1 // SPDX-License-Identifier: GPL-2.0-only
2 /******************************************************************************
3 *******************************************************************************
5 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
6 ** Copyright (C) 2004-2011 Red Hat, Inc. All rights reserved.
9 *******************************************************************************
10 ******************************************************************************/
12 #include "dlm_internal.h"
13 #include "lockspace.h"
20 #include "requestqueue.h"
24 /* If the start for which we're re-enabling locking (seq) has been superseded
25 by a newer stop (ls_recover_seq), we need to leave locking disabled.
27 We suspend dlm_recv threads here to avoid the race where dlm_recv a) sees
28 locking stopped and b) adds a message to the requestqueue, but dlm_recoverd
29 enables locking and clears the requestqueue between a and b. */
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);
40 /* unblocks processes waiting to enter the dlm */
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);
65 * This list of root rsb's will be the basis of most of the recovery
69 dlm_create_root_list(ls);
72 * Add or remove nodes from the lockspace's ls_nodes list.
75 error = dlm_recover_members(ls, rv, &neg);
77 log_rinfo(ls, "dlm_recover_members error %d", error);
81 dlm_recover_dir_nodeid(ls);
83 ls->ls_recover_dir_sent_res = 0;
84 ls->ls_recover_dir_sent_msg = 0;
85 ls->ls_recover_locks_in = 0;
87 dlm_set_recover_status(ls, DLM_RS_NODES);
89 error = dlm_recover_members_wait(ls);
91 log_rinfo(ls, "dlm_recover_members_wait error %d", error);
98 * Rebuild our own share of the directory by collecting from all other
99 * nodes their master rsb names that hash to us.
102 error = dlm_recover_directory(ls);
104 log_rinfo(ls, "dlm_recover_directory error %d", error);
108 dlm_set_recover_status(ls, DLM_RS_DIR);
110 error = dlm_recover_directory_wait(ls);
112 log_rinfo(ls, "dlm_recover_directory_wait error %d", error);
116 log_rinfo(ls, "dlm_recover_directory %u out %u messages",
117 ls->ls_recover_dir_sent_res, ls->ls_recover_dir_sent_msg);
120 * We may have outstanding operations that are waiting for a reply from
121 * a failed node. Mark these to be resent after recovery. Unlock and
122 * cancel ops can just be completed.
125 dlm_recover_waiters_pre(ls);
127 error = dlm_recovery_stopped(ls);
131 if (neg || dlm_no_directory(ls)) {
133 * Clear lkb's for departed nodes.
136 dlm_recover_purge(ls);
139 * Get new master nodeid's for rsb's that were mastered on
143 error = dlm_recover_masters(ls);
145 log_rinfo(ls, "dlm_recover_masters error %d", error);
150 * Send our locks on remastered rsb's to the new masters.
153 error = dlm_recover_locks(ls);
155 log_rinfo(ls, "dlm_recover_locks error %d", error);
159 dlm_set_recover_status(ls, DLM_RS_LOCKS);
161 error = dlm_recover_locks_wait(ls);
163 log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
167 log_rinfo(ls, "dlm_recover_locks %u in",
168 ls->ls_recover_locks_in);
171 * Finalize state in master rsb's now that all locks can be
172 * checked. This includes conversion resolution and lvb
176 dlm_recover_rsbs(ls);
179 * Other lockspace members may be going through the "neg" steps
180 * while also adding us to the lockspace, in which case they'll
181 * be doing the recover_locks (RS_LOCKS) barrier.
183 dlm_set_recover_status(ls, DLM_RS_LOCKS);
185 error = dlm_recover_locks_wait(ls);
187 log_rinfo(ls, "dlm_recover_locks_wait error %d", error);
192 dlm_release_root_list(ls);
195 * Purge directory-related requests that are saved in requestqueue.
196 * All dir requests from before recovery are invalid now due to the dir
197 * rebuild and will be resent by the requesting nodes.
200 dlm_purge_requestqueue(ls);
202 dlm_set_recover_status(ls, DLM_RS_DONE);
204 error = dlm_recover_done_wait(ls);
206 log_rinfo(ls, "dlm_recover_done_wait error %d", error);
210 dlm_clear_members_gone(ls);
212 dlm_adjust_timeouts(ls);
214 dlm_callback_resume(ls);
216 error = enable_locking(ls, rv->seq);
218 log_rinfo(ls, "enable_locking error %d", error);
222 error = dlm_process_requestqueue(ls);
224 log_rinfo(ls, "dlm_process_requestqueue error %d", error);
228 error = dlm_recover_waiters_post(ls);
230 log_rinfo(ls, "dlm_recover_waiters_post error %d", error);
234 dlm_recover_grant(ls);
236 log_rinfo(ls, "dlm_recover %llu generation %u done: %u ms",
237 (unsigned long long)rv->seq, ls->ls_generation,
238 jiffies_to_msecs(jiffies - start));
239 mutex_unlock(&ls->ls_recoverd_active);
241 dlm_lsop_recover_done(ls);
245 dlm_release_root_list(ls);
246 log_rinfo(ls, "dlm_recover %llu error %d",
247 (unsigned long long)rv->seq, error);
248 mutex_unlock(&ls->ls_recoverd_active);
252 /* The dlm_ls_start() that created the rv we take here may already have been
253 stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
256 static void do_ls_recovery(struct dlm_ls *ls)
258 struct dlm_recover *rv = NULL;
260 spin_lock(&ls->ls_recover_lock);
261 rv = ls->ls_recover_args;
262 ls->ls_recover_args = NULL;
263 if (rv && ls->ls_recover_seq == rv->seq)
264 clear_bit(LSFL_RECOVER_STOP, &ls->ls_flags);
265 spin_unlock(&ls->ls_recover_lock);
274 static int dlm_recoverd(void *arg)
278 ls = dlm_find_lockspace_local(arg);
280 log_print("dlm_recoverd: no lockspace %p", arg);
284 down_write(&ls->ls_in_recovery);
285 set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
286 wake_up(&ls->ls_recover_lock_wait);
290 * We call kthread_should_stop() after set_current_state().
291 * This is because it works correctly if kthread_stop() is
292 * called just before set_current_state().
294 set_current_state(TASK_INTERRUPTIBLE);
295 if (kthread_should_stop()) {
296 set_current_state(TASK_RUNNING);
299 if (!test_bit(LSFL_RECOVER_WORK, &ls->ls_flags) &&
300 !test_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
301 if (kthread_should_stop())
305 set_current_state(TASK_RUNNING);
307 if (test_and_clear_bit(LSFL_RECOVER_DOWN, &ls->ls_flags)) {
308 down_write(&ls->ls_in_recovery);
309 set_bit(LSFL_RECOVER_LOCK, &ls->ls_flags);
310 wake_up(&ls->ls_recover_lock_wait);
313 if (test_and_clear_bit(LSFL_RECOVER_WORK, &ls->ls_flags))
317 if (test_bit(LSFL_RECOVER_LOCK, &ls->ls_flags))
318 up_write(&ls->ls_in_recovery);
320 dlm_put_lockspace(ls);
324 int dlm_recoverd_start(struct dlm_ls *ls)
326 struct task_struct *p;
329 p = kthread_run(dlm_recoverd, ls, "dlm_recoverd");
333 ls->ls_recoverd_task = p;
337 void dlm_recoverd_stop(struct dlm_ls *ls)
339 kthread_stop(ls->ls_recoverd_task);
342 void dlm_recoverd_suspend(struct dlm_ls *ls)
344 wake_up(&ls->ls_wait_general);
345 mutex_lock(&ls->ls_recoverd_active);
348 void dlm_recoverd_resume(struct dlm_ls *ls)
350 mutex_unlock(&ls->ls_recoverd_active);