Commit | Line | Data |
---|---|---|
e7fd4179 DT |
1 | /****************************************************************************** |
2 | ******************************************************************************* | |
3 | ** | |
4 | ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
3ae1acf9 | 5 | ** Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved. |
e7fd4179 DT |
6 | ** |
7 | ** This copyrighted material is made available to anyone wishing to use, | |
8 | ** modify, copy, or redistribute it subject to the terms and conditions | |
9 | ** of the GNU General Public License v.2. | |
10 | ** | |
11 | ******************************************************************************* | |
12 | ******************************************************************************/ | |
13 | ||
14 | #include "dlm_internal.h" | |
15 | #include "lockspace.h" | |
16 | #include "member.h" | |
17 | #include "dir.h" | |
18 | #include "ast.h" | |
19 | #include "recover.h" | |
20 | #include "lowcomms.h" | |
21 | #include "lock.h" | |
22 | #include "requestqueue.h" | |
23 | #include "recoverd.h" | |
24 | ||
25 | ||
26 | /* If the start for which we're re-enabling locking (seq) has been superseded | |
c36258b5 DT |
27 | by a newer stop (ls_recover_seq), we need to leave locking disabled. |
28 | ||
29 | We suspend dlm_recv threads here to avoid the race where dlm_recv a) sees | |
30 | locking stopped and b) adds a message to the requestqueue, but dlm_recoverd | |
31 | enables locking and clears the requestqueue between a and b. */ | |
e7fd4179 DT |
32 | |
33 | static int enable_locking(struct dlm_ls *ls, uint64_t seq) | |
34 | { | |
35 | int error = -EINTR; | |
36 | ||
c36258b5 DT |
37 | down_write(&ls->ls_recv_active); |
38 | ||
e7fd4179 DT |
39 | spin_lock(&ls->ls_recover_lock); |
40 | if (ls->ls_recover_seq == seq) { | |
41 | set_bit(LSFL_RUNNING, &ls->ls_flags); | |
c36258b5 | 42 | /* unblocks processes waiting to enter the dlm */ |
e7fd4179 DT |
43 | up_write(&ls->ls_in_recovery); |
44 | error = 0; | |
45 | } | |
46 | spin_unlock(&ls->ls_recover_lock); | |
c36258b5 DT |
47 | |
48 | up_write(&ls->ls_recv_active); | |
e7fd4179 DT |
49 | return error; |
50 | } | |
51 | ||
52 | static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv) | |
53 | { | |
54 | unsigned long start; | |
55 | int error, neg = 0; | |
56 | ||
57adf7ee | 57 | log_debug(ls, "recover %llx", (unsigned long long)rv->seq); |
e7fd4179 | 58 | |
90135925 | 59 | mutex_lock(&ls->ls_recoverd_active); |
e7fd4179 DT |
60 | |
61 | /* | |
62 | * Suspending and resuming dlm_astd ensures that no lkb's from this ls | |
63 | * will be processed by dlm_astd during recovery. | |
64 | */ | |
65 | ||
66 | dlm_astd_suspend(); | |
67 | dlm_astd_resume(); | |
68 | ||
69 | /* | |
85f0379a DT |
70 | * Free non-master tossed rsb's. Master rsb's are kept on toss |
71 | * list and put on root list to be included in resdir recovery. | |
e7fd4179 DT |
72 | */ |
73 | ||
85f0379a | 74 | dlm_clear_toss_list(ls); |
e7fd4179 DT |
75 | |
76 | /* | |
85f0379a DT |
77 | * This list of root rsb's will be the basis of most of the recovery |
78 | * routines. | |
e7fd4179 DT |
79 | */ |
80 | ||
85f0379a | 81 | dlm_create_root_list(ls); |
e7fd4179 DT |
82 | |
83 | /* | |
84 | * Add or remove nodes from the lockspace's ls_nodes list. | |
85 | * Also waits for all nodes to complete dlm_recover_members. | |
86 | */ | |
87 | ||
88 | error = dlm_recover_members(ls, rv, &neg); | |
89 | if (error) { | |
8ec68867 | 90 | log_debug(ls, "recover_members failed %d", error); |
e7fd4179 DT |
91 | goto fail; |
92 | } | |
93 | start = jiffies; | |
94 | ||
95 | /* | |
96 | * Rebuild our own share of the directory by collecting from all other | |
97 | * nodes their master rsb names that hash to us. | |
98 | */ | |
99 | ||
100 | error = dlm_recover_directory(ls); | |
101 | if (error) { | |
8ec68867 | 102 | log_debug(ls, "recover_directory failed %d", error); |
e7fd4179 DT |
103 | goto fail; |
104 | } | |
105 | ||
e7fd4179 DT |
106 | /* |
107 | * Wait for all nodes to complete directory rebuild. | |
108 | */ | |
109 | ||
110 | error = dlm_recover_directory_wait(ls); | |
111 | if (error) { | |
8ec68867 | 112 | log_debug(ls, "recover_directory_wait failed %d", error); |
e7fd4179 DT |
113 | goto fail; |
114 | } | |
115 | ||
116 | /* | |
117 | * We may have outstanding operations that are waiting for a reply from | |
118 | * a failed node. Mark these to be resent after recovery. Unlock and | |
119 | * cancel ops can just be completed. | |
120 | */ | |
121 | ||
122 | dlm_recover_waiters_pre(ls); | |
123 | ||
124 | error = dlm_recovery_stopped(ls); | |
125 | if (error) | |
126 | goto fail; | |
127 | ||
128 | if (neg || dlm_no_directory(ls)) { | |
129 | /* | |
130 | * Clear lkb's for departed nodes. | |
131 | */ | |
132 | ||
133 | dlm_purge_locks(ls); | |
134 | ||
135 | /* | |
136 | * Get new master nodeid's for rsb's that were mastered on | |
137 | * departed nodes. | |
138 | */ | |
139 | ||
140 | error = dlm_recover_masters(ls); | |
141 | if (error) { | |
8ec68867 | 142 | log_debug(ls, "recover_masters failed %d", error); |
e7fd4179 DT |
143 | goto fail; |
144 | } | |
145 | ||
146 | /* | |
147 | * Send our locks on remastered rsb's to the new masters. | |
148 | */ | |
149 | ||
150 | error = dlm_recover_locks(ls); | |
151 | if (error) { | |
8ec68867 | 152 | log_debug(ls, "recover_locks failed %d", error); |
e7fd4179 DT |
153 | goto fail; |
154 | } | |
155 | ||
156 | error = dlm_recover_locks_wait(ls); | |
157 | if (error) { | |
8ec68867 | 158 | log_debug(ls, "recover_locks_wait failed %d", error); |
e7fd4179 DT |
159 | goto fail; |
160 | } | |
161 | ||
162 | /* | |
163 | * Finalize state in master rsb's now that all locks can be | |
164 | * checked. This includes conversion resolution and lvb | |
165 | * settings. | |
166 | */ | |
167 | ||
168 | dlm_recover_rsbs(ls); | |
91c0dc93 DT |
169 | } else { |
170 | /* | |
171 | * Other lockspace members may be going through the "neg" steps | |
172 | * while also adding us to the lockspace, in which case they'll | |
4b77f2c9 | 173 | * be doing the recover_locks (RS_LOCKS) barrier. |
91c0dc93 DT |
174 | */ |
175 | dlm_set_recover_status(ls, DLM_RS_LOCKS); | |
4b77f2c9 DT |
176 | |
177 | error = dlm_recover_locks_wait(ls); | |
178 | if (error) { | |
8ec68867 | 179 | log_debug(ls, "recover_locks_wait failed %d", error); |
4b77f2c9 DT |
180 | goto fail; |
181 | } | |
e7fd4179 DT |
182 | } |
183 | ||
184 | dlm_release_root_list(ls); | |
185 | ||
2896ee37 DT |
186 | /* |
187 | * Purge directory-related requests that are saved in requestqueue. | |
188 | * All dir requests from before recovery are invalid now due to the dir | |
189 | * rebuild and will be resent by the requesting nodes. | |
190 | */ | |
191 | ||
192 | dlm_purge_requestqueue(ls); | |
193 | ||
e7fd4179 DT |
194 | dlm_set_recover_status(ls, DLM_RS_DONE); |
195 | error = dlm_recover_done_wait(ls); | |
196 | if (error) { | |
8ec68867 | 197 | log_debug(ls, "recover_done_wait failed %d", error); |
e7fd4179 DT |
198 | goto fail; |
199 | } | |
200 | ||
201 | dlm_clear_members_gone(ls); | |
202 | ||
3ae1acf9 DT |
203 | dlm_adjust_timeouts(ls); |
204 | ||
e7fd4179 DT |
205 | error = enable_locking(ls, rv->seq); |
206 | if (error) { | |
8ec68867 | 207 | log_debug(ls, "enable_locking failed %d", error); |
e7fd4179 DT |
208 | goto fail; |
209 | } | |
210 | ||
211 | error = dlm_process_requestqueue(ls); | |
212 | if (error) { | |
8ec68867 | 213 | log_debug(ls, "process_requestqueue failed %d", error); |
e7fd4179 DT |
214 | goto fail; |
215 | } | |
216 | ||
217 | error = dlm_recover_waiters_post(ls); | |
218 | if (error) { | |
8ec68867 | 219 | log_debug(ls, "recover_waiters_post failed %d", error); |
e7fd4179 DT |
220 | goto fail; |
221 | } | |
222 | ||
223 | dlm_grant_after_purge(ls); | |
224 | ||
225 | dlm_astd_wake(); | |
226 | ||
57adf7ee RK |
227 | log_debug(ls, "recover %llx done: %u ms", |
228 | (unsigned long long)rv->seq, | |
e7fd4179 | 229 | jiffies_to_msecs(jiffies - start)); |
90135925 | 230 | mutex_unlock(&ls->ls_recoverd_active); |
e7fd4179 DT |
231 | |
232 | return 0; | |
233 | ||
234 | fail: | |
235 | dlm_release_root_list(ls); | |
57adf7ee RK |
236 | log_debug(ls, "recover %llx error %d", |
237 | (unsigned long long)rv->seq, error); | |
90135925 | 238 | mutex_unlock(&ls->ls_recoverd_active); |
e7fd4179 DT |
239 | return error; |
240 | } | |
241 | ||
2cdc98aa DT |
242 | /* The dlm_ls_start() that created the rv we take here may already have been |
243 | stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP | |
244 | flag set. */ | |
245 | ||
e7fd4179 DT |
246 | static void do_ls_recovery(struct dlm_ls *ls) |
247 | { | |
248 | struct dlm_recover *rv = NULL; | |
249 | ||
250 | spin_lock(&ls->ls_recover_lock); | |
251 | rv = ls->ls_recover_args; | |
252 | ls->ls_recover_args = NULL; | |
2cdc98aa DT |
253 | if (rv && ls->ls_recover_seq == rv->seq) |
254 | clear_bit(LSFL_RECOVERY_STOP, &ls->ls_flags); | |
e7fd4179 DT |
255 | spin_unlock(&ls->ls_recover_lock); |
256 | ||
257 | if (rv) { | |
258 | ls_recover(ls, rv); | |
259 | kfree(rv->nodeids); | |
d44e0fc7 | 260 | kfree(rv->new); |
e7fd4179 DT |
261 | kfree(rv); |
262 | } | |
263 | } | |
264 | ||
265 | static int dlm_recoverd(void *arg) | |
266 | { | |
267 | struct dlm_ls *ls; | |
268 | ||
269 | ls = dlm_find_lockspace_local(arg); | |
5f88f1ea DT |
270 | if (!ls) { |
271 | log_print("dlm_recoverd: no lockspace %p", arg); | |
272 | return -1; | |
273 | } | |
e7fd4179 DT |
274 | |
275 | while (!kthread_should_stop()) { | |
276 | set_current_state(TASK_INTERRUPTIBLE); | |
277 | if (!test_bit(LSFL_WORK, &ls->ls_flags)) | |
278 | schedule(); | |
279 | set_current_state(TASK_RUNNING); | |
280 | ||
281 | if (test_and_clear_bit(LSFL_WORK, &ls->ls_flags)) | |
282 | do_ls_recovery(ls); | |
283 | } | |
284 | ||
285 | dlm_put_lockspace(ls); | |
286 | return 0; | |
287 | } | |
288 | ||
289 | void dlm_recoverd_kick(struct dlm_ls *ls) | |
290 | { | |
291 | set_bit(LSFL_WORK, &ls->ls_flags); | |
292 | wake_up_process(ls->ls_recoverd_task); | |
293 | } | |
294 | ||
295 | int dlm_recoverd_start(struct dlm_ls *ls) | |
296 | { | |
297 | struct task_struct *p; | |
298 | int error = 0; | |
299 | ||
300 | p = kthread_run(dlm_recoverd, ls, "dlm_recoverd"); | |
301 | if (IS_ERR(p)) | |
302 | error = PTR_ERR(p); | |
303 | else | |
304 | ls->ls_recoverd_task = p; | |
305 | return error; | |
306 | } | |
307 | ||
308 | void dlm_recoverd_stop(struct dlm_ls *ls) | |
309 | { | |
310 | kthread_stop(ls->ls_recoverd_task); | |
311 | } | |
312 | ||
313 | void dlm_recoverd_suspend(struct dlm_ls *ls) | |
314 | { | |
f6db1b8e | 315 | wake_up(&ls->ls_wait_general); |
90135925 | 316 | mutex_lock(&ls->ls_recoverd_active); |
e7fd4179 DT |
317 | } |
318 | ||
319 | void dlm_recoverd_resume(struct dlm_ls *ls) | |
320 | { | |
90135925 | 321 | mutex_unlock(&ls->ls_recoverd_active); |
e7fd4179 DT |
322 | } |
323 |