blob: a3f7de7f3a8f969b9500875203297f4b8d6872ef [file] [log] [blame]
David Teiglande7fd4172006-01-18 09:30:29 +00001/******************************************************************************
2*******************************************************************************
3**
4** Copyright (C) 2005 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"
David Teiglande7fd4172006-01-18 09:30:29 +000018#include "rcom.h"
19#include "config.h"
20
21/*
22 * Following called by dlm_recoverd thread
23 */
24
25static void add_ordered_member(struct dlm_ls *ls, struct dlm_member *new)
26{
27 struct dlm_member *memb = NULL;
28 struct list_head *tmp;
29 struct list_head *newlist = &new->list;
30 struct list_head *head = &ls->ls_nodes;
31
32 list_for_each(tmp, head) {
33 memb = list_entry(tmp, struct dlm_member, list);
34 if (new->nodeid < memb->nodeid)
35 break;
36 }
37
38 if (!memb)
39 list_add_tail(newlist, head);
40 else {
41 /* FIXME: can use list macro here */
42 newlist->prev = tmp->prev;
43 newlist->next = tmp;
44 tmp->prev->next = newlist;
45 tmp->prev = newlist;
46 }
47}
48
49static int dlm_add_member(struct dlm_ls *ls, int nodeid)
50{
51 struct dlm_member *memb;
52 int w;
53
David Teigland90135922006-01-20 08:47:07 +000054 memb = kzalloc(sizeof(struct dlm_member), GFP_KERNEL);
David Teiglande7fd4172006-01-18 09:30:29 +000055 if (!memb)
56 return -ENOMEM;
57
58 w = dlm_node_weight(ls->ls_name, nodeid);
59 if (w < 0)
60 return w;
61
62 memb->nodeid = nodeid;
63 memb->weight = w;
64 add_ordered_member(ls, memb);
65 ls->ls_num_nodes++;
66 return 0;
67}
68
69static void dlm_remove_member(struct dlm_ls *ls, struct dlm_member *memb)
70{
71 list_move(&memb->list, &ls->ls_nodes_gone);
72 ls->ls_num_nodes--;
73}
74
75static int dlm_is_member(struct dlm_ls *ls, int nodeid)
76{
77 struct dlm_member *memb;
78
79 list_for_each_entry(memb, &ls->ls_nodes, list) {
80 if (memb->nodeid == nodeid)
David Teigland90135922006-01-20 08:47:07 +000081 return 1;
David Teiglande7fd4172006-01-18 09:30:29 +000082 }
David Teigland90135922006-01-20 08:47:07 +000083 return 0;
David Teiglande7fd4172006-01-18 09:30:29 +000084}
85
86int dlm_is_removed(struct dlm_ls *ls, int nodeid)
87{
88 struct dlm_member *memb;
89
90 list_for_each_entry(memb, &ls->ls_nodes_gone, list) {
91 if (memb->nodeid == nodeid)
David Teigland90135922006-01-20 08:47:07 +000092 return 1;
David Teiglande7fd4172006-01-18 09:30:29 +000093 }
David Teigland90135922006-01-20 08:47:07 +000094 return 0;
David Teiglande7fd4172006-01-18 09:30:29 +000095}
96
97static void clear_memb_list(struct list_head *head)
98{
99 struct dlm_member *memb;
100
101 while (!list_empty(head)) {
102 memb = list_entry(head->next, struct dlm_member, list);
103 list_del(&memb->list);
104 kfree(memb);
105 }
106}
107
108void dlm_clear_members(struct dlm_ls *ls)
109{
110 clear_memb_list(&ls->ls_nodes);
111 ls->ls_num_nodes = 0;
112}
113
114void dlm_clear_members_gone(struct dlm_ls *ls)
115{
116 clear_memb_list(&ls->ls_nodes_gone);
117}
118
119static void make_member_array(struct dlm_ls *ls)
120{
121 struct dlm_member *memb;
122 int i, w, x = 0, total = 0, all_zero = 0, *array;
123
124 kfree(ls->ls_node_array);
125 ls->ls_node_array = NULL;
126
127 list_for_each_entry(memb, &ls->ls_nodes, list) {
128 if (memb->weight)
129 total += memb->weight;
130 }
131
132 /* all nodes revert to weight of 1 if all have weight 0 */
133
134 if (!total) {
135 total = ls->ls_num_nodes;
136 all_zero = 1;
137 }
138
139 ls->ls_total_weight = total;
140
141 array = kmalloc(sizeof(int) * total, GFP_KERNEL);
142 if (!array)
143 return;
144
145 list_for_each_entry(memb, &ls->ls_nodes, list) {
146 if (!all_zero && !memb->weight)
147 continue;
148
149 if (all_zero)
150 w = 1;
151 else
152 w = memb->weight;
153
154 DLM_ASSERT(x < total, printk("total %d x %d\n", total, x););
155
156 for (i = 0; i < w; i++)
157 array[x++] = memb->nodeid;
158 }
159
160 ls->ls_node_array = array;
161}
162
163/* send a status request to all members just to establish comms connections */
164
David Teiglandf6db1b82006-08-08 17:06:07 -0500165static int ping_members(struct dlm_ls *ls)
David Teiglande7fd4172006-01-18 09:30:29 +0000166{
167 struct dlm_member *memb;
David Teiglandf6db1b82006-08-08 17:06:07 -0500168 int error = 0;
169
170 list_for_each_entry(memb, &ls->ls_nodes, list) {
171 error = dlm_recovery_stopped(ls);
172 if (error)
173 break;
174 error = dlm_rcom_status(ls, memb->nodeid);
175 if (error)
176 break;
177 }
178 if (error)
David Teiglandfaa0f262006-08-08 17:08:42 -0500179 log_debug(ls, "ping_members aborted %d last nodeid %d",
180 error, ls->ls_recover_nodeid);
David Teiglandf6db1b82006-08-08 17:06:07 -0500181 return error;
David Teiglande7fd4172006-01-18 09:30:29 +0000182}
183
184int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out)
185{
186 struct dlm_member *memb, *safe;
187 int i, error, found, pos = 0, neg = 0, low = -1;
188
189 /* move departed members from ls_nodes to ls_nodes_gone */
190
191 list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) {
David Teigland90135922006-01-20 08:47:07 +0000192 found = 0;
David Teiglande7fd4172006-01-18 09:30:29 +0000193 for (i = 0; i < rv->node_count; i++) {
194 if (memb->nodeid == rv->nodeids[i]) {
David Teigland90135922006-01-20 08:47:07 +0000195 found = 1;
David Teiglande7fd4172006-01-18 09:30:29 +0000196 break;
197 }
198 }
199
200 if (!found) {
201 neg++;
202 dlm_remove_member(ls, memb);
203 log_debug(ls, "remove member %d", memb->nodeid);
204 }
205 }
206
207 /* add new members to ls_nodes */
208
209 for (i = 0; i < rv->node_count; i++) {
210 if (dlm_is_member(ls, rv->nodeids[i]))
211 continue;
212 dlm_add_member(ls, rv->nodeids[i]);
213 pos++;
214 log_debug(ls, "add member %d", rv->nodeids[i]);
215 }
216
217 list_for_each_entry(memb, &ls->ls_nodes, list) {
218 if (low == -1 || memb->nodeid < low)
219 low = memb->nodeid;
220 }
221 ls->ls_low_nodeid = low;
222
223 make_member_array(ls);
224 dlm_set_recover_status(ls, DLM_RS_NODES);
225 *neg_out = neg;
226
David Teiglandf6db1b82006-08-08 17:06:07 -0500227 error = ping_members(ls);
228 if (error)
229 goto out;
David Teiglande7fd4172006-01-18 09:30:29 +0000230
231 error = dlm_recover_members_wait(ls);
David Teiglandf6db1b82006-08-08 17:06:07 -0500232 out:
233 log_debug(ls, "total members %d error %d", ls->ls_num_nodes, error);
David Teiglande7fd4172006-01-18 09:30:29 +0000234 return error;
235}
236
237/*
238 * Following called from lockspace.c
239 */
240
241int dlm_ls_stop(struct dlm_ls *ls)
242{
243 int new;
244
245 /*
246 * A stop cancels any recovery that's in progress (see RECOVERY_STOP,
247 * dlm_recovery_stopped()) and prevents any new locks from being
248 * processed (see RUNNING, dlm_locking_stopped()).
249 */
250
251 spin_lock(&ls->ls_recover_lock);
252 set_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
253 new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags);
254 ls->ls_recover_seq++;
255 spin_unlock(&ls->ls_recover_lock);
256
257 /*
258 * This in_recovery lock does two things:
259 *
260 * 1) Keeps this function from returning until all threads are out
261 * of locking routines and locking is truely stopped.
262 * 2) Keeps any new requests from being processed until it's unlocked
263 * when recovery is complete.
264 */
265
266 if (new)
267 down_write(&ls->ls_in_recovery);
268
269 /*
270 * The recoverd suspend/resume makes sure that dlm_recoverd (if
271 * running) has noticed the clearing of RUNNING above and quit
272 * processing the previous recovery. This will be true for all nodes
273 * before any nodes start the new recovery.
274 */
275
276 dlm_recoverd_suspend(ls);
277 ls->ls_recover_status = 0;
278 dlm_recoverd_resume(ls);
279 return 0;
280}
281
282int dlm_ls_start(struct dlm_ls *ls)
283{
284 struct dlm_recover *rv = NULL, *rv_old;
285 int *ids = NULL;
286 int error, count;
287
David Teigland90135922006-01-20 08:47:07 +0000288 rv = kzalloc(sizeof(struct dlm_recover), GFP_KERNEL);
David Teiglande7fd4172006-01-18 09:30:29 +0000289 if (!rv)
290 return -ENOMEM;
David Teiglande7fd4172006-01-18 09:30:29 +0000291
292 error = count = dlm_nodeid_list(ls->ls_name, &ids);
293 if (error <= 0)
294 goto fail;
295
296 spin_lock(&ls->ls_recover_lock);
297
298 /* the lockspace needs to be stopped before it can be started */
299
300 if (!dlm_locking_stopped(ls)) {
301 spin_unlock(&ls->ls_recover_lock);
302 log_error(ls, "start ignored: lockspace running");
303 error = -EINVAL;
304 goto fail;
305 }
306
307 rv->nodeids = ids;
308 rv->node_count = count;
309 rv->seq = ++ls->ls_recover_seq;
310 rv_old = ls->ls_recover_args;
311 ls->ls_recover_args = rv;
312 spin_unlock(&ls->ls_recover_lock);
313
314 if (rv_old) {
315 kfree(rv_old->nodeids);
316 kfree(rv_old);
317 }
318
319 dlm_recoverd_kick(ls);
320 return 0;
321
322 fail:
323 kfree(rv);
324 kfree(ids);
325 return error;
326}
327