blob: 3780caf7ae0c239776951724c8d951dca5aef5dd [file] [log] [blame]
David Teiglande7fd4172006-01-18 09:30:29 +00001/******************************************************************************
2*******************************************************************************
3**
4** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
David Teigland60f98d12011-11-02 14:30:58 -05005** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved.
David Teiglande7fd4172006-01-18 09:30:29 +00006**
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26/* If the start for which we're re-enabling locking (seq) has been superseded
David Teiglandc36258b2007-09-27 15:53:38 -050027 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. */
David Teiglande7fd4172006-01-18 09:30:29 +000032
33static int enable_locking(struct dlm_ls *ls, uint64_t seq)
34{
35 int error = -EINTR;
36
David Teiglandc36258b2007-09-27 15:53:38 -050037 down_write(&ls->ls_recv_active);
38
David Teiglande7fd4172006-01-18 09:30:29 +000039 spin_lock(&ls->ls_recover_lock);
40 if (ls->ls_recover_seq == seq) {
41 set_bit(LSFL_RUNNING, &ls->ls_flags);
David Teiglandc36258b2007-09-27 15:53:38 -050042 /* unblocks processes waiting to enter the dlm */
David Teiglande7fd4172006-01-18 09:30:29 +000043 up_write(&ls->ls_in_recovery);
44 error = 0;
45 }
46 spin_unlock(&ls->ls_recover_lock);
David Teiglandc36258b2007-09-27 15:53:38 -050047
48 up_write(&ls->ls_recv_active);
David Teiglande7fd4172006-01-18 09:30:29 +000049 return error;
50}
51
52static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
53{
54 unsigned long start;
55 int error, neg = 0;
56
David Teiglandf95a34c2011-10-14 12:34:58 -050057 log_debug(ls, "dlm_recover %llx", (unsigned long long)rv->seq);
David Teiglande7fd4172006-01-18 09:30:29 +000058
David Teigland90135922006-01-20 08:47:07 +000059 mutex_lock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +000060
David Teigland23e8e1a2011-04-05 13:16:24 -050061 dlm_callback_suspend(ls);
David Teiglande7fd4172006-01-18 09:30:29 +000062
63 /*
David Teigland85f03792008-01-16 13:02:31 -060064 * Free non-master tossed rsb's. Master rsb's are kept on toss
65 * list and put on root list to be included in resdir recovery.
66 */
67
68 dlm_clear_toss_list(ls);
69
70 /*
David Teiglande7fd4172006-01-18 09:30:29 +000071 * This list of root rsb's will be the basis of most of the recovery
72 * routines.
73 */
74
75 dlm_create_root_list(ls);
76
77 /*
David Teiglande7fd4172006-01-18 09:30:29 +000078 * Add or remove nodes from the lockspace's ls_nodes list.
David Teiglande7fd4172006-01-18 09:30:29 +000079 */
80
81 error = dlm_recover_members(ls, rv, &neg);
82 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -050083 log_debug(ls, "dlm_recover_members error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +000084 goto fail;
85 }
David Teiglandf95a34c2011-10-14 12:34:58 -050086
87 dlm_set_recover_status(ls, DLM_RS_NODES);
88
89 error = dlm_recover_members_wait(ls);
90 if (error) {
91 log_debug(ls, "dlm_recover_members_wait error %d", error);
92 goto fail;
93 }
94
David Teiglande7fd4172006-01-18 09:30:29 +000095 start = jiffies;
96
97 /*
98 * Rebuild our own share of the directory by collecting from all other
99 * nodes their master rsb names that hash to us.
100 */
101
102 error = dlm_recover_directory(ls);
103 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500104 log_debug(ls, "dlm_recover_directory error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000105 goto fail;
106 }
107
David Teiglandf95a34c2011-10-14 12:34:58 -0500108 dlm_set_recover_status(ls, DLM_RS_DIR);
David Teiglande7fd4172006-01-18 09:30:29 +0000109
110 error = dlm_recover_directory_wait(ls);
111 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500112 log_debug(ls, "dlm_recover_directory_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000113 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) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500142 log_debug(ls, "dlm_recover_masters error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000143 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) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500152 log_debug(ls, "dlm_recover_locks error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000153 goto fail;
154 }
155
David Teiglandf95a34c2011-10-14 12:34:58 -0500156 dlm_set_recover_status(ls, DLM_RS_LOCKS);
157
David Teiglande7fd4172006-01-18 09:30:29 +0000158 error = dlm_recover_locks_wait(ls);
159 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500160 log_debug(ls, "dlm_recover_locks_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000161 goto fail;
162 }
163
164 /*
165 * Finalize state in master rsb's now that all locks can be
166 * checked. This includes conversion resolution and lvb
167 * settings.
168 */
169
170 dlm_recover_rsbs(ls);
David Teigland91c0dc92006-10-31 11:56:01 -0600171 } else {
172 /*
173 * Other lockspace members may be going through the "neg" steps
174 * while also adding us to the lockspace, in which case they'll
David Teigland4b77f2c2006-11-01 09:31:48 -0600175 * be doing the recover_locks (RS_LOCKS) barrier.
David Teigland91c0dc92006-10-31 11:56:01 -0600176 */
177 dlm_set_recover_status(ls, DLM_RS_LOCKS);
David Teigland4b77f2c2006-11-01 09:31:48 -0600178
179 error = dlm_recover_locks_wait(ls);
180 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500181 log_debug(ls, "dlm_recover_locks_wait error %d", error);
David Teigland4b77f2c2006-11-01 09:31:48 -0600182 goto fail;
183 }
David Teiglande7fd4172006-01-18 09:30:29 +0000184 }
185
186 dlm_release_root_list(ls);
187
David Teigland2896ee32006-11-27 11:31:22 -0600188 /*
189 * Purge directory-related requests that are saved in requestqueue.
190 * All dir requests from before recovery are invalid now due to the dir
191 * rebuild and will be resent by the requesting nodes.
192 */
193
194 dlm_purge_requestqueue(ls);
195
David Teiglande7fd4172006-01-18 09:30:29 +0000196 dlm_set_recover_status(ls, DLM_RS_DONE);
David Teiglandf95a34c2011-10-14 12:34:58 -0500197
David Teiglande7fd4172006-01-18 09:30:29 +0000198 error = dlm_recover_done_wait(ls);
199 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500200 log_debug(ls, "dlm_recover_done_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000201 goto fail;
202 }
203
204 dlm_clear_members_gone(ls);
205
David Teigland3ae1acf2007-05-18 08:59:31 -0500206 dlm_adjust_timeouts(ls);
207
David Teigland23e8e1a2011-04-05 13:16:24 -0500208 dlm_callback_resume(ls);
209
David Teiglande7fd4172006-01-18 09:30:29 +0000210 error = enable_locking(ls, rv->seq);
211 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500212 log_debug(ls, "enable_locking error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000213 goto fail;
214 }
215
216 error = dlm_process_requestqueue(ls);
217 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500218 log_debug(ls, "dlm_process_requestqueue error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000219 goto fail;
220 }
221
222 error = dlm_recover_waiters_post(ls);
223 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500224 log_debug(ls, "dlm_recover_waiters_post error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000225 goto fail;
226 }
227
228 dlm_grant_after_purge(ls);
229
David Teigland60f98d12011-11-02 14:30:58 -0500230 log_debug(ls, "dlm_recover %llx generation %u done: %u ms",
231 (unsigned long long)rv->seq, ls->ls_generation,
David Teiglande7fd4172006-01-18 09:30:29 +0000232 jiffies_to_msecs(jiffies - start));
David Teigland90135922006-01-20 08:47:07 +0000233 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000234
David Teigland60f98d12011-11-02 14:30:58 -0500235 dlm_lsop_recover_done(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000236 return 0;
237
238 fail:
239 dlm_release_root_list(ls);
David Teiglandf95a34c2011-10-14 12:34:58 -0500240 log_debug(ls, "dlm_recover %llx error %d",
Ryusuke Konishi57adf7e2006-11-29 09:33:48 -0500241 (unsigned long long)rv->seq, error);
David Teigland90135922006-01-20 08:47:07 +0000242 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000243 return error;
244}
245
David Teigland2cdc98a2006-10-31 11:56:08 -0600246/* The dlm_ls_start() that created the rv we take here may already have been
247 stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
248 flag set. */
249
David Teiglande7fd4172006-01-18 09:30:29 +0000250static void do_ls_recovery(struct dlm_ls *ls)
251{
252 struct dlm_recover *rv = NULL;
253
254 spin_lock(&ls->ls_recover_lock);
255 rv = ls->ls_recover_args;
256 ls->ls_recover_args = NULL;
David Teigland2cdc98a2006-10-31 11:56:08 -0600257 if (rv && ls->ls_recover_seq == rv->seq)
258 clear_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
David Teiglande7fd4172006-01-18 09:30:29 +0000259 spin_unlock(&ls->ls_recover_lock);
260
261 if (rv) {
262 ls_recover(ls, rv);
David Teigland60f98d12011-11-02 14:30:58 -0500263 kfree(rv->nodes);
David Teiglande7fd4172006-01-18 09:30:29 +0000264 kfree(rv);
265 }
266}
267
268static int dlm_recoverd(void *arg)
269{
270 struct dlm_ls *ls;
271
272 ls = dlm_find_lockspace_local(arg);
David Teigland5f88f1e2006-08-24 14:47:20 -0500273 if (!ls) {
274 log_print("dlm_recoverd: no lockspace %p", arg);
275 return -1;
276 }
David Teiglande7fd4172006-01-18 09:30:29 +0000277
278 while (!kthread_should_stop()) {
279 set_current_state(TASK_INTERRUPTIBLE);
280 if (!test_bit(LSFL_WORK, &ls->ls_flags))
281 schedule();
282 set_current_state(TASK_RUNNING);
283
284 if (test_and_clear_bit(LSFL_WORK, &ls->ls_flags))
285 do_ls_recovery(ls);
286 }
287
288 dlm_put_lockspace(ls);
289 return 0;
290}
291
292void dlm_recoverd_kick(struct dlm_ls *ls)
293{
294 set_bit(LSFL_WORK, &ls->ls_flags);
295 wake_up_process(ls->ls_recoverd_task);
296}
297
298int dlm_recoverd_start(struct dlm_ls *ls)
299{
300 struct task_struct *p;
301 int error = 0;
302
303 p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
304 if (IS_ERR(p))
305 error = PTR_ERR(p);
306 else
307 ls->ls_recoverd_task = p;
308 return error;
309}
310
311void dlm_recoverd_stop(struct dlm_ls *ls)
312{
313 kthread_stop(ls->ls_recoverd_task);
314}
315
316void dlm_recoverd_suspend(struct dlm_ls *ls)
317{
David Teiglandf6db1b82006-08-08 17:06:07 -0500318 wake_up(&ls->ls_wait_general);
David Teigland90135922006-01-20 08:47:07 +0000319 mutex_lock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000320}
321
322void dlm_recoverd_resume(struct dlm_ls *ls)
323{
David Teigland90135922006-01-20 08:47:07 +0000324 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000325}
326