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 * lnet/libcfs/watchdog.c
38 * Author: Jacob Berkman <jacob@clusterfs.com>
41 #define DEBUG_SUBSYSTEM S_LNET
43 #include <libcfs/kp30.h>
44 #include <libcfs/libcfs.h>
45 #include "tracefile.h"
48 cfs_timer_t lcw_timer; /* kernel timer */
49 struct list_head lcw_list;
50 struct timeval lcw_last_touched;
53 void (*lcw_callback)(pid_t, void *);
57 cfs_duration_t lcw_time; /* time until watchdog fires, jiffies */
68 * The dispatcher will complete lcw_start_completion when it starts,
69 * and lcw_stop_completion when it exits.
70 * Wake lcw_event_waitq to signal timer callback dispatches.
72 static struct completion lcw_start_completion;
73 static struct completion lcw_stop_completion;
74 static wait_queue_head_t lcw_event_waitq;
77 * Set this and wake lcw_event_waitq to stop the dispatcher.
82 static unsigned long lcw_flags = 0;
85 * Number of outstanding watchdogs.
86 * When it hits 1, we start the dispatcher.
87 * When it hits 0, we stop the distpatcher.
89 static __u32 lcw_refcount = 0;
90 static DECLARE_MUTEX(lcw_refcount_sem);
93 * List of timers that have fired that need their callbacks run by the
96 static spinlock_t lcw_pending_timers_lock = SPIN_LOCK_UNLOCKED; /* BH lock! */
97 static struct list_head lcw_pending_timers = \
98 LIST_HEAD_INIT(lcw_pending_timers);
100 #ifdef HAVE_TASKLIST_LOCK
102 lcw_dump(struct lc_watchdog *lcw)
107 read_lock(&tasklist_lock);
108 tsk = find_task_by_pid(lcw->lcw_pid);
111 CWARN("Process %d was not found in the task list; "
112 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
113 } else if (tsk != lcw->lcw_task) {
114 CWARN("The current process %d did not set the watchdog; "
115 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
117 libcfs_debug_dumpstack(tsk);
120 read_unlock(&tasklist_lock);
125 lcw_dump(struct lc_watchdog *lcw)
127 CERROR("unable to dump stack because of missing export\n");
131 static void lcw_cb(unsigned long data)
133 struct lc_watchdog *lcw = (struct lc_watchdog *)data;
137 if (lcw->lcw_state != LC_WATCHDOG_ENABLED) {
142 lcw->lcw_state = LC_WATCHDOG_EXPIRED;
144 /* NB this warning should appear on the console, but may not get into
145 * the logs since we're running in a softirq handler */
147 CWARN("Watchdog triggered for pid %d: it was inactive for %lds\n",
148 (int)lcw->lcw_pid, cfs_duration_sec(lcw->lcw_time));
151 spin_lock_bh(&lcw_pending_timers_lock);
153 if (list_empty(&lcw->lcw_list)) {
154 list_add(&lcw->lcw_list, &lcw_pending_timers);
155 wake_up(&lcw_event_waitq);
158 spin_unlock_bh(&lcw_pending_timers_lock);
163 static int is_watchdog_fired(void)
167 if (test_bit(LCW_FLAG_STOP, &lcw_flags))
170 spin_lock_bh(&lcw_pending_timers_lock);
171 rc = !list_empty(&lcw_pending_timers);
172 spin_unlock_bh(&lcw_pending_timers_lock);
176 static int lcw_dispatch_main(void *data)
180 struct lc_watchdog *lcw;
184 cfs_daemonize("lc_watchdogd");
186 SIGNAL_MASK_LOCK(current, flags);
187 sigfillset(¤t->blocked);
189 SIGNAL_MASK_UNLOCK(current, flags);
191 complete(&lcw_start_completion);
194 wait_event_interruptible(lcw_event_waitq, is_watchdog_fired());
195 CDEBUG(D_INFO, "Watchdog got woken up...\n");
196 if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
197 CDEBUG(D_INFO, "LCW_FLAG_STOP was set, shutting down...\n");
199 spin_lock_bh(&lcw_pending_timers_lock);
200 rc = !list_empty(&lcw_pending_timers);
201 spin_unlock_bh(&lcw_pending_timers_lock);
203 CERROR("pending timers list was not empty at "
204 "time of watchdog dispatch shutdown\n");
209 spin_lock_bh(&lcw_pending_timers_lock);
210 while (!list_empty(&lcw_pending_timers)) {
212 lcw = list_entry(lcw_pending_timers.next,
215 list_del_init(&lcw->lcw_list);
216 spin_unlock_bh(&lcw_pending_timers_lock);
218 CDEBUG(D_INFO, "found lcw for pid %d: inactive for "
219 "%lds\n", (int)lcw->lcw_pid,
220 cfs_duration_sec(lcw->lcw_time));
222 if (lcw->lcw_state != LC_WATCHDOG_DISABLED)
223 lcw->lcw_callback(lcw->lcw_pid, lcw->lcw_data);
225 spin_lock_bh(&lcw_pending_timers_lock);
227 spin_unlock_bh(&lcw_pending_timers_lock);
230 complete(&lcw_stop_completion);
235 static void lcw_dispatch_start(void)
240 LASSERT(lcw_refcount == 1);
242 init_completion(&lcw_stop_completion);
243 init_completion(&lcw_start_completion);
244 init_waitqueue_head(&lcw_event_waitq);
246 CDEBUG(D_INFO, "starting dispatch thread\n");
247 rc = kernel_thread(lcw_dispatch_main, NULL, 0);
249 CERROR("error spawning watchdog dispatch thread: %d\n", rc);
253 wait_for_completion(&lcw_start_completion);
254 CDEBUG(D_INFO, "watchdog dispatcher initialization complete.\n");
259 static void lcw_dispatch_stop(void)
262 LASSERT(lcw_refcount == 0);
264 CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
266 set_bit(LCW_FLAG_STOP, &lcw_flags);
267 wake_up(&lcw_event_waitq);
269 wait_for_completion(&lcw_stop_completion);
271 CDEBUG(D_INFO, "watchdog dispatcher has shut down.\n");
276 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
277 void (*callback)(pid_t, void *),
280 struct lc_watchdog *lcw = NULL;
283 LIBCFS_ALLOC(lcw, sizeof(*lcw));
285 CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
286 RETURN(ERR_PTR(-ENOMEM));
289 lcw->lcw_task = cfs_current();
290 lcw->lcw_pid = cfs_curproc_pid();
291 lcw->lcw_time = cfs_time_seconds(timeout_ms) / 1000;
292 lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
293 lcw->lcw_data = data;
294 lcw->lcw_state = LC_WATCHDOG_DISABLED;
296 INIT_LIST_HEAD(&lcw->lcw_list);
298 lcw->lcw_timer.function = lcw_cb;
299 lcw->lcw_timer.data = (unsigned long)lcw;
300 lcw->lcw_timer.expires = jiffies + lcw->lcw_time;
301 init_timer(&lcw->lcw_timer);
303 down(&lcw_refcount_sem);
304 if (++lcw_refcount == 1)
305 lcw_dispatch_start();
306 up(&lcw_refcount_sem);
308 /* Keep this working in case we enable them by default */
309 if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
310 do_gettimeofday(&lcw->lcw_last_touched);
311 add_timer(&lcw->lcw_timer);
316 EXPORT_SYMBOL(lc_watchdog_add);
318 static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
320 struct timeval newtime;
321 struct timeval timediff;
323 do_gettimeofday(&newtime);
324 if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
325 cfs_timeval_sub(&newtime, &lcw->lcw_last_touched, &timediff);
326 CWARN("Expired watchdog for pid %d %s after %lu.%.4lus\n",
330 timediff.tv_usec / 100);
332 lcw->lcw_last_touched = newtime;
335 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
338 LASSERT(lcw != NULL);
340 spin_lock_bh(&lcw_pending_timers_lock);
341 list_del_init(&lcw->lcw_list);
342 spin_unlock_bh(&lcw_pending_timers_lock);
344 lcw_update_time(lcw, "touched");
345 lcw->lcw_state = LC_WATCHDOG_ENABLED;
347 mod_timer(&lcw->lcw_timer, jiffies +
348 cfs_time_seconds(timeout_ms) / 1000);
352 EXPORT_SYMBOL(lc_watchdog_touch_ms);
354 /* deprecated - use above instead */
355 void lc_watchdog_touch(struct lc_watchdog *lcw)
357 lc_watchdog_touch_ms(lcw, cfs_duration_sec(lcw->lcw_time) * 1000);
359 EXPORT_SYMBOL(lc_watchdog_touch);
361 void lc_watchdog_disable(struct lc_watchdog *lcw)
364 LASSERT(lcw != NULL);
366 spin_lock_bh(&lcw_pending_timers_lock);
367 if (!list_empty(&lcw->lcw_list))
368 list_del_init(&lcw->lcw_list);
369 spin_unlock_bh(&lcw_pending_timers_lock);
371 lcw_update_time(lcw, "disabled");
372 lcw->lcw_state = LC_WATCHDOG_DISABLED;
376 EXPORT_SYMBOL(lc_watchdog_disable);
378 void lc_watchdog_delete(struct lc_watchdog *lcw)
381 LASSERT(lcw != NULL);
383 del_timer(&lcw->lcw_timer);
385 lcw_update_time(lcw, "deleted");
387 spin_lock_bh(&lcw_pending_timers_lock);
388 if (!list_empty(&lcw->lcw_list))
389 list_del_init(&lcw->lcw_list);
390 spin_unlock_bh(&lcw_pending_timers_lock);
392 down(&lcw_refcount_sem);
393 if (--lcw_refcount == 0)
395 up(&lcw_refcount_sem);
397 LIBCFS_FREE(lcw, sizeof(*lcw));
401 EXPORT_SYMBOL(lc_watchdog_delete);
404 * Provided watchdog handlers
407 void lc_watchdog_dumplog(pid_t pid, void *data)
409 libcfs_debug_dumplog_internal((void *)((unsigned long)pid));
411 EXPORT_SYMBOL(lc_watchdog_dumplog);
413 #else /* !defined(WITH_WATCHDOG) */
415 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
416 void (*callback)(pid_t pid, void *),
419 static struct lc_watchdog watchdog;
422 EXPORT_SYMBOL(lc_watchdog_add);
424 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
427 EXPORT_SYMBOL(lc_watchdog_touch_ms);
429 void lc_watchdog_touch(struct lc_watchdog *lcw)
432 EXPORT_SYMBOL(lc_watchdog_touch);
434 void lc_watchdog_disable(struct lc_watchdog *lcw)
437 EXPORT_SYMBOL(lc_watchdog_disable);
439 void lc_watchdog_delete(struct lc_watchdog *lcw)
442 EXPORT_SYMBOL(lc_watchdog_delete);