xref: /openbmc/linux/fs/dlm/member.c (revision d8eabc37)
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_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 	memset(line, 0, sizeof(line));
73 
74 	if (array) {
75 		for (i = 0; i < array_size; i++) {
76 			if (!array[i].nodeid)
77 				continue;
78 
79 			ret = snprintf(line + pos, len - pos, " %d:%d",
80 				       array[i].slot, array[i].nodeid);
81 			if (ret >= len - pos)
82 				break;
83 			pos += ret;
84 		}
85 	} else if (ro0) {
86 		for (i = 0; i < num_slots; i++) {
87 			ret = snprintf(line + pos, len - pos, " %d:%d",
88 				       ro0[i].ro_slot, ro0[i].ro_nodeid);
89 			if (ret >= len - pos)
90 				break;
91 			pos += ret;
92 		}
93 	}
94 
95 	log_rinfo(ls, "generation %u slots %d%s", gen, num_slots, line);
96 }
97 
98 int dlm_slots_copy_in(struct dlm_ls *ls)
99 {
100 	struct dlm_member *memb;
101 	struct dlm_rcom *rc = ls->ls_recover_buf;
102 	struct rcom_config *rf = (struct rcom_config *)rc->rc_buf;
103 	struct rcom_slot *ro0, *ro;
104 	int our_nodeid = dlm_our_nodeid();
105 	int i, num_slots;
106 	uint32_t gen;
107 
108 	if (!dlm_slots_version(&rc->rc_header))
109 		return -1;
110 
111 	gen = le32_to_cpu(rf->rf_generation);
112 	if (gen <= ls->ls_generation) {
113 		log_error(ls, "dlm_slots_copy_in gen %u old %u",
114 			  gen, ls->ls_generation);
115 	}
116 	ls->ls_generation = gen;
117 
118 	num_slots = le16_to_cpu(rf->rf_num_slots);
119 	if (!num_slots)
120 		return -1;
121 
122 	ro0 = (struct rcom_slot *)(rc->rc_buf + sizeof(struct rcom_config));
123 
124 	for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
125 		ro->ro_nodeid = le32_to_cpu(ro->ro_nodeid);
126 		ro->ro_slot = le16_to_cpu(ro->ro_slot);
127 	}
128 
129 	log_slots(ls, gen, num_slots, ro0, NULL, 0);
130 
131 	list_for_each_entry(memb, &ls->ls_nodes, list) {
132 		for (i = 0, ro = ro0; i < num_slots; i++, ro++) {
133 			if (ro->ro_nodeid != memb->nodeid)
134 				continue;
135 			memb->slot = ro->ro_slot;
136 			memb->slot_prev = memb->slot;
137 			break;
138 		}
139 
140 		if (memb->nodeid == our_nodeid) {
141 			if (ls->ls_slot && ls->ls_slot != memb->slot) {
142 				log_error(ls, "dlm_slots_copy_in our slot "
143 					  "changed %d %d", ls->ls_slot,
144 					  memb->slot);
145 				return -1;
146 			}
147 
148 			if (!ls->ls_slot)
149 				ls->ls_slot = memb->slot;
150 		}
151 
152 		if (!memb->slot) {
153 			log_error(ls, "dlm_slots_copy_in nodeid %d no slot",
154 				   memb->nodeid);
155 			return -1;
156 		}
157 	}
158 
159 	return 0;
160 }
161 
162 /* for any nodes that do not support slots, we will not have set memb->slot
163    in wait_status_all(), so memb->slot will remain -1, and we will not
164    assign slots or set ls_num_slots here */
165 
166 int dlm_slots_assign(struct dlm_ls *ls, int *num_slots, int *slots_size,
167 		     struct dlm_slot **slots_out, uint32_t *gen_out)
168 {
169 	struct dlm_member *memb;
170 	struct dlm_slot *array;
171 	int our_nodeid = dlm_our_nodeid();
172 	int array_size, max_slots, i;
173 	int need = 0;
174 	int max = 0;
175 	int num = 0;
176 	uint32_t gen = 0;
177 
178 	/* our own memb struct will have slot -1 gen 0 */
179 
180 	list_for_each_entry(memb, &ls->ls_nodes, list) {
181 		if (memb->nodeid == our_nodeid) {
182 			memb->slot = ls->ls_slot;
183 			memb->generation = ls->ls_generation;
184 			break;
185 		}
186 	}
187 
188 	list_for_each_entry(memb, &ls->ls_nodes, list) {
189 		if (memb->generation > gen)
190 			gen = memb->generation;
191 
192 		/* node doesn't support slots */
193 
194 		if (memb->slot == -1)
195 			return -1;
196 
197 		/* node needs a slot assigned */
198 
199 		if (!memb->slot)
200 			need++;
201 
202 		/* node has a slot assigned */
203 
204 		num++;
205 
206 		if (!max || max < memb->slot)
207 			max = memb->slot;
208 
209 		/* sanity check, once slot is assigned it shouldn't change */
210 
211 		if (memb->slot_prev && memb->slot && memb->slot_prev != memb->slot) {
212 			log_error(ls, "nodeid %d slot changed %d %d",
213 				  memb->nodeid, memb->slot_prev, memb->slot);
214 			return -1;
215 		}
216 		memb->slot_prev = memb->slot;
217 	}
218 
219 	array_size = max + need;
220 	array = kcalloc(array_size, sizeof(*array), GFP_NOFS);
221 	if (!array)
222 		return -ENOMEM;
223 
224 	num = 0;
225 
226 	/* fill in slots (offsets) that are used */
227 
228 	list_for_each_entry(memb, &ls->ls_nodes, list) {
229 		if (!memb->slot)
230 			continue;
231 
232 		if (memb->slot > array_size) {
233 			log_error(ls, "invalid slot number %d", memb->slot);
234 			kfree(array);
235 			return -1;
236 		}
237 
238 		array[memb->slot - 1].nodeid = memb->nodeid;
239 		array[memb->slot - 1].slot = memb->slot;
240 		num++;
241 	}
242 
243 	/* assign new slots from unused offsets */
244 
245 	list_for_each_entry(memb, &ls->ls_nodes, list) {
246 		if (memb->slot)
247 			continue;
248 
249 		for (i = 0; i < array_size; i++) {
250 			if (array[i].nodeid)
251 				continue;
252 
253 			memb->slot = i + 1;
254 			memb->slot_prev = memb->slot;
255 			array[i].nodeid = memb->nodeid;
256 			array[i].slot = memb->slot;
257 			num++;
258 
259 			if (!ls->ls_slot && memb->nodeid == our_nodeid)
260 				ls->ls_slot = memb->slot;
261 			break;
262 		}
263 
264 		if (!memb->slot) {
265 			log_error(ls, "no free slot found");
266 			kfree(array);
267 			return -1;
268 		}
269 	}
270 
271 	gen++;
272 
273 	log_slots(ls, gen, num, NULL, array, array_size);
274 
275 	max_slots = (dlm_config.ci_buffer_size - sizeof(struct dlm_rcom) -
276 		     sizeof(struct rcom_config)) / sizeof(struct rcom_slot);
277 
278 	if (num > max_slots) {
279 		log_error(ls, "num_slots %d exceeds max_slots %d",
280 			  num, max_slots);
281 		kfree(array);
282 		return -1;
283 	}
284 
285 	*gen_out = gen;
286 	*slots_out = array;
287 	*slots_size = array_size;
288 	*num_slots = num;
289 	return 0;
290 }
291 
292 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
293 {
294 	struct dlm_member *memb = NULL;
295 	struct list_head *tmp;
296 	struct list_head *newlist = &new->list;
297 	struct list_head *head = &ls->ls_nodes;
298 
299 	list_for_each(tmp, head) {
300 		memb = list_entry(tmp, struct dlm_member, list);
301 		if (new->nodeid < memb->nodeid)
302 			break;
303 	}
304 
305 	if (!memb)
306 		list_add_tail(newlist, head);
307 	else {
308 		/* FIXME: can use list macro here */
309 		newlist->prev = tmp->prev;
310 		newlist->next = tmp;
311 		tmp->prev->next = newlist;
312 		tmp->prev = newlist;
313 	}
314 }
315 
316 static int dlm_add_member(struct dlm_ls *ls, struct dlm_config_node *node)
317 {
318 	struct dlm_member *memb;
319 	int error;
320 
321 	memb = kzalloc(sizeof(*memb), GFP_NOFS);
322 	if (!memb)
323 		return -ENOMEM;
324 
325 	error = dlm_lowcomms_connect_node(node->nodeid);
326 	if (error < 0) {
327 		kfree(memb);
328 		return error;
329 	}
330 
331 	memb->nodeid = node->nodeid;
332 	memb->weight = node->weight;
333 	memb->comm_seq = node->comm_seq;
334 	add_ordered_member(ls, memb);
335 	ls->ls_num_nodes++;
336 	return 0;
337 }
338 
339 static struct dlm_member *find_memb(struct list_head *head, int nodeid)
340 {
341 	struct dlm_member *memb;
342 
343 	list_for_each_entry(memb, head, list) {
344 		if (memb->nodeid == nodeid)
345 			return memb;
346 	}
347 	return NULL;
348 }
349 
350 int dlm_is_member(struct dlm_ls *ls, int nodeid)
351 {
352 	if (find_memb(&ls->ls_nodes, nodeid))
353 		return 1;
354 	return 0;
355 }
356 
357 int dlm_is_removed(struct dlm_ls *ls, int nodeid)
358 {
359 	if (find_memb(&ls->ls_nodes_gone, nodeid))
360 		return 1;
361 	return 0;
362 }
363 
364 static void clear_memb_list(struct list_head *head)
365 {
366 	struct dlm_member *memb;
367 
368 	while (!list_empty(head)) {
369 		memb = list_entry(head->next, struct dlm_member, list);
370 		list_del(&memb->list);
371 		kfree(memb);
372 	}
373 }
374 
375 void dlm_clear_members(struct dlm_ls *ls)
376 {
377 	clear_memb_list(&ls->ls_nodes);
378 	ls->ls_num_nodes = 0;
379 }
380 
381 void dlm_clear_members_gone(struct dlm_ls *ls)
382 {
383 	clear_memb_list(&ls->ls_nodes_gone);
384 }
385 
386 static void make_member_array(struct dlm_ls *ls)
387 {
388 	struct dlm_member *memb;
389 	int i, w, x = 0, total = 0, all_zero = 0, *array;
390 
391 	kfree(ls->ls_node_array);
392 	ls->ls_node_array = NULL;
393 
394 	list_for_each_entry(memb, &ls->ls_nodes, list) {
395 		if (memb->weight)
396 			total += memb->weight;
397 	}
398 
399 	/* all nodes revert to weight of 1 if all have weight 0 */
400 
401 	if (!total) {
402 		total = ls->ls_num_nodes;
403 		all_zero = 1;
404 	}
405 
406 	ls->ls_total_weight = total;
407 	array = kmalloc_array(total, sizeof(*array), GFP_NOFS);
408 	if (!array)
409 		return;
410 
411 	list_for_each_entry(memb, &ls->ls_nodes, list) {
412 		if (!all_zero && !memb->weight)
413 			continue;
414 
415 		if (all_zero)
416 			w = 1;
417 		else
418 			w = memb->weight;
419 
420 		DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
421 
422 		for (i = 0; i < w; i++)
423 			array[x++] = memb->nodeid;
424 	}
425 
426 	ls->ls_node_array = array;
427 }
428 
429 /* send a status request to all members just to establish comms connections */
430 
431 static int ping_members(struct dlm_ls *ls)
432 {
433 	struct dlm_member *memb;
434 	int error = 0;
435 
436 	list_for_each_entry(memb, &ls->ls_nodes, list) {
437 		error = dlm_recovery_stopped(ls);
438 		if (error)
439 			break;
440 		error = dlm_rcom_status(ls, memb->nodeid, 0);
441 		if (error)
442 			break;
443 	}
444 	if (error)
445 		log_rinfo(ls, "ping_members aborted %d last nodeid %d",
446 			  error, ls->ls_recover_nodeid);
447 	return error;
448 }
449 
450 static void dlm_lsop_recover_prep(struct dlm_ls *ls)
451 {
452 	if (!ls->ls_ops || !ls->ls_ops->recover_prep)
453 		return;
454 	ls->ls_ops->recover_prep(ls->ls_ops_arg);
455 }
456 
457 static void dlm_lsop_recover_slot(struct dlm_ls *ls, struct dlm_member *memb)
458 {
459 	struct dlm_slot slot;
460 	uint32_t seq;
461 	int error;
462 
463 	if (!ls->ls_ops || !ls->ls_ops->recover_slot)
464 		return;
465 
466 	/* if there is no comms connection with this node
467 	   or the present comms connection is newer
468 	   than the one when this member was added, then
469 	   we consider the node to have failed (versus
470 	   being removed due to dlm_release_lockspace) */
471 
472 	error = dlm_comm_seq(memb->nodeid, &seq);
473 
474 	if (!error && seq == memb->comm_seq)
475 		return;
476 
477 	slot.nodeid = memb->nodeid;
478 	slot.slot = memb->slot;
479 
480 	ls->ls_ops->recover_slot(ls->ls_ops_arg, &slot);
481 }
482 
483 void dlm_lsop_recover_done(struct dlm_ls *ls)
484 {
485 	struct dlm_member *memb;
486 	struct dlm_slot *slots;
487 	int i, num;
488 
489 	if (!ls->ls_ops || !ls->ls_ops->recover_done)
490 		return;
491 
492 	num = ls->ls_num_nodes;
493 	slots = kcalloc(num, sizeof(*slots), GFP_KERNEL);
494 	if (!slots)
495 		return;
496 
497 	i = 0;
498 	list_for_each_entry(memb, &ls->ls_nodes, list) {
499 		if (i == num) {
500 			log_error(ls, "dlm_lsop_recover_done bad num %d", num);
501 			goto out;
502 		}
503 		slots[i].nodeid = memb->nodeid;
504 		slots[i].slot = memb->slot;
505 		i++;
506 	}
507 
508 	ls->ls_ops->recover_done(ls->ls_ops_arg, slots, num,
509 				 ls->ls_slot, ls->ls_generation);
510  out:
511 	kfree(slots);
512 }
513 
514 static struct dlm_config_node *find_config_node(struct dlm_recover *rv,
515 						int nodeid)
516 {
517 	int i;
518 
519 	for (i = 0; i < rv->nodes_count; i++) {
520 		if (rv->nodes[i].nodeid == nodeid)
521 			return &rv->nodes[i];
522 	}
523 	return NULL;
524 }
525 
526 int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
527 {
528 	struct dlm_member *memb, *safe;
529 	struct dlm_config_node *node;
530 	int i, error, neg = 0, low = -1;
531 
532 	/* previously removed members that we've not finished removing need to
533 	   count as a negative change so the "neg" recovery steps will happen */
534 
535 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
536 		log_rinfo(ls, "prev removed member %d", memb->nodeid);
537 		neg++;
538 	}
539 
540 	/* move departed members from ls_nodes to ls_nodes_gone */
541 
542 	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
543 		node = find_config_node(rv, memb->nodeid);
544 		if (node && !node->new)
545 			continue;
546 
547 		if (!node) {
548 			log_rinfo(ls, "remove member %d", memb->nodeid);
549 		} else {
550 			/* removed and re-added */
551 			log_rinfo(ls, "remove member %d comm_seq %u %u",
552 				  memb->nodeid, memb->comm_seq, node->comm_seq);
553 		}
554 
555 		neg++;
556 		list_move(&memb->list, &ls->ls_nodes_gone);
557 		ls->ls_num_nodes--;
558 		dlm_lsop_recover_slot(ls, memb);
559 	}
560 
561 	/* add new members to ls_nodes */
562 
563 	for (i = 0; i < rv->nodes_count; i++) {
564 		node = &rv->nodes[i];
565 		if (dlm_is_member(ls, node->nodeid))
566 			continue;
567 		dlm_add_member(ls, node);
568 		log_rinfo(ls, "add member %d", node->nodeid);
569 	}
570 
571 	list_for_each_entry(memb, &ls->ls_nodes, list) {
572 		if (low == -1 || memb->nodeid < low)
573 			low = memb->nodeid;
574 	}
575 	ls->ls_low_nodeid = low;
576 
577 	make_member_array(ls);
578 	*neg_out = neg;
579 
580 	error = ping_members(ls);
581 	if (!error || error == -EPROTO) {
582 		/* new_lockspace() may be waiting to know if the config
583 		   is good or bad */
584 		ls->ls_members_result = error;
585 		complete(&ls->ls_members_done);
586 	}
587 
588 	log_rinfo(ls, "dlm_recover_members %d nodes", ls->ls_num_nodes);
589 	return error;
590 }
591 
592 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
593    dlm_ls_start() is called on any of them to start the new recovery. */
594 
595 int dlm_ls_stop(struct dlm_ls *ls)
596 {
597 	int new;
598 
599 	/*
600 	 * Prevent dlm_recv from being in the middle of something when we do
601 	 * the stop.  This includes ensuring dlm_recv isn't processing a
602 	 * recovery message (rcom), while dlm_recoverd is aborting and
603 	 * resetting things from an in-progress recovery.  i.e. we want
604 	 * dlm_recoverd to abort its recovery without worrying about dlm_recv
605 	 * processing an rcom at the same time.  Stopping dlm_recv also makes
606 	 * it easy for dlm_receive_message() to check locking stopped and add a
607 	 * message to the requestqueue without races.
608 	 */
609 
610 	down_write(&ls->ls_recv_active);
611 
612 	/*
613 	 * Abort any recovery that's in progress (see RECOVER_STOP,
614 	 * dlm_recovery_stopped()) and tell any other threads running in the
615 	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
616 	 */
617 
618 	spin_lock(&ls->ls_recover_lock);
619 	set_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
620 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
621 	ls->ls_recover_seq++;
622 	spin_unlock(&ls->ls_recover_lock);
623 
624 	/*
625 	 * Let dlm_recv run again, now any normal messages will be saved on the
626 	 * requestqueue for later.
627 	 */
628 
629 	up_write(&ls->ls_recv_active);
630 
631 	/*
632 	 * This in_recovery lock does two things:
633 	 * 1) Keeps this function from returning until all threads are out
634 	 *    of locking routines and locking is truly stopped.
635 	 * 2) Keeps any new requests from being processed until it's unlocked
636 	 *    when recovery is complete.
637 	 */
638 
639 	if (new) {
640 		set_bit(LSFL_RECOVER_DOWN, &ls->ls_flags);
641 		wake_up_process(ls->ls_recoverd_task);
642 		wait_event(ls->ls_recover_lock_wait,
643 			   test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags));
644 	}
645 
646 	/*
647 	 * The recoverd suspend/resume makes sure that dlm_recoverd (if
648 	 * running) has noticed RECOVER_STOP above and quit processing the
649 	 * previous recovery.
650 	 */
651 
652 	dlm_recoverd_suspend(ls);
653 
654 	spin_lock(&ls->ls_recover_lock);
655 	kfree(ls->ls_slots);
656 	ls->ls_slots = NULL;
657 	ls->ls_num_slots = 0;
658 	ls->ls_slots_size = 0;
659 	ls->ls_recover_status = 0;
660 	spin_unlock(&ls->ls_recover_lock);
661 
662 	dlm_recoverd_resume(ls);
663 
664 	if (!ls->ls_recover_begin)
665 		ls->ls_recover_begin = jiffies;
666 
667 	dlm_lsop_recover_prep(ls);
668 	return 0;
669 }
670 
671 int dlm_ls_start(struct dlm_ls *ls)
672 {
673 	struct dlm_recover *rv, *rv_old;
674 	struct dlm_config_node *nodes = NULL;
675 	int error, count;
676 
677 	rv = kzalloc(sizeof(*rv), GFP_NOFS);
678 	if (!rv)
679 		return -ENOMEM;
680 
681 	error = dlm_config_nodes(ls->ls_name, &nodes, &count);
682 	if (error < 0)
683 		goto fail_rv;
684 
685 	spin_lock(&ls->ls_recover_lock);
686 
687 	/* the lockspace needs to be stopped before it can be started */
688 
689 	if (!dlm_locking_stopped(ls)) {
690 		spin_unlock(&ls->ls_recover_lock);
691 		log_error(ls, "start ignored: lockspace running");
692 		error = -EINVAL;
693 		goto fail;
694 	}
695 
696 	rv->nodes = nodes;
697 	rv->nodes_count = count;
698 	rv->seq = ++ls->ls_recover_seq;
699 	rv_old = ls->ls_recover_args;
700 	ls->ls_recover_args = rv;
701 	spin_unlock(&ls->ls_recover_lock);
702 
703 	if (rv_old) {
704 		log_error(ls, "unused recovery %llx %d",
705 			  (unsigned long long)rv_old->seq, rv_old->nodes_count);
706 		kfree(rv_old->nodes);
707 		kfree(rv_old);
708 	}
709 
710 	set_bit(LSFL_RECOVER_WORK, &ls->ls_flags);
711 	wake_up_process(ls->ls_recoverd_task);
712 	return 0;
713 
714  fail:
715 	kfree(nodes);
716  fail_rv:
717 	kfree(rv);
718 	return error;
719 }
720 
721