Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
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" | |
e7fd4179 DT |
18 | #include "rcom.h" |
19 | #include "config.h" | |
20 | ||
21 | /* | |
22 | * Following called by dlm_recoverd thread | |
23 | */ | |
24 | ||
25 | static 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 | ||
49 | static int dlm_add_member(struct dlm_ls *ls, int nodeid) | |
50 | { | |
51 | struct dlm_member *memb; | |
52 | int w; | |
53 | ||
90135925 | 54 | memb = kzalloc(sizeof(struct dlm_member), GFP_KERNEL); |
e7fd4179 DT |
55 | 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 | ||
69 | static 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 | ||
75 | static 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) | |
90135925 | 81 | return 1; |
e7fd4179 | 82 | } |
90135925 | 83 | return 0; |
e7fd4179 DT |
84 | } |
85 | ||
86 | int 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) | |
90135925 | 92 | return 1; |
e7fd4179 | 93 | } |
90135925 | 94 | return 0; |
e7fd4179 DT |
95 | } |
96 | ||
97 | static 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 | ||
108 | void dlm_clear_members(struct dlm_ls *ls) | |
109 | { | |
110 | clear_memb_list(&ls->ls_nodes); | |
111 | ls->ls_num_nodes = 0; | |
112 | } | |
113 | ||
114 | void dlm_clear_members_gone(struct dlm_ls *ls) | |
115 | { | |
116 | clear_memb_list(&ls->ls_nodes_gone); | |
117 | } | |
118 | ||
119 | static 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 | ||
165 | static void ping_members(struct dlm_ls *ls) | |
166 | { | |
167 | struct dlm_member *memb; | |
168 | list_for_each_entry(memb, &ls->ls_nodes, list) | |
169 | dlm_rcom_status(ls, memb->nodeid); | |
170 | } | |
171 | ||
172 | int dlm_recover_members(struct dlm_ls *ls, struct dlm_recover *rv, int *neg_out) | |
173 | { | |
174 | struct dlm_member *memb, *safe; | |
175 | int i, error, found, pos = 0, neg = 0, low = -1; | |
176 | ||
177 | /* move departed members from ls_nodes to ls_nodes_gone */ | |
178 | ||
179 | list_for_each_entry_safe(memb, safe, &ls->ls_nodes, list) { | |
90135925 | 180 | found = 0; |
e7fd4179 DT |
181 | for (i = 0; i < rv->node_count; i++) { |
182 | if (memb->nodeid == rv->nodeids[i]) { | |
90135925 | 183 | found = 1; |
e7fd4179 DT |
184 | break; |
185 | } | |
186 | } | |
187 | ||
188 | if (!found) { | |
189 | neg++; | |
190 | dlm_remove_member(ls, memb); | |
191 | log_debug(ls, "remove member %d", memb->nodeid); | |
192 | } | |
193 | } | |
194 | ||
195 | /* add new members to ls_nodes */ | |
196 | ||
197 | for (i = 0; i < rv->node_count; i++) { | |
198 | if (dlm_is_member(ls, rv->nodeids[i])) | |
199 | continue; | |
200 | dlm_add_member(ls, rv->nodeids[i]); | |
201 | pos++; | |
202 | log_debug(ls, "add member %d", rv->nodeids[i]); | |
203 | } | |
204 | ||
205 | list_for_each_entry(memb, &ls->ls_nodes, list) { | |
206 | if (low == -1 || memb->nodeid < low) | |
207 | low = memb->nodeid; | |
208 | } | |
209 | ls->ls_low_nodeid = low; | |
210 | ||
211 | make_member_array(ls); | |
212 | dlm_set_recover_status(ls, DLM_RS_NODES); | |
213 | *neg_out = neg; | |
214 | ||
215 | ping_members(ls); | |
216 | ||
217 | error = dlm_recover_members_wait(ls); | |
218 | log_debug(ls, "total members %d", ls->ls_num_nodes); | |
219 | return error; | |
220 | } | |
221 | ||
222 | /* | |
223 | * Following called from lockspace.c | |
224 | */ | |
225 | ||
226 | int dlm_ls_stop(struct dlm_ls *ls) | |
227 | { | |
228 | int new; | |
229 | ||
230 | /* | |
231 | * A stop cancels any recovery that's in progress (see RECOVERY_STOP, | |
232 | * dlm_recovery_stopped()) and prevents any new locks from being | |
233 | * processed (see RUNNING, dlm_locking_stopped()). | |
234 | */ | |
235 | ||
236 | spin_lock(&ls->ls_recover_lock); | |
237 | set_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); | |
238 | new = test_and_clear_bit(LSFL_RUNNING, &ls->ls_flags); | |
239 | ls->ls_recover_seq++; | |
240 | spin_unlock(&ls->ls_recover_lock); | |
241 | ||
242 | /* | |
243 | * This in_recovery lock does two things: | |
244 | * | |
245 | * 1) Keeps this function from returning until all threads are out | |
246 | * of locking routines and locking is truely stopped. | |
247 | * 2) Keeps any new requests from being processed until it's unlocked | |
248 | * when recovery is complete. | |
249 | */ | |
250 | ||
251 | if (new) | |
252 | down_write(&ls->ls_in_recovery); | |
253 | ||
254 | /* | |
255 | * The recoverd suspend/resume makes sure that dlm_recoverd (if | |
256 | * running) has noticed the clearing of RUNNING above and quit | |
257 | * processing the previous recovery. This will be true for all nodes | |
258 | * before any nodes start the new recovery. | |
259 | */ | |
260 | ||
261 | dlm_recoverd_suspend(ls); | |
262 | ls->ls_recover_status = 0; | |
263 | dlm_recoverd_resume(ls); | |
264 | return 0; | |
265 | } | |
266 | ||
267 | int dlm_ls_start(struct dlm_ls *ls) | |
268 | { | |
269 | struct dlm_recover *rv = NULL, *rv_old; | |
270 | int *ids = NULL; | |
271 | int error, count; | |
272 | ||
90135925 | 273 | rv = kzalloc(sizeof(struct dlm_recover), GFP_KERNEL); |
e7fd4179 DT |
274 | if (!rv) |
275 | return -ENOMEM; | |
e7fd4179 DT |
276 | |
277 | error = count = dlm_nodeid_list(ls->ls_name, &ids); | |
278 | if (error <= 0) | |
279 | goto fail; | |
280 | ||
281 | spin_lock(&ls->ls_recover_lock); | |
282 | ||
283 | /* the lockspace needs to be stopped before it can be started */ | |
284 | ||
285 | if (!dlm_locking_stopped(ls)) { | |
286 | spin_unlock(&ls->ls_recover_lock); | |
287 | log_error(ls, "start ignored: lockspace running"); | |
288 | error = -EINVAL; | |
289 | goto fail; | |
290 | } | |
291 | ||
292 | rv->nodeids = ids; | |
293 | rv->node_count = count; | |
294 | rv->seq = ++ls->ls_recover_seq; | |
295 | rv_old = ls->ls_recover_args; | |
296 | ls->ls_recover_args = rv; | |
297 | spin_unlock(&ls->ls_recover_lock); | |
298 | ||
299 | if (rv_old) { | |
300 | kfree(rv_old->nodeids); | |
301 | kfree(rv_old); | |
302 | } | |
303 | ||
304 | dlm_recoverd_kick(ls); | |
305 | return 0; | |
306 | ||
307 | fail: | |
308 | kfree(rv); | |
309 | kfree(ids); | |
310 | return error; | |
311 | } | |
312 |