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26 /* If the start for which we're re-enabling locking (seq) has been superseded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);
61 dlm_callback_suspend(ls);
64 * 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.
68 dlm_clear_toss_list(ls);
71 * This list of root rsb's will be the basis of most of the recovery
75 dlm_create_root_list(ls);
78 * Add or remove nodes from the lockspace's ls_nodes list.
79 * Also waits for all nodes to complete dlm_recover_members.
82 error = dlm_recover_members(ls, rv, &neg);
84 log_debug(ls, "recover_members failed %d", error);
90 * Rebuild our own share of the directory by collecting from all other
91 * nodes their master rsb names that hash to us.
94 error = dlm_recover_directory(ls);
96 log_debug(ls, "recover_directory failed %d", error);
101 * Wait for all nodes to complete directory rebuild.
104 error = dlm_recover_directory_wait(ls);
106 log_debug(ls, "recover_directory_wait failed %d", error);
111 * We may have outstanding operations that are waiting for a reply from
112 * a failed node. Mark these to be resent after recovery. Unlock and
113 * cancel ops can just be completed.
116 dlm_recover_waiters_pre(ls);
118 error = dlm_recovery_stopped(ls);
122 if (neg || dlm_no_directory(ls)) {
124 * Clear lkb's for departed nodes.
130 * Get new master nodeid's for rsb's that were mastered on
134 error = dlm_recover_masters(ls);
136 log_debug(ls, "recover_masters failed %d", error);
141 * Send our locks on remastered rsb's to the new masters.
144 error = dlm_recover_locks(ls);
146 log_debug(ls, "recover_locks failed %d", error);
150 error = dlm_recover_locks_wait(ls);
152 log_debug(ls, "recover_locks_wait failed %d", error);
157 * Finalize state in master rsb's now that all locks can be
158 * checked. This includes conversion resolution and lvb
162 dlm_recover_rsbs(ls);
165 * Other lockspace members may be going through the "neg" steps
166 * while also adding us to the lockspace, in which case they'll
167 * be doing the recover_locks (RS_LOCKS) barrier.
169 dlm_set_recover_status(ls, DLM_RS_LOCKS);
171 error = dlm_recover_locks_wait(ls);
173 log_debug(ls, "recover_locks_wait failed %d", error);
178 dlm_release_root_list(ls);
181 * Purge directory-related requests that are saved in requestqueue.
182 * All dir requests from before recovery are invalid now due to the dir
183 * rebuild and will be resent by the requesting nodes.
186 dlm_purge_requestqueue(ls);
188 dlm_set_recover_status(ls, DLM_RS_DONE);
189 error = dlm_recover_done_wait(ls);
191 log_debug(ls, "recover_done_wait failed %d", error);
195 dlm_clear_members_gone(ls);
197 dlm_adjust_timeouts(ls);
199 dlm_callback_resume(ls);
201 error = enable_locking(ls, rv->seq);
203 log_debug(ls, "enable_locking failed %d", error);
207 error = dlm_process_requestqueue(ls);
209 log_debug(ls, "process_requestqueue failed %d", error);
213 error = dlm_recover_waiters_post(ls);
215 log_debug(ls, "recover_waiters_post failed %d", error);
219 dlm_grant_after_purge(ls);
221 log_debug(ls, "recover %llx done: %u ms",
222 (unsigned long long)rv->seq,
223 jiffies_to_msecs(jiffies - start));
224 mutex_unlock(&ls->ls_recoverd_active);
229 dlm_release_root_list(ls);
230 log_debug(ls, "recover %llx error %d",
231 (unsigned long long)rv->seq, error);
232 mutex_unlock(&ls->ls_recoverd_active);
236 /* The dlm_ls_start() that created the rv we take here may already have been
237 stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
240 static void do_ls_recovery(struct dlm_ls *ls)
242 struct dlm_recover *rv = NULL;
244 spin_lock(&ls->ls_recover_lock);
245 rv = ls->ls_recover_args;
246 ls->ls_recover_args = NULL;
247 if (rv && ls->ls_recover_seq == rv->seq)
248 clear_bit(LSFL_RECOVERY_STOP, &ls->ls_flags);
249 spin_unlock(&ls->ls_recover_lock);
259 static int dlm_recoverd(void *arg)
263 ls = dlm_find_lockspace_local(arg);
265 log_print("dlm_recoverd: no lockspace %p", arg);
269 while (!kthread_should_stop()) {
270 set_current_state(TASK_INTERRUPTIBLE);
271 if (!test_bit(LSFL_WORK, &ls->ls_flags))
273 set_current_state(TASK_RUNNING);
275 if (test_and_clear_bit(LSFL_WORK, &ls->ls_flags))
279 dlm_put_lockspace(ls);
283 void dlm_recoverd_kick(struct dlm_ls *ls)
285 set_bit(LSFL_WORK, &ls->ls_flags);
286 wake_up_process(ls->ls_recoverd_task);
289 int dlm_recoverd_start(struct dlm_ls *ls)
291 struct task_struct *p;
294 p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
298 ls->ls_recoverd_task = p;
302 void dlm_recoverd_stop(struct dlm_ls *ls)
304 kthread_stop(ls->ls_recoverd_task);
307 void dlm_recoverd_suspend(struct dlm_ls *ls)
309 wake_up(&ls->ls_wait_general);
310 mutex_lock(&ls->ls_recoverd_active);
313 void dlm_recoverd_resume(struct dlm_ls *ls)
315 mutex_unlock(&ls->ls_recoverd_active);