]>
git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blob - fs/dlm/recoverd.c
1 /******************************************************************************
2 *******************************************************************************
4 ** Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
5 ** Copyright (C) 2004-2011 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26 /* If the start for which we're re-enabling locking (seq) has been superseded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
);
44 clear_bit(LSFL_RECOVER_LOCK
, &ls
->ls_flags
);
47 spin_unlock(&ls
->ls_recover_lock
);
49 up_write(&ls
->ls_recv_active
);
53 static int ls_recover(struct dlm_ls
*ls
, struct dlm_recover
*rv
)
58 log_rinfo(ls
, "dlm_recover %llu", (unsigned long long)rv
->seq
);
60 mutex_lock(&ls
->ls_recoverd_active
);
62 dlm_callback_suspend(ls
);
67 * This list of root rsb's will be the basis of most of the recovery
71 dlm_create_root_list(ls
);
74 * Add or remove nodes from the lockspace's ls_nodes list.
77 error
= dlm_recover_members(ls
, rv
, &neg
);
79 log_rinfo(ls
, "dlm_recover_members error %d", error
);
83 dlm_recover_dir_nodeid(ls
);
85 ls
->ls_recover_dir_sent_res
= 0;
86 ls
->ls_recover_dir_sent_msg
= 0;
87 ls
->ls_recover_locks_in
= 0;
89 dlm_set_recover_status(ls
, DLM_RS_NODES
);
91 error
= dlm_recover_members_wait(ls
);
93 log_rinfo(ls
, "dlm_recover_members_wait error %d", error
);
100 * Rebuild our own share of the directory by collecting from all other
101 * nodes their master rsb names that hash to us.
104 error
= dlm_recover_directory(ls
);
106 log_rinfo(ls
, "dlm_recover_directory error %d", error
);
110 dlm_set_recover_status(ls
, DLM_RS_DIR
);
112 error
= dlm_recover_directory_wait(ls
);
114 log_rinfo(ls
, "dlm_recover_directory_wait error %d", error
);
118 log_rinfo(ls
, "dlm_recover_directory %u out %u messages",
119 ls
->ls_recover_dir_sent_res
, ls
->ls_recover_dir_sent_msg
);
122 * We may have outstanding operations that are waiting for a reply from
123 * a failed node. Mark these to be resent after recovery. Unlock and
124 * cancel ops can just be completed.
127 dlm_recover_waiters_pre(ls
);
129 error
= dlm_recovery_stopped(ls
);
133 if (neg
|| dlm_no_directory(ls
)) {
135 * Clear lkb's for departed nodes.
138 dlm_recover_purge(ls
);
141 * Get new master nodeid's for rsb's that were mastered on
145 error
= dlm_recover_masters(ls
);
147 log_rinfo(ls
, "dlm_recover_masters error %d", error
);
152 * Send our locks on remastered rsb's to the new masters.
155 error
= dlm_recover_locks(ls
);
157 log_rinfo(ls
, "dlm_recover_locks error %d", error
);
161 dlm_set_recover_status(ls
, DLM_RS_LOCKS
);
163 error
= dlm_recover_locks_wait(ls
);
165 log_rinfo(ls
, "dlm_recover_locks_wait error %d", error
);
169 log_rinfo(ls
, "dlm_recover_locks %u in",
170 ls
->ls_recover_locks_in
);
173 * Finalize state in master rsb's now that all locks can be
174 * checked. This includes conversion resolution and lvb
178 dlm_recover_rsbs(ls
);
181 * Other lockspace members may be going through the "neg" steps
182 * while also adding us to the lockspace, in which case they'll
183 * be doing the recover_locks (RS_LOCKS) barrier.
185 dlm_set_recover_status(ls
, DLM_RS_LOCKS
);
187 error
= dlm_recover_locks_wait(ls
);
189 log_rinfo(ls
, "dlm_recover_locks_wait error %d", error
);
194 dlm_release_root_list(ls
);
197 * Purge directory-related requests that are saved in requestqueue.
198 * All dir requests from before recovery are invalid now due to the dir
199 * rebuild and will be resent by the requesting nodes.
202 dlm_purge_requestqueue(ls
);
204 dlm_set_recover_status(ls
, DLM_RS_DONE
);
206 error
= dlm_recover_done_wait(ls
);
208 log_rinfo(ls
, "dlm_recover_done_wait error %d", error
);
212 dlm_clear_members_gone(ls
);
214 dlm_adjust_timeouts(ls
);
216 dlm_callback_resume(ls
);
218 error
= enable_locking(ls
, rv
->seq
);
220 log_rinfo(ls
, "enable_locking error %d", error
);
224 error
= dlm_process_requestqueue(ls
);
226 log_rinfo(ls
, "dlm_process_requestqueue error %d", error
);
230 error
= dlm_recover_waiters_post(ls
);
232 log_rinfo(ls
, "dlm_recover_waiters_post error %d", error
);
236 dlm_recover_grant(ls
);
238 log_rinfo(ls
, "dlm_recover %llu generation %u done: %u ms",
239 (unsigned long long)rv
->seq
, ls
->ls_generation
,
240 jiffies_to_msecs(jiffies
- start
));
241 mutex_unlock(&ls
->ls_recoverd_active
);
243 dlm_lsop_recover_done(ls
);
247 dlm_release_root_list(ls
);
248 log_rinfo(ls
, "dlm_recover %llu error %d",
249 (unsigned long long)rv
->seq
, error
);
250 mutex_unlock(&ls
->ls_recoverd_active
);
254 /* The dlm_ls_start() that created the rv we take here may already have been
255 stopped via dlm_ls_stop(); in that case we need to leave the RECOVERY_STOP
258 static void do_ls_recovery(struct dlm_ls
*ls
)
260 struct dlm_recover
*rv
= NULL
;
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
);
276 static int dlm_recoverd(void *arg
)
280 ls
= dlm_find_lockspace_local(arg
);
282 log_print("dlm_recoverd: no lockspace %p", arg
);
286 down_write(&ls
->ls_in_recovery
);
287 set_bit(LSFL_RECOVER_LOCK
, &ls
->ls_flags
);
288 wake_up(&ls
->ls_recover_lock_wait
);
292 * We call kthread_should_stop() after set_current_state().
293 * This is because it works correctly if kthread_stop() is
294 * called just before set_current_state().
296 set_current_state(TASK_INTERRUPTIBLE
);
297 if (kthread_should_stop()) {
298 set_current_state(TASK_RUNNING
);
301 if (!test_bit(LSFL_RECOVER_WORK
, &ls
->ls_flags
) &&
302 !test_bit(LSFL_RECOVER_DOWN
, &ls
->ls_flags
)) {
303 if (kthread_should_stop())
307 set_current_state(TASK_RUNNING
);
309 if (test_and_clear_bit(LSFL_RECOVER_DOWN
, &ls
->ls_flags
)) {
310 down_write(&ls
->ls_in_recovery
);
311 set_bit(LSFL_RECOVER_LOCK
, &ls
->ls_flags
);
312 wake_up(&ls
->ls_recover_lock_wait
);
315 if (test_and_clear_bit(LSFL_RECOVER_WORK
, &ls
->ls_flags
))
319 if (test_bit(LSFL_RECOVER_LOCK
, &ls
->ls_flags
))
320 up_write(&ls
->ls_in_recovery
);
322 dlm_put_lockspace(ls
);
326 int dlm_recoverd_start(struct dlm_ls
*ls
)
328 struct task_struct
*p
;
331 p
= kthread_run(dlm_recoverd
, ls
, "dlm_recoverd");
335 ls
->ls_recoverd_task
= p
;
339 void dlm_recoverd_stop(struct dlm_ls
*ls
)
341 kthread_stop(ls
->ls_recoverd_task
);
344 void dlm_recoverd_suspend(struct dlm_ls
*ls
)
346 wake_up(&ls
->ls_wait_general
);
347 mutex_lock(&ls
->ls_recoverd_active
);
350 void dlm_recoverd_resume(struct dlm_ls
*ls
)
352 mutex_unlock(&ls
->ls_recoverd_active
);