1 /******************************************************************************
2 *******************************************************************************
3 **
4 **  Copyright (C) 2005-2009 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 
add_ordered_member(struct dlm_ls * ls,struct dlm_member * new)22 static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
23 {
24 	struct dlm_member *memb = NULL;
25 	struct list_head *tmp;
26 	struct list_head *newlist = &new->list;
27 	struct list_head *head = &ls->ls_nodes;
28 
29 	list_for_each(tmp, head) {
30 		memb = list_entry(tmp, struct dlm_member, list);
31 		if (new->nodeid < memb->nodeid)
32 			break;
33 	}
34 
35 	if (!memb)
36 		list_add_tail(newlist, head);
37 	else {
38 		/* FIXME: can use list macro here */
39 		newlist->prev = tmp->prev;
40 		newlist->next = tmp;
41 		tmp->prev->next = newlist;
42 		tmp->prev = newlist;
43 	}
44 }
45 
dlm_add_member(struct dlm_ls * ls,int nodeid)46 static int dlm_add_member(struct dlm_ls *ls, int nodeid)
47 {
48 	struct dlm_member *memb;
49 	int w, error;
50 
51 	memb = kzalloc(sizeof(struct dlm_member), GFP_NOFS);
52 	if (!memb)
53 		return -ENOMEM;
54 
55 	w = dlm_node_weight(ls->ls_name, nodeid);
56 	if (w < 0) {
57 		kfree(memb);
58 		return w;
59 	}
60 
61 	error = dlm_lowcomms_connect_node(nodeid);
62 	if (error < 0) {
63 		kfree(memb);
64 		return error;
65 	}
66 
67 	memb->nodeid = nodeid;
68 	memb->weight = w;
69 	add_ordered_member(ls, memb);
70 	ls->ls_num_nodes++;
71 	return 0;
72 }
73 
dlm_remove_member(struct dlm_ls * ls,struct dlm_member * memb)74 static void dlm_remove_member(struct dlm_ls *ls, struct dlm_member *memb)
75 {
76 	list_move(&memb->list, &ls->ls_nodes_gone);
77 	ls->ls_num_nodes--;
78 }
79 
dlm_is_member(struct dlm_ls * ls,int nodeid)80 int dlm_is_member(struct dlm_ls *ls, int nodeid)
81 {
82 	struct dlm_member *memb;
83 
84 	list_for_each_entry(memb, &ls->ls_nodes, list) {
85 		if (memb->nodeid == nodeid)
86 			return 1;
87 	}
88 	return 0;
89 }
90 
dlm_is_removed(struct dlm_ls * ls,int nodeid)91 int dlm_is_removed(struct dlm_ls *ls, int nodeid)
92 {
93 	struct dlm_member *memb;
94 
95 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
96 		if (memb->nodeid == nodeid)
97 			return 1;
98 	}
99 	return 0;
100 }
101 
clear_memb_list(struct list_head * head)102 static void clear_memb_list(struct list_head *head)
103 {
104 	struct dlm_member *memb;
105 
106 	while (!list_empty(head)) {
107 		memb = list_entry(head->next, struct dlm_member, list);
108 		list_del(&memb->list);
109 		kfree(memb);
110 	}
111 }
112 
dlm_clear_members(struct dlm_ls * ls)113 void dlm_clear_members(struct dlm_ls *ls)
114 {
115 	clear_memb_list(&ls->ls_nodes);
116 	ls->ls_num_nodes = 0;
117 }
118 
dlm_clear_members_gone(struct dlm_ls * ls)119 void dlm_clear_members_gone(struct dlm_ls *ls)
120 {
121 	clear_memb_list(&ls->ls_nodes_gone);
122 }
123 
make_member_array(struct dlm_ls * ls)124 static void make_member_array(struct dlm_ls *ls)
125 {
126 	struct dlm_member *memb;
127 	int i, w, x = 0, total = 0, all_zero = 0, *array;
128 
129 	kfree(ls->ls_node_array);
130 	ls->ls_node_array = NULL;
131 
132 	list_for_each_entry(memb, &ls->ls_nodes, list) {
133 		if (memb->weight)
134 			total += memb->weight;
135 	}
136 
137 	/* all nodes revert to weight of 1 if all have weight 0 */
138 
139 	if (!total) {
140 		total = ls->ls_num_nodes;
141 		all_zero = 1;
142 	}
143 
144 	ls->ls_total_weight = total;
145 
146 	array = kmalloc(sizeof(int) * total, GFP_NOFS);
147 	if (!array)
148 		return;
149 
150 	list_for_each_entry(memb, &ls->ls_nodes, list) {
151 		if (!all_zero && !memb->weight)
152 			continue;
153 
154 		if (all_zero)
155 			w = 1;
156 		else
157 			w = memb->weight;
158 
159 		DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
160 
161 		for (i = 0; i < w; i++)
162 			array[x++] = memb->nodeid;
163 	}
164 
165 	ls->ls_node_array = array;
166 }
167 
168 /* send a status request to all members just to establish comms connections */
169 
ping_members(struct dlm_ls * ls)170 static int ping_members(struct dlm_ls *ls)
171 {
172 	struct dlm_member *memb;
173 	int error = 0;
174 
175 	list_for_each_entry(memb, &ls->ls_nodes, list) {
176 		error = dlm_recovery_stopped(ls);
177 		if (error)
178 			break;
179 		error = dlm_rcom_status(ls, memb->nodeid);
180 		if (error)
181 			break;
182 	}
183 	if (error)
184 		log_debug(ls, "ping_members aborted %d last nodeid %d",
185 			  error, ls->ls_recover_nodeid);
186 	return error;
187 }
188 
dlm_recover_members(struct dlm_ls * ls,struct dlm_recover * rv,int * neg_out)189 int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
190 {
191 	struct dlm_member *memb, *safe;
192 	int i, error, found, pos = 0, neg = 0, low = -1;
193 
194 	/* previously removed members that we've not finished removing need to
195 	   count as a negative change so the "neg" recovery steps will happen */
196 
197 	list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
198 		log_debug(ls, "prev removed member %d", memb->nodeid);
199 		neg++;
200 	}
201 
202 	/* move departed members from ls_nodes to ls_nodes_gone */
203 
204 	list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
205 		found = 0;
206 		for (i = 0; i < rv->node_count; i++) {
207 			if (memb->nodeid == rv->nodeids[i]) {
208 				found = 1;
209 				break;
210 			}
211 		}
212 
213 		if (!found) {
214 			neg++;
215 			dlm_remove_member(ls, memb);
216 			log_debug(ls, "remove member %d", memb->nodeid);
217 		}
218 	}
219 
220 	/* Add an entry to ls_nodes_gone for members that were removed and
221 	   then added again, so that previous state for these nodes will be
222 	   cleared during recovery. */
223 
224 	for (i = 0; i < rv->new_count; i++) {
225 		if (!dlm_is_member(ls, rv->new[i]))
226 			continue;
227 		log_debug(ls, "new nodeid %d is a re-added member", rv->new[i]);
228 
229 		memb = kzalloc(sizeof(struct dlm_member), GFP_NOFS);
230 		if (!memb)
231 			return -ENOMEM;
232 		memb->nodeid = rv->new[i];
233 		list_add_tail(&memb->list, &ls->ls_nodes_gone);
234 		neg++;
235 	}
236 
237 	/* add new members to ls_nodes */
238 
239 	for (i = 0; i < rv->node_count; i++) {
240 		if (dlm_is_member(ls, rv->nodeids[i]))
241 			continue;
242 		dlm_add_member(ls, rv->nodeids[i]);
243 		pos++;
244 		log_debug(ls, "add member %d", rv->nodeids[i]);
245 	}
246 
247 	list_for_each_entry(memb, &ls->ls_nodes, list) {
248 		if (low == -1 || memb->nodeid < low)
249 			low = memb->nodeid;
250 	}
251 	ls->ls_low_nodeid = low;
252 
253 	make_member_array(ls);
254 	dlm_set_recover_status(ls, DLM_RS_NODES);
255 	*neg_out = neg;
256 
257 	error = ping_members(ls);
258 	if (!error || error == -EPROTO) {
259 		/* new_lockspace() may be waiting to know if the config
260 		   is good or bad */
261 		ls->ls_members_result = error;
262 		complete(&ls->ls_members_done);
263 	}
264 	if (error)
265 		goto out;
266 
267 	error = dlm_recover_members_wait(ls);
268  out:
269 	log_debug(ls, "total members %d error %d", ls->ls_num_nodes, error);
270 	return error;
271 }
272 
273 /* Userspace guarantees that dlm_ls_stop() has completed on all nodes before
274    dlm_ls_start() is called on any of them to start the new recovery. */
275 
dlm_ls_stop(struct dlm_ls * ls)276 int dlm_ls_stop(struct dlm_ls *ls)
277 {
278 	int new;
279 
280 	/*
281 	 * Prevent dlm_recv from being in the middle of something when we do
282 	 * the stop.  This includes ensuring dlm_recv isn't processing a
283 	 * recovery message (rcom), while dlm_recoverd is aborting and
284 	 * resetting things from an in-progress recovery.  i.e. we want
285 	 * dlm_recoverd to abort its recovery without worrying about dlm_recv
286 	 * processing an rcom at the same time.  Stopping dlm_recv also makes
287 	 * it easy for dlm_receive_message() to check locking stopped and add a
288 	 * message to the requestqueue without races.
289 	 */
290 
291 	down_write(&ls->ls_recv_active);
292 
293 	/*
294 	 * Abort any recovery that's in progress (see RECOVERY_STOP,
295 	 * dlm_recovery_stopped()) and tell any other threads running in the
296 	 * dlm to quit any processing (see RUNNING, dlm_locking_stopped()).
297 	 */
298 
299 	spin_lock(&ls->ls_recover_lock);
300 	set_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
301 	new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
302 	ls->ls_recover_seq++;
303 	spin_unlock(&ls->ls_recover_lock);
304 
305 	/*
306 	 * Let dlm_recv run again, now any normal messages will be saved on the
307 	 * requestqueue for later.
308 	 */
309 
310 	up_write(&ls->ls_recv_active);
311 
312 	/*
313 	 * This in_recovery lock does two things:
314 	 * 1) Keeps this function from returning until all threads are out
315 	 *    of locking routines and locking is truly stopped.
316 	 * 2) Keeps any new requests from being processed until it's unlocked
317 	 *    when recovery is complete.
318 	 */
319 
320 	if (new)
321 		down_write(&ls->ls_in_recovery);
322 
323 	/*
324 	 * The recoverd suspend/resume makes sure that dlm_recoverd (if
325 	 * running) has noticed RECOVERY_STOP above and quit processing the
326 	 * previous recovery.
327 	 */
328 
329 	dlm_recoverd_suspend(ls);
330 	ls->ls_recover_status = 0;
331 	dlm_recoverd_resume(ls);
332 
333 	if (!ls->ls_recover_begin)
334 		ls->ls_recover_begin = jiffies;
335 	return 0;
336 }
337 
dlm_ls_start(struct dlm_ls * ls)338 int dlm_ls_start(struct dlm_ls *ls)
339 {
340 	struct dlm_recover *rv = NULL, *rv_old;
341 	int *ids = NULL, *new = NULL;
342 	int error, ids_count = 0, new_count = 0;
343 
344 	rv = kzalloc(sizeof(struct dlm_recover), GFP_NOFS);
345 	if (!rv)
346 		return -ENOMEM;
347 
348 	error = dlm_nodeid_list(ls->ls_name, &ids, &ids_count,
349 				&new, &new_count);
350 	if (error < 0)
351 		goto fail;
352 
353 	spin_lock(&ls->ls_recover_lock);
354 
355 	/* the lockspace needs to be stopped before it can be started */
356 
357 	if (!dlm_locking_stopped(ls)) {
358 		spin_unlock(&ls->ls_recover_lock);
359 		log_error(ls, "start ignored: lockspace running");
360 		error = -EINVAL;
361 		goto fail;
362 	}
363 
364 	rv->nodeids = ids;
365 	rv->node_count = ids_count;
366 	rv->new = new;
367 	rv->new_count = new_count;
368 	rv->seq = ++ls->ls_recover_seq;
369 	rv_old = ls->ls_recover_args;
370 	ls->ls_recover_args = rv;
371 	spin_unlock(&ls->ls_recover_lock);
372 
373 	if (rv_old) {
374 		log_error(ls, "unused recovery %llx %d",
375 			  (unsigned long long)rv_old->seq, rv_old->node_count);
376 		kfree(rv_old->nodeids);
377 		kfree(rv_old->new);
378 		kfree(rv_old);
379 	}
380 
381 	dlm_recoverd_kick(ls);
382 	return 0;
383 
384  fail:
385 	kfree(rv);
386 	kfree(ids);
387 	kfree(new);
388 	return error;
389 }
390 
391