1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2004-2007 Red Hat, Inc. All rights reserved.
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.
11 *******************************************************************************
12 ******************************************************************************/
14 #include "dlm_internal.h"
15 #include "lockspace.h"
22 #include "requestqueue.h"
26 /* If the start for which we're re-enabling locking (seq) has been superseded
27 by a newer stop (ls_recover_seq), we need to leave locking disabled.
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. */
33 static int enable_locking(struct dlm_ls
*ls
, uint64_t seq
)
37 down_write(&ls
->ls_recv_active
);
39 spin_lock(&ls
->ls_recover_lock
);
40 if (ls
->ls_recover_seq
== seq
) {
41 set_bit(LSFL_RUNNING
, &ls
->ls_flags
);
42 /* unblocks processes waiting to enter the dlm */
43 up_write(&ls
->ls_in_recovery
);
46 spin_unlock(&ls
->ls_recover_lock
);
48 up_write(&ls
->ls_recv_active
);
52 static int ls_recover(struct dlm_ls
*ls
, struct dlm_recover
*rv
)
57 log_debug(ls
, "recover %llx", (unsigned long long)rv
->seq
);
59 mutex_lock(&ls
->ls_recoverd_active
);
62 * Suspending and resuming dlm_astd ensures that no lkb's from this ls
63 * will be processed by dlm_astd during recovery.
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.
74 dlm_clear_toss_list(ls
);
77 * This list of root rsb's will be the basis of most of the recovery
81 dlm_create_root_list(ls
);
84 * Add or remove nodes from the lockspace's ls_nodes list.
85 * Also waits for all nodes to complete dlm_recover_members.
88 error
= dlm_recover_members(ls
, rv
, &neg
);
90 log_debug(ls
, "recover_members failed %d", error
);
96 * Rebuild our own share of the directory by collecting from all other
97 * nodes their master rsb names that hash to us.
100 error
= dlm_recover_directory(ls
);
102 log_debug(ls
, "recover_directory failed %d", error
);
107 * Wait for all nodes to complete directory rebuild.
110 error
= dlm_recover_directory_wait(ls
);
112 log_debug(ls
, "recover_directory_wait failed %d", error
);
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.
122 dlm_recover_waiters_pre(ls
);
124 error
= dlm_recovery_stopped(ls
);
128 if (neg
|| dlm_no_directory(ls
)) {
130 * Clear lkb's for departed nodes.
136 * Get new master nodeid's for rsb's that were mastered on
140 error
= dlm_recover_masters(ls
);
142 log_debug(ls
, "recover_masters failed %d", error
);
147 * Send our locks on remastered rsb's to the new masters.
150 error
= dlm_recover_locks(ls
);
152 log_debug(ls
, "recover_locks failed %d", error
);
156 error
= dlm_recover_locks_wait(ls
);
158 log_debug(ls
, "recover_locks_wait failed %d", error
);
163 * Finalize state in master rsb's now that all locks can be
164 * checked. This includes conversion resolution and lvb
168 dlm_recover_rsbs(ls
);
171 * Other lockspace members may be going through the "neg" steps
172 * while also adding us to the lockspace, in which case they'll
173 * be doing the recover_locks (RS_LOCKS) barrier.
175 dlm_set_recover_status(ls
, DLM_RS_LOCKS
);
177 error
= dlm_recover_locks_wait(ls
);
179 log_debug(ls
, "recover_locks_wait failed %d", error
);
184 dlm_release_root_list(ls
);
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.
192 dlm_purge_requestqueue(ls
);
194 dlm_set_recover_status(ls
, DLM_RS_DONE
);
195 error
= dlm_recover_done_wait(ls
);
197 log_debug(ls
, "recover_done_wait failed %d", error
);
201 dlm_clear_members_gone(ls
);
203 dlm_adjust_timeouts(ls
);
205 error
= enable_locking(ls
, rv
->seq
);
207 log_debug(ls
, "enable_locking failed %d", error
);
211 error
= dlm_process_requestqueue(ls
);
213 log_debug(ls
, "process_requestqueue failed %d", error
);
217 error
= dlm_recover_waiters_post(ls
);
219 log_debug(ls
, "recover_waiters_post failed %d", error
);
223 dlm_grant_after_purge(ls
);
227 log_debug(ls
, "recover %llx done: %u ms",
228 (unsigned long long)rv
->seq
,
229 jiffies_to_msecs(jiffies
- start
));
230 mutex_unlock(&ls
->ls_recoverd_active
);
235 dlm_release_root_list(ls
);
236 log_debug(ls
, "recover %llx error %d",
237 (unsigned long long)rv
->seq
, error
);
238 mutex_unlock(&ls
->ls_recoverd_active
);
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
246 static void do_ls_recovery(struct dlm_ls
*ls
)
248 struct dlm_recover
*rv
= NULL
;
250 spin_lock(&ls
->ls_recover_lock
);
251 rv
= ls
->ls_recover_args
;
252 ls
->ls_recover_args
= NULL
;
253 if (rv
&& ls
->ls_recover_seq
== rv
->seq
)
254 clear_bit(LSFL_RECOVERY_STOP
, &ls
->ls_flags
);
255 spin_unlock(&ls
->ls_recover_lock
);
265 static int dlm_recoverd(void *arg
)
269 ls
= dlm_find_lockspace_local(arg
);
271 log_print("dlm_recoverd: no lockspace %p", arg
);
275 while (!kthread_should_stop()) {
276 set_current_state(TASK_INTERRUPTIBLE
);
277 if (!test_bit(LSFL_WORK
, &ls
->ls_flags
))
279 set_current_state(TASK_RUNNING
);
281 if (test_and_clear_bit(LSFL_WORK
, &ls
->ls_flags
))
285 dlm_put_lockspace(ls
);
289 void dlm_recoverd_kick(struct dlm_ls
*ls
)
291 set_bit(LSFL_WORK
, &ls
->ls_flags
);
292 wake_up_process(ls
->ls_recoverd_task
);
295 int dlm_recoverd_start(struct dlm_ls
*ls
)
297 struct task_struct
*p
;
300 p
= kthread_run(dlm_recoverd
, ls
, "dlm_recoverd");
304 ls
->ls_recoverd_task
= p
;
308 void dlm_recoverd_stop(struct dlm_ls
*ls
)
310 kthread_stop(ls
->ls_recoverd_task
);
313 void dlm_recoverd_suspend(struct dlm_ls
*ls
)
315 wake_up(&ls
->ls_wait_general
);
316 mutex_lock(&ls
->ls_recoverd_active
);
319 void dlm_recoverd_resume(struct dlm_ls
*ls
)
321 mutex_unlock(&ls
->ls_recoverd_active
);