1 /* -*- mode: c; c-basic-offset: 8; indent-tabs-mode: nil; -*-
2 * vim:expandtab:shiftwidth=8:tabstop=8:
4 * Copyright (C) 2004 Cluster File Systems, Inc.
5 * Author: Jacob Berkman <jacob@clusterfs.com>
7 * This file is part of Lustre, http://www.lustre.org.
9 * Lustre is free software; you can redistribute it and/or
10 * modify it under the terms of version 2 of the GNU General Public
11 * License as published by the Free Software Foundation.
13 * Lustre is distributed in the hope that it will be useful,
14 * but WITHOUT ANY WARRANTY; without even the implied warranty of
15 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
16 * GNU General Public License for more details.
18 * You should have received a copy of the GNU General Public License
19 * along with Lustre; if not, write to the Free Software
20 * Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
23 #define DEBUG_SUBSYSTEM S_LNET
25 #include <libcfs/kp30.h>
26 #include <libcfs/libcfs.h>
27 #include "tracefile.h"
30 cfs_timer_t lcw_timer; /* kernel timer */
31 struct list_head lcw_list;
32 struct timeval lcw_last_touched;
35 void (*lcw_callback)(pid_t, void *);
39 cfs_duration_t lcw_time; /* time until watchdog fires, jiffies */
50 * The dispatcher will complete lcw_start_completion when it starts,
51 * and lcw_stop_completion when it exits.
52 * Wake lcw_event_waitq to signal timer callback dispatches.
54 static struct completion lcw_start_completion;
55 static struct completion lcw_stop_completion;
56 static wait_queue_head_t lcw_event_waitq;
59 * Set this and wake lcw_event_waitq to stop the dispatcher.
64 static unsigned long lcw_flags = 0;
67 * Number of outstanding watchdogs.
68 * When it hits 1, we start the dispatcher.
69 * When it hits 0, we stop the distpatcher.
71 static __u32 lcw_refcount = 0;
72 static DECLARE_MUTEX(lcw_refcount_sem);
75 * List of timers that have fired that need their callbacks run by the
78 static spinlock_t lcw_pending_timers_lock = SPIN_LOCK_UNLOCKED; /* BH lock! */
79 static struct list_head lcw_pending_timers = \
80 LIST_HEAD_INIT(lcw_pending_timers);
82 #ifdef HAVE_TASKLIST_LOCK
84 lcw_dump(struct lc_watchdog *lcw)
89 read_lock(&tasklist_lock);
90 tsk = find_task_by_pid(lcw->lcw_pid);
93 CWARN("Process %d was not found in the task list; "
94 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
95 } else if (tsk != lcw->lcw_task) {
96 CWARN("The current process %d did not set the watchdog; "
97 "watchdog callback may be incomplete\n", (int)lcw->lcw_pid);
99 libcfs_debug_dumpstack(tsk);
102 read_unlock(&tasklist_lock);
107 lcw_dump(struct lc_watchdog *lcw)
109 CERROR("unable to dump stack because of missing export\n");
113 static void lcw_cb(unsigned long data)
115 struct lc_watchdog *lcw = (struct lc_watchdog *)data;
119 if (lcw->lcw_state != LC_WATCHDOG_ENABLED) {
124 lcw->lcw_state = LC_WATCHDOG_EXPIRED;
126 /* NB this warning should appear on the console, but may not get into
127 * the logs since we're running in a softirq handler */
129 CWARN("Watchdog triggered for pid %d: it was inactive for %lds\n",
130 (int)lcw->lcw_pid, cfs_duration_sec(lcw->lcw_time));
133 spin_lock_bh(&lcw_pending_timers_lock);
135 if (list_empty(&lcw->lcw_list)) {
136 list_add(&lcw->lcw_list, &lcw_pending_timers);
137 wake_up(&lcw_event_waitq);
140 spin_unlock_bh(&lcw_pending_timers_lock);
145 static int is_watchdog_fired(void)
149 if (test_bit(LCW_FLAG_STOP, &lcw_flags))
152 spin_lock_bh(&lcw_pending_timers_lock);
153 rc = !list_empty(&lcw_pending_timers);
154 spin_unlock_bh(&lcw_pending_timers_lock);
158 static int lcw_dispatch_main(void *data)
162 struct lc_watchdog *lcw;
166 cfs_daemonize("lc_watchdogd");
168 SIGNAL_MASK_LOCK(current, flags);
169 sigfillset(¤t->blocked);
171 SIGNAL_MASK_UNLOCK(current, flags);
173 complete(&lcw_start_completion);
176 wait_event_interruptible(lcw_event_waitq, is_watchdog_fired());
177 CDEBUG(D_INFO, "Watchdog got woken up...\n");
178 if (test_bit(LCW_FLAG_STOP, &lcw_flags)) {
179 CDEBUG(D_INFO, "LCW_FLAG_STOP was set, shutting down...\n");
181 spin_lock_bh(&lcw_pending_timers_lock);
182 rc = !list_empty(&lcw_pending_timers);
183 spin_unlock_bh(&lcw_pending_timers_lock);
185 CERROR("pending timers list was not empty at "
186 "time of watchdog dispatch shutdown\n");
191 spin_lock_bh(&lcw_pending_timers_lock);
192 while (!list_empty(&lcw_pending_timers)) {
194 lcw = list_entry(lcw_pending_timers.next,
197 list_del_init(&lcw->lcw_list);
198 spin_unlock_bh(&lcw_pending_timers_lock);
200 CDEBUG(D_INFO, "found lcw for pid %d: inactive for "
201 "%lds\n", (int)lcw->lcw_pid,
202 cfs_duration_sec(lcw->lcw_time));
204 if (lcw->lcw_state != LC_WATCHDOG_DISABLED)
205 lcw->lcw_callback(lcw->lcw_pid, lcw->lcw_data);
207 spin_lock_bh(&lcw_pending_timers_lock);
209 spin_unlock_bh(&lcw_pending_timers_lock);
212 complete(&lcw_stop_completion);
217 static void lcw_dispatch_start(void)
222 LASSERT(lcw_refcount == 1);
224 init_completion(&lcw_stop_completion);
225 init_completion(&lcw_start_completion);
226 init_waitqueue_head(&lcw_event_waitq);
228 CDEBUG(D_INFO, "starting dispatch thread\n");
229 rc = kernel_thread(lcw_dispatch_main, NULL, 0);
231 CERROR("error spawning watchdog dispatch thread: %d\n", rc);
235 wait_for_completion(&lcw_start_completion);
236 CDEBUG(D_INFO, "watchdog dispatcher initialization complete.\n");
241 static void lcw_dispatch_stop(void)
244 LASSERT(lcw_refcount == 0);
246 CDEBUG(D_INFO, "trying to stop watchdog dispatcher.\n");
248 set_bit(LCW_FLAG_STOP, &lcw_flags);
249 wake_up(&lcw_event_waitq);
251 wait_for_completion(&lcw_stop_completion);
253 CDEBUG(D_INFO, "watchdog dispatcher has shut down.\n");
258 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
259 void (*callback)(pid_t, void *),
262 struct lc_watchdog *lcw = NULL;
265 LIBCFS_ALLOC(lcw, sizeof(*lcw));
267 CDEBUG(D_INFO, "Could not allocate new lc_watchdog\n");
268 RETURN(ERR_PTR(-ENOMEM));
271 lcw->lcw_task = cfs_current();
272 lcw->lcw_pid = cfs_curproc_pid();
273 lcw->lcw_time = cfs_time_seconds(timeout_ms) / 1000;
274 lcw->lcw_callback = (callback != NULL) ? callback : lc_watchdog_dumplog;
275 lcw->lcw_data = data;
276 lcw->lcw_state = LC_WATCHDOG_DISABLED;
278 INIT_LIST_HEAD(&lcw->lcw_list);
280 lcw->lcw_timer.function = lcw_cb;
281 lcw->lcw_timer.data = (unsigned long)lcw;
282 lcw->lcw_timer.expires = jiffies + lcw->lcw_time;
283 init_timer(&lcw->lcw_timer);
285 down(&lcw_refcount_sem);
286 if (++lcw_refcount == 1)
287 lcw_dispatch_start();
288 up(&lcw_refcount_sem);
290 /* Keep this working in case we enable them by default */
291 if (lcw->lcw_state == LC_WATCHDOG_ENABLED) {
292 do_gettimeofday(&lcw->lcw_last_touched);
293 add_timer(&lcw->lcw_timer);
298 EXPORT_SYMBOL(lc_watchdog_add);
300 static void lcw_update_time(struct lc_watchdog *lcw, const char *message)
302 struct timeval newtime;
303 struct timeval timediff;
305 do_gettimeofday(&newtime);
306 if (lcw->lcw_state == LC_WATCHDOG_EXPIRED) {
307 cfs_timeval_sub(&newtime, &lcw->lcw_last_touched, &timediff);
308 CWARN("Expired watchdog for pid %d %s after %lu.%.4lus\n",
312 timediff.tv_usec / 100);
314 lcw->lcw_last_touched = newtime;
317 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
320 LASSERT(lcw != NULL);
322 spin_lock_bh(&lcw_pending_timers_lock);
323 list_del_init(&lcw->lcw_list);
324 spin_unlock_bh(&lcw_pending_timers_lock);
326 lcw_update_time(lcw, "touched");
327 lcw->lcw_state = LC_WATCHDOG_ENABLED;
329 mod_timer(&lcw->lcw_timer, jiffies +
330 cfs_time_seconds(timeout_ms) / 1000);
334 EXPORT_SYMBOL(lc_watchdog_touch_ms);
336 /* deprecated - use above instead */
337 void lc_watchdog_touch(struct lc_watchdog *lcw)
339 lc_watchdog_touch_ms(lcw, cfs_duration_sec(lcw->lcw_time) * 1000);
341 EXPORT_SYMBOL(lc_watchdog_touch);
343 void lc_watchdog_disable(struct lc_watchdog *lcw)
346 LASSERT(lcw != NULL);
348 spin_lock_bh(&lcw_pending_timers_lock);
349 if (!list_empty(&lcw->lcw_list))
350 list_del_init(&lcw->lcw_list);
351 spin_unlock_bh(&lcw_pending_timers_lock);
353 lcw_update_time(lcw, "disabled");
354 lcw->lcw_state = LC_WATCHDOG_DISABLED;
358 EXPORT_SYMBOL(lc_watchdog_disable);
360 void lc_watchdog_delete(struct lc_watchdog *lcw)
363 LASSERT(lcw != NULL);
365 del_timer(&lcw->lcw_timer);
367 lcw_update_time(lcw, "deleted");
369 spin_lock_bh(&lcw_pending_timers_lock);
370 if (!list_empty(&lcw->lcw_list))
371 list_del_init(&lcw->lcw_list);
372 spin_unlock_bh(&lcw_pending_timers_lock);
374 down(&lcw_refcount_sem);
375 if (--lcw_refcount == 0)
377 up(&lcw_refcount_sem);
379 LIBCFS_FREE(lcw, sizeof(*lcw));
383 EXPORT_SYMBOL(lc_watchdog_delete);
386 * Provided watchdog handlers
389 void lc_watchdog_dumplog(pid_t pid, void *data)
391 libcfs_debug_dumplog_internal((void *)((unsigned long)pid));
393 EXPORT_SYMBOL(lc_watchdog_dumplog);
395 #else /* !defined(WITH_WATCHDOG) */
397 struct lc_watchdog *lc_watchdog_add(int timeout_ms,
398 void (*callback)(pid_t pid, void *),
401 static struct lc_watchdog watchdog;
404 EXPORT_SYMBOL(lc_watchdog_add);
406 void lc_watchdog_touch_ms(struct lc_watchdog *lcw, int timeout_ms)
409 EXPORT_SYMBOL(lc_watchdog_touch_ms);
411 void lc_watchdog_touch(struct lc_watchdog *lcw)
414 EXPORT_SYMBOL(lc_watchdog_touch);
416 void lc_watchdog_disable(struct lc_watchdog *lcw)
419 EXPORT_SYMBOL(lc_watchdog_disable);
421 void lc_watchdog_delete(struct lc_watchdog *lcw)
424 EXPORT_SYMBOL(lc_watchdog_delete);