log_debug(ls, "recover_conversion %x set gr %d",
844 				  lkb->lkb_id, grmode);
845 			lkb->lkb_grmode = grmode;
846 		}
847 	}
848 }
849 
850 /* We've become the new master for this rsb and waiting/converting locks may
851    need to be granted in dlm_recover_grant() due to locks that may have
852    existed from a removed node. */
853 
854 static void recover_grant(struct dlm_rsb *r)
855 {
856 	if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
857 		rsb_set_flag(r, RSB_RECOVER_GRANT);
858 }
859 
860 void dlm_recover_rsbs(struct dlm_ls *ls)
861 {
862 	struct dlm_rsb *r;
863 	unsigned int count = 0;
864 
865 	down_read(&ls->ls_root_sem);
866 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
867 		lock_rsb(r);
868 		if (is_master(r)) {
869 			if (rsb_flag(r, RSB_RECOVER_CONVERT))
870 				recover_conversion(r);
871 
872 			/* recover lvb before granting locks so the updated
873 			   lvb/VALNOTVALID is presented in the completion */
874 			recover_lvb(r);
875 
876 			if (rsb_flag(r, RSB_NEW_MASTER2))
877 				recover_grant(r);
878 			count++;
879 		} else {
880 			rsb_clear_flag(r, RSB_VALNOTVALID);
881 		}
882 		rsb_clear_flag(r, RSB_RECOVER_CONVERT);
883 		rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
884 		rsb_clear_flag(r, RSB_NEW_MASTER2);
885 		unlock_rsb(r);
886 	}
887 	up_read(&ls->ls_root_sem);
888 
889 	if (count)
890 		log_rinfo(ls, "dlm_recover_rsbs %d done", count);
891 }
892 
893 /* Create a single list of all root rsb's to be used during recovery */
894 
895 int dlm_create_root_list(struct dlm_ls *ls)
896 {
897 	struct rb_node *n;
898 	struct dlm_rsb *r;
899 	int i, error = 0;
900 
901 	down_write(&ls->ls_root_sem);
902 	if (!list_empty(&ls->ls_root_list)) {
903 		log_error(ls, "root list not empty");
904 		error = -EINVAL;
905 		goto out;
906 	}
907 
908 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
909 		spin_lock(&ls->ls_rsbtbl[i].lock);
910 		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
911 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
912 			list_add(&r->res_root_list, &ls->ls_root_list);
913 			dlm_hold_rsb(r);
914 		}
915 
916 		if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss))
917 			log_error(ls, "dlm_create_root_list toss not empty");
918 		spin_unlock(&ls->ls_rsbtbl[i].lock);
919 	}
920  out:
921 	up_write(&ls->ls_root_sem);
922 	return error;
923 }
924 
925 void dlm_release_root_list(struct dlm_ls *ls)
926 {
927 	struct dlm_rsb *r, *safe;
928 
929 	down_write(&ls->ls_root_sem);
930 	list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
931 		list_del_init(&r->res_root_list);
932 		dlm_put_rsb(r);
933 	}
934 	up_write(&ls->ls_root_sem);
935 }
936 
937 void dlm_clear_toss(struct dlm_ls *ls)
938 {
939 	struct rb_node *n, *next;
940 	struct dlm_rsb *r;
941 	unsigned int count = 0;
942 	int i;
943 
944 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
945 		spin_lock(&ls->ls_rsbtbl[i].lock);
946 		for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) {
947 			next = rb_next(n);
948 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
949 			rb_erase(n, &ls->ls_rsbtbl[i].toss);
950 			dlm_free_rsb(r);
951 			count++;
952 		}
953 		spin_unlock(&ls->ls_rsbtbl[i].lock);
954 	}
955 
956 	if (count)
957 		log_rinfo(ls, "dlm_clear_toss %u done", count);
958 }
959 
960