blob: fd677c8c3d3b041396817bc930bde7e5e5b0be87 [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 Teigland3ae1acf2007-05-18 08:59:31 -05005** Copyright (C) 2004-2007 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26/* If the start for which we're re-enabling locking (seq) has been superseded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
Ryusuke Konishi57adf7e2006-11-29 09:33:48 -050057 log_debug(ls, "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
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 /*
David Teigland85f03792008-01-16 13:02:31 -060070 * 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.
72 */
73
74 dlm_clear_toss_list(ls);
75
76 /*
David Teiglande7fd4172006-01-18 09:30:29 +000077 * This list of root rsb's will be the basis of most of the recovery
78 * routines.
79 */
80
81 dlm_create_root_list(ls);
82
83 /*
David Teiglande7fd4172006-01-18 09:30:29 +000084 * 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) {
David Teigland8ec68862007-01-09 09:38:39 -060090 log_debug(ls, "recover_members failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +000091 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) {
David Teigland8ec68862007-01-09 09:38:39 -0600102 log_debug(ls, "recover_directory failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000103 goto fail;
104 }
105
106 /*
David Teiglande7fd4172006-01-18 09:30:29 +0000107 * Wait for all nodes to complete directory rebuild.
108 */
109
110 error = dlm_recover_directory_wait(ls);
111 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600112 log_debug(ls, "recover_directory_wait failed %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 Teigland8ec68862007-01-09 09:38:39 -0600142 log_debug(ls, "recover_masters failed %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 Teigland8ec68862007-01-09 09:38:39 -0600152 log_debug(ls, "recover_locks failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000153 goto fail;
154 }
155
156 error = dlm_recover_locks_wait(ls);
157 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600158 log_debug(ls, "recover_locks_wait failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000159 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);
David Teigland91c0dc92006-10-31 11:56:01 -0600169 } 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
David Teigland4b77f2c2006-11-01 09:31:48 -0600173 * be doing the recover_locks (RS_LOCKS) barrier.
David Teigland91c0dc92006-10-31 11:56:01 -0600174 */
175 dlm_set_recover_status(ls, DLM_RS_LOCKS);
David Teigland4b77f2c2006-11-01 09:31:48 -0600176
177 error = dlm_recover_locks_wait(ls);
178 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600179 log_debug(ls, "recover_locks_wait failed %d", error);
David Teigland4b77f2c2006-11-01 09:31:48 -0600180 goto fail;
181 }
David Teiglande7fd4172006-01-18 09:30:29 +0000182 }
183
184 dlm_release_root_list(ls);
185
David Teigland2896ee32006-11-27 11:31:22 -0600186 /*
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
David Teiglande7fd4172006-01-18 09:30:29 +0000194 dlm_set_recover_status(ls, DLM_RS_DONE);
195 error = dlm_recover_done_wait(ls);
196 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600197 log_debug(ls, "recover_done_wait failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000198 goto fail;
199 }
200
201 dlm_clear_members_gone(ls);
202
David Teigland3ae1acf2007-05-18 08:59:31 -0500203 dlm_adjust_timeouts(ls);
204
David Teiglande7fd4172006-01-18 09:30:29 +0000205 error = enable_locking(ls, rv->seq);
206 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600207 log_debug(ls, "enable_locking failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000208 goto fail;
209 }
210
211 error = dlm_process_requestqueue(ls);
212 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600213 log_debug(ls, "process_requestqueue failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000214 goto fail;
215 }
216
217 error = dlm_recover_waiters_post(ls);
218 if (error) {
David Teigland8ec68862007-01-09 09:38:39 -0600219 log_debug(ls, "recover_waiters_post failed %d", error);
David Teiglande7fd4172006-01-18 09:30:29 +0000220 goto fail;
221 }
222
223 dlm_grant_after_purge(ls);
224
225 dlm_astd_wake();
226
Ryusuke Konishi57adf7e2006-11-29 09:33:48 -0500227 log_debug(ls, "recover %llx done: %u ms",
228 (unsigned long long)rv->seq,
David Teiglande7fd4172006-01-18 09:30:29 +0000229 jiffies_to_msecs(jiffies - start));
David Teigland90135922006-01-20 08:47:07 +0000230 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000231
232 return 0;
233
234 fail:
235 dlm_release_root_list(ls);
Ryusuke Konishi57adf7e2006-11-29 09:33:48 -0500236 log_debug(ls, "recover %llx error %d",
237 (unsigned long long)rv->seq, error);
David Teigland90135922006-01-20 08:47:07 +0000238 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000239 return error;
240}
241
David Teigland2cdc98a2006-10-31 11:56:08 -0600242/* 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
David Teiglande7fd4172006-01-18 09:30:29 +0000246static 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;
David Teigland2cdc98a2006-10-31 11:56:08 -0600253 if (rv && ls->ls_recover_seq == rv->seq)
254 clear_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
David Teiglande7fd4172006-01-18 09:30:29 +0000255 spin_unlock(&ls->ls_recover_lock);
256
257 if (rv) {
258 ls_recover(ls, rv);
259 kfree(rv->nodeids);
David Teiglandd44e0fc2008-03-18 14:22:11 -0500260 kfree(rv->new);
David Teiglande7fd4172006-01-18 09:30:29 +0000261 kfree(rv);
262 }
263}
264
265static int dlm_recoverd(void *arg)
266{
267 struct dlm_ls *ls;
268
269 ls = dlm_find_lockspace_local(arg);
David Teigland5f88f1e2006-08-24 14:47:20 -0500270 if (!ls) {
271 log_print("dlm_recoverd: no lockspace %p", arg);
272 return -1;
273 }
David Teiglande7fd4172006-01-18 09:30:29 +0000274
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
289void 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
295int 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
308void dlm_recoverd_stop(struct dlm_ls *ls)
309{
310 kthread_stop(ls->ls_recoverd_task);
311}
312
313void dlm_recoverd_suspend(struct dlm_ls *ls)
314{
David Teiglandf6db1b82006-08-08 17:06:07 -0500315 wake_up(&ls->ls_wait_general);
David Teigland90135922006-01-20 08:47:07 +0000316 mutex_lock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000317}
318
319void dlm_recoverd_resume(struct dlm_ls *ls)
320{
David Teigland90135922006-01-20 08:47:07 +0000321 mutex_unlock(&ls->ls_recoverd_active);
David Teiglande7fd4172006-01-18 09:30:29 +0000322}
323