fs/dlm/recoverd.c

31 static int enable_locking(struct dlm_ls *ls, uint64_t seq)
35 	down_write(&ls->ls_recv_active);
37 	spin_lock(&ls->ls_recover_lock);
38 	if (ls->ls_recover_seq == seq) {
39 		set_bit(LSFL_RUNNING, &ls->ls_flags);
41 		up_write(&ls->ls_in_recovery);
42 		clear_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
45 	spin_unlock(&ls->ls_recover_lock);
47 	up_write(&ls->ls_recv_active);
51 static int ls_recover(struct dlm_ls *ls, struct dlm_recover *rv)
56 	log_rinfo(ls, "dlm_recover %llu", (unsigned long long)rv->seq);
58 	mutex_lock(&ls->ls_recoverd_active);
60 	dlm_callback_suspend(ls);
62 	dlm_clear_toss(ls);
69 	dlm_create_root_list(ls);
79 	error = dlm_recover_members(ls, rv, &neg);
81 		log_rinfo(ls, "dlm_recover_members error %d", error);
85 	dlm_recover_dir_nodeid(ls);
87 	ls->ls_recover_dir_sent_res = 0;
88 	ls->ls_recover_dir_sent_msg = 0;
89 	ls->ls_recover_locks_in = 0;
91 	dlm_set_recover_status(ls, DLM_RS_NODES);
93 	error = dlm_recover_members_wait(ls, rv->seq);
95 		log_rinfo(ls, "dlm_recover_members_wait error %d", error);
106 	error = dlm_recover_directory(ls, rv->seq);
108 		log_rinfo(ls, "dlm_recover_directory error %d", error);
112 	dlm_set_recover_status(ls, DLM_RS_DIR);
114 	error = dlm_recover_directory_wait(ls, rv->seq);
116 		log_rinfo(ls, "dlm_recover_directory_wait error %d", error);
120 	log_rinfo(ls, "dlm_recover_directory %u out %u messages",
121 		  ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
129 	dlm_recover_waiters_pre(ls);
131 	if (dlm_recovery_stopped(ls)) {
136 	if (neg || dlm_no_directory(ls)) {
141 		dlm_recover_purge(ls);
148 		error = dlm_recover_masters(ls, rv->seq);
150 			log_rinfo(ls, "dlm_recover_masters error %d", error);
158 		error = dlm_recover_locks(ls, rv->seq);
160 			log_rinfo(ls, "dlm_recover_locks error %d", error);
164 		dlm_set_recover_status(ls, DLM_RS_LOCKS);
166 		error = dlm_recover_locks_wait(ls, rv->seq);
168 			log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
172 		log_rinfo(ls, "dlm_recover_locks %u in",
173 			  ls->ls_recover_locks_in);
181 		dlm_recover_rsbs(ls);
188 		dlm_set_recover_status(ls, DLM_RS_LOCKS);
190 		error = dlm_recover_locks_wait(ls, rv->seq);
192 			log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
197 	dlm_release_root_list(ls);
205 	dlm_purge_requestqueue(ls);
207 	dlm_set_recover_status(ls, DLM_RS_DONE);
209 	error = dlm_recover_done_wait(ls, rv->seq);
211 		log_rinfo(ls, "dlm_recover_done_wait error %d", error);
215 	dlm_clear_members_gone(ls);
217 	dlm_callback_resume(ls);
219 	error = enable_locking(ls, rv->seq);
221 		log_rinfo(ls, "enable_locking error %d", error);
225 	error = dlm_process_requestqueue(ls);
227 		log_rinfo(ls, "dlm_process_requestqueue error %d", error);
231 	error = dlm_recover_waiters_post(ls);
233 		log_rinfo(ls, "dlm_recover_waiters_post error %d", error);
237 	dlm_recover_grant(ls);
239 	log_rinfo(ls, "dlm_recover %llu generation %u done: %u ms",
240 		  (unsigned long long)rv->seq, ls->ls_generation,
242 	mutex_unlock(&ls->ls_recoverd_active);
247 	dlm_release_root_list(ls);
248 	mutex_unlock(&ls->ls_recoverd_active);
257 static void do_ls_recovery(struct dlm_ls *ls)
262 	spin_lock(&ls->ls_recover_lock);
263 	rv = ls->ls_recover_args;
264 	ls->ls_recover_args = NULL;
265 	if (rv && ls->ls_recover_seq == rv->seq)
266 		clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
267 	spin_unlock(&ls->ls_recover_lock);
270 		error = ls_recover(ls, rv);
273 			ls->ls_recovery_result = 0;
274 			complete(&ls->ls_recovery_done);
276 			dlm_lsop_recover_done(ls);
282 			log_rinfo(ls, "%s %llu interrupted and should be queued to run again",
286 			log_rinfo(ls, "%s %llu error %d", __func__,
290 			ls->ls_recovery_result = error;
291 			complete(&ls->ls_recovery_done);
302 	struct dlm_ls *ls;
304 	ls = dlm_find_lockspace_local(arg);
305 	if (!ls) {
310 	down_write(&ls->ls_in_recovery);
311 	set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
312 	wake_up(&ls->ls_recover_lock_wait);
325 		if (!test_bit(LSFL_RECOVER_WORK, &ls->ls_flags) &&
326 		    !test_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
333 		if (test_and_clear_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
334 			down_write(&ls->ls_in_recovery);
335 			set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
336 			wake_up(&ls->ls_recover_lock_wait);
339 		if (test_and_clear_bit(LSFL_RECOVER_WORK, &ls->ls_flags))
340 			do_ls_recovery(ls);
343 	if (test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags))
344 		up_write(&ls->ls_in_recovery);
346 	dlm_put_lockspace(ls);
350 int dlm_recoverd_start(struct dlm_ls *ls)
355 	p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
359                 ls->ls_recoverd_task = p;
363 void dlm_recoverd_stop(struct dlm_ls *ls)
365 	kthread_stop(ls->ls_recoverd_task);
368 void dlm_recoverd_suspend(struct dlm_ls *ls)
370 	wake_up(&ls->ls_wait_general);
371 	mutex_lock(&ls->ls_recoverd_active);
374 void dlm_recoverd_resume(struct dlm_ls *ls)
376 	mutex_unlock(&ls->ls_recoverd_active);