xref: /linux/fs/dlm/member.c (revision c4c11dd160a8cc98f402c4e12f94b1572e822ffd)
1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) 2005-2011 Red Hat, Inc.  All rights reserved.
5 **
6 **  This copyrighted material is made available to anyone wishing to use,
7 **  modify, copy, or redistribute it subject to the terms and conditions
8 **  of the GNU General Public License v.2.
9 **
10 *******************************************************************************
11 ******************************************************************************/
12 
13 #include "dlm_internal.h"
14 #include "lockspace.h"
15 #include "member.h"
16 #include "recoverd.h"
17 #include "recover.h"
18 #include "rcom.h"
19 #include "config.h"
20 #include "lowcomms.h"
21 
22 int dlm_slots_version(struct dlm_header *h)
23 {
24 	if ((h->h_version & 0x0000FFFF) < DLM_HEADER_SLOTS)
25 		return 0;
26 	return 1;
27 }
28 
29 void dlm_slot_save(struct dlm_ls *ls, struct dlm_rcom *rc,
30 		   struct dlm_member *memb)
31 {
32 	struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
33 
34 	if (!dlm_slots_version(&rc->rc_header))
35 		return;
36 
37 	memb->slot = le16_to_cpu(rf->rf_our_slot);
38 	memb->generation = le32_to_cpu(rf->rf_generation);
39 }
40 
41 void dlm_slots_copy_out(struct dlm_ls *ls, struct dlm_rcom *rc)
42 {
43 	struct dlm_slot *slot;
44 	struct rcom_slot *ro;
45 	int i;
46 
47 	ro = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
48 
49 	/* ls_slots array is sparse, but not rcom_slots */
50 
51 	for (i = 0; i < ls->ls_slots_size; i++) {
52 		slot = &ls->ls_slots[i];
53 		if (!slot->nodeid)
54 			continue;
55 		ro->ro_nodeid = cpu_to_le32(slot->nodeid);
56 		ro->ro_slot = cpu_to_le16(slot->slot);
57 		ro++;
58 	}
59 }
60 
61 #define SLOT_DEBUG_LINE 128
62 
63 static void log_debug_slots(struct dlm_ls *ls, uint32_t gen, int num_slots,
64 			    struct rcom_slot *ro0, struct dlm_slot *array,
65 			    int array_size)
66 {
67 	char line[SLOT_DEBUG_LINE];
68 	int len = SLOT_DEBUG_LINE - 1;
69 	int pos = 0;
70 	int ret, i;
71 
72 	if (!dlm_config.ci_log_debug)
73 		return;
74 
75 	memset(line, 0, sizeof(line));
76 
77 	if (array) {
78 		for (i = 0; i < array_size; i++) {
79 			if (!array[i].nodeid)
80 				continue;
81 
82 			ret = snprintf(line + pos, len - pos, " %d:%d",
83 				       array[i].slot, array[i].nodeid);
84 			if (ret >= len - pos)
85 				break;
86 			pos += ret;
87 		}
88 	} else if (ro0) {
89 		for (i = 0; i < num_slots; i++) {
90 			ret = snprintf(line + pos, len - pos, " %d:%d",
91 				       ro0[i].ro_slot, ro0[i].ro_nodeid);
92 			if (ret >= len - pos)
93 				break;
94 			pos += ret;
95 		}
96 	}
97 
98 	log_debug(ls, "generation %u slots %d%s", gen, num_slots, line);
99 }
100 
101 int dlm_slots_copy_in(struct dlm_ls *ls)
102 {
103 	struct dlm_member *memb;
104 	struct dlm_rcom *rc = ls->ls_recover_buf;
105 	struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
106 	struct rcom_slot *ro0, *ro;
107 	int our_nodeid = dlm_our_nodeid();
108 	int i, num_slots;
109 	uint32_t gen;
110 
111 	if (!dlm_slots_version(&rc->rc_header))
112 		return -1;
113 
114 	gen = le32_to_cpu(rf->rf_generation);
115 	if (gen <= ls->ls_generation) {
116 		log_error(ls, "dlm_slots_copy_in gen %u old %u",
117 			  gen, ls->ls_generation);
118 	}
119 	ls->ls_generation = gen;
120 
121 	num_slots = le16_to_cpu(rf->rf_num_slots);
122 	if (!num_slots)
123 		return -1;
124 
125 	ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
126 
127 	for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
128 		ro->ro_nodeid = le32_to_cpu(ro->ro_nodeid);
129 		ro->ro_slot = le16_to_cpu(ro->ro_slot);
130 	}
131 
132 	log_debug_slots(ls, gen, num_slots, ro0, NULL, 0);
133 
134 	list_for_each_entry(memb, &ls->ls_nodes, list) {
135 		for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
136 			if (ro->ro_nodeid != memb->nodeid)
137 				continue;
138 			memb->slot = ro->ro_slot;
139 			memb->slot_prev = memb->slot;
140 			break;
141 		}
142 
143 		if (memb->nodeid == our_nodeid) {
144 			if (ls->ls_slot && ls->ls_slot != memb->slot) {
145 				log_error(ls, "dlm_slots_copy_in our slot "
146 					  "changed %d %d", ls->ls_slot,
147 					  memb->slot);
148 				return -1;
149 			}
150 
151 			if (!ls->ls_slot)
152 				ls->ls_slot = memb->slot;
153 		}
154 
155 		if (!memb->slot) {
156 			log_error(ls, "dlm_slots_copy_in nodeid %d no slot",
157 				   memb->nodeid);
158 			return -1;
159 		}
160 	}
161 
162 	return 0;
163 }
164 
165 /* for any nodes that do not support slots, we will not have set memb->slot
166    in wait_status_all(), so memb->slot will remain -1, and we will not
167    assign slots or set ls_num_slots here */
168 
169 int dlm_slots_assign(struct dlm_ls *ls, int *num_slots, int *slots_size,
170 		     struct dlm_slot **slots_out, uint32_t *gen_out)
171 {
172 	struct dlm_member *memb;
173 	struct dlm_slot *array;
174 	int our_nodeid = dlm_our_nodeid();
175 	int array_size, max_slots, i;
176 	int need = 0;
177 	int max = 0;
178 	int num = 0;
179 	uint32_t gen = 0;
180 
181 	/* our own memb struct will have slot -1 gen 0 */
182 
183 	list_for_each_entry(memb, &ls->ls_nodes, list) {
184 		if (memb->nodeid == our_nodeid) {
185 			memb->slot = ls->ls_slot;
186 			memb->generation = ls->ls_generation;
187 			break;
188 		}
189 	}
190 
191 	list_for_each_entry(memb, &ls->ls_nodes, list) {
192 		if (memb->generation > gen)
193 			gen = memb->generation;
194 
195 		/* node doesn't support slots */
196 
197 		if (memb->slot == -1)
198 			return -1;
199 
200 		/* node needs a slot assigned */
201 
202 		if (!memb->slot)
203 			need++;
204 
205 		/* node has a slot assigned */
206 
207 		num++;
208 
209 		if (!max || max < memb->slot)
210 			max = memb->slot;
211 
212 		/* sanity check, once slot is assigned it shouldn't change */
213 
214 		if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) {
215 			log_error(ls, "nodeid %d slot changed %d %d",
216 				  memb->nodeid, memb->slot_prev, memb->slot);
217 			return -1;
218 		}
219 		memb->slot_prev = memb->slot;
220 	}
221 
222 	array_size = max + need;
223 
224 	array = kzalloc(array_size * sizeof(struct dlm_slot), GFP_NOFS);
225 	if (!array)
226 		return -ENOMEM;
227 
228 	num = 0;
229 
230 	/* fill in slots (offsets) that are used */
231 
232 	list_for_each_entry(memb, &ls->ls_nodes, list) {
233 		if (!memb->slot)
234 			continue;
235 
236 		if (memb->slot > array_size) {
237 			log_error(ls, "invalid slot number %d", memb->slot);
238 			kfree(array);
239 			return -1;
240 		}
241 
242 		array[memb->slot - 1].nodeid = memb->nodeid;
243 		array[memb->slot - 1].slot = memb->slot;
244 		num++;
245 	}
246 
247 	/* assign new slots from unused offsets */
248 
249 	list_for_each_entry(memb, &ls->ls_nodes, list) {
250 		if (memb->slot)
251 			continue;
252 
253 		for (i = 0; i < array_size; i++) {
254 			if (array[i].nodeid)
255 				continue;
256 
257 			memb->slot = i + 1;
258 			memb->slot_prev = memb->slot;
259 			array[i].nodeid = memb->nodeid;
260 			array[i].slot = memb->slot;
261 			num++;
262 
263 			if (!ls->ls_slot && memb->nodeid == our_nodeid)
264 				ls->ls_slot = memb->slot;
265 			break;
266 		}
267 
268 		if (!memb->slot) {
269 			log_error(ls, "no free slot found");
270 			kfree(array);
271 			return -1;
272 		}
273 	}
274 
275 	gen++;
276 
277 	log_debug_slots(ls, gen, num, NULL, array, array_size);
278 
279 	max_slots = (dlm_config.ci_buffer_size - sizeof(struct dlm_rcom) -
280 		     sizeof(struct rcom_config)) / sizeof(struct rcom_slot);
281 
282 	if (num > max_slots) {
283 		log_error(ls, "num_slots %d exceeds max_slots %d",
284 			  num, max_slots);
285 		kfree(array);
286 		return -1;
287 	}
288 
289 	*gen_out = gen;
290 	*slots_out = array;
291 	*slots_size = array_size;
292 	*num_slots = num;
293 	return 0;
294 }
295 
296 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
297 {
298 	struct dlm_member *memb = NULL;
299 	struct list_head *tmp;
300 	struct list_head *newlist = &new->list;
301 	struct list_head *head = &ls->ls_nodes;
302 
303 	list_for_each(tmp, head) {
304 		memb = list_entry(tmp, struct dlm_member, list);
305 		if (new->nodeid < memb->nodeid)
306 			break;
307 	}
308 
309 	if (!memb)
310 		list_add_tail(newlist, head);
311 	else {
312 		/* FIXME: can use list macro here */
313 		newlist->prev = tmp->prev;
314 		newlist->next = tmp;
315 		tmp->prev->next = newlist;
316 		tmp->prev = newlist;
317 	}
318 }
319 
320 static int dlm_add_member(struct dlm_ls *ls, struct dlm_config_node *node)
321 {
322 	struct dlm_member *memb;
323 	int error;
324 
325 	memb = kzalloc(sizeof(struct dlm_member), GFP_NOFS);
326 	if (!memb)
327 		return -ENOMEM;
328 
329 	error = dlm_lowcomms_connect_node(node->nodeid);
330 	if (error < 0) {
331 		kfree(memb);
332 		return error;
333 	}
334 
335 	memb->nodeid = node->nodeid;
336 	memb->weight = node->weight;
337 	memb->comm_seq = node->comm_seq;
338 	add_ordered_member(ls, memb);
339 	ls->ls_num_nodes++;
340 	return 0;
341 }
342 
343 static struct dlm_member *find_memb(struct list_head *head, int nodeid)
344 {
345 	struct dlm_member *memb;
346 
347 	list_for_each_entry(memb, head, list) {
348 		if (memb->nodeid == nodeid)
349 			return memb;
350 	}
351 	return NULL;
352 }
353 
354 int dlm_is_member(struct dlm_ls *ls, int nodeid)
355 {
356 	if (find_memb(&ls->ls_nodes, nodeid))
357 		return 1;
358 	return 0;
359 }
360 
361 int dlm_is_removed(struct dlm_ls *ls, int nodeid)
362 {
363 	if (find_memb(&ls->ls_nodes_gone, nodeid))
364 		return 1;
365 	return 0;
366 }
367 
368 static void clear_memb_list(struct list_head *head)
369 {
370 	struct dlm_member *memb;
371 
372 	while (!list_empty(head)) {
373 		memb = list_entry(head->next, struct dlm_member, list);
374 		list_del(&memb->list);
375 		kfree(memb);
376 	}
377 }
378 
379 void dlm_clear_members(struct dlm_ls *ls)
380 {
381 	clear_memb_list(&ls->ls_nodes);
382 	ls->ls_num_nodes = 0;
383 }
384 
385 void dlm_clear_members_gone(struct dlm_ls *ls)
386 {
387 	clear_memb_list(&ls->ls_nodes_gone);
388 }
389 
390 static void make_member_array(struct dlm_ls *ls)
391 {
392 	struct dlm_member *memb;
393 	int i, w, x = 0, total = 0, all_zero = 0, *array;
394 
395 	kfree(ls->ls_node_array);
396 	ls->ls_node_array = NULL;
397 
398 	list_for_each_entry(memb, &ls->ls_nodes, list) {
399 		if (memb->weight)
400 			total += memb->weight;
401 	}
402 
403 	/* all nodes revert to weight of 1 if all have weight 0 */
404 
405 	if (!total) {
406 		total = ls->ls_num_nodes;
407 		all_zero = 1;
408 	}
409 
410 	ls->ls_total_weight = total;
411 
412 	array = kmalloc(sizeof(int) * total, GFP_NOFS);
413 	if (!array)
414 		return;
415 
416 	list_for_each_entry(memb, &ls->ls_nodes, list) {
417 		if (!all_zero && !memb->weight)
418 			continue;
419 
420 		if (all_zero)
421 			w = 1;
422 		else
423 			w = memb->weight;
424 
425 		DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
426 
427 		for (i = 0; i < w; i++)
428 			array[x++] = memb->nodeid;
429 	}
430 
431 	ls->ls_node_array = array;
432 }
433 
434 /* send a status request to all members just to establish comms connections */
435 
436 static int ping_members(struct dlm_ls *ls)
437 {
438 	struct dlm_member *memb;
439 	int error = 0;
440 
441 	list_for_each_entry(memb, &ls->ls_nodes, list) {
442 		error = dlm_recovery_stopped(ls);
443 		if (error)
444 			break;
445 		error = dlm_rcom_status(ls, memb->nodeid, 0);
446 		if (error)
447 			break;
448 	}
449 	if (error)
450 		log_debug(ls, "ping_members aborted %d last nodeid %d",
451 			  error, ls->ls_recover_nodeid);
452 	return error;
453 }
454 
455 static void dlm_lsop_recover_prep(struct dlm_ls *ls)
456 {
457 	if (!ls->ls_ops || !ls->ls_ops->recover_prep)
458 		return;
459 	ls->ls_ops->recover_prep(ls->ls_ops_arg);
460 }
461 
462 static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
463 {
464 	struct dlm_slot slot;
465 	uint32_t seq;
466 	int error;
467 
468 	if (!ls->ls_ops || !ls->ls_ops->recover_slot)
469 		return;
470 
471 	/* if there is no comms connection with this node
472 	   or the present comms connection is newer
473 	   than the one when this member was added, then
474 	   we consider the node to have failed (versus
475 	   being removed due to dlm_release_lockspace) */
476 
477 	error = dlm_comm_seq(memb->nodeid, &seq);
478 
479 	if (!error && seq == memb->comm_seq)
480 		return;
481 
482 	slot.nodeid = memb->nodeid;
483 	slot.slot = memb->slot;
484 
485 	ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot);
486 }
487 
488 void dlm_lsop_recover_done(struct dlm_ls *ls)
489 {
490 	struct dlm_member *memb;
491 	struct dlm_slot *slots;
492 	int i, num;
493 
494 	if (!ls->ls_ops || !ls->ls_ops->recover_done)
495 		return;
496 
497 	num = ls->ls_num_nodes;
498 
499 	slots = kzalloc(num * sizeof(struct dlm_slot), GFP_KERNEL);
500 	if (!slots)
501 		return;
502 
503 	i = 0;
504 	list_for_each_entry(memb, &ls->ls_nodes, list) {
505 		if (i == num) {
506 			log_error(ls, "dlm_lsop_recover_done bad num %d", num);
507 			goto out;
508 		}
509 		slots[i].nodeid = memb->nodeid;
510 		slots[i].slot = memb->slot;
511 		i++;
512 	}
513 
514 	ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num,
515 				 ls->ls_slot, ls->ls_generation);
516  out:
517 	kfree(slots);
518 }
519 
520 static struct dlm_config_node *find_config_node(struct dlm_recover *rv,
521 						int nodeid)
522 {
523 	int i;
524 
525 	for (i = 0; i < rv->nodes_count; i++) {
526 		if (rv->nodes[i].nodeid == nodeid)
527 			return &rv->nodes[i];
528 	}
529 	return NULL;
530 }
531 
532 int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
533 {
534 	struct dlm_member *memb, *safe;
535 	struct dlm_config_node *node;
536 	int i, error, neg = 0, low = -1;
537 
538 	/* previously removed members that we've not finished removing need to
539 	   count as a negative change so the "neg" recovery steps will happen */
540 
541 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
542 		log_debug(ls, "prev removed member %d", memb->nodeid);
543 		neg++;
544 	}
545 
546 	/* move departed members from ls_nodes to ls_nodes_gone */
547 
548 	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
549 		node = find_config_node(rv, memb->nodeid);
550 		if (node && !node->new)
551 			continue;
552 
553 		if (!node) {
554 			log_debug(ls, "remove member %d", memb->nodeid);
555 		} else {
556 			/* removed and re-added */
557 			log_debug(ls, "remove member %d comm_seq %u %u",
558 				  memb->nodeid, memb->comm_seq, node->comm_seq);
559 		}
560 
561 		neg++;
562 		list_move(&memb->list, &ls->ls_nodes_gone);
563 		ls->ls_num_nodes--;
564 		dlm_lsop_recover_slot(ls, memb);
565 	}
566 
567 	/* add new members to ls_nodes */
568 
569 	for (i = 0; i < rv->nodes_count; i++) {
570 		node = &rv->nodes[i];
571 		if (dlm_is_member(ls, node->nodeid))
572 			continue;
573 		dlm_add_member(ls, node);
574 		log_debug(ls, "add member %d", node->nodeid);
575 	}
576 
577 	list_for_each_entry(memb, &ls->ls_nodes, list) {
578 		if (low == -1 || memb->nodeid < low)
579 			low = memb->nodeid;
580 	}
581 	ls->ls_low_nodeid = low;
582 
583 	make_member_array(ls);
584 	*neg_out = neg;
585 
586 	error = ping_members(ls);
587 	if (!error || error == -EPROTO) {
588 		/* new_lockspace() may be waiting to know if the config
589 		   is good or bad */
590 		ls->ls_members_result = error;
591 		complete(&ls->ls_members_done);
592 	}
593 
594 	log_debug(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes);
595 	return error;
596 }
597 
598 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
599    dlm_ls_start() is called on any of them to start the new recovery. */
600 
601 int dlm_ls_stop(struct dlm_ls *ls)
602 {
603 	int new;
604 
605 	/*
606 	 * Prevent dlm_recv from being in the middle of something when we do
607 	 * the stop.  This includes ensuring dlm_recv isn't processing a
608 	 * recovery message (rcom), while dlm_recoverd is aborting and
609 	 * resetting things from an in-progress recovery.  i.e. we want
610 	 * dlm_recoverd to abort its recovery without worrying about dlm_recv
611 	 * processing an rcom at the same time.  Stopping dlm_recv also makes
612 	 * it easy for dlm_receive_message() to check locking stopped and add a
613 	 * message to the requestqueue without races.
614 	 */
615 
616 	down_write(&ls->ls_recv_active);
617 
618 	/*
619 	 * Abort any recovery that's in progress (see RECOVER_STOP,
620 	 * dlm_recovery_stopped()) and tell any other threads running in the
621 	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
622 	 */
623 
624 	spin_lock(&ls->ls_recover_lock);
625 	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
626 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
627 	ls->ls_recover_seq++;
628 	spin_unlock(&ls->ls_recover_lock);
629 
630 	/*
631 	 * Let dlm_recv run again, now any normal messages will be saved on the
632 	 * requestqueue for later.
633 	 */
634 
635 	up_write(&ls->ls_recv_active);
636 
637 	/*
638 	 * This in_recovery lock does two things:
639 	 * 1) Keeps this function from returning until all threads are out
640 	 *    of locking routines and locking is truly stopped.
641 	 * 2) Keeps any new requests from being processed until it's unlocked
642 	 *    when recovery is complete.
643 	 */
644 
645 	if (new) {
646 		set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags);
647 		wake_up_process(ls->ls_recoverd_task);
648 		wait_event(ls->ls_recover_lock_wait,
649 			   test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
650 	}
651 
652 	/*
653 	 * The recoverd suspend/resume makes sure that dlm_recoverd (if
654 	 * running) has noticed RECOVER_STOP above and quit processing the
655 	 * previous recovery.
656 	 */
657 
658 	dlm_recoverd_suspend(ls);
659 
660 	spin_lock(&ls->ls_recover_lock);
661 	kfree(ls->ls_slots);
662 	ls->ls_slots = NULL;
663 	ls->ls_num_slots = 0;
664 	ls->ls_slots_size = 0;
665 	ls->ls_recover_status = 0;
666 	spin_unlock(&ls->ls_recover_lock);
667 
668 	dlm_recoverd_resume(ls);
669 
670 	if (!ls->ls_recover_begin)
671 		ls->ls_recover_begin = jiffies;
672 
673 	dlm_lsop_recover_prep(ls);
674 	return 0;
675 }
676 
677 int dlm_ls_start(struct dlm_ls *ls)
678 {
679 	struct dlm_recover *rv = NULL, *rv_old;
680 	struct dlm_config_node *nodes;
681 	int error, count;
682 
683 	rv = kzalloc(sizeof(struct dlm_recover), GFP_NOFS);
684 	if (!rv)
685 		return -ENOMEM;
686 
687 	error = dlm_config_nodes(ls->ls_name, &nodes, &count);
688 	if (error < 0)
689 		goto fail;
690 
691 	spin_lock(&ls->ls_recover_lock);
692 
693 	/* the lockspace needs to be stopped before it can be started */
694 
695 	if (!dlm_locking_stopped(ls)) {
696 		spin_unlock(&ls->ls_recover_lock);
697 		log_error(ls, "start ignored: lockspace running");
698 		error = -EINVAL;
699 		goto fail;
700 	}
701 
702 	rv->nodes = nodes;
703 	rv->nodes_count = count;
704 	rv->seq = ++ls->ls_recover_seq;
705 	rv_old = ls->ls_recover_args;
706 	ls->ls_recover_args = rv;
707 	spin_unlock(&ls->ls_recover_lock);
708 
709 	if (rv_old) {
710 		log_error(ls, "unused recovery %llx %d",
711 			  (unsigned long long)rv_old->seq, rv_old->nodes_count);
712 		kfree(rv_old->nodes);
713 		kfree(rv_old);
714 	}
715 
716 	set_bit(LSFL_RECOVER_WORK, &ls->ls_flags);
717 	wake_up_process(ls->ls_recoverd_task);
718 	return 0;
719 
720  fail:
721 	kfree(rv);
722 	kfree(nodes);
723 	return error;
724 }
725 
726