xref: /openbmc/linux/fs/dlm/recover.c (revision ffcdf473)
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(&ls->ls_recover_lock);
78 	status = ls->ls_recover_status;
79 	spin_unlock(&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(&ls->ls_recover_lock);
91 	_set_recover_status(ls, status);
92 	spin_unlock(&ls->ls_recover_lock);
93 }
94 
95 static int wait_status_all(struct dlm_ls *ls, uint32_t wait_status,
96 			   int save_slots)
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);
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)
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);
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)
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);
161 		if (!error)
162 			dlm_set_recover_status(ls, status_all);
163 	} else
164 		error = wait_status_low(ls, status_all, 0);
165 
166 	return error;
167 }
168 
169 int dlm_recover_members_wait(struct dlm_ls *ls)
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);
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(&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(&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, DLM_RSF_NEED_SLOTS);
203 		if (error)
204 			goto out;
205 
206 		dlm_slots_copy_in(ls);
207 	}
208  out:
209 	return error;
210 }
211 
212 int dlm_recover_directory_wait(struct dlm_ls *ls)
213 {
214 	return wait_status(ls, DLM_RS_DIR);
215 }
216 
217 int dlm_recover_locks_wait(struct dlm_ls *ls)
218 {
219 	return wait_status(ls, DLM_RS_LOCKS);
220 }
221 
222 int dlm_recover_done_wait(struct dlm_ls *ls)
223 {
224 	return wait_status(ls, DLM_RS_DONE);
225 }
226 
227 /*
228  * The recover_list contains all the rsb's for which we've requested the new
229  * master nodeid.  As replies are returned from the resource directories the
230  * rsb's are removed from the list.  When the list is empty we're done.
231  *
232  * The recover_list is later similarly used for all rsb's for which we've sent
233  * new lkb's and need to receive new corresponding lkid's.
234  *
235  * We use the address of the rsb struct as a simple local identifier for the
236  * rsb so we can match an rcom reply with the rsb it was sent for.
237  */
238 
239 static int recover_list_empty(struct dlm_ls *ls)
240 {
241 	int empty;
242 
243 	spin_lock(&ls->ls_recover_list_lock);
244 	empty = list_empty(&ls->ls_recover_list);
245 	spin_unlock(&ls->ls_recover_list_lock);
246 
247 	return empty;
248 }
249 
250 static void recover_list_add(struct dlm_rsb *r)
251 {
252 	struct dlm_ls *ls = r->res_ls;
253 
254 	spin_lock(&ls->ls_recover_list_lock);
255 	if (list_empty(&r->res_recover_list)) {
256 		list_add_tail(&r->res_recover_list, &ls->ls_recover_list);
257 		ls->ls_recover_list_count++;
258 		dlm_hold_rsb(r);
259 	}
260 	spin_unlock(&ls->ls_recover_list_lock);
261 }
262 
263 static void recover_list_del(struct dlm_rsb *r)
264 {
265 	struct dlm_ls *ls = r->res_ls;
266 
267 	spin_lock(&ls->ls_recover_list_lock);
268 	list_del_init(&r->res_recover_list);
269 	ls->ls_recover_list_count--;
270 	spin_unlock(&ls->ls_recover_list_lock);
271 
272 	dlm_put_rsb(r);
273 }
274 
275 static void recover_list_clear(struct dlm_ls *ls)
276 {
277 	struct dlm_rsb *r, *s;
278 
279 	spin_lock(&ls->ls_recover_list_lock);
280 	list_for_each_entry_safe(r, s, &ls->ls_recover_list, res_recover_list) {
281 		list_del_init(&r->res_recover_list);
282 		r->res_recover_locks_count = 0;
283 		dlm_put_rsb(r);
284 		ls->ls_recover_list_count--;
285 	}
286 
287 	if (ls->ls_recover_list_count != 0) {
288 		log_error(ls, "warning: recover_list_count %d",
289 			  ls->ls_recover_list_count);
290 		ls->ls_recover_list_count = 0;
291 	}
292 	spin_unlock(&ls->ls_recover_list_lock);
293 }
294 
295 static int recover_idr_empty(struct dlm_ls *ls)
296 {
297 	int empty = 1;
298 
299 	spin_lock(&ls->ls_recover_idr_lock);
300 	if (ls->ls_recover_list_count)
301 		empty = 0;
302 	spin_unlock(&ls->ls_recover_idr_lock);
303 
304 	return empty;
305 }
306 
307 static int recover_idr_add(struct dlm_rsb *r)
308 {
309 	struct dlm_ls *ls = r->res_ls;
310 	int rv;
311 
312 	idr_preload(GFP_NOFS);
313 	spin_lock(&ls->ls_recover_idr_lock);
314 	if (r->res_id) {
315 		rv = -1;
316 		goto out_unlock;
317 	}
318 	rv = idr_alloc(&ls->ls_recover_idr, r, 1, 0, GFP_NOWAIT);
319 	if (rv < 0)
320 		goto out_unlock;
321 
322 	r->res_id = rv;
323 	ls->ls_recover_list_count++;
324 	dlm_hold_rsb(r);
325 	rv = 0;
326 out_unlock:
327 	spin_unlock(&ls->ls_recover_idr_lock);
328 	idr_preload_end();
329 	return rv;
330 }
331 
332 static void recover_idr_del(struct dlm_rsb *r)
333 {
334 	struct dlm_ls *ls = r->res_ls;
335 
336 	spin_lock(&ls->ls_recover_idr_lock);
337 	idr_remove(&ls->ls_recover_idr, r->res_id);
338 	r->res_id = 0;
339 	ls->ls_recover_list_count--;
340 	spin_unlock(&ls->ls_recover_idr_lock);
341 
342 	dlm_put_rsb(r);
343 }
344 
345 static struct dlm_rsb *recover_idr_find(struct dlm_ls *ls, uint64_t id)
346 {
347 	struct dlm_rsb *r;
348 
349 	spin_lock(&ls->ls_recover_idr_lock);
350 	r = idr_find(&ls->ls_recover_idr, (int)id);
351 	spin_unlock(&ls->ls_recover_idr_lock);
352 	return r;
353 }
354 
355 static void recover_idr_clear(struct dlm_ls *ls)
356 {
357 	struct dlm_rsb *r;
358 	int id;
359 
360 	spin_lock(&ls->ls_recover_idr_lock);
361 
362 	idr_for_each_entry(&ls->ls_recover_idr, r, id) {
363 		idr_remove(&ls->ls_recover_idr, id);
364 		r->res_id = 0;
365 		r->res_recover_locks_count = 0;
366 		ls->ls_recover_list_count--;
367 
368 		dlm_put_rsb(r);
369 	}
370 
371 	if (ls->ls_recover_list_count != 0) {
372 		log_error(ls, "warning: recover_list_count %d",
373 			  ls->ls_recover_list_count);
374 		ls->ls_recover_list_count = 0;
375 	}
376 	spin_unlock(&ls->ls_recover_idr_lock);
377 }
378 
379 
380 /* Master recovery: find new master node for rsb's that were
381    mastered on nodes that have been removed.
382 
383    dlm_recover_masters
384    recover_master
385    dlm_send_rcom_lookup            ->  receive_rcom_lookup
386                                        dlm_dir_lookup
387    receive_rcom_lookup_reply       <-
388    dlm_recover_master_reply
389    set_new_master
390    set_master_lkbs
391    set_lock_master
392 */
393 
394 /*
395  * Set the lock master for all LKBs in a lock queue
396  * If we are the new master of the rsb, we may have received new
397  * MSTCPY locks from other nodes already which we need to ignore
398  * when setting the new nodeid.
399  */
400 
401 static void set_lock_master(struct list_head *queue, int nodeid)
402 {
403 	struct dlm_lkb *lkb;
404 
405 	list_for_each_entry(lkb, queue, lkb_statequeue) {
406 		if (!test_bit(DLM_IFL_MSTCPY_BIT, &lkb->lkb_iflags)) {
407 			lkb->lkb_nodeid = nodeid;
408 			lkb->lkb_remid = 0;
409 		}
410 	}
411 }
412 
413 static void set_master_lkbs(struct dlm_rsb *r)
414 {
415 	set_lock_master(&r->res_grantqueue, r->res_nodeid);
416 	set_lock_master(&r->res_convertqueue, r->res_nodeid);
417 	set_lock_master(&r->res_waitqueue, r->res_nodeid);
418 }
419 
420 /*
421  * Propagate the new master nodeid to locks
422  * The NEW_MASTER flag tells dlm_recover_locks() which rsb's to consider.
423  * The NEW_MASTER2 flag tells recover_lvb() and recover_grant() which
424  * rsb's to consider.
425  */
426 
427 static void set_new_master(struct dlm_rsb *r)
428 {
429 	set_master_lkbs(r);
430 	rsb_set_flag(r, RSB_NEW_MASTER);
431 	rsb_set_flag(r, RSB_NEW_MASTER2);
432 }
433 
434 /*
435  * We do async lookups on rsb's that need new masters.  The rsb's
436  * waiting for a lookup reply are kept on the recover_list.
437  *
438  * Another node recovering the master may have sent us a rcom lookup,
439  * and our dlm_master_lookup() set it as the new master, along with
440  * NEW_MASTER so that we'll recover it here (this implies dir_nodeid
441  * equals our_nodeid below).
442  */
443 
444 static int recover_master(struct dlm_rsb *r, unsigned int *count)
445 {
446 	struct dlm_ls *ls = r->res_ls;
447 	int our_nodeid, dir_nodeid;
448 	int is_removed = 0;
449 	int error;
450 
451 	if (is_master(r))
452 		return 0;
453 
454 	is_removed = dlm_is_removed(ls, r->res_nodeid);
455 
456 	if (!is_removed && !rsb_flag(r, RSB_NEW_MASTER))
457 		return 0;
458 
459 	our_nodeid = dlm_our_nodeid();
460 	dir_nodeid = dlm_dir_nodeid(r);
461 
462 	if (dir_nodeid == our_nodeid) {
463 		if (is_removed) {
464 			r->res_master_nodeid = our_nodeid;
465 			r->res_nodeid = 0;
466 		}
467 
468 		/* set master of lkbs to ourself when is_removed, or to
469 		   another new master which we set along with NEW_MASTER
470 		   in dlm_master_lookup */
471 		set_new_master(r);
472 		error = 0;
473 	} else {
474 		recover_idr_add(r);
475 		error = dlm_send_rcom_lookup(r, dir_nodeid);
476 	}
477 
478 	(*count)++;
479 	return error;
480 }
481 
482 /*
483  * All MSTCPY locks are purged and rebuilt, even if the master stayed the same.
484  * This is necessary because recovery can be started, aborted and restarted,
485  * causing the master nodeid to briefly change during the aborted recovery, and
486  * change back to the original value in the second recovery.  The MSTCPY locks
487  * may or may not have been purged during the aborted recovery.  Another node
488  * with an outstanding request in waiters list and a request reply saved in the
489  * requestqueue, cannot know whether it should ignore the reply and resend the
490  * request, or accept the reply and complete the request.  It must do the
491  * former if the remote node purged MSTCPY locks, and it must do the later if
492  * the remote node did not.  This is solved by always purging MSTCPY locks, in
493  * which case, the request reply would always be ignored and the request
494  * resent.
495  */
496 
497 static int recover_master_static(struct dlm_rsb *r, unsigned int *count)
498 {
499 	int dir_nodeid = dlm_dir_nodeid(r);
500 	int new_master = dir_nodeid;
501 
502 	if (dir_nodeid == dlm_our_nodeid())
503 		new_master = 0;
504 
505 	dlm_purge_mstcpy_locks(r);
506 	r->res_master_nodeid = dir_nodeid;
507 	r->res_nodeid = new_master;
508 	set_new_master(r);
509 	(*count)++;
510 	return 0;
511 }
512 
513 /*
514  * Go through local root resources and for each rsb which has a master which
515  * has departed, get the new master nodeid from the directory.  The dir will
516  * assign mastery to the first node to look up the new master.  That means
517  * we'll discover in this lookup if we're the new master of any rsb's.
518  *
519  * We fire off all the dir lookup requests individually and asynchronously to
520  * the correct dir node.
521  */
522 
523 int dlm_recover_masters(struct dlm_ls *ls)
524 {
525 	struct dlm_rsb *r;
526 	unsigned int total = 0;
527 	unsigned int count = 0;
528 	int nodir = dlm_no_directory(ls);
529 	int error;
530 
531 	log_rinfo(ls, "dlm_recover_masters");
532 
533 	down_read(&ls->ls_root_sem);
534 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
535 		if (dlm_recovery_stopped(ls)) {
536 			up_read(&ls->ls_root_sem);
537 			error = -EINTR;
538 			goto out;
539 		}
540 
541 		lock_rsb(r);
542 		if (nodir)
543 			error = recover_master_static(r, &count);
544 		else
545 			error = recover_master(r, &count);
546 		unlock_rsb(r);
547 		cond_resched();
548 		total++;
549 
550 		if (error) {
551 			up_read(&ls->ls_root_sem);
552 			goto out;
553 		}
554 	}
555 	up_read(&ls->ls_root_sem);
556 
557 	log_rinfo(ls, "dlm_recover_masters %u of %u", count, total);
558 
559 	error = dlm_wait_function(ls, &recover_idr_empty);
560  out:
561 	if (error)
562 		recover_idr_clear(ls);
563 	return error;
564 }
565 
566 int dlm_recover_master_reply(struct dlm_ls *ls, struct dlm_rcom *rc)
567 {
568 	struct dlm_rsb *r;
569 	int ret_nodeid, new_master;
570 
571 	r = recover_idr_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_idr_del(r);
591 
592 	if (recover_idr_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 {
619 	struct dlm_lkb *lkb;
620 	int error = 0;
621 
622 	list_for_each_entry(lkb, head, lkb_statequeue) {
623 	   	error = dlm_send_rcom_lock(r, lkb);
624 		if (error)
625 			break;
626 		r->res_recover_locks_count++;
627 	}
628 
629 	return error;
630 }
631 
632 static int recover_locks(struct dlm_rsb *r)
633 {
634 	int error = 0;
635 
636 	lock_rsb(r);
637 
638 	DLM_ASSERT(!r->res_recover_locks_count, dlm_dump_rsb(r););
639 
640 	error = recover_locks_queue(r, &r->res_grantqueue);
641 	if (error)
642 		goto out;
643 	error = recover_locks_queue(r, &r->res_convertqueue);
644 	if (error)
645 		goto out;
646 	error = recover_locks_queue(r, &r->res_waitqueue);
647 	if (error)
648 		goto out;
649 
650 	if (r->res_recover_locks_count)
651 		recover_list_add(r);
652 	else
653 		rsb_clear_flag(r, RSB_NEW_MASTER);
654  out:
655 	unlock_rsb(r);
656 	return error;
657 }
658 
659 int dlm_recover_locks(struct dlm_ls *ls)
660 {
661 	struct dlm_rsb *r;
662 	int error, count = 0;
663 
664 	down_read(&ls->ls_root_sem);
665 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
666 		if (is_master(r)) {
667 			rsb_clear_flag(r, RSB_NEW_MASTER);
668 			continue;
669 		}
670 
671 		if (!rsb_flag(r, RSB_NEW_MASTER))
672 			continue;
673 
674 		if (dlm_recovery_stopped(ls)) {
675 			error = -EINTR;
676 			up_read(&ls->ls_root_sem);
677 			goto out;
678 		}
679 
680 		error = recover_locks(r);
681 		if (error) {
682 			up_read(&ls->ls_root_sem);
683 			goto out;
684 		}
685 
686 		count += r->res_recover_locks_count;
687 	}
688 	up_read(&ls->ls_root_sem);
689 
690 	log_rinfo(ls, "dlm_recover_locks %d out", count);
691 
692 	error = dlm_wait_function(ls, &recover_list_empty);
693  out:
694 	if (error)
695 		recover_list_clear(ls);
696 	return error;
697 }
698 
699 void dlm_recovered_lock(struct dlm_rsb *r)
700 {
701 	DLM_ASSERT(rsb_flag(r, RSB_NEW_MASTER), dlm_dump_rsb(r););
702 
703 	r->res_recover_locks_count--;
704 	if (!r->res_recover_locks_count) {
705 		rsb_clear_flag(r, RSB_NEW_MASTER);
706 		recover_list_del(r);
707 	}
708 
709 	if (recover_list_empty(r->res_ls))
710 		wake_up(&r->res_ls->ls_wait_general);
711 }
712 
713 /*
714  * The lvb needs to be recovered on all master rsb's.  This includes setting
715  * the VALNOTVALID flag if necessary, and determining the correct lvb contents
716  * based on the lvb's of the locks held on the rsb.
717  *
718  * RSB_VALNOTVALID is set in two cases:
719  *
720  * 1. we are master, but not new, and we purged an EX/PW lock held by a
721  * failed node (in dlm_recover_purge which set RSB_RECOVER_LVB_INVAL)
722  *
723  * 2. we are a new master, and there are only NL/CR locks left.
724  * (We could probably improve this by only invaliding in this way when
725  * the previous master left uncleanly.  VMS docs mention that.)
726  *
727  * The LVB contents are only considered for changing when this is a new master
728  * of the rsb (NEW_MASTER2).  Then, the rsb's lvb is taken from any lkb with
729  * mode > CR.  If no lkb's exist with mode above CR, the lvb contents are taken
730  * from the lkb with the largest lvb sequence number.
731  */
732 
733 static void recover_lvb(struct dlm_rsb *r)
734 {
735 	struct dlm_lkb *big_lkb = NULL, *iter, *high_lkb = NULL;
736 	uint32_t high_seq = 0;
737 	int lock_lvb_exists = 0;
738 	int lvblen = r->res_ls->ls_lvblen;
739 
740 	if (!rsb_flag(r, RSB_NEW_MASTER2) &&
741 	    rsb_flag(r, RSB_RECOVER_LVB_INVAL)) {
742 		/* case 1 above */
743 		rsb_set_flag(r, RSB_VALNOTVALID);
744 		return;
745 	}
746 
747 	if (!rsb_flag(r, RSB_NEW_MASTER2))
748 		return;
749 
750 	/* we are the new master, so figure out if VALNOTVALID should
751 	   be set, and set the rsb lvb from the best lkb available. */
752 
753 	list_for_each_entry(iter, &r->res_grantqueue, lkb_statequeue) {
754 		if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
755 			continue;
756 
757 		lock_lvb_exists = 1;
758 
759 		if (iter->lkb_grmode > DLM_LOCK_CR) {
760 			big_lkb = iter;
761 			goto setflag;
762 		}
763 
764 		if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
765 			high_lkb = iter;
766 			high_seq = iter->lkb_lvbseq;
767 		}
768 	}
769 
770 	list_for_each_entry(iter, &r->res_convertqueue, lkb_statequeue) {
771 		if (!(iter->lkb_exflags & DLM_LKF_VALBLK))
772 			continue;
773 
774 		lock_lvb_exists = 1;
775 
776 		if (iter->lkb_grmode > DLM_LOCK_CR) {
777 			big_lkb = iter;
778 			goto setflag;
779 		}
780 
781 		if (((int)iter->lkb_lvbseq - (int)high_seq) >= 0) {
782 			high_lkb = iter;
783 			high_seq = iter->lkb_lvbseq;
784 		}
785 	}
786 
787  setflag:
788 	if (!lock_lvb_exists)
789 		goto out;
790 
791 	/* lvb is invalidated if only NL/CR locks remain */
792 	if (!big_lkb)
793 		rsb_set_flag(r, RSB_VALNOTVALID);
794 
795 	if (!r->res_lvbptr) {
796 		r->res_lvbptr = dlm_allocate_lvb(r->res_ls);
797 		if (!r->res_lvbptr)
798 			goto out;
799 	}
800 
801 	if (big_lkb) {
802 		r->res_lvbseq = big_lkb->lkb_lvbseq;
803 		memcpy(r->res_lvbptr, big_lkb->lkb_lvbptr, lvblen);
804 	} else if (high_lkb) {
805 		r->res_lvbseq = high_lkb->lkb_lvbseq;
806 		memcpy(r->res_lvbptr, high_lkb->lkb_lvbptr, lvblen);
807 	} else {
808 		r->res_lvbseq = 0;
809 		memset(r->res_lvbptr, 0, lvblen);
810 	}
811  out:
812 	return;
813 }
814 
815 /* All master rsb's flagged RECOVER_CONVERT need to be looked at.  The locks
816    converting PR->CW or CW->PR need to have their lkb_grmode set. */
817 
818 static void recover_conversion(struct dlm_rsb *r)
819 {
820 	struct dlm_ls *ls = r->res_ls;
821 	struct dlm_lkb *lkb;
822 	int grmode = -1;
823 
824 	list_for_each_entry(lkb, &r->res_grantqueue, lkb_statequeue) {
825 		if (lkb->lkb_grmode == DLM_LOCK_PR ||
826 		    lkb->lkb_grmode == DLM_LOCK_CW) {
827 			grmode = lkb->lkb_grmode;
828 			break;
829 		}
830 	}
831 
832 	list_for_each_entry(lkb, &r->res_convertqueue, lkb_statequeue) {
833 		if (lkb->lkb_grmode != DLM_LOCK_IV)
834 			continue;
835 		if (grmode == -1) {
836 			log_debug(ls, "recover_conversion %x set gr to rq %d",
837 				  lkb->lkb_id, lkb->lkb_rqmode);
838 			lkb->lkb_grmode = lkb->lkb_rqmode;
839 		} else {
840 			log_debug(ls, "recover_conversion %x set gr %d",
841 				  lkb->lkb_id, grmode);
842 			lkb->lkb_grmode = grmode;
843 		}
844 	}
845 }
846 
847 /* We've become the new master for this rsb and waiting/converting locks may
848    need to be granted in dlm_recover_grant() due to locks that may have
849    existed from a removed node. */
850 
851 static void recover_grant(struct dlm_rsb *r)
852 {
853 	if (!list_empty(&r->res_waitqueue) || !list_empty(&r->res_convertqueue))
854 		rsb_set_flag(r, RSB_RECOVER_GRANT);
855 }
856 
857 void dlm_recover_rsbs(struct dlm_ls *ls)
858 {
859 	struct dlm_rsb *r;
860 	unsigned int count = 0;
861 
862 	down_read(&ls->ls_root_sem);
863 	list_for_each_entry(r, &ls->ls_root_list, res_root_list) {
864 		lock_rsb(r);
865 		if (is_master(r)) {
866 			if (rsb_flag(r, RSB_RECOVER_CONVERT))
867 				recover_conversion(r);
868 
869 			/* recover lvb before granting locks so the updated
870 			   lvb/VALNOTVALID is presented in the completion */
871 			recover_lvb(r);
872 
873 			if (rsb_flag(r, RSB_NEW_MASTER2))
874 				recover_grant(r);
875 			count++;
876 		} else {
877 			rsb_clear_flag(r, RSB_VALNOTVALID);
878 		}
879 		rsb_clear_flag(r, RSB_RECOVER_CONVERT);
880 		rsb_clear_flag(r, RSB_RECOVER_LVB_INVAL);
881 		rsb_clear_flag(r, RSB_NEW_MASTER2);
882 		unlock_rsb(r);
883 	}
884 	up_read(&ls->ls_root_sem);
885 
886 	if (count)
887 		log_rinfo(ls, "dlm_recover_rsbs %d done", count);
888 }
889 
890 /* Create a single list of all root rsb's to be used during recovery */
891 
892 int dlm_create_root_list(struct dlm_ls *ls)
893 {
894 	struct rb_node *n;
895 	struct dlm_rsb *r;
896 	int i, error = 0;
897 
898 	down_write(&ls->ls_root_sem);
899 	if (!list_empty(&ls->ls_root_list)) {
900 		log_error(ls, "root list not empty");
901 		error = -EINVAL;
902 		goto out;
903 	}
904 
905 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
906 		spin_lock(&ls->ls_rsbtbl[i].lock);
907 		for (n = rb_first(&ls->ls_rsbtbl[i].keep); n; n = rb_next(n)) {
908 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
909 			list_add(&r->res_root_list, &ls->ls_root_list);
910 			dlm_hold_rsb(r);
911 		}
912 
913 		if (!RB_EMPTY_ROOT(&ls->ls_rsbtbl[i].toss))
914 			log_error(ls, "dlm_create_root_list toss not empty");
915 		spin_unlock(&ls->ls_rsbtbl[i].lock);
916 	}
917  out:
918 	up_write(&ls->ls_root_sem);
919 	return error;
920 }
921 
922 void dlm_release_root_list(struct dlm_ls *ls)
923 {
924 	struct dlm_rsb *r, *safe;
925 
926 	down_write(&ls->ls_root_sem);
927 	list_for_each_entry_safe(r, safe, &ls->ls_root_list, res_root_list) {
928 		list_del_init(&r->res_root_list);
929 		dlm_put_rsb(r);
930 	}
931 	up_write(&ls->ls_root_sem);
932 }
933 
934 void dlm_clear_toss(struct dlm_ls *ls)
935 {
936 	struct rb_node *n, *next;
937 	struct dlm_rsb *r;
938 	unsigned int count = 0;
939 	int i;
940 
941 	for (i = 0; i < ls->ls_rsbtbl_size; i++) {
942 		spin_lock(&ls->ls_rsbtbl[i].lock);
943 		for (n = rb_first(&ls->ls_rsbtbl[i].toss); n; n = next) {
944 			next = rb_next(n);
945 			r = rb_entry(n, struct dlm_rsb, res_hashnode);
946 			rb_erase(n, &ls->ls_rsbtbl[i].toss);
947 			dlm_free_rsb(r);
948 			count++;
949 		}
950 		spin_unlock(&ls->ls_rsbtbl[i].lock);
951 	}
952 
953 	if (count)
954 		log_rinfo(ls, "dlm_clear_toss %u done", count);
955 }
956 
957