/*
- * Bacula thread watchdog routine. General routine that monitors
- * the daemon and signals a thread if it is blocked on a BSOCK
- * too long. This prevents catastropic long waits -- generally
- * due to Windows "hanging" the app.
+ * Bacula thread watchdog routine. General routine that
+ * allows setting a watchdog timer with a callback that is
+ * called when the timer goes off.
*
* Kern Sibbald, January MMII
*
*/
/*
- Copyright (C) 2000-2003 Kern Sibbald and John Walker
+ Copyright (C) 2000-2005 Kern Sibbald
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License as
#include "bacula.h"
#include "jcr.h"
-/* This breaks Kern's #include rules, but I don't want to put it into bacula.h
- * until it has been discussed with him */
-#include "bsd_queue.h"
-
/* Exported globals */
-time_t watchdog_time; /* this has granularity of SLEEP_TIME */
+time_t watchdog_time = 0; /* this has granularity of SLEEP_TIME */
+time_t watchdog_sleep_time = 60; /* examine things every 60 seconds */
-#define SLEEP_TIME 1 /* examine things every second */
+/* Locals */
+static pthread_mutex_t timer_mutex = PTHREAD_MUTEX_INITIALIZER;
+static pthread_cond_t timer = PTHREAD_COND_INITIALIZER;
/* Forward referenced functions */
-static void *watchdog_thread(void *arg);
+extern "C" void *watchdog_thread(void *arg);
+
+static void wd_lock();
+static void wd_unlock();
/* Static globals */
-static pthread_mutex_t mutex = PTHREAD_MUTEX_INITIALIZER;
-static pthread_cond_t timer = PTHREAD_COND_INITIALIZER;
-static int quit;
+static bool quit = false;;
static bool wd_is_init = false;
+static brwlock_t lock; /* watchdog lock */
-/* Forward referenced callback functions */
-static void callback_child_timer(watchdog_t *self);
-static void callback_thread_timer(watchdog_t *self);
static pthread_t wd_tid;
+static dlist *wd_queue;
+static dlist *wd_inactive;
-/* Static globals */
-static TAILQ_HEAD(/* no struct */, s_watchdog_t) wd_queue =
- TAILQ_HEAD_INITIALIZER(wd_queue);
-static TAILQ_HEAD(/* no struct */, s_watchdog_t) wd_inactive =
- TAILQ_HEAD_INITIALIZER(wd_inactive);
-
-/*
+/*
* Start watchdog thread
*
* Returns: 0 on success
- * errno on failure
+ * errno on failure
*/
int start_watchdog(void)
{
int stat;
+ watchdog_t *dummy = NULL;
+ int errstat;
- Dmsg0(200, "Initialising NicB-hacked watchdog thread\n");
+ if (wd_is_init) {
+ return 0;
+ }
+ Dmsg0(800, "Initialising NicB-hacked watchdog thread\n");
watchdog_time = time(NULL);
- quit = FALSE;
+
+ if ((errstat=rwl_init(&lock)) != 0) {
+ Emsg1(M_ABORT, 0, _("Unable to initialize watchdog lock. ERR=%s\n"),
+ strerror(errstat));
+ }
+ wd_queue = New(dlist(dummy, &dummy->link));
+ wd_inactive = New(dlist(dummy, &dummy->link));
+
if ((stat = pthread_create(&wd_tid, NULL, watchdog_thread, NULL)) != 0) {
return stat;
}
return 0;
}
+/*
+ * Wake watchdog timer thread so that it walks the
+ * queue and adjusts its wait time (or exits).
+ */
+static void ping_watchdog()
+{
+ P(timer_mutex);
+ pthread_cond_signal(&timer);
+ V(timer_mutex);
+}
+
/*
* Terminate the watchdog thread
*
* Returns: 0 on success
- * errno on failure
+ * errno on failure
*/
int stop_watchdog(void)
{
int stat;
- watchdog_t *p, *n;
+ watchdog_t *p;
if (!wd_is_init) {
- Emsg0(M_ABORT, 0, "BUG! stop_watchdog called before start_watchdog\n");
+ return 0;
}
- Dmsg0(200, "Sending stop signal to NicB-hacked watchdog thread\n");
- P(mutex);
- quit = true;
- stat = pthread_cond_signal(&timer);
- V(mutex);
-
+ quit = true; /* notify watchdog thread to stop */
wd_is_init = false;
+ ping_watchdog();
stat = pthread_join(wd_tid, NULL);
- TAILQ_FOREACH_SAFE(p, &wd_queue, qe, n) {
- TAILQ_REMOVE(&wd_queue, p, qe);
+ while (!wd_queue->empty()) {
+ void *item = wd_queue->first();
+ wd_queue->remove(item);
+ p = (watchdog_t *)item;
if (p->destructor != NULL) {
- p->destructor(p);
+ p->destructor(p);
}
free(p);
}
+ delete wd_queue;
+ wd_queue = NULL;
- TAILQ_FOREACH_SAFE(p, &wd_inactive, qe, n) {
- TAILQ_REMOVE(&wd_inactive, p, qe);
+ while (!wd_inactive->empty()) {
+ void *item = wd_inactive->first();
+ wd_inactive->remove(item);
+ p = (watchdog_t *)item;
if (p->destructor != NULL) {
- p->destructor(p);
+ p->destructor(p);
}
free(p);
}
+ delete wd_inactive;
+ wd_inactive = NULL;
+ rwl_destroy(&lock);
return stat;
}
-watchdog_t *watchdog_new(void)
+watchdog_t *new_watchdog(void)
{
- watchdog_t *wd = (watchdog_t *) malloc(sizeof(watchdog_t));
+ watchdog_t *wd = (watchdog_t *)malloc(sizeof(watchdog_t));
if (!wd_is_init) {
- Emsg0(M_ABORT, 0, "BUG! watchdog_new called before start_watchdog\n");
+ start_watchdog();
}
if (wd == NULL) {
bool register_watchdog(watchdog_t *wd)
{
if (!wd_is_init) {
- Emsg0(M_ABORT, 0, "BUG! register_watchdog called before start_watchdog\n");
+ Emsg0(M_ABORT, 0, _("BUG! register_watchdog called before start_watchdog\n"));
}
if (wd->callback == NULL) {
- Emsg1(M_ABORT, 0, "BUG! Watchdog %p has NULL callback\n", wd);
+ Emsg1(M_ABORT, 0, _("BUG! Watchdog %p has NULL callback\n"), wd);
}
if (wd->interval == 0) {
- Emsg1(M_ABORT, 0, "BUG! Watchdog %p has zero interval\n", wd);
+ Emsg1(M_ABORT, 0, _("BUG! Watchdog %p has zero interval\n"), wd);
}
- P(mutex);
+ wd_lock();
wd->next_fire = watchdog_time + wd->interval;
- TAILQ_INSERT_TAIL(&wd_queue, wd, qe);
- Dmsg3(200, "Registered watchdog %p, interval %d%s\n",
- wd, wd->interval, wd->one_shot ? " one shot" : "");
- V(mutex);
+ wd_queue->append(wd);
+ Dmsg3(800, "Registered watchdog %p, interval %d%s\n",
+ wd, wd->interval, wd->one_shot ? " one shot" : "");
+ wd_unlock();
+ ping_watchdog();
return false;
}
-bool unregister_watchdog_unlocked(watchdog_t *wd)
+bool unregister_watchdog(watchdog_t *wd)
{
- watchdog_t *p, *n;
+ watchdog_t *p;
+ bool ok = false;
if (!wd_is_init) {
- Emsg0(M_ABORT, 0, "BUG! unregister_watchdog_unlocked called before start_watchdog\n");
+ Emsg0(M_ABORT, 0, _("BUG! unregister_watchdog_unlocked called before start_watchdog\n"));
}
- TAILQ_FOREACH_SAFE(p, &wd_queue, qe, n) {
+ wd_lock();
+ foreach_dlist(p, wd_queue) {
if (wd == p) {
- TAILQ_REMOVE(&wd_queue, wd, qe);
- Dmsg1(200, "Unregistered watchdog %p\n", wd);
- return true;
+ wd_queue->remove(wd);
+ Dmsg1(800, "Unregistered watchdog %p\n", wd);
+ ok = true;
+ goto get_out;
}
}
- TAILQ_FOREACH_SAFE(p, &wd_inactive, qe, n) {
+ foreach_dlist(p, wd_inactive) {
if (wd == p) {
- TAILQ_REMOVE(&wd_inactive, wd, qe);
- Dmsg1(200, "Unregistered inactive watchdog %p\n", wd);
- return true;
+ wd_inactive->remove(wd);
+ Dmsg1(800, "Unregistered inactive watchdog %p\n", wd);
+ ok = true;
+ goto get_out;
}
}
- Dmsg1(200, "Failed to unregister watchdog %p\n", wd);
+ Dmsg1(800, "Failed to unregister watchdog %p\n", wd);
- return false;
+get_out:
+ wd_unlock();
+ ping_watchdog();
+ return ok;
}
-bool unregister_watchdog(watchdog_t *wd)
+/*
+ * This is the thread that walks the watchdog queue
+ * and when a queue item fires, the callback is
+ * invoked. If it is a one shot, the queue item
+ * is moved to the inactive queue.
+ */
+extern "C" void *watchdog_thread(void *arg)
{
- bool ret;
+ struct timespec timeout;
+ struct timeval tv;
+ struct timezone tz;
+ time_t next_time;
+
+ Dmsg0(800, "NicB-reworked watchdog thread entered\n");
+
+ while (!quit) {
+ watchdog_t *p;
+
+ /*
+ *
+ * NOTE. lock_jcr_chain removed, but the message below
+ * was left until we are sure there are no deadlocks.
+ *
+ * We lock the jcr chain here because a good number of the
+ * callback routines lock the jcr chain. We need to lock
+ * it here *before* the watchdog lock because the SD message
+ * thread first locks the jcr chain, then when closing the
+ * job locks the watchdog chain. If the two threads do not
+ * lock in the same order, we get a deadlock -- each holds
+ * the other's needed lock.
+ */
+ wd_lock();
+
+walk_list:
+ watchdog_time = time(NULL);
+ next_time = watchdog_time + watchdog_sleep_time;
+ foreach_dlist(p, wd_queue) {
+ if (p->next_fire <= watchdog_time) {
+ /* Run the callback */
+ p->callback(p);
+
+ /* Reschedule (or move to inactive list if it's a one-shot timer) */
+ if (p->one_shot) {
+ wd_queue->remove(p);
+ wd_inactive->append(p);
+ goto walk_list;
+ } else {
+ p->next_fire = watchdog_time + p->interval;
+ }
+ }
+ if (p->next_fire < next_time) {
+ next_time = p->next_fire;
+ }
+ }
+ wd_unlock();
+
+ /*
+ * Wait sleep time or until someone wakes us
+ */
+ gettimeofday(&tv, &tz);
+ timeout.tv_nsec = tv.tv_usec * 1000;
+ timeout.tv_sec = tv.tv_sec + next_time - time(NULL);
+ while (timeout.tv_nsec >= 1000000000) {
+ timeout.tv_nsec -= 1000000000;
+ timeout.tv_sec++;
+ }
- if (!wd_is_init) {
- Emsg0(M_ABORT, 0, "BUG! unregister_watchdog called before start_watchdog\n");
+ Dmsg1(1900, "pthread_cond_timedwait %d\n", timeout.tv_sec - tv.tv_sec);
+ /* Note, this unlocks mutex during the sleep */
+ P(timer_mutex);
+ pthread_cond_timedwait(&timer, &timer_mutex, &timeout);
+ V(timer_mutex);
}
- P(mutex);
- ret = unregister_watchdog_unlocked(wd);
- V(mutex);
-
- return ret;
+ Dmsg0(800, "NicB-reworked watchdog thread exited\n");
+ return NULL;
}
-static void *watchdog_thread(void *arg)
+/*
+ * Watchdog lock, this can be called multiple times by the same
+ * thread without blocking, but must be unlocked the number of
+ * times it was locked.
+ */
+static void wd_lock()
{
- Dmsg0(200, "NicB-reworked watchdog thread entered\n");
-
- while (true) {
- watchdog_t *p, *n;
-
- P(mutex);
- if (quit) {
- V(mutex);
- break;
- }
-
- watchdog_time = time(NULL);
-
- TAILQ_FOREACH_SAFE(p, &wd_queue, qe, n) {
- if (p->next_fire < watchdog_time) {
- /* Run the callback */
- p->callback(p);
-
- /* Reschedule (or move to inactive list if it's a one-shot timer) */
- if (p->one_shot) {
- TAILQ_REMOVE(&wd_queue, p, qe);
- TAILQ_INSERT_TAIL(&wd_inactive, p, qe);
- } else {
- p->next_fire = watchdog_time + p->interval;
- }
- }
- }
- V(mutex);
- bmicrosleep(SLEEP_TIME, 0);
+ int errstat;
+ if ((errstat=rwl_writelock(&lock)) != 0) {
+ Emsg1(M_ABORT, 0, _("rwl_writelock failure. ERR=%s\n"),
+ strerror(errstat));
}
+}
- Dmsg0(200, "NicB-reworked watchdog thread exited\n");
-
- return NULL;
+/*
+ * Unlock the watchdog. This can be called multiple times by the
+ * same thread up to the number of times that thread called
+ * wd_ lock()/
+ */
+static void wd_unlock()
+{
+ int errstat;
+ if ((errstat=rwl_writeunlock(&lock)) != 0) {
+ Emsg1(M_ABORT, 0, _("rwl_writeunlock failure. ERR=%s\n"),
+ strerror(errstat));
+ }
}