blob: 32f9f8926ec3f17a3a69433402e8a0fecb07a1f6 [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26/* If the start for which we're re-enabling locking (seq) has been superseded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);
David Teigland475f2302012-08-02 11:08:21 -050044 clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
David Teiglande7fd4172006-01-18 09:30:29 +000045 error = 0;
46 }
47 spin_unlock(&ls->ls_recover_lock);
David Teiglandc36258b2007-09-27 15:53:38 -050048
49 up_write(&ls->ls_recv_active);
David Teiglande7fd4172006-01-18 09:30:29 +000050 return error;
51}
52
53static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
54{
55 unsigned long start;
56 int error, neg = 0;
57
David Teigland6d40c4a2012-04-23 16:36:01 -050058 log_debug(ls, "dlm_recover %llu", (unsigned long long)rv->seq);
David Teiglande7fd4172006-01-18 09:30:29 +000059
David Teigland90135922006-01-20 08:47:07 +000060 mutex_lock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +000061
David Teigland23e8e1a2011-04-05 13:16:24 -050062 dlm_callback_suspend(ls);
David Teiglande7fd4172006-01-18 09:30:29 +000063
David Teiglandc04fecb2012-05-10 10:18:07 -050064 dlm_clear_toss(ls);
David Teigland85f03792008-01-16 13:02:31 -060065
66 /*
David Teiglande7fd4172006-01-18 09:30:29 +000067 * This list of root rsb's will be the basis of most of the recovery
68 * routines.
69 */
70
71 dlm_create_root_list(ls);
72
73 /*
David Teiglande7fd4172006-01-18 09:30:29 +000074 * Add or remove nodes from the lockspace's ls_nodes list.
David Teiglande7fd4172006-01-18 09:30:29 +000075 */
76
77 error = dlm_recover_members(ls, rv, &neg);
78 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -050079 log_debug(ls, "dlm_recover_members error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +000080 goto fail;
81 }
David Teiglandf95a34c2011-10-14 12:34:58 -050082
David Teiglandc04fecb2012-05-10 10:18:07 -050083 dlm_recover_dir_nodeid(ls);
84
85 ls->ls_recover_dir_sent_res = 0;
86 ls->ls_recover_dir_sent_msg = 0;
David Teigland48756472012-04-26 15:54:29 -050087 ls->ls_recover_locks_in = 0;
88
David Teiglandf95a34c2011-10-14 12:34:58 -050089 dlm_set_recover_status(ls, DLM_RS_NODES);
90
91 error = dlm_recover_members_wait(ls);
92 if (error) {
93 log_debug(ls, "dlm_recover_members_wait error %d", error);
94 goto fail;
95 }
96
David Teiglande7fd4172006-01-18 09:30:29 +000097 start = jiffies;
98
99 /*
100 * Rebuild our own share of the directory by collecting from all other
101 * nodes their master rsb names that hash to us.
102 */
103
104 error = dlm_recover_directory(ls);
105 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500106 log_debug(ls, "dlm_recover_directory error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000107 goto fail;
108 }
109
David Teiglandf95a34c2011-10-14 12:34:58 -0500110 dlm_set_recover_status(ls, DLM_RS_DIR);
David Teiglande7fd4172006-01-18 09:30:29 +0000111
112 error = dlm_recover_directory_wait(ls);
113 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500114 log_debug(ls, "dlm_recover_directory_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000115 goto fail;
116 }
117
David Teiglandc04fecb2012-05-10 10:18:07 -0500118 log_debug(ls, "dlm_recover_directory %u out %u messages",
119 ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
120
David Teiglande7fd4172006-01-18 09:30:29 +0000121 /*
122 * We may have outstanding operations that are waiting for a reply from
123 * a failed node. Mark these to be resent after recovery. Unlock and
124 * cancel ops can just be completed.
125 */
126
127 dlm_recover_waiters_pre(ls);
128
129 error = dlm_recovery_stopped(ls);
130 if (error)
131 goto fail;
132
133 if (neg || dlm_no_directory(ls)) {
134 /*
135 * Clear lkb's for departed nodes.
136 */
137
David Teigland48756472012-04-26 15:54:29 -0500138 dlm_recover_purge(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000139
140 /*
141 * Get new master nodeid's for rsb's that were mastered on
142 * departed nodes.
143 */
144
145 error = dlm_recover_masters(ls);
146 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500147 log_debug(ls, "dlm_recover_masters error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000148 goto fail;
149 }
150
151 /*
152 * Send our locks on remastered rsb's to the new masters.
153 */
154
155 error = dlm_recover_locks(ls);
156 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500157 log_debug(ls, "dlm_recover_locks error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000158 goto fail;
159 }
160
David Teiglandf95a34c2011-10-14 12:34:58 -0500161 dlm_set_recover_status(ls, DLM_RS_LOCKS);
162
David Teiglande7fd4172006-01-18 09:30:29 +0000163 error = dlm_recover_locks_wait(ls);
164 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500165 log_debug(ls, "dlm_recover_locks_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000166 goto fail;
167 }
168
David Teigland48756472012-04-26 15:54:29 -0500169 log_debug(ls, "dlm_recover_locks %u in",
170 ls->ls_recover_locks_in);
171
David Teiglande7fd4172006-01-18 09:30:29 +0000172 /*
173 * Finalize state in master rsb's now that all locks can be
174 * checked. This includes conversion resolution and lvb
175 * settings.
176 */
177
178 dlm_recover_rsbs(ls);
David Teigland91c0dc92006-10-31 11:56:01 -0600179 } else {
180 /*
181 * Other lockspace members may be going through the "neg" steps
182 * while also adding us to the lockspace, in which case they'll
David Teigland4b77f2c2006-11-01 09:31:48 -0600183 * be doing the recover_locks (RS_LOCKS) barrier.
David Teigland91c0dc92006-10-31 11:56:01 -0600184 */
185 dlm_set_recover_status(ls, DLM_RS_LOCKS);
David Teigland4b77f2c2006-11-01 09:31:48 -0600186
187 error = dlm_recover_locks_wait(ls);
188 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500189 log_debug(ls, "dlm_recover_locks_wait error %d", error);
David Teigland4b77f2c2006-11-01 09:31:48 -0600190 goto fail;
191 }
David Teiglande7fd4172006-01-18 09:30:29 +0000192 }
193
194 dlm_release_root_list(ls);
195
David Teigland2896ee32006-11-27 11:31:22 -0600196 /*
197 * Purge directory-related requests that are saved in requestqueue.
198 * All dir requests from before recovery are invalid now due to the dir
199 * rebuild and will be resent by the requesting nodes.
200 */
201
202 dlm_purge_requestqueue(ls);
203
David Teiglande7fd4172006-01-18 09:30:29 +0000204 dlm_set_recover_status(ls, DLM_RS_DONE);
David Teiglandf95a34c2011-10-14 12:34:58 -0500205
David Teiglande7fd4172006-01-18 09:30:29 +0000206 error = dlm_recover_done_wait(ls);
207 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500208 log_debug(ls, "dlm_recover_done_wait error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000209 goto fail;
210 }
211
212 dlm_clear_members_gone(ls);
213
David Teigland3ae1acf2007-05-18 08:59:31 -0500214 dlm_adjust_timeouts(ls);
215
David Teigland23e8e1a2011-04-05 13:16:24 -0500216 dlm_callback_resume(ls);
217
David Teiglande7fd4172006-01-18 09:30:29 +0000218 error = enable_locking(ls, rv->seq);
219 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500220 log_debug(ls, "enable_locking error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000221 goto fail;
222 }
223
224 error = dlm_process_requestqueue(ls);
225 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500226 log_debug(ls, "dlm_process_requestqueue error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000227 goto fail;
228 }
229
230 error = dlm_recover_waiters_post(ls);
231 if (error) {
David Teiglandf95a34c2011-10-14 12:34:58 -0500232 log_debug(ls, "dlm_recover_waiters_post error %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000233 goto fail;
234 }
235
David Teigland48756472012-04-26 15:54:29 -0500236 dlm_recover_grant(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000237
David Teigland6d40c4a2012-04-23 16:36:01 -0500238 log_debug(ls, "dlm_recover %llu generation %u done: %u ms",
David Teigland60f98d12011-11-02 14:30:58 -0500239 (unsigned long long)rv->seq, ls->ls_generation,
David Teiglande7fd4172006-01-18 09:30:29 +0000240 jiffies_to_msecs(jiffies - start));
David Teigland90135922006-01-20 08:47:07 +0000241 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000242
David Teigland60f98d12011-11-02 14:30:58 -0500243 dlm_lsop_recover_done(ls);
David Teiglande7fd4172006-01-18 09:30:29 +0000244 return 0;
245
246 fail:
247 dlm_release_root_list(ls);
David Teigland6d40c4a2012-04-23 16:36:01 -0500248 log_debug(ls, "dlm_recover %llu error %d",
Ryusuke Konishi57adf7e2006-11-29 09:33:48 -0500249 (unsigned long long)rv->seq, error);
David Teigland90135922006-01-20 08:47:07 +0000250 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000251 return error;
252}
253
David Teigland2cdc98a2006-10-31 11:56:08 -0600254/* The dlm_ls_start() that created the rv we take here may already have been
255 stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
256 flag set. */
257
David Teiglande7fd4172006-01-18 09:30:29 +0000258static void do_ls_recovery(struct dlm_ls *ls)
259{
260 struct dlm_recover *rv = NULL;
261
262 spin_lock(&ls->ls_recover_lock);
263 rv = ls->ls_recover_args;
264 ls->ls_recover_args = NULL;
David Teigland2cdc98a2006-10-31 11:56:08 -0600265 if (rv && ls->ls_recover_seq == rv->seq)
David Teigland475f2302012-08-02 11:08:21 -0500266 clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
David Teiglande7fd4172006-01-18 09:30:29 +0000267 spin_unlock(&ls->ls_recover_lock);
268
269 if (rv) {
270 ls_recover(ls, rv);
David Teigland60f98d12011-11-02 14:30:58 -0500271 kfree(rv->nodes);
David Teiglande7fd4172006-01-18 09:30:29 +0000272 kfree(rv);
273 }
274}
275
276static int dlm_recoverd(void *arg)
277{
278 struct dlm_ls *ls;
279
280 ls = dlm_find_lockspace_local(arg);
David Teigland5f88f1e2006-08-24 14:47:20 -0500281 if (!ls) {
282 log_print("dlm_recoverd: no lockspace %p", arg);
283 return -1;
284 }
David Teiglande7fd4172006-01-18 09:30:29 +0000285
David Teigland475f2302012-08-02 11:08:21 -0500286 down_write(&ls->ls_in_recovery);
287 set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
288 wake_up(&ls->ls_recover_lock_wait);
289
David Teiglande7fd4172006-01-18 09:30:29 +0000290 while (!kthread_should_stop()) {
291 set_current_state(TASK_INTERRUPTIBLE);
David Teigland475f2302012-08-02 11:08:21 -0500292 if (!test_bit(LSFL_RECOVER_WORK, &ls->ls_flags) &&
293 !test_bit(LSFL_RECOVER_DOWN, &ls->ls_flags))
David Teiglande7fd4172006-01-18 09:30:29 +0000294 schedule();
295 set_current_state(TASK_RUNNING);
296
David Teigland475f2302012-08-02 11:08:21 -0500297 if (test_and_clear_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
298 down_write(&ls->ls_in_recovery);
299 set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
300 wake_up(&ls->ls_recover_lock_wait);
301 }
302
303 if (test_and_clear_bit(LSFL_RECOVER_WORK, &ls->ls_flags))
David Teiglande7fd4172006-01-18 09:30:29 +0000304 do_ls_recovery(ls);
305 }
306
David Teigland475f2302012-08-02 11:08:21 -0500307 if (test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags))
308 up_write(&ls->ls_in_recovery);
309
David Teiglande7fd4172006-01-18 09:30:29 +0000310 dlm_put_lockspace(ls);
311 return 0;
312}
313
David Teiglande7fd4172006-01-18 09:30:29 +0000314int dlm_recoverd_start(struct dlm_ls *ls)
315{
316 struct task_struct *p;
317 int error = 0;
318
319 p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
320 if (IS_ERR(p))
321 error = PTR_ERR(p);
322 else
323 ls->ls_recoverd_task = p;
324 return error;
325}
326
327void dlm_recoverd_stop(struct dlm_ls *ls)
328{
329 kthread_stop(ls->ls_recoverd_task);
330}
331
332void dlm_recoverd_suspend(struct dlm_ls *ls)
333{
David Teiglandf6db1b82006-08-08 17:06:07 -0500334 wake_up(&ls->ls_wait_general);
David Teigland90135922006-01-20 08:47:07 +0000335 mutex_lock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000336}
337
338void dlm_recoverd_resume(struct dlm_ls *ls)
339{
David Teigland90135922006-01-20 08:47:07 +0000340 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000341}
342