1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
6 * DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER.
8 * This program is free software; you can redistribute it and/or modify
9 * it under the terms of the GNU General Public License version 2 only,
10 * as published by the Free Software Foundation.
12 * This program is distributed in the hope that it will be useful, but
13 * WITHOUT ANY WARRANTY; without even the implied warranty of
14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
15 * General Public License version 2 for more details (a copy is included
16 * in the LICENSE file that accompanied this code).
18 * You should have received a copy of the GNU General Public License
19 * version 2 along with this program; If not, see
20 * http://www.sun.com/software/products/lustre/docs/GPLv2.pdf
22 * Please contact Sun Microsystems, Inc., 4150 Network Circle, Santa Clara,
23 * CA 95054 USA or visit www.sun.com if you need additional information or
29 * Copyright 2008 Sun Microsystems, Inc. All rights reserved
30 * Use is subject to license terms.
33 * This file is part of Lustre, http://www.lustre.org/
34 * Lustre is a trademark of Sun Microsystems, Inc.
36 * libcfs/libcfs/watchdog.c
38 * Author: Jacob Berkman <jacob@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_LNET
43 #include <libcfs/libcfs.h>
44 #include "tracefile.h"
47 cfs_timer_t lcw_timer; /* kernel timer */
48 struct list_head lcw_list;
49 struct timeval lcw_last_touched;
52 void (*lcw_callback)(pid_t, void *);
56 cfs_duration_t lcw_time; /* time until watchdog fires, jiffies */
67 * The dispatcher will complete lcw_start_completion when it starts,
68 * and lcw_stop_completion when it exits.
69 * Wake lcw_event_waitq to signal timer callback dispatches.
71 static struct completion lcw_start_completion;
72 static struct completion lcw_stop_completion;
73 static wait_queue_head_t lcw_event_waitq;
76 * Set this and wake lcw_event_waitq to stop the dispatcher.
81 static unsigned long lcw_flags = 0;
84 * Number of outstanding watchdogs.
85 * When it hits 1, we start the dispatcher.
86 * When it hits 0, we stop the distpatcher.
88 static __u32 lcw_refcount = 0;
89 static DECLARE_MUTEX(lcw_refcount_sem);
92 * List of timers that have fired that need their callbacks run by the
95 static spinlock_t lcw_pending_timers_lock = SPIN_LOCK_UNLOCKED; /* BH lock! */
96 static struct list_head lcw_pending_timers = \
97 LIST_HEAD_INIT(lcw_pending_timers);
99 #ifdef HAVE_TASKLIST_LOCK
101 lcw_dump(struct lc_watchdog *lcw)
106 read_lock(&tasklist_lock);
107 tsk = find_task_by_pid(lcw->lcw_pid);
110 CWARN("Process %d was not found in the task list; "
111 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
112 } else if (tsk != lcw->lcw_task) {
113 CWARN("The current process %d did not set the watchdog; "
114 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
116 libcfs_debug_dumpstack(tsk);
119 read_unlock(&tasklist_lock);
124 lcw_dump(struct lc_watchdog *lcw)
126 CERROR("unable to dump stack because of missing export\n");
130 static void lcw_cb(unsigned long data)
132 struct lc_watchdog *lcw = (struct lc_watchdog *)data;
136 if (lcw->lcw_state != LC_WATCHDOG_ENABLED) {
141 lcw->lcw_state = LC_WATCHDOG_EXPIRED;
143 /* NB this warning should appear on the console, but may not get into
144 * the logs since we're running in a softirq handler */
146 CWARN("Watchdog triggered for pid %d: it was inactive for %lds\n",
147 (int)lcw->lcw_pid, cfs_duration_sec(lcw->lcw_time));
150 spin_lock_bh(&lcw_pending_timers_lock);
152 if (list_empty(&lcw->lcw_list)) {
153 list_add(&lcw->lcw_list, &lcw_pending_timers);
154 wake_up(&lcw_event_waitq);
157 spin_unlock_bh(&lcw_pending_timers_lock);
162 static int is_watchdog_fired(void)
166 if (test_bit(LCW_FLAG_STOP, &lcw_flags))
169 spin_lock_bh(&lcw_pending_timers_lock);
170 rc = !list_empty(&lcw_pending_timers);
171 spin_unlock_bh(&lcw_pending_timers_lock);
175 static int lcw_dispatch_main(void *data)
179 struct lc_watchdog *lcw;
183 cfs_daemonize("lc_watchdogd");
185 SIGNAL_MASK_LOCK(current, flags);
186 sigfillset(¤t->blocked);
188 SIGNAL_MASK_UNLOCK(current, flags);
190 complete(&lcw_start_completion);
193 wait_event_interruptible(lcw_event_waitq, is_watchdog_fired());
194 CDEBUG(D_INFO, "Watchdog got woken up...\n");
195 if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
196 CDEBUG(D_INFO, "LCW_FLAG_STOP was set, shutting down...\n");
198 spin_lock_bh(&lcw_pending_timers_lock);
199 rc = !list_empty(&lcw_pending_timers);
200 spin_unlock_bh(&lcw_pending_timers_lock);
202 CERROR("pending timers list was not empty at "
203 "time of watchdog dispatch shutdown\n");
208 spin_lock_bh(&lcw_pending_timers_lock);
209 while (!list_empty(&lcw_pending_timers)) {
211 lcw = list_entry(lcw_pending_timers.next,
214 list_del_init(&lcw->lcw_list);
215 spin_unlock_bh(&lcw_pending_timers_lock);
217 CDEBUG(D_INFO, "found lcw for pid %d: inactive for "
218 "%lds\n", (int)lcw->lcw_pid,
219 cfs_duration_sec(lcw->lcw_time));
221 if (lcw->lcw_state != LC_WATCHDOG_DISABLED)
222 lcw->lcw_callback(lcw->lcw_pid, lcw->lcw_data);
224 spin_lock_bh(&lcw_pending_timers_lock);
226 spin_unlock_bh(&lcw_pending_timers_lock);
229 complete(&lcw_stop_completion);
234 static void lcw_dispatch_start(void)
239 LASSERT(lcw_refcount == 1);
241 init_completion(&lcw_stop_completion);
242 init_completion(&lcw_start_completion);
243 init_waitqueue_head(&lcw_event_waitq);
245 CDEBUG(D_INFO, "starting dispatch thread\n");
246 rc = kernel_thread(lcw_dispatch_main, NULL, 0);
248 CERROR("error spawning watchdog dispatch thread: %d\n", rc);
252 wait_for_completion(&lcw_start_completion);
253 CDEBUG(D_INFO, "watchdog dispatcher initialization complete.\n");
258 static void lcw_dispatch_stop(void)
261 LASSERT(lcw_refcount == 0);
263 CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
265 set_bit(LCW_FLAG_STOP, &lcw_flags);
266 wake_up(&lcw_event_waitq);
268 wait_for_completion(&lcw_stop_completion);
270 CDEBUG(D_INFO, "watchdog dispatcher has shut down.\n");
275 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
276 void (*callback)(pid_t, void *),
279 struct lc_watchdog *lcw = NULL;
282 LIBCFS_ALLOC(lcw, sizeof(*lcw));
284 CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
285 RETURN(ERR_PTR(-ENOMEM));
288 lcw->lcw_task = cfs_current();
289 lcw->lcw_pid = cfs_curproc_pid();
290 lcw->lcw_time = cfs_time_seconds(timeout_ms) / 1000;
291 lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
292 lcw->lcw_data = data;
293 lcw->lcw_state = LC_WATCHDOG_DISABLED;
295 INIT_LIST_HEAD(&lcw->lcw_list);
297 lcw->lcw_timer.function = lcw_cb;
298 lcw->lcw_timer.data = (unsigned long)lcw;
299 lcw->lcw_timer.expires = jiffies + lcw->lcw_time;
300 init_timer(&lcw->lcw_timer);
302 down(&lcw_refcount_sem);
303 if (++lcw_refcount == 1)
304 lcw_dispatch_start();
305 up(&lcw_refcount_sem);
307 /* Keep this working in case we enable them by default */
308 if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
309 do_gettimeofday(&lcw->lcw_last_touched);
310 add_timer(&lcw->lcw_timer);
315 EXPORT_SYMBOL(lc_watchdog_add);
317 static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
319 struct timeval newtime;
320 struct timeval timediff;
322 do_gettimeofday(&newtime);
323 if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
324 cfs_timeval_sub(&newtime, &lcw->lcw_last_touched, &timediff);
325 CWARN("Expired watchdog for pid %d %s after %lu.%.4lus\n",
329 timediff.tv_usec / 100);
331 lcw->lcw_last_touched = newtime;
334 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
337 LASSERT(lcw != NULL);
339 spin_lock_bh(&lcw_pending_timers_lock);
340 list_del_init(&lcw->lcw_list);
341 spin_unlock_bh(&lcw_pending_timers_lock);
343 lcw_update_time(lcw, "touched");
344 lcw->lcw_state = LC_WATCHDOG_ENABLED;
346 mod_timer(&lcw->lcw_timer, jiffies +
347 cfs_time_seconds(timeout_ms) / 1000);
351 EXPORT_SYMBOL(lc_watchdog_touch_ms);
353 /* deprecated - use above instead */
354 void lc_watchdog_touch(struct lc_watchdog *lcw)
356 lc_watchdog_touch_ms(lcw, cfs_duration_sec(lcw->lcw_time) * 1000);
358 EXPORT_SYMBOL(lc_watchdog_touch);
360 void lc_watchdog_disable(struct lc_watchdog *lcw)
363 LASSERT(lcw != NULL);
365 spin_lock_bh(&lcw_pending_timers_lock);
366 if (!list_empty(&lcw->lcw_list))
367 list_del_init(&lcw->lcw_list);
368 spin_unlock_bh(&lcw_pending_timers_lock);
370 lcw_update_time(lcw, "disabled");
371 lcw->lcw_state = LC_WATCHDOG_DISABLED;
375 EXPORT_SYMBOL(lc_watchdog_disable);
377 void lc_watchdog_delete(struct lc_watchdog *lcw)
380 LASSERT(lcw != NULL);
382 del_timer(&lcw->lcw_timer);
384 lcw_update_time(lcw, "deleted");
386 spin_lock_bh(&lcw_pending_timers_lock);
387 if (!list_empty(&lcw->lcw_list))
388 list_del_init(&lcw->lcw_list);
389 spin_unlock_bh(&lcw_pending_timers_lock);
391 down(&lcw_refcount_sem);
392 if (--lcw_refcount == 0)
394 up(&lcw_refcount_sem);
396 LIBCFS_FREE(lcw, sizeof(*lcw));
400 EXPORT_SYMBOL(lc_watchdog_delete);
403 * Provided watchdog handlers
406 void lc_watchdog_dumplog(pid_t pid, void *data)
408 libcfs_debug_dumplog_internal((void *)((unsigned long)pid));
410 EXPORT_SYMBOL(lc_watchdog_dumplog);
412 #else /* !defined(WITH_WATCHDOG) */
414 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
415 void (*callback)(pid_t pid, void *),
418 static struct lc_watchdog watchdog;
421 EXPORT_SYMBOL(lc_watchdog_add);
423 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
426 EXPORT_SYMBOL(lc_watchdog_touch_ms);
428 void lc_watchdog_touch(struct lc_watchdog *lcw)
431 EXPORT_SYMBOL(lc_watchdog_touch);
433 void lc_watchdog_disable(struct lc_watchdog *lcw)
436 EXPORT_SYMBOL(lc_watchdog_disable);
438 void lc_watchdog_delete(struct lc_watchdog *lcw)
441 EXPORT_SYMBOL(lc_watchdog_delete);